Thanh Tri Lam
[2024] APO 53
•19 December 2024
IP AUSTRALIA
AUSTRALIAN PATENT OFFICE
Thanh Tri Lam [2024] APO 53
Patent Application: 2023208068
Title:Machine learning based optimal control system for magnetic continuous variable transmission
Patent Applicant: Thanh Tri Lam
Delegate:Greg Powell
Decision Date: 19 December 2024
Hearing Date: Written submissions due on 31 October 2024
Catchwords: PATENTS – section 97(1) – re-examination – examiner objection –clear enough and complete enough disclosure – undue burden – specification does not contain enough information to enable performance of invention across full width of the claims – no patentable subject matter identified in the specification – application refused – best method and support not considered.
Representation: Applicant did not make submissions
IP AUSTRALIA
AUSTRALIAN PATENT OFFICE
Patent Application: 2023208068
Title:Machine learning based optimal control system for magnetic continuous variable transmission
Patent Applicant: Thanh Tri Lam
Date of Decision: 19 December 2024
DECISION
The specification does not satisfy the requirements of s40(2)(a).
I refuse the application.
REASONS FOR DECISION
Background
Patent application 2023208068 (the present application) was filed by Thanh Tri Lam (the applicant) on 24 July 2023. The present application claims priority from Australian provisional applications 2023900824, filed on 24 March 2023 and 2023902148, filed on 4 July 2023.
Following acceptance, the applicant filed amendments on 3 October 2023 (the October amendments) proposing changes to the description, claims and drawings. The applicant also filed further amendments on 5 and 11 December 2023, but consideration of these amendments was held in abeyance until the October amendments had been finally dealt with. The October amendments were allowed on 3 January 2024.
I will note that, in my opinion, the October amendments introduced features which could not be said to have been clearly and unambiguously disclosed in the application as filed. While the October amendment was not allowable, they have, nevertheless, been allowed and the features added by the amendments are now part of the specification I am considering.
An internal review following allowance of the October amendments identified grounds for re–examining. A first re–examination report (the first report) was issued on 28 February 2024 stating that the 5 and 11 December amendments were not allowable. Notwithstanding this, the first report stated that the specification did not contain a complete enough disclosure, and the claims lacked support. Opinion on the novelty and inventiveness of the claimed invention was reserved. The applicant responded to the first report on 25 April 2024 (applicant’s first response) by way of written submissions and proposed amendments to the claims. A second re–examination report (the second report) issued on 31 May 2024 stating that the amendments were not allowable, and maintained the objections (and reservations) of the first report, as well as raising the objection that the specification did not disclose a best method of performing the invention. The second report noted that the applicant had had “conversations with supervising examiners and the assistant general manager”[1], and it was the examiner’s understanding that the applicant “may consider requesting a hearing on the objections in [the second] report”[2].
[1] Second report at page 6
[2] See ibid
A response to the second report was filed on 30 July 2024 with further submissions and amendments to the claims. On 23 August 2024 a third re–examination report (the third report) issued which stated that the amendments filed with this response were also not allowable, and, once again, concluded, after discussion of the written submissions, that these submissions did not overcome the objections in the first report. In the third report, the examiner stated:
“Whilst you have not yet requested to be heard, re-examination of this application has clearly reached an impasse between yourself and the examination section concerned. This is also the case for re-examination of 2023214228, which has very similar re-examination issues as this application. Consequently, the Commissioner will now refer re-examination of this application and of 2023214228 to a hearing officer to resolve the re-examination issues for both applications, which may include refusing either application or disagreeing with the re-examination objections on either application.”[3]
[3] Third report page 13
Consequently, the applicant was invited to file written submissions but, in spite of a further reminder, filed none. As such I will consider the material already on file.
Amendments
As noted above, the examiner has stated that the amendments proposed by the applicant during re-examination are not allowable. I agree. The proposed amendments attempt to delete many features from the claims such that the claims as proposed to be amended do not fall within the scope of the claims prior to amendment, contrary to the requirement of s102(2)(a). Therefore, this decision is in relation to the specification as amended on 3 January 2024 (i.e. containing the October amendments), and any reference to the specification in my decision is to the specification as amended.
The invention as described
The invention relates to what it calls a “Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission”[4] (MLBOCS for M-CVT). The specification states that the invention is “particularly appropriate for harvesting wind or wave energy as power supplied by waves or winds is variable in a wide range”.[5]
[4] Specification at [0003]
[5] See ibid
The specification states that a M-CVT has:
“a stationary member referred to as a stator and three rotors, which are a Mechanical Power Input Rotor (MPIR), an Electrical Power Input Rotor (EPIR) and a Mechanical Power Output Rotor (MPOR), to be either coaxial or noncoaxial.”[6]
Typically, the MPOR is connected to an electrical generator so that the rotation of the MPIR caused by waves or wind is used to generate electricity.
[6] Ibid at [0004]
Arrangements of rotors are shown in the figures. Figures 1(a), (b) and (c) are as follows:
Referring to figures 1(a) and (b), which is said to be a coaxial M-CVT, the specification states that, of the MPIR, the MPOR and the EPIR, there are two pole pair rotors 301, 303 and a pole piece rotor 302. The pole pair rotors have several pole pairs, with each rotor having a different number of pole pairs. The pole piece rotor has several ferromagnetic pole pieces, where the number of pole pieces equals the total number of pole pairs present in the pole pair rotors.
Figure 1(c), purports to show what is referred to as a non-coaxial M-CVT, also called a “bevelled M-CVT”. The specification states that, in a bevelled M-CVT, the rotors 338, 340, 344 are cones and at least two of the three rotors are non-coaxial. Assuming, for the non-coaxial arrangement, that the black and white strips represent respective pole pairs, it is not clear from the specification what numerical relationship exists between pole pairs 330 and what I assume to be the pole piece rotor 344, or whether the pole piece rotor has pole pieces at all. The specification goes no further in explaining figure 1(c) than I have set out above.
While noting that the variable rotating (electro)magnetic field created jointly by the MPIR and EPIR pair, interacts with the magnetic field of the MPOR to rotate the MPOR, the specification states that the EPIR “integrates” an electric motor/generator which works either as a motor or a generator depending on the mechanical power inputs and outputs. When the input speed of the MPIR is low the EPIR works as a motor and is rotated by electrical power to ensure the MPOR rotates at the desired speed; that is, by controlling its rotational speed, the EPIR effectively changes the gear ratio. Similarly, when input power is excessive, the EPIR operates as a generator converting excess mechanical energy into electricity, while keeping the MPOR rotating at the desired speed.
The specification then describes the MLBOCS, stating that this system is used to control the M–CVT in order to obtain optimal power or energy output. The specification states:
“The MLBOCS integrates a Deep Learning Optimization Algorithm (DLOA) which is used to solve the optimization problem for obtaining optimal solutions. The optimization problem, of which the Objective Function is the total power or energy output of the MCVT, is the Mathematical Model of the M-CVT. The combination of both the optimization problem (the Mathematical Model) and a Deep Learning based algorithm, which is the Deep Learning Optimization Algorithm (DLOA), to solve the optimization problem using Artificial Neural Network (ANN) in order to obtain optimal solutions with regards to time steps, is called the Numerical Model of the M-CVT. The MLBOCS composes of the Numerical Model of the M-CVT and Controllers which require Control Parameters derived from the optimal solutions of the optimization problem (the Mathematical Model). As a result, power outputs of the M-CVT are optimally controlled.”[7]
[7] Ibid at [0006]
The specification then explains that the states of the M-CVT are expressed by “a set of time history M-CVT-State-Data”[8], which data is obtained from a system of sensors. The state data and, therefore, the sensors required are very extensive. While a full list of sensors and data is given in claim 1 (set out below), as a small example, the specification states that the sensors monitor various properties of the rotors (MPIR, EPIR, MPOR) such as torque, rotational speed, acceleration of rotational speed, power consumed/generated (for the EPIR) over time.
[8] Ibid at [0007]
The specification states that the MLBOCS contains a Master Optimal Control System (MOCS) which is a control software with DLOA integrated into it. The DLOA is said to be trained by input data. This input data is said to be the time history M-CVT-State-Data collected from different sensors over time, but it does not have to be all the data that has been collected. The specification also states that some of the input data could be data that is derived from the collected data.
Discussing the construction of the mathematical model, the specification states:
“The Mathematical Model of the M-CVT (the optimization problem) applied to find maximum power or energy output is an optimization problem. Assuming that mechanical power outputs of the M-CVT is the function MOut(t(I)), of which t(I) is time step and I is index. If the M-CVT includes Energy Storages used as regulators, the function of power output of the Energy Storages (SOut(t(I))) is also taken into account. In addition, power consumed by the Master Optimal Control System (MOCS), including power used for changing gear ratios by rotating the EPIR, is expressed with the Function EConsumed(t(I)). Furthermore, power generated by the EPIR in the case it works as a generator is expressed with the function EGenerated(t(I)). Thus, we have the following functions of total power output and total energy output at the time step t(I):
(1)Total power output of the M-CVT at the time step t(I) is PowerOut(t(I)) = MOut(t(I)) + SOut(t(I)) - EConsumed(t(I)) + EGenerated(t(I)), where “I” is the index of the time step t(I).
(2)Total energy output of the M-CVT over a time interval Ti(I), which equals (t(I+1)- t(I)), is the function EnergyOut(Ti(I)) = (Ti(I)) x (PowerOut(t(I))) where Time Interval is the period of time between two adjacent time steps t(I) and t(I+1). Total energy output of the M-CVT over multiple Time Intervals is the sum of each total energy output of the M-CVT over each Time Interval.
(3)Total energy output of the M-CVT over a period of time, which is the sum of multiple consecutive time interval Ti(I), is the function TotalEnergyOut() where TotalEnergyOut() equals to the sum of multiple EnergyOut(Ti(I)).
Variables of the above functions are time (t(I) and varieties of variables represented in series of time history Input Data.”[9]
[9] Ibid at [0010]–[0011]
The specification further states that:
“if the M-CVT is connected with a Generator, the mathematical model of the DLOA is to solve an optimization problem which maximizes either power or energy output by maximizing the Objective Function of the M-CVT using Deep Learning Optimizers such as Gradient Descent or Conjugate Gradient.”[10]
[10] Ibid at [0012]
In this circumstance, the objective function could be one of the three noted above or another function representing the required mechanical inputs over a period of time of a device rotated by the M-CVT.
Similarly, the optimisation problem could be one of minimising the loss or cost of the objective function, so that the least amount of power output from the M-CVT is lost such that the device rotated by the M-CVT receives as much of that power output.
According to the specification, machine learning or deep learning are required to solve the optimisation problem given the complex interactions between devices of the M-CVT, while being subjected by variable mechanical power inputs and temperatures.
The output data of the MOCS (which will be used by the control policy of the MLBOCS) is various predictions of various parameters at a particular time step, or at various points in time, at which the objective function is optimal. The output data has a set of predicted Optimal Control Parameters (POCP) which are deployed for the control policy to apply to control (in an unspecified way) the controllable devices of the M-CVT, such as the EPIR, the clutch (optionally), regulating systems, etc. Essentially, the optimal solution which is predicted to occur at a predicted time step (or various time steps, which may or may not be regularly spaced) is obtained, where the predicted data derived from the optimal solution reflects the next state of the controllable components or portions of the M-CVT. The control parameters, which are used to move the M-CVT to its optimal state, are derived from this predicted data.
It can be the case that the MOCS can be started without any prior learning or optimisation at that point (starting with a simple control policy) while sets of data are being collected and the DLOA is trained. Once the DLOA is trained, an interactive process can occur as the DLOA processes newly collected data to find a more optimal state.
Finally, the specification states that the DLOA is composed of:
“popular deep learning algorithms used to solve the optimization problem (the Mathematical Model of the M-CVT) to find out optimal solutions with respect to optimal time steps. The deep learning algorithms includes artificial neural network (ANN)s with an input layer, multi deep layers and an output layer.”[11]
The specification does not provide any deep learning algorithm as they “are popular”[12].
[11] Ibid at [0016]
[12] See ibid
The invention as claimed
The specification ends with 8 claims. The entire claim set is set out in the Annex to this decision. Independent claim 1 is directed to the MLBOCS. Claim 1, which is lengthy, is as follows:
“1. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) comprising:
·a Magnetic Continuous Variable Transmission (M-CVT); wherein:
othe M-CVT comprising:
§a Mechanical Power Input Rotor (MPIR); wherein the MPIR receives mechanical power inputs;
§a Mechanical Power Output Rotor (MPOR); wherein the MPOR delivers mechanical power outputs;
§an Electrical Power Input Rotor (EPIR) comprising a motor-generator; wherein:
·the motor-generator is controllably operated by a source of variable electric power;
·the motor-generator is able to work relying on mechanical power inputs and mechanical power outputs for its dual functions of driving a device and generating electricity; wherein:
othe motor-generator works as an electric motor together with the MPIR to make the MPOR delivering controllable mechanical power outputs;
othe motor-generator works as a generator to generate electricity from excessed mechanical power inputs while maintaining the MPOR delivering controllable mechanical power outputs;
§wherein:
·among the MPIR, the MPOR and the EPIR, there are the followings rotors:
oa (first) Pole Pair Rotor, a (second) Pole Pair Rotor and a Pole Piece Rotor;
·wherein the pole pair rotors:
oeach Pole Pair Rotor has a number of pole pairs;
othe number of pole pairs of the (first) Pole Pair Rotor and the number of pole pairs of the (second) Pole Pair Rotor should be different;
othe pole pairs are putted together in a number of strips of pole pairs; wherein each strip is either continuous or noncontinuous;
·and wherein the Pole Piece Rotor:
othe Pole Piece Rotor has a number of ferromagnetic pole pieces;
othe ferromagnetic pole pieces are putted together in a number of strips of pole pieces; wherein each strip is either continuous or noncontinuous;
othe number of ferromagnetic pole pieces equals to total pole pairs of both the (first) Pole Pair Rotor and the (second) Pole Pair Rotor;
ototal strips of pole pairs of the two pole pair rotors equal to the number of strips of pole pieces;
§and wherein:
·the MPOR has a MPOR-Magnetic Field created by its pole pairs or pole pieces;
·a variable rotating (electro)magnetic field created jointly by the MPIR and the EPIR interacts with the MPOR-Magnetic Field making the MPOR rotated;
§and wherein the EPIR of the M-CVT is controlled by a control system to work as either an electric motor or a generator depending on mechanical power inputs of the MPIR and mechanical power outputs of the MPOR, including rotational directions; wherein:
·the control system controls electric currents supplied by a source to rotate the EPIR for changing gear ratios in order to maintain required mechanical power outputs of the M-CVT;
·the control system controls electric currents generated by the EPIR while maintaining required mechanical power outputs of the M-CVT;
·a Machine Learning Based Optimal Control System (MLBOCS) comprising:
oa system of M-CVT-State-Sensors;
oa Master Optimal Control System (MOCS);
owherein the system of M-CVT-State-Sensors comprising:
§a system of MPIR-State-Sensors further comprising:
·a system of MPIR-Torque-Sensors used to monitor torques of the MPIR;
·a system of MPIR-Speed-Sensors used to monitor rotational speeds of the MPIR;
·an optional system of MPIR-Accelerators used to monitor rotational accelerations of the MPIR;
·a system of MPIR-Pulse-State-Sensors used to monitor MPIR-Pulse-States;
§a system of MPOR-State-Sensors further comprising:
·a system of MPOR-Torque-Sensors used to monitor torques of the MPOR;
·a system of MPOR-Speed-Sensors used to monitor rotational speeds of the MPOR;
·an optional system of MPOR-Accelerators used to monitor rotational accelerations of the MPOR;
§a system of EPIR-State-Sensors further comprising:
·a system of EPIR-Torque-Sensors used to monitor torques of the EPIR;
·a system of EPIR-Speed-Sensors used to monitor rotational speeds of the EPIR;
·an optional system of EPIR-Accelerators used to monitor rotational accelerations of the EPIR;
·a system of EPIR-Power Consumed-Sensors used to monitor time history variations of a number of electric currents supplied to a number of coils of the EPIR working as an electric motor;
·a system of EPIR-Power Generated-Sensors used to monitor electrical power or energy generated by the EPIR working as a generator;
§a system of M-CVT-Mechanical Power Input-Sensors used to monitor mechanical power input of the M-CVT;
§a system of M-CVT-Mechanical Power Output-Sensors used to monitor mechanical power output of the M-CVT;
§an optional system of M-CVT-Clutch-State-Sensors used to monitor M-CVTClutch-States;
§An optional system of M-CVT-RS-State-Sensors comprising:
·a system of M-CVT-MPRS-State-Sensors used to monitor M-CVT-MPRS-States;
·a system of M-CVT-EPRS-State-Sensors used to monitor system of M-CVT-EPRS-States;
§a system of M-CVT-Power Consumed-Control System-Sensors used to monitor electrical energy consumed for controlling the M-CVT; wherein the M-CVT-Power Consumed-Control System-Sensors comprising a system of EPIR-Power Consumed-Sensors;
§a system of M-CVT-Electrical Power Output-Sensors used to collect time history data of power output of the M-CVT;
§an optional system of Temperature-State-Sensors;
oand wherein the Master Optimal Control System (MOCS) comprising:
§a set of time history Input Data, known as the M-CVT-State-Data;
§a set of time history Output Data; wherein:
·the Output Data comprising a set of time history Predicted Optimal Control Parameters (POCP);
·the POCP is used to construct the Control Policy of the MOCS;
·the Control Policy is used to control operations of controllable devices of the M-CVT;
§a Controlling Software; wherein:
·the Controlling Software reads the set of Input Data;
·the Controlling Software generates the set of Output Data;
·the Controlling Software instructs the controllable devices of the M-CVT to progress to new states specified by the set of Predicted Optimal Control Parameters (POCP);
·the Controlling Software comprising the Deep Learning Optimization Algorithm (DLOA) using the set of Input Data and generating the set of Output Data;
§a Deep Learning Optimization Algorithm (DLOA); wherein the DLOA comprising an algorithm of Deep Learning using Artificial Neural Networks to solve an Optimization Problem; wherein:
·the Deep Learning Optimizers applied can be any kind of popular Optimizers such as Gradient Descent or Conjugate Gradient;
·the Optimization Problem is a Mathematical Model of the Multiphysics System of the M-CVT;
·the Multiphysics System incorporates complex interactions between a number of multi physical portions of the M-CVT such as temperatures, magnetic fields, Regulating Systems and motions of mechanical devices;
·the Objective Function of the Optimization Problem of the M-CVT is optionally chosen from the following:
opower output of the M-CVT expressed as the function Power-Out(t(I+1)); wherein t(I) is the time step at the time indexed (I) and t(I+1) is the time step at the time indexed (I+1);
oenergy output of the M-CVT expressed as the function Energy-Out(Ti(I)); wherein Ti(I) is the time interval between the time step t(I) and the time step t(I+1);
ototal energy output of the M-CVT over a period of time covering multiple time intervals, expressed as the function Total-Energy-Out();
oDesired Function() minus Power-Out(t(I+1)) or Energy-Out(Ti(I)) or Total-Energy-Out(); wherein the Desired Function() expresses time history mechanical power inputs required by a device rotated by the MPOR;
·the solutions of the Optimization Problem are used to construct the set of Predicted Optimal Control Parameters (POCP);
§and wherein the Deep Learning Optimization Algorithm (DLOA) further optionally comprising one or more of the following strategies:
·an optional Variable Time Interval Method applied for optimizing the Objective Function with variable time intervals;
·an optional Preselected Time Interval Method applied for optimizing the Objective Function with preselected time intervals;
·an optional Interactive Learning Control Method (ILCM) applied for learning and optimizing phases of the Deep Learning Optimization Algorithm (DLOA);
·an optional Untrained Initialized Method (UIM) applied for the initialization of the Master Optimal Control System (MOCS) with an untrained Deep Learning Optimization Algorithm (DLOA);
§and wherein the set of time history Input Data comprising:
·a set of time history MPIR-State-Data further comprising:
oa set of time history MPIR-Torque-Data obtained from the system of MPIR-Torque-Sensors;
oa set of time history MPIR-Speed-Data obtained from the system of MPIR-Speed-Sensors;
oa set of time history MPIR-Acceleration-Data obtained from the system of MPIR-Accelerators;
oa set of time history MPIR-Pulse-State-Data obtained from the system of MPIR-Pulse-State-Sensors;
·a set of time history MPOR-State-Data further comprising:
oa set of time history MPOR-Torque-Data obtained from the system of MPOR-Torque-Sensors;
oa set of time history MPOR-Speed-Data obtained from the system of MPOR-Speed-Sensors;
oa set of time history MPOR-Acceleration-Data obtained from the system of MPOR-Accelerators;
·a set of time history EPIR-State-Data further comprising:
oa set of time history EPIR-Torque-Data obtained from the system of EPIR-Torque-Sensors;
oa set of time history EPIR-Speed-Data obtained from the system of EPIR-Speed-Sensors;
oa set of time history EPIR-Acceleration-Data obtained from the system of EPIR-Accelerators;
oa set of time history EPIR-Power Consumed-Data obtained from the system of EPIR-Power Consumed-Sensors;
oa set of time history EPIR-Power Generated-Data obtained from the system of EPIR-Power Generated-Sensors;
·a set of time history M-CVT-Mechanical Power Input-Data obtained from the system of M-CVT-Mechanical Power Input-Sensors;
·a set of time history M-CVT-Mechanical Power Output-Data obtained from the system of M-CVT-Mechanical Power Output-Sensors;
·an optional set of time history M-CVT-Clutch-State-Data obtained from the optional system of M-CVT-Clutch-State-Sensors;
·an optional set of time history M-CVT-RS-State-Data further comprising:
oan optional set of time history M-CVT-MPRS-State-Data further comprising a set of time history M-CVT-MPRS-Control Parameter-Data;
oan optional set of time history M-CVT-EPRS-State-Data further comprising a set of time history M-CVT-EPRS-Control Parameter-Data;
·a set of time history M-CVT-Power Consumed-Control System-Data further comprising the set of EPIR-Power Consumed-Data;
·a set of time history M-CVT-Electrical Power Output-Data;
·an optional set of time history Temperature-State-Data;
·an optional set of time history Variable-Time Interval-Data;
§and wherein the set of time history Output Data comprising:
·a set of time history Predicted Optimal Time Step-Data;
·a set of predicted M-CVT-Mechanical Power Input-Data;
·a set of predicted M-CVT-Mechanical Power Output-Data;
·an optional set of predicted M-CVT-Clutch-State-Data;
·a set of predicted MPIR-Pulse-State-Data;
·a set of predicted M-CVT-Electrical Power Output-Data;
·a set of predicted Optimal Control Parameter (POCP)s applied at the Predicted Optimal Time Step; wherein the POCPs comprising:
oa set of predicted EPIR-Power Consumed-Data and a set of predicted EPIR-Power Generated-Data containing parameters for controlling the EPIR at the Predicted Optimal Time Step; wherein:
§the EPIR is switched between states of working as an electric motor or a generator;
§as an electric motor, using the predicted EPIR-Power Consumed-Data, the EPIR is variably rotated by handling the controllable electric currents consumed by the EPIR in order to handle required rotations of the MPOR;
§as a generator, while mechanical input power (MIP) being excessed, the EPIR is controlled for both handling required rotations of the MPOR and generating electricity using the excessed MIP;
oan optional set of time history M-CVT-Clutch-Control Parameter-Data and an optional set of predicted MPIR-Pulse-State-Data containing parameters for controlling the optional M-CVT-Clutch at the Predicted Optimal Time Step;
oa set of predicted M-CVT-MPRS-Control Parameter-Data and a set of predicted M-CVT-EPRS-Control Parameter-Data containing parameters for controlling the M-CVT-MPRSs and the M-CVT-EPRSs at the Predicted Optimal Time Step.”
The claim essentially defines what I have set out above as the described invention. That is, in a particular M-CVT system, a control system that makes use of a multitude of data sets and derives a predicted optimal set of parameters for various points in time.
Examination
First report
As already noted earlier in this decision, in the first report of 28 February 2024, the examiner raised an objection stating that the specification did not contain a complete enough disclosure, and the claims lacked support. The examiner stated that the claims appeared to be a mere list of technical features, with a claimed desired result of an optimal control system, without the specification providing sufficient technical details to enable a person skilled in the art (PSA) to achieve the result.
Specifically, the examiner stated that the description lacked detail as to the construction of the M-CVT as the specification omitted details of its construction, operational arrangement of its components, and the composition or the operating characteristics, of the components. For example, the examiner stated that the juxtaposition of the rotors when in a coaxial or non-coaxial position was not disclosed, and that the specification did not provide enough information to allow the PSA to determine the size of gaps between rotors to “create the best interactions between magnetic fields of the MPIR, the EPIR and the MPOR.”[13] The examiner also noted that it was not clear how any stator associated with the EPIR was fabricated, or how it was positioned relative to each of the other rotors to achieve a motor or generator effect.
[13] Ibid at [0004]
The examiner also stated that the detail surrounding the state sensors was limited and did not explain how the readings from all the sensors would be used given the different data rates for each sensor, the location of the sensors or how they communicated their readings. The examiner also referred to a lack of details around the process of training the DLOA with respect to the labelling of ground truth data or hyperparameters. Lastly, the examiner referred to a lack of direction of how to construct a control policy from the POCPs, stating that the PSA would have no guidance on how to proceed with the construction of a control policy from the POCPs, which would necessitate further invention.
Response to first report
The applicant responded (applicant’s first response) stating that the M-CVT and control system used were conventional. As such, in the applicant’s opinion, details about the physical structure of the M-CVT would be computed by design engineers and it was not reasonable to provide such estimates. The applicant stated that the invention was an algorithm/software. The applicant modelled a conventional M-CVT to obtain a model, which was a function, then optimised the function using a conventional optimisation method (deep learning algorithms) to obtain an optimal solution which is outputted for use by a conventional control system.
The applicant stated that their model of the M-CVT was a numerical model, being the set of time history state data (whose structure was defined in the description) forming a non-continuous multivariable function encompassing complex interactions between the components of the M-CVT because the time history was a record of the operational states of the M-CVT as the components interacted. The applicant stated inter alia:
“Once the model (or the function) of the M-CVT is said to be trained, using deep learning, … the model become an activation function (a neural network model). So, the activation function is actually the model of the M-CVT. When the (trained) model is said optimized, using deep learning, … the activation function is optimized and the activation function, which is the (trained) model, is also the objective function of the optimization problem. So, the objective function, which is the (trained) model of the M-CVT and which is multivariable, expresses operational states of the MCVT. Optimizing the (trained) model, (which is the activation function as well as the objective function), using deep learning, means to search for optimal states of the M-CVT which is the optimal solution. As there is only one possibility for the process of using conventional deep learning algorithms for optimization from training to obtaining the optimal solution, this process is not necessary to be presented in details for all steps of the process.”[14]
[14] Applicant’s first response pages 2-3
The applicant further submitted that features such as sensor data rate and size of gaps would be within the knowledge of design engineers and the appropriate sensors and M-CVT designs would be selected and computed as appropriate.
Second report
In their second report, after noting that the amendments filed by the applicant were not allowable, the examiner addressed the applicant’s arguments. Noting the applicant’s statement that the invention was algorithm, the examiner stated that there were no details of the algorithm. Regarding the applicant’s statement that their model was the set of time history M-CVT-State-Data, the examiner noted that, firstly, the specification did not contain such a statement and, secondly, a model would be “a single entity e.g. a single mathematical expression or a set of interrelated expressions”[15] rather than disjointed data sets. The examiner also noted that the applicant had misunderstood their point about the specification not explaining how the readings from all the sensors would be used given the different data rates for each sensor, and this still needed to be addressed.
[15] Second report page 3
The examiner noted a contradiction between the applicant’s statement that the model of the M-CVT was the set of time history M-CVT-State-Data and the applicant’s statement that the model of the M-CVT was an activation function (a neural network model). The examiner also noted a lack of disclosure of a trained neural network being a model. Further contradictions were pointed out by the examiner between the statement that the neural network activation function (which is deep learning) was the model which was invented and various statements that explanations on deep learning algorithms or optimizations using deep learning algorithms were not relevant to the invention. As no information about deep learning had been provided (because it was apparently not relevant), and the neural network model developed by deep learning was the invention, it was the examiner’s opinion that the PSA faced an undue or excessive burden to perform the invention as claimed.
Having noted the applicant’s statement that details about the physical structure of the M-CVT would be computed by design engineers and it was not reasonable to provide such estimates, the examiner then stated that, as the specification did not provide at least one example, or specific range, of (for example) suitable rotor gap(s), it did not disclose a method of putting the invention into effect. The consequence was that the specification did not satisfy s40(2)(aa) which required the disclosure of best method known to the applicant of performing the invention.
Response to second report
The applicant responded (applicant’s second response) stating that their software only had the task of performing optimisation, and criticised the examiner for requiring the control software to carry out other tasks such as evaluating and handling the set of output data, and controlling the components of the M-CVT.
The applicant also stated that there was no relation between the gaps/rotational operations of the M-CVT and the equations governing the rotation of the rotors of the M-CVT. As such, there was no need to provide examples of rotor gaps as they were not relevant to the invention.
With respect to the model, the applicant stated that a function can be expressed as either a set of data or a mathematical expression. As the set of input data comprised several subsets of data versus time, the set of input data was already a function versus time. The applicant also disagreed with the characterisation of the data sets as “disjointed”, noting that (i) speed, velocity and acceleration were interrelated as derivatives, (ii) pulse data was “the set of pulse states of the rotor MPIR …, as explained in my description [which I note is not explained], is the collection of speeds of the MPIR when the speeds equal zero”[16], torque was related to speed and mechanical input/output power, and (iii) the mechanical input power of the MPIR plus the mechanical input power of the EPIR (converted from the electric power input of the EPIR) equalled to mechanical output power of the MPOR (minus energy lost by heat/friction).
[16] Applicant’s second response page 3
The applicant stated that the time history input data could be said to be an initial model expressed as a non-continuous function which is then further transformed by using artificial neural networks using deep learning algorithms into a trained model which was a linear continuous function being the model of the M-CVT. The applicant stated that:
“instead of forming the mathematical model of the M-CVT using mathematical transformations basing on equations and formulars [sic], I was using a numerical transformation to form the mathematical model of the M-CVT by using the ANN”[17],
and
“forming such kind of the mathematical model from a set of input data without using mathematical transformations basing on equations and formulars [sic] is one of the main reasons that AI/ANN theories are developed:
Mathematicians did want to find out a way to form a mathematical function expressing interrelations between the subsets of the set of input data. In other words, the mathematicians found out the way to transform a set of data to a mathematical function.”[18]In the applicant’s opinion, this also addressed the issue with contradictions raised by the examiner.
[17] Ibid page 5
[18] See ibid
On the question of data rates, the applicant stated that the manufacturers of the sensors would make data collection as easy as possible, and the PSA would know that they needed to select sensors that had a data rate that would match with the desired rotational speeds of the rotors. The applicant also noted that using variable time intervals in the process of optimisation would adapt to varying input data and data rates.
Third report
In the third report, the examiner maintained the view that the complete specification did not enable the PSA to perform the invention as claimed because the complete specification did not provide sufficient technical details regarding some of the features recited in the claims. The examiner once again noted the lack of significant details as to the construction or operational arrangement of the components, their composition, or their operating characteristics of the M-CVT. The examiner also noted again the lack of disclosure with respect to the material from which a stator could be fabricated and positioned relative to each of the rotors, in order to achieve a motor or generator effect with the EPIR.
The examiner referred once more to a lack of details around the process of training the DLOA with respect to the labelling of ground truth data or hyperparameters, noting that the absence of this information placed an undue burden on the PSA trying to achieve the desired result. The examiner again referred to a lack of direction of how to construct a control policy from the POCPs, and how the control policy is actually used to control the controllable devices. The examiner stated that the PSA would have no guidance on how to proceed with the construction of a control policy from the POCPs, which would necessitate further invention. The examiner stated:
“Claims 1-8 currently appear to be a mere list of technical features put together claiming a desired result without the specification sufficiently explaining how the result is to be achieved. Particularly, the claims appear to be no more than speculative, given the absence of concrete/definite technical elements or steps in your specification that tie all the recited features together in such a way as to enable a person skilled in the art to successfully perform your invention.”[19]
[19] Third report, page 5
The examiner’s responses to the applicant’s arguments were as follows (underlining and italics in original):
“Argument 0: hooking irrelevant tasks for rejecting my application.
You submit:
My software claimed has only one task: performing optimization using a ready-made set of input data to obtain a set of output data. The (optimization) software does not embrace everything. It excludes the tasks of data collection, data preparations, handling the set of data and so on. It does not include how a control software evaluates the set of output data, how the control software handles the set of output data, how the control software controls the components of the M-CVT.
Essentially, you are arguing that your claimed invention merely uses an already trained deep learning algorithm to simply optimise a numerical model. That is, the software is abstracted from the actual device and merely requires a black box with data inputs and data outputs that are then applied to an optimisation process.
The specification clearly states:
‘Firstly, the MLBOCS for M-CVT starts working with a simple control policy which does not need to include optimization. In this case, although the energy generated is not optimized, the control system can be configured to be able to work without optimization while data are being collected and the DLOA is being trained. Once the DLOA is fully trained, it is ready to work fully.’ [ see for example, Description: Page 14 ]
Therefore, the training the DLOA in the context of an actual functional M-CVT device is a necessary part of the invention and requires more than simply a black box as you argue.
Argument 1: the gaps between the rotors
The original objection raised included: (28 February 2024)
The description also makes broad general statements such as: ‘In addition, gaps between the MPIR, the EPIR and the MPOR are small enough in order to create the best interactions between magnetic fields of the MPIR, the EPIR and the MPOR’ [see Description: Page 2]; or ‘It is particular preferred if the mechanical input power is harvested from waves or winds’. However, it is unclear how this is achieved in practice by the MCVT, especially in the context of the other operating components, such as the mechanical power output rotor. For example, the specification provides no further information on the dimensions of the gaps between the rotors. The range of potential gap dimensions (possibly ranging from say 0.1mm or 10cm) is so huge that the person skilled in the art would need to carry out a vast number of experiments in order to determine a ‘small enough’ rotor gap for which your invention will work as claimed. The absence of a specific rotor gap dimension in your specification also means that that the person skilled in the art is left in doubt regarding the tolerances to work towards when milling the various components of each rotor.
In your previous response submission you stated: (25 April 2024)
The gaps you said, like gaps of an electric motor, depend on many factors and ARE COMPUTED BY DESIGN ENGINEERS basing on rotational speeds, distribution of magnets, properties of magnets, properties of bearings, tolerance of manufacturing, how large the device is and properties of rotational drive shafts, and properties of the rotors such as weight. It is not reasonable to provide an estimation of measurements of the device while the measurements MUST BE COMPUTED and CAN BE COMPUTED by any design engineer with ease.
In your current response submission you argue: (31 July 2024)
THERE IS NO RELATION between the gaps and rotational operations of the M-CVT. THERE IS also NO RELATION between the gaps and ANY control system used to control the MCVT. In other words, rotational operations and controlling the M-CVT are completely independent with the gaps. No matter how the gaps are, the software and the control system can still work well. What I proposed can work (‘putting the invention into effect’) well without any requirement about the gaps as they are not relevant.
... Furthermore, my claim does not mention anything related to gaps.
I note that your two statements are contradictory. In the first, you argue the gaps are part of the invention, but are well-known and hence could easily be calculated by an engineer. However, in the second, you argue the gaps are irrelevant to the operation of the M-CVT.
Furthermore, the applicant submits:
I am proving this argument using mathematics and mechanical mathematics with the governing equations of motions. The equations govern rotations of the rotors of the M-CVT, in the Appendix 1 and the Appendix 2 below. The three rotors rotate and transmit mechanical power between them without anything related to the gaps: It means the gaps are not relevant.
I disagree. Your reference to general equations such as the Governing and Willis equations does not provide evidence that the gaps are not relevant to your invention. These equations do not form part of the specification as originally filed. Hence, your reference to these equations merely provides common general knowledge as to how a general rotor operates and not evidence that the gaps are not a [sic] necessary features for the working of the invention.
Specifically, and most importantly, the focus of the objection is in relation to the statement in your specification that states:
‘In addition, gaps between the MPIR, the EPIR and the MPOR are small enough in order to create the best interactions between magnetic fields of the MPIR, the EPIR and the MPOR.’ [ see Description: Page 2 ]
This statement makes it clear that the gaps are not an optional feature of the invention, but rather are part of your invention that is to be optimised.
Furthermore, your invention is defined in the specification (both in the description and claims) to include: a control system (MLBOCS) and the M-CVT. The above mentioned statement makes it clear the M-CVT components of the MPIR, EPIR and MPOR interact and require a gap small enough to create the best interaction, which is then controlled by MLBOCS to optimise power output [ see Description: Page 2, 14 ].
Hence, based on the above and the fact that the entire specification details an optimisation process, it is logical and consistent to conclude that the optimisation of the gaps (smallest gap to produce the best interaction) between the MPIR, EPIR and MPOR is a necessary part of the invention.
Therefore, the objection in relation to these gaps is maintained.
Argument 2: the set of time history input data is a model or not to be the model of the M-CVT?
The original objection raised included: (28 February 2024)
‘At argument 1.2 in your response, you have stated that the set of time history M-CVT-State-Data is a model of the M-CVT. Firstly, your specification as filed does not indicate that the model is M-CVT state data. Secondly, a person skilled in the art would understand a model to be a single entity e.g. a single mathematical expression or a set of interrelated expressions. However, the set of time history M-CVT State-Data as defined in your specification is a set of data regarding torque, acceleration, speed, current, energy/power generated or consumed etc, each obtained from respective sensors. Other than the fact that these pieces of data are from the same M-CVT, these are disjoint sets’
The specification provides no specific details as to how such a so-called model integrates any of the various component interactions of the M-CVT. Instead it merely discusses collecting the set of time history data and providing this data as input to a generic optimisation (deep learning algorithm) to obtain an optimal solution.
In summary, the applicant submits:
There is nothing wrong if I say my model is expressed with a set of data.
The set of input data of mine comprises several subsets of time history data, each subset contains data versus time, meaning that each subset is already a function versus time. Thus, the set of input data is already a FUNCTION, at least versus time.
The subsets comply with mechanical rules for rotations and power conversion that skilled persons should know...
From (1) to (8), it is clear that all the subsets of data of the set of input data, as a whole, have INTERRELATIONS BETWEEN THEM, mechanically, and also versus time. It is clear that the set of input data is a ‘set of interrelated expressions’. Thus, the set of input data is a model that I already featured it, defined its characteristics, formed its contains and structured it with multi-INTERRELATIONS, although anything related to fundamental knowledges
Putting aside the multi-INTERRELATIONS I explained, when the set of input data is collected while the M-CVT is operating, as I explained many times, it already represented operational states of the M-CVT. The operational states themselves are already a model as defined in physics. While the data is collecting, components and subsystems of the M-CVT are interacting together creating the data which already contain the interactions and relations of the subsets of the set of input data.
Your argument here is essentially that no further details are required as it is all well-known and hence such details are easily derivable by a person skilled in the art.
However, this also means that you are arguing your invention is merely using well-known mechanical and power equations to construct well-known input and output interactions between collected data that is merely applied to a well-known optimisation process via well-known deep learning techniques.
Hence, on the balance of probabilities, I see no ingenuity in terms of the model at all. Further discussion, in terms of clear and complete disclosure is made below in relation to your Arguments 4-5.
Argument 3: everything related to sensors you raised in the reports
The original objection raised included: (28 February 2024)
The description outlines that the MPIR, EPIR and MPOR comprise various state sensors to collect mechanical and electrical power input/output, temperature, acceleration, torque, electric current, and speed data. However, as the person skilled in the art would understand, these sensors have vastly different data rates (i.e. the number of readings per unit time varies dramatically for these sensors and the person skilled in the art then faces the problem of how many readings of each sensor to use, e.g. how many readings of the acceleration sensor, which typically changes very frequently, should be used vis-a-vis readings from the temperature sensor, which typically changes much more slowly? In other words, are all the readings from all sensors to be used or is there some temporal filtering involved? If there is no temporal filtering involved, the person skilled in the art would face the problem of how to apply these unequal readings per unit time/step as input into the deep learning algorithm. Moreover, there are no details provided on the actual sensor type, the location of the sensors (on or around the MPIR, EPIR and MPOR) or details regarding sensor data communication means in the context of the MCVT.
In summary you raised the following arguments:
All types of sensors referred in my application, and most of any other types of sensors, have only one method to collect data, that the data is read by a computer via either a wireless or a cable. Collecting data from a sensor is always guided carefully by all manufacturers who also make sure that all users can use their sensors to collect data with ease.
In addition, any other points you raised relating to sensors can also be handled easily by skilled persons for the same reasons explained above.
Thus, the software can work with any data rates of sensors which provide data at any time. If the data rates are faster, the accuracies of the software are better or vice versa.
It is clear that either the Variable Time Interval Method or the further refined fixed time step is used to handle convergence, including convergence rates, of the process of optimization to adapt with varying input data, including your ‘plurality of data sources of such widely varying rates’. It is obvious that, in order to adapt with more refined variables of a function, any method of time step integrations used need to have time interval refined. This is also a fundamental knowledge. So, the ‘widely varying rates’ you said and broader and the wide range of varying data of my set of input data working together are handled in my software by using refiner time steps as explained.
You are essentially arguing that the sensors and their implementation are well-known and easily derivable by a person skilled in the art. I agree, the sensors as claimed and described in the specification are merely generic; and no details in relation to their type, location, and configuration are discussed anywhere within the specification [see Description: particularly Pages 4-9]. Hence, on the balance of probabilities, I see no ingenuity in the sensors as input devices or their implementation, that is, they are merely an arrangement of well-known generic sensors that a person skilled in the art could easily select and place at the required locations and configure in order to collect data and handle any data rate issues that may arise.
Argument 3: the role of the motor-generator
Your submission in relation to the ‘role of the motor-generator’ is in response to an objection raised in the previous exam report, specifically, the removal of specific features:
‘The role of the motor-generator being controllably operated by a source of variable electric power is completely omitted from the claim(s) filed with the amendment request’
However, as noted above and explained in detail below at objection item 3, your current proposed amendments are not allowed.
The current objection is based on your claims filed on 3 October 2023. In those claims, the features related to ‘the role of the motor-generator being controllably operated by a source of variable electric power’ are included and therefore your arguments are not relevant to the current objection raised.
Nevertheless, for the sake of completeness, I note that your arguments seem to have completely missed the point of that previous objection. That is, the basis of that objection was about your removal of features in relation to the control system and the control of the motor-general, and hence your arguments in relation to the equivalence of mechanical power and electrical power is not relevant at all. In removing the features of the control system and its role in controlling the motor-generator you are broadening the scope of the claims, which is not allowed.
Argument 4: how did I present in my description about forming the model of the M-CVT
In summary, you submitted:
‘time history input data, is regarded to be an initialized model.
The initialized model is expressed with a noncontinuous function presented with the set of input data having interrelations between them.
The initialized model is transformed to an artificial neural network (ANN) model, which is called the trained model, with a training process, using the deep learning algorithms referred in my claims.
At the completion of training the ANN, the ANN model (the trained model) is obtained together with the optimal solutions of the optimization.
The trained model is a linear continuous function which expresses the model of the M-CVT.’
You seem to be suggesting the so-called ‘initialised model’ is a numerical model to model the behaviour of the M-CVT in terms of mechanical and power equations (that you refer to in your ‘argument 2: items 1-9’. However, your specification does not define this model, except to say it has collected input data and output data from various sensors. And your response, argues the construction of this model is merely common general knowledge that could be easily derived by a person skilled in the art. Further, you seem to be arguing that you can define the M-CVT interactions in terms of specific closed form equations, however, this implies that no form of modelling is actually required, since these equations provide the specific interactions between the input and output of the system. Hence, all that remains is a system that simply collects input and output data.
Argument 5: further points regarding to handling input data and selecting an appropriate deep learning algorithm.
The specification merely states that the ‘deep learning algorithm is trained with input data’ [ see Description: Pages 9, 14 ]. The specification provides no further details about the training process or algorithms used and your arguments merely infer that this information could [sic] is common general knowledge that is easily derivable by a person skilled in the art.
You submitted in response to this:
The task of handling input data and other common tasks of using ANN, such as the tasks listed below, are common in solving any problems using ANN. In other words, any skilled person has to encounter the task listed below in solving any problem using ANN. Thus, the skilled person needs to understand such kind of common knowledge in order to implement my software. 2) The listed tasks are fundamental in using ANN.
In essence, you are arguing that the deep learning algorithm and its training is merely common general knowledge and a person skilled in the art could easily solve these details.
However, you then state:
The wide range of general M-CVTs with general operational conditions makes the task of determining an appropriate deep learning algorithm and associated hyperparameters to be specified (specifically) is impossible. There is NO appropriate deep learning algorithm and associated hyperparameters for the wide range of configurations of the M-CVTs, each further with a wide range of operational conditions while my software is recommended for any M-CVT with any operational conditions.
Now, in contradiction to your previous statement, you are arguing that it is impossible to provide such details. Here, you are essentially arguing that a person skilled in the art cannot implement a best method of the invention as it requires further trial and error or invention to determine such details, which actually supports the maintaining of the objection raised against your invention as claimed.
In conclusion, your response submissions, merely argue that these features are common general knowledge that a person skilled in the art could easily derive without ingenuity, which I do not find convincing as discussed above.
Therefore, for the above reasons, I maintain that your complete specification does not provide a clear enough and complete enough disclosure of the invention defined by Claims 1-8 (as filed on 3 October 2023).”[20]
[20] Third report pages 5–10
The examiner also maintained the support objection for these same reasons.
s40(2)(a) – Clear enough and complete enough disclosure
The requirement for clear enough and complete enough disclosure was introduced into the Patents Act 1990 (the Act) as part of the Intellectual Property Laws Amendment (Raising the Bar) Act 2012 (RTB) reforms. Specifically, s40(2)(a) reads as follows:
“(2) A complete specification must:
(a)disclose the invention in a manner which is clear enough and complete enough for the invention to be performed by a person skilled in the relevant art;…”
As indicated in Encompass Corporation Pty Ltd v InfoTrack Pty Ltd[21] the requirements of s40(2)(a) equate to enablement of the invention; I note that this was undisturbed in the appeal to Encompass[22]. As explained in the Explanatory Memorandum to the RTB legislation at item 8, enablement amounts to a requirement that “…sufficient information must be provided to enable the whole width of the claimed invention to be performed by the skilled person without undue burden, or the need for further invention”.
[21] [2018] FCA 421 (Encompass) at [167]
[22] Encompass Corporation Pty Ltd v InfoTrack Pty Ltd [2019] FCAFC 161
The nature of s40(2)(a) was considered in some detail by Dr Barker in CSR Building Products Limited v United States Gypsum Company[23], including an extensive consideration of a number of UK and EPO decisions relevant to an understanding of this part of the Act.
[23] [2015] APO 72 (CSR)
After this consideration Dr Barker provided a test for s40(2)(a):
“In order to decide whether a specification provides a disclosure as required by section 40(2), it is necessary to:
(i)construe the claims to determine the scope of invention as claimed,
(ii)construe the description to determine what it discloses to the person skilled in the art, and
(iii)decide whether the specification provides an enabling disclosure of all the things that fall within the scope of the claims.”[24]
[24] CSR at [95]
In Evolva SA[25], Dr McCaffery provided some further analysis and consideration of UK and EPO decisions relevant to the question of s40(2)(a). After having done so, Dr McCaffery expanded on the third point from the test in CSR as follows:
“Does the specification provide an enabling disclosure of all the things that fall within the scope of the claims, and in particular:
(i)Is it plausible that the invention can be worked across the full scope of the claim?
(ii)Can the invention be performed across the full scope of the claim without undue burden?”[26]
[25] [2017] APO 57; 133 IPR 147 (Evolva)
[26] Evolva at [45]
I note that the above approach was approved by Justice Burley in Cytec Industries Inc. v Nalco Company[27]. It was also adopted by Justice Burley in TCT Group Pty Ltd v Polaris IP Pty Ltd[28], in the context of determining priority dates, albeit without reference to the sub-test from Evolva.
[27] [2021] FCA 970; 162 IPR 202 at [143] to [146]
[28] [2022] FCA 1493; 170 IPR 313 at [154]
Consideration
The examiner’s position is correct. While I acknowledge that it is permissible to have regard to the common general knowledge when seeking to implement an invention, the specification is also not a puzzle to be solved by the PSA. In the present case, the high-level diagrams and operational aspects on the apparatus and control system leaves out pieces of the puzzle detailing how the system is programmed or performs a particular task.
Construction of the M-CVT – Gaps
The claims contain no limitation on the gap between the various rotors of the M-CVT. At first glance, it would seem reasonable to say that the gaps to be used in this M-CVT would be something that the PSA would be able to ascertain through routine trial and error. However, as noted by the examiner, the gaps are said by the specification to be such that they create the “best interactions between magnetic fields of the MPIR, the EPIR and the MPOR”[29]. Moreover, I note that what is considered “best” is not explored by the specification.
[29] Specification at [0004]
While it could be said that the “best” interactions are ones where the strongest possible magnetic field exists between the magnets of the pole pair rotors and the pole pieces, it is not at all established by the specification that this must be the case for the system which is to be controlled by the described MLBOCS. As the examiner noted, the “range of potential gap dimensions (possibly ranging from say 0.1mm or 10cm) is so huge that the person skilled in the art would need to carry out a vast number of experiments in order to determine a ‘small enough’ rotor gap for which your invention will work as claimed.”[30] Moreover, assuming a strong magnetic field is what is required, it may be the case that the thickness of the pole pair magnets and/or pole pieces may have a role to play.
[30] Third report at page 4
However, the specification does not discuss such things. To my mind, it cannot be said that the specification provides an enabling disclosure of all possible arrangements of gaps and magnet/pole piece thicknesses to achieve the “best” interaction within the claimed MLBOCS-controlled system. The level of experimentation required is such that the invention could not be performed across the full scope of the claim without undue burden.
Construction of the M-CVT – Stator
During re-examination, the examiner stated that the disclosure did not indicate the presence of a stator associated with the rotors. The examiner also noted that the material from which the stator was to be fabricated and how it was to be positioned relative to the rotors in order to achieve a motor or generator effect was not disclosed. As to the first point, the examiner was incorrect since, as noted at [9] above, the disclosure indicated the presence of a stator. However, I note that (i) the mention of a stator was introduced by the October amendments and (ii) the description of this stator goes nowhere after this. As to the second point, while the material from which the stator is made could be said to be well within the skill of the PSA, the examiner was correct when they stated that how the stator was to be positioned relative to the rotors was not discussed. To my mind there remains much mystery around the stator that must be present for the motor/generator function to be achieved.
Throughout the specification the applicant is coy as to which rotor is the MPIR, which is the MPOR and which is the EPIR. Looking at figure 1(a) and (b) shown at [10] above, assuming that the three rings represent the rotors, if a stator is to be present it is difficult to see how it could be positioned anywhere other than either around the outside of the outermost rotor 301, or inside the innermost rotor 303. Following basic motor/generator design, that stator position would mean that the EPIR is either the outmost pole pair rotor or the innermost pole pair rotor. This would then seem to suggest that the EPIR cannot be the pole piece rotor, and the stator is another piece which carries coils. However, the specification, when describing the state sensors, states that the system of EPIR-Power Consumed-Sensors monitors the variation in:
“a number of electric currents supplied to a number of coils of the EPIR for the case that the EPIR works as an electric motor”.[31] (my emphasis)
This suggests that it is the EPIR that carries the coils. How that is to be achieved for the arrangement shown in figures 1(a) and (b) is not discussed. Moreover, the arrangement of the EPIR carrying the coils would also seem to rule out another possible configuration that might have occurred to the PSA, being that of a separate rotor/stator component which is connected to the EPIR and either rotates the EPIR when it operates as a motor, or is rotated by the EPIR when it operates as a generator. The coils being located on the EPIR makes this impossible.
[31] Specification at [0007](3)(a)IV
Moreover, when it comes to the arrangement of pole pair rotors and the pole piece rotor shown in figure 1(c), the situation is even more dire. It is difficult to envisage how what might be termed a “conventional” stator could be incorporated into this arrangement, let alone an arrangement where the EPIR carries coils.
As such, to the extent that a stator is described (which is more or less that one is present), it would require undue experimentation by the PSA to find a location for a stator that achieves the function of allowing the coil-carrying EPIR to operate as a motor and generator. There is simply no enablement.
Training the DLOA
As noted by the examiner, the specification provides little detail as to this algorithm beyond stating that it is performed by an artificial neural network and making reference to optimisation methods such as gradient descent or conjugate gradient. Furthermore, as expressed by the examiner in each report, no guidance is provided as to how to select and label ground truth data, as well as the hyperparameters for training the DLOA.
Throughout the re-examination, the applicant maintained that their invention was an algorithm but supplied no details. The justification for this was that such algorithms were “popular”. However, I note that the specification makes a number of statements with respect to the DLOA. In addition to the statement I have cited at [14] above, other relevant statements are as follows:
“The MLBOCS also contains a Master Optimal Control System (MOCS) which is a Control Software with a Deep Learning Optimization Algorithm (DLOA) integrated. The MOCS includes the followings:
(1)Input Data, which are in time histories, are collected from different types of Sensors at a series of time. Some of these data may be derived from the collected data basing on mechanical or physical relations. The Input Data … are used to train the DLOA …
…
It is optional to either include or exclude time interval in computation of the Deep Learning Optimization Algorithm (DLOA). Time interval, which equal to (t(I+1)-t(I)), is a period of time between two adjacent time steps which are t(I) and t(I+1) (where i is the index of the time step). If the DLOA optimises energy output, which reflects total energy generated over a period of time (time interval), then time intervals are required for the computation. Otherwise, if the DLOA optimises power output, time intervals are not required for the computation although the time interval does actually exist as the time history Input Data are values at the series of time. Thus, the DLOA can apply either fixed or variable time intervals. The Variable Time Interval is recommended to be used for maximizing energy output and controls. The Variable-Time Interval-Data is obtained from outputs of the DLOA for every time step: the DLOA determines (predicts) how long the predicted optimal time interval of the Predicted Optimal Time Step is with the condition that, at the end of the predicted optimal time interval, the energy output obtained over the predicted optimal time interval is maximized.
(2)Output data, which are obtained from the optimal solution of the Objective Function at a Predicted Optimal Time Step (POTS) t(I+1) …”
(3)…
(4)… The DLOA deploys the following DLOA-Additional Strategies:
I.Variable Time Interval Method: The optimization problem is solved (or the Objective Function is optimized) at a series of time, which are called time steps (..., t(I-2), t(I-1), t(I), t(I+1), …), in which ‘I’ is index, to obtained optimal solutions. The time steps where the Objective Function is optimal is called the optimal time steps (..., t(I-2), t(I-1), t(I), t(I+1), …). A period of time between two adjacent time steps is called a time interval … There are two options, which are related to time intervals, to solve the optimization problem. The first option is to treat the time interval (and time step) as Parameters: the time step is predefined and, as a result, the time interval is fixed … The second option, which is the Variable Time Interval Method, is to treat the time step and, as a result, the time interval to be variable. In this case, values of both the time step and the time interval are derived from the optimal solution of the optimization problem. As a result, values of the time intervals vary with respect to time … The time interval derived from the optimal time step, which is obtained using Variable Time Interval Method, is called the optimal time interval. So, the purpose of the optimization problem is to obtained [sic] the optimal solution which is occurred at the Predicted Optimal Time Step. Data derived from the optimal solution are predicted data which reflect the next state of the components or portions of the M-CVT. Control Parameters, which are used to progress (or control) the M-CVT to the next state, are derived from the predicted data. The predicted optimal time interval can be obtained as follow: Firstly, the time (t) is also treated as a variable of the Objective Function. As the Input Data reflect values at the series of time, the value of these series of time are also used in training the DLOA with respect to the variable time (t) of the Objective Function. The optimal solution of the Objective Function provides the value of the Predicted Optimal Time Step, from which the predicted optimal time interval is derived …
…
IV.Interactive Learning Control Method (ILCM):
As Data collected from the M-CVT-State-Sensors become more and more, the DLOA can learn newly collected Data and find optimal solutions interactively.
V.Untrained Initialized Method (UIM): The Master Optimal Control System of the MLBOCS for M-CVT can be started without prior learning. Firstly, the MLBOCS for M-CVT starts working with a simple control policy which does not need to include optimization. In this case, although the energy generated is not optimized, the control system can be configured to be able to work without optimization while data are being collected and the DLOA is being trained. Once the DLOA is fully trained, it is ready to work fully. In this case, Interactive Learning Control Method (ILCM) should be integrated.”[32]
“Why it is an optimization problem applied to maximize power output? Depending on devices connected to the output of the M-CVT, requirements for mechanical outputs may varies [sic]. For example, if the M-CVT is connected with a Generator, the mathematical model of the DLOA is to solve an optimization problem which maximizes either power or energy output by maximizing the Objective Function of the M-CVT using Deep Learning Optimizers such as Gradient Descent or Conjugate Gradient.”[33]
“How is the Deep Learning Optimization Algorithm (DLOA)? The DLOA is composed of:
(1) popular deep learning algorithms used to solve the optimization problem (the Mathematical Model of the M-CVT) to find out optimal solutions with respect to optimal time steps. The deep learning algorithms includes [sic] artificial neural network (ANN)s with an input layer, multi deep layers and an output layer. As deep learning algorithms are popular, they are not presented in this document. The deep learning algorithms are used to find (predicted) approximated functions of the power or energy outputs of the M-CVT. In other words, the deep learning algorithm is used to solve the optimization problem which is the Mathematical Model of the M-CVT.
(2) enhancements of the deep learning algorithm by applying the DLOA-Additional Strategies.”[34]
[32] Ibid at [0008]
[33] Ibid at [0012]
[34] Ibid at [0016]
I accept that there are known techniques in the art for developing algorithms. However, as I have already noted, the specification is not a puzzle for the PSA to solve. English is clearly not the applicant’s first language, and there may well be a language barrier here. However, while the applicant has supplied a lot of information about how the DLOA may operate, what is missing, in my opinion (and as was noted by the examiner), is how that algorithm is created. The applicant has effectively said “Here is a lot of data about the operation of a M-CVT, now use AI to create the mathematical model for the DLOA, and/or the DLOA itself.” While I accept that this is a somewhat facetious statement, it captures the fact that the mathematical model is missing. It also does not assist when the applicant equates the mathematical model and the optimisation problem, since an optimisation problem is traditionally defined by (an) objective function(s) and constraints.
The specification states that the objective functions are:
“(1) PowerOut(t(I)).
(2) EnergyOut(Ti(I)).
(3) TotalEnergyOut().
(4)DOutput() minus (PowerOut(t(I)) or EnergyOut(Ti(I)) or TotalEnergyOut()) where DOutput() is a required time history variable mechanical inputs (for best performance) of a device rotated by the M-CVT”[35],
but the constraints are not elucidated.
[35] Ibid at [0012]
As the examiner noted in the third report, the applicant had indicated that:
“The task of handling input data and other common tasks of using ANN, such as the tasks listed below, are common in solving any problems using ANN. In other words, any skilled person has to encounter the task listed below in solving any problem using ANN. Thus, the skilled person needs to understand such kind of common knowledge in order to implement my software.”[36]
[36] Applicant’s second response page 8
The “tasks listed below” were stated by the applicant to be:
“determining an appropriate deep learning algorithm and associated hyperparameters – especially given that most of the inputs will be sequential time series data, • determining whether to use labelled training data, • determining how much training data is required to achieve acceptable reliability and avoid overfitting as well as local optimums, • determining how to ‘preprocess’ and ‘clean’ training data to reduce noise, bias and imbalance, • determining how to evaluate and deploy the machine learning models once trained, particularly given the modelling of chaotic systems such as the water surface state, in which the inherent margin of error in the sensors providing the inputs would result in drastically different outputs regardless of the accuracy of the model itself, • determining how to scale the model with new data in the form of ILCM or additional sensor data.”[37]
[37] Ibid page 9
The applicant then stated:
“The wide range of general M-CVTs with general operational conditions makes the task of determining an appropriate deep learning algorithm and associated hyperparameters to be specified (specifically) is impossible. There is NO appropriate deep learning algorithm and associated hyperparameters for the wide range of configurations of the M-CVTs, each further with a wide range of operational conditions while my software is recommended for any M-CVT with any operational conditions.”[38]
[38] See ibid
Somewhat contrary to the examiner, I do not think that these two passages represent the applicant saying that the determination of an appropriate DLOA is, at the same time, (i) within the skill of the PSA and (ii) impossible to do. Rather, in these two passages, I think the applicant is trying to say that (i) the PSA would be able to solve whatever “tasks” were required to be done when practically implementing the invention for a particular M-CVT connecting two particular mechanical devices; and (ii) it was impossible for the applicant to provide more specific information, because the invention covered all possible practical applications, and the information will be different depending on the particular application. However, it is clear to me that the applicant has not provided any example of how this could be done for even one particular system, e.g., a wind turbine power generator, so it is not clear how one practical application needs to be adapted to a different practical application.
To my mind, the burden placed upon the PSA to determine an appropriate DLOA is so large as to be unreasonable. The detail provided in the specification is so high-level that I can discern no real guidance in the specification to direct the PSA on how to go about their task of moving from collected data from various sensors to a trained model via a DLOA. This feature lacks clear enough and complete enough disclosure.
Master Optimal Control System (MOCS)
In the third report, the examiner stated:
“The claims state that time history Output Data comprising a set of time history predicted Optimal Control Parameter (POCP)s is used to construct the control policy. However, your specification does not provide direction on how to construct a control policy from the POCPs and in the absence of this information, the person skilled in the art would have no clue on how to proceed with the construction of a control policy from the POCPs, hence necessitating further invention by the person skilled in the art. Your claimed invention also states that the control policy is used to control operations of controllable devices of the M-CVT. However, the specification does not provide any detail whatsoever on how the control policy is actually used to control the operations of the controllable devices of the M-CVT as claimed. In the absence of this information, the person skilled in the art cannot determine how to actually control the operations of the controllable device using the control policy, hence needing further invention by the person skilled in the art.”[39]
[39] Third report page 5
I agree with the examiner. The PSA has no guidance in the specification as to how to achieve this activity.
I note that the applicant stated that the software which is their invention:
“does not embrace everything. It excludes the tasks of data collection, data preparations, handling the set of data and so on. It does not include how a control software evaluates the set of output data, how the control software handles the set of output data, how the control software controls the components of the M-CVT.”[40]
In this passage the applicant appears to be arguing that there is no requirement to explain how the control policy is constructed or how the M-CVT is controlled because it does not form part of their invention and that, as such, how to do these things is not explained in the specification.
[40] Applicant’s second response page 1
However, I note that claim 1 requires that the MOCS comprise inter alia:
§“a set of time history Output Data; wherein:
· the Output Data comprising a set of time history Predicted Optimal Control Parameters (POCP);
· the POCP is used to construct the Control Policy of the MOCS;
· the Control Policy is used to control operations of controllable devices of the M-CVT;
…
§a Controlling Software; wherein:
…
· the Controlling Software instructs the controllable devices of the M-CVT to progress to new states specified by the set of Predicted Optimal Control Parameters (POCP)”. (my emphasis)
Clearly, the invention includes the constructing of a control policy and using that policy to control the M-CVT. That is, it appears to me that the invention is required to undertake tasks which the applicant says (i) it does not do and (ii) the specification does not explain how to do.
It clearly follows that the PSA is left in the position to attempt to achieve a result with no guidance as to how it can be achieved. As with the other features above, this places an unreasonable burden on the PSA. The inevitable conclusion is that this feature is not enabled.
Conclusion
The specification does not satisfy the requirements of s40(2)(a). There is no clear enough and complete enough disclosure of:
(a)Construction of the M-CVT, specifically the layout of the rotors with respect to gaps between components and the positioning of the stator(s) so that the EPIR can operate as a motor/generator;
(b)Training of the DLOA; and
(c)Creation of the MOCS, specifically the construction of the control policy from the POCPs, and the use of the control policy to control the M-CVT.
At this point the question which arises is whether there is any utility in continuing re-examination. The broad nature of the specification is such that I cannot see any allowable amendment which could result in a patentable invention. I also note that the applicant did not access any of the letters requesting submissions prior to the due date, and has not accessed them from then until the date of this decision. Such lack of action appears to indicate that the applicant has lost interest in prosecuting this application. This lack of interest and the fact that I cannot see any technical features that could address the issues I have found suggest that there is no benefit in continuing re-examination. Accordingly, I will refuse the application.
Incidental – best method and support
Given this finding, there is no need for me to address the examiner’s objection that the specification does not provide a best method, as required by s40(2)(aa). That being said, it seems to me that, given what is required by the claims, the examiner’s observations are prima facie correct.
In the same way, there is no need for me to address the examiner’s objection that the claimed invention is not supported, as required by s40(3). Nevertheless, noting that a failure under s40(2)(a) to enable the invention must mean that the claims extend to subject-matter which still not be at the disposal of the PSA, the examiner’s support observation is prima facie correct.
Greg Powell
Delegate of the Commissioner of Patents
Annex
1. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) comprising:
·a Magnetic Continuous Variable Transmission (M-CVT); wherein:
othe M-CVT comprising:
§a Mechanical Power Input Rotor (MPIR); wherein the MPIR receives mechanical power inputs;
§a Mechanical Power Output Rotor (MPOR); wherein the MPOR delivers mechanical power outputs;
§an Electrical Power Input Rotor (EPIR) comprising a motor-generator; wherein:
·the motor-generator is controllably operated by a source of variable electric power;
·the motor-generator is able to work relying on mechanical power inputs and mechanical power outputs for its dual functions of driving a device and generating electricity; wherein:
othe motor-generator works as an electric motor together with the MPIR to make the MPOR delivering controllable mechanical power outputs;
othe motor-generator works as a generator to generate electricity from excessed mechanical power inputs while maintaining the MPOR delivering controllable mechanical power outputs;
§wherein:
·among the MPIR, the MPOR and the EPIR, there are the followings rotors:
oa (first) Pole Pair Rotor, a (second) Pole Pair Rotor and a Pole Piece Rotor;
·wherein the pole pair rotors:
oeach Pole Pair Rotor has a number of pole pairs;
othe number of pole pairs of the (first) Pole Pair Rotor and the number of pole pairs of the (second) Pole Pair Rotor should be different;
othe pole pairs are putted together in a number of strips of pole pairs;
owherein each strip is either continuous or noncontinuous;
·and wherein the Pole Piece Rotor:
othe Pole Piece Rotor has a number of ferromagnetic pole pieces;
othe ferromagnetic pole pieces are putted together in a number of strips of pole pieces; wherein each strip is either continuous or noncontinuous;
othe number of ferromagnetic pole pieces equals to total pole pairs of both the (first) Pole Pair Rotor and the (second) Pole Pair Rotor;
ototal strips of pole pairs of the two pole pair rotors equal to the number of strips of pole pieces;
§and wherein:
·the MPOR has a MPOR-Magnetic Field created by its pole pairs or pole pieces;
·a variable rotating (electro)magnetic field created jointly by the MPIR and the EPIR interacts with the MPOR-Magnetic Field making the MPOR rotated;
§and wherein the EPIR of the M-CVT is controlled by a control system to work as either an electric motor or a generator depending on mechanical power inputs of the MPIR and mechanical power outputs of the MPOR, including rotational directions; wherein:
·the control system controls electric currents supplied by a source to rotate the EPIR for changing gear ratios in order to maintain required mechanical power outputs of the M-CVT;
·the control system controls electric currents generated by the EPIR while maintaining required mechanical power outputs of the M-CVT;
·a Machine Learning Based Optimal Control System (MLBOCS) comprising:
oa system of M-CVT-State-Sensors;
oa Master Optimal Control System (MOCS);
owherein the system of M-CVT-State-Sensors comprising:
§a system of MPIR-State-Sensors further comprising:
·a system of MPIR-Torque-Sensors used to monitor torques of the MPIR;
·a system of MPIR-Speed-Sensors used to monitor rotational speeds of the MPIR;
·an optional system of MPIR-Accelerators used to monitor rotational accelerations of the MPIR;
·a system of MPIR-Pulse-State-Sensors used to monitor MPIR-Pulse-States;
§a system of MPOR-State-Sensors further comprising:
·a system of MPOR-Torque-Sensors used to monitor torques of the MPOR;
·a system of MPOR-Speed-Sensors used to monitor rotational speeds of the MPOR;
·an optional system of MPOR-Accelerators used to monitor rotational accelerations of the MPOR;
§a system of EPIR-State-Sensors further comprising:
·a system of EPIR-Torque-Sensors used to monitor torques of the EPIR;
·a system of EPIR-Speed-Sensors used to monitor rotational speeds of the EPIR;
·an optional system of EPIR-Accelerators used to monitor rotational accelerations of the EPIR;
·a system of EPIR-Power Consumed-Sensors used to monitor time history variations of a number of electric currents supplied to a number of coils of the EPIR working as an electric motor;
·a system of EPIR-Power Generated-Sensors used to monitor electrical power or energy generated by the EPIR working as a generator;
§a system of M-CVT-Mechanical Power Input-Sensors used to monitor mechanical power input of the M-CVT;
§a system of M-CVT-Mechanical Power Output-Sensors used to monitor mechanical power output of the M-CVT;
§an optional system of M-CVT-Clutch-State-Sensors used to monitor M-CVTClutch-States;
§An optional system of M-CVT-RS-State-Sensors comprising:
·a system of M-CVT-MPRS-State-Sensors used to monitor M-CVT-MPRSStates;
·a system of M-CVT-EPRS-State-Sensors used to monitor system of M-CVTEPRS-States;
§a system of M-CVT-Power Consumed-Control System-Sensors used to monitor electrical energy consumed for controlling the M-CVT; wherein the M-CVTPowerConsumed-Control System-Sensors comprising a system of EPIR-PowerConsumed-Sensors;
§a system of M-CVT-Electrical Power Output-Sensors used to collect time history data of power output of the M-CVT;
§an optional system of Temperature-State-Sensors;
oand wherein the Master Optimal Control System (MOCS) comprising:
§a set of time history Input Data, known as the M-CVT-State-Data;
§a set of time history Output Data; wherein:
·the Output Data comprising a set of time history Predicted Optimal Control Parameters (POCP);
·the POCP is used to construct the Control Policy of the MOCS;
·the Control Policy is used to control operations of controllable devices of the M-CVT;
§a Controlling Software; wherein:
·the Controlling Software reads the set of Input Data;
·the Controlling Software generates the set of Output Data;
·the Controlling Software instructs the controllable devices of the M-CVT to progress to new states specified by the set of Predicted Optimal Control Parameters (POCP);
·the Controlling Software comprising the Deep Learning Optimization Algorithm (DLOA) using the set of Input Data and generating the set of Output Data;
§a Deep Learning Optimization Algorithm (DLOA); wherein the DLOA comprising an algorithm of Deep Learning using Artificial Neural Networks to solve an Optimization Problem; wherein:
·the Deep Learning Optimizers applied can be any kind of popular Optimizers such as Gradient Descent or Conjugate Gradient;
·the Optimization Problem is a Mathematical Model of the Multiphysics System of the M-CVT;
·the Multiphysics System incorporates complex interactions between a number of multi physical portions of the M-CVT such as temperatures, magnetic fields, Regulating Systems and motions of mechanical devices;
·the Objective Function of the Optimization Problem of the M-CVT is optionally chosen from the following:
opower output of the M-CVT expressed as the function Power-Out(t(I+1)); wherein t(I) is the time step at the time indexed (I) and t(I+1) is the time step at the time indexed (I+1);
oenergy output of the M-CVT expressed as the function Energy-Out(Ti(I)); wherein Ti(I) is the time interval between the time step t(I) and the time step t(I+1);
ototal energy output of the M-CVT over a period of time covering multiple time intervals, expressed as the function Total-Energy-Out();
oDesired Function() minus Power-Out(t(I+1)) or Energy-Out(Ti(I)) or Total-Energy-Out(); wherein the Desired Function() expresses time history mechanical power inputs required by a device rotated by the MPOR;
·the solutions of the Optimization Problem are used to construct the set of Predicted Optimal Control Parameters (POCP);
§and wherein the Deep Learning Optimization Algorithm (DLOA) further optionally comprising one or more of the following strategies:
·an optional Variable Time Interval Method applied for optimizing the Objective Function with variable time intervals;
·an optional Preselected Time Interval Method applied for optimizing the Objective Function with preselected time intervals;
·an optional Interactive Learning Control Method (ILCM) applied for learning and optimizing phases of the Deep Learning Optimization Algorithm (DLOA);
·an optional Untrained Initialized Method (UIM) applied for the initialization of the Master Optimal Control System (MOCS) with an untrained Deep Learning Optimization Algorithm (DLOA);
§and wherein the set of time history Input Data comprising:
·a set of time history MPIR-State-Data further comprising:
oa set of time history MPIR-Torque-Data obtained from the system of MPIR-Torque-Sensors;
oa set of time history MPIR-Speed-Data obtained from the system of MPIR-Speed-Sensors;
oa set of time history MPIR-Acceleration-Data obtained from the system of MPIR-Accelerators;
oa set of time history MPIR-Pulse-State-Data obtained from the system of MPIR-Pulse-State-Sensors;
·a set of time history MPOR-State-Data further comprising:
oa set of time history MPOR-Torque-Data obtained from the system of MPOR-Torque-Sensors;
oa set of time history MPOR-Speed-Data obtained from the system of MPOR-Speed-Sensors;
oa set of time history MPOR-Acceleration-Data obtained from the system of MPOR-Accelerators;
·a set of time history EPIR-State-Data further comprising:
oa set of time history EPIR-Torque-Data obtained from the system of EPIR-Torque-Sensors;
oa set of time history EPIR-Speed-Data obtained from the system of EPIR-Speed-Sensors;
oa set of time history EPIR-Acceleration-Data obtained from the system of EPIR-Accelerators;
oa set of time history EPIR-Power Consumed-Data obtained from the system of EPIR-Power Consumed-Sensors;
oa set of time history EPIR-Power Generated-Data obtained from the system of EPIR-Power Generated-Sensors;
·a set of time history M-CVT-Mechanical Power Input-Data obtained from the system of M-CVT-Mechanical Power Input-Sensors;
·a set of time history M-CVT-Mechanical Power Output-Data obtained from the system of M-CVT-Mechanical Power Output-Sensors;
·an optional set of time history M-CVT-Clutch-State-Data obtained from the optional system of M-CVT-Clutch-State-Sensors;
·an optional set of time history M-CVT-RS-State-Data further comprising:
oan optional set of time history M-CVT-MPRS-State-Data further comprising a set of time history M-CVT-MPRS-Control Parameter-Data;
oan optional set of time history M-CVT-EPRS-State-Data further comprising a set of time history M-CVT-EPRS-Control Parameter-Data;
·a set of time history M-CVT-Power Consumed-Control System-Data further comprising the set of EPIR-Power Consumed-Data;
·a set of time history M-CVT-Electrical Power Output-Data;
·an optional set of time history Temperature-State-Data;
·an optional set of time history Variable-Time Interval-Data;
§and wherein the set of time history Output Data comprising:
·a set of time history Predicted Optimal Time Step-Data;
·a set of predicted M-CVT-Mechanical Power Input-Data;
·a set of predicted M-CVT-Mechanical Power Output-Data;
·an optional set of predicted M-CVT-Clutch-State-Data;
·a set of predicted MPIR-Pulse-State-Data;
·a set of predicted M-CVT-Electrical Power Output-Data;
·a set of predicted Optimal Control Parameter (POCP)s applied at the Predicted Optimal Time Step; wherein the POCPs comprising:
oa set of predicted EPIR-Power Consumed-Data and a set of predicted EPIR-Power Generated-Data containing parameters for controlling the EPIR at the Predicted Optimal Time Step; wherein:
§the EPIR is switched between states of working as an electric motor or a generator;
§as an electric motor, using the predicted EPIR-Power Consumed-Data, the EPIR is variably rotated by handling the controllable electric currents consumed by the EPIR in order to handle required rotations of the MPOR;
§as a generator, while mechanical input power (MIP) being excessed, the EPIR is controlled for both handling required rotations of the MPOR and generating electricity using the excessed MIP;
oan optional set of time history M-CVT-Clutch-Control Parameter-Data and an optional set of predicted MPIR-Pulse-State-Data containing parameters for controlling the optional M-CVT-Clutch at the Predicted Optimal Time Step;
oa set of predicted M-CVT-MPRS-Control Parameter-Data and a set of predicted M-CVT-EPRS-Control Parameter-Data containing parameters for controlling the M-CVT-MPRSs and the M-CVT-EPRSs at the Predicted Optimal Time Step.
2. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) according to claim 1;
·wherein its Magnetic Continuous Variable Transmission (M-CVT) is further simplified to have only two physical rotors; wherein:
·the pole piece rotor is excluded while the (first) pole pair rotor and the (second) pole pair rotor are remained to be included; wherein:
omagnetic interactions between the pole piece rotor, the (first) pole pair rotor and the (second) pole pair rotor are replaced with controllable magnetic interactions between the (first) pole pair rotor and the (second) pole pair rotor; wherein the controllable magnetic interactions are created by a (controllable) rotating electromagnetic field established between the (first) pole pair rotor and the (second) pole pair rotor; wherein the rotating electromagnetic field has a rotational vector, comprising directions and amplitudes of rotational speeds, to be used as the rotational vector of the EPIR for the control policy;
oa Pole Pair Rotor among the (first) pole pair rotor and the (second) pole pair rotor is a MPIR; wherein the MPIR receives mechanical power inputs;
othe other Pole Pair Rotor is an EPIR-MPOR; wherein:
§the EPIR-MPOR is combined from an EPIR and a MPOR; wherein:
§the EPIR-MPOR has all features of the EPIR and the MPOR;
§the EPIR-MPOR outputs mechanical power in conjunctions with its controllable rotations assisted by the MPIR and by the motor-generator comprised in the EPIR;
§the motor-generator of the EPIR-MPOR converts excessed mechanical power inputs to electricity while maintaining mechanical power outputs of the EPIR-MPOR;
·all remaining features of the MLBOCS for M-CVT remain unchanged.
3. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) according to claim 1; wherein its Magnetic Continuous Variable Transmission (M-CVT) is further enhanced to have two MPIRs and an EPIR-MPOR; wherein:
·the M-CVT comprises a pair of MPIRs and an EPIR-MPOR; wherein:
oeach MPIR receives mechanical power inputs separately;
othe EPIR-MPOR is combined from an EPIR and a MPOR; wherein:
§the EPIR-MPOR has all features of the EPIR and the MPOR;
§the EPIR-MPOR outputs mechanical power in conjunctions with its controllable rotations assisted by the MPIR and by the motor-generator comprised in the EPIR;
§the motor-generator of the EPIR-MPOR converts excessed mechanical power inputs to electricity while maintaining mechanical power outputs of the EPIRMPOR;
oand wherein the M-CVT is used to combine two sources of mechanical power inputs via the pair of MPIRs to create controllable mechanical power outputs via the EPIR-MPOR;
·all remaining features of the MLBOCS for M-CVT remain unchanged.
4. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) according to claim 1; wherein its Magnetic Continuous Variable Transmission (M-CVT) is further enhanced to have a MPIR and two EPIR-MPORs; wherein:
·the M-CVT comprises a MPIR and a pair of EPIR-MPORs; wherein:
othe MPIR receives mechanical power inputs;
oeach EPIR-MPOR is combined from an EPIR and a MPOR; wherein:
§the EPIR-MPOR has all features of the EPIR and the MPOR;
§the EPIR-MPOR outputs mechanical power in conjunctions with its controllable rotations assisted by the MPIR and by the motor-generator comprised in the EPIR;
§the motor-generator of the EPIR-MPOR converts excessed mechanical power inputs to electricity while maintaining mechanical power outputs of the EPIRMPOR;
othe M-CVT is used to distribute mechanical power from a source of inputs via the MPIR to obtain two controllable sources of mechanical power outputs separately via the pair of EPIR-MPORs using the two separate motor-generators comprised in the pair of EPIR-MPORs;
·all remaining features of the MLBOCS for M-CVT remain unchanged.
5. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) according to any one of claims from 1 to 4 further comprising a Multiple Contained M-CVT (MCM-CVT); wherein:
·the MCM-CVT comprises a number of M-CVTs; wherein:
orotors of each M-CVT are coaxial and the M-CVTs are coaxial;
othe most inner M-CVT is contained by an outer M-CVT;
othe outer M-CVT becomes the next inner M-CVT;
othe next inner M-CVT is contained by the next outer M-CVT;
othe process is continued until reaching the most next outer M-CVT;
othe most next outer M-CVT is the largest one;
ogear ratios of the MCM-CVT equal to multiplications of all gear ratios of the MCVTs;
oeach M-CVT of each pair of adjacent M-CVTs shares a common rotor; wherein:
othe common rotor is the MPOR of a M-CVT of the pair;
othe common rotor is the MPIR of the other M-CVT of the pair;
oeach M-CVT has systems of sensors and sets of data like the M-CVT comprised in the claims from 1 to 4;
·all remaining features of the MLBOCS for M-CVT remain unchanged.
6. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) according to any one of claims from 1 to 5 further comprising a Multiple Bevelled M-CVT (MBM-CVT); wherein:
·the MBM-CVT comprises a number of M-CVTs; wherein:
oat least one of the M-CVTs has noncoaxial rotors;
oeach rotor of each M-CVT have magnetic contacts with at least another rotor of the same M-CVT;
oeach M-CVT has magnetic contacts with at least one of the remaining M-CVTs;
otorque capacity of the MBM-CVT might be less in comparison with coaxial M-CVTs;
oat least a MPIR rotor of a M-CVT is used for mechanical power input and at least a MPOR of another M-CVT is used for mechanical power output of the MBM-CVT;
oeach M-CVT has systems of sensors and sets of data like the M-CVT comprised in the claims from 1 to 5;
·all remaining features of the MLBOCS for M-CVT remain unchanged.
7. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) according to any one of claims from 1 to 6 with further enhancements for multiple mechanical power inputs and outputs; wherein:
·A number of MPIRs of a number of M-CVTs are used for mechanical power inputs from a number of different external sources;
·A number of MPORs of a number of M-CVTs are used for mechanical power outputs to a number of different external devices;
·A number of M-CVTs do not have EPIRs if these have all three rotors used for either mechanical power inputs or outputs;
·each M-CVT has systems of sensors and sets of data like the M-CVT comprised in the claims from 1 to 6;
·all remaining features of the MLBOCS for M-CVT remain unchanged.
8. A Machine Learning Based Optimal Control System for Magnetic Continuous Variable Transmission (MLBOCS for M-CVT) according to any one of claims from 1 to 7 applied for transmitting energy harvested from waves or winds.
0
6
0