Method for predicting performance of driving motor for vehicle and optimizing design parameters by using AI

Abstract

An artificial intelligence (AI)-based motor development system for optimizing a drive motor design for a vehicle according to the present disclosure includes: a predicted AI model generation section configured to generate a predicted AI model by fitting a polynomial curve to a noise peak prediction based on a modified motor design variable obtained from a motor computer-aided design (CAD) map; and predicting the performance improvement of the motor, including noise / vibration / acoustic vibration roughness (NVH). The system further includes a design parameter optimization AI model generation section configured to optimize a motor design parameter size from a design parameter optimization proposal AI model obtained by any of reinforcement learning, Q-learning, and particle swarm optimization (PSO) using the predictive AI model as a feature extractor for a target motor performance improvement.

Description

Technical Field

[0001] The present disclosure relates to the design of a vehicle drive motor. More specifically, the present disclosure relates to a method for predicting the performance of a vehicle drive motor and optimizing the motor design using artificial intelligence (AI). Background Art

[0002] Typically, motor computer-aided design (CAD) is used to develop a motor for a vehicle (ie, a drive motor).

[0003] For example, a method for developing an electric motor using CAD includes determining design parameters of the electric motor (e.g., factors of the stator-rotor assembly), which may be configurable in the CAD drawing of the electric motor. The method also includes case-by-case analysis of all conditions for correlation between the determined design parameters. Furthermore, the method may include predicting, based on the analysis results, improvements to the motor's performance, including noise / vibration / harshness (NVH), through simulation. The motor development is then completed by applying these results to the dimensions of the motor's design parameters.

[0004] Therefore, the method used for motor CAD development requires the generation and analysis of simulation analysis models of CAD motor drawings. To do this, it is important to determine the motor design parameters, which are both analysis parameters and design parameters used to predict the target performance of the motor.

[0005] However, in the motor CAD development method, specific units that are determined to be changeable based on the motor designer's experience are determined as motor design parameters. During the generation of the analysis model, all parameter combinations that reflect the correlation between the motor design parameters are often not included.

[0006] This limitation arises because electric motor CAD development methods, due to time and cost constraints, cannot generate analytical models that reflect all combinations of motor design parameters. Due to these limitations, the impact of changing certain design parameters is inevitably determined based on the motor designer's experience. Summary of the Invention

[0007] This disclosure provides an AI model for optimizing motor design parameters, which can provide motor design parameters for achieving target performance. This is achieved by: generating a motor performance prediction AI model through AI learning based on motor design parameters and motor performance data from motor CAD drawings; extracting features from the motor performance prediction AI model; and applying reinforcement learning to improve the target motor performance.

[0008] A method for using artificial intelligence (AI) to predict the performance of a vehicle drive motor and optimize its design parameters includes obtaining data on motor design parameters and motor performance of the drive motor installed in the vehicle. The method also includes generating a motor performance prediction AI model using an automated machine learning (AutoML) unit based on the obtained motor design parameter and motor performance data. The method also includes applying an evolutionary algorithm to the motor performance prediction AI model and generating a motor design parameter optimization AI model using reinforcement learning.

[0009] In addition, the target data of the motor performance prediction AI model can be motor performance, and is obtained by analyzing motor design parameters.

[0010] Additionally, combinations of motor design parameters may be received through design of experiments (DOE).

[0011] In addition, the input data of the motor performance prediction AI model can be motor design parameters. The motor design parameters can include any one or more of slots, teeth, stator teeth, magnetic bridges, magnets, and center columns.

[0012] In addition, the output data of the motor performance prediction AI model can be motor performance, and the motor performance can include any one or more of noise / vibration / harshness (NVH), torque, torque ripple and flux.

[0013] In addition, the motor design parameter optimization AI model can use the motor performance (which is the output data of the motor performance prediction AI model) as input data, and the motor design parameters (which are the input data of the motor performance prediction AI model) as output data.

[0014] In addition, the motor design parameter optimization AI model can adopt any one or more of reinforcement learning, Q-learning, and particle swarm optimization (PSO).

[0015] In addition, the motor design parameter optimization AI model can calculate multiple combinations of optimized motor design parameters and prioritize the multiple combinations of optimized motor design parameters under the constraints of motor performance.

[0016] In addition, in the motor design parameter optimization AI model, when the goal of improving motor performance is to reduce NVH, the minimum torque change can be set as the limiting condition for power performance.

[0017] In addition, during the optimization process of the motor design parameter optimization AI model, the noise level can be predicted from the machine learning (ML) model for the noise level.

[0018] A method for optimizing vehicle drive motor designs using an AI-based motor development system according to the present disclosure includes generating an optimized set of motor design parameter dimensions based on a target motor performance improvement. This is achieved by using an AI performance prediction model that responds to changes in design parameters of a vehicle motor (i.e., an electric vehicle motor) and an AI design parameter optimization proposal model that uses the AI ​​performance prediction model as a feature extractor.

[0019] By using experimental radiation noise data and motor CAD drawing data, NVH performance, magnetic force, torque ripple and motor torque performance data can be marked according to the design parameter changes of analysis and simulation to form a performance prediction AI model for performance prediction.

[0020] By using a performance prediction model based on design parameter variations, many design parameter combinations with high impact probabilities that can achieve the desired target performance can be proposed.

[0021] The Shapley Additive Explanation Pair (SHAP) function, which ranks features in descending order of importance, can be applied to ensure that the motor developer's domain knowledge is consistent with the prediction model. This allows the performance prediction AI model to serve as a feature extractor for the design parameter model and provide multiple combinations of optimized design parameters for achieving the target performance.

[0022] The experimental and analytical results of the input data are labeled in a two-step process. Specifically, in the first step, the performance prediction model is predicted for various combinations of design parameter changes. In the second step, the performance prediction AI model from the first step is used as a feature extractor. Many optimal design parameter combinations that achieve the target performance can be proposed using reinforcement learning.

[0023] The performance of all motor design parameter combinations can be verified and practical design parameters can be proposed. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above and other objects, features and other advantages of the present disclosure will be more clearly understood from the following detailed description taken in conjunction with the accompanying drawings.

[0025] Figure 1 FIG. 1 is a configuration diagram of a method for predicting performance and optimizing design of a vehicle drive motor according to an embodiment of the present disclosure.

[0026] Figure 2 The difference in prediction result data of the performance prediction AI model for each vehicle segment according to an embodiment of the present disclosure is shown.

[0027] Figure 3 is a step-by-step conceptual diagram of a method for predicting performance and optimizing design of a vehicle drive motor according to an embodiment of the present disclosure.

[0028] Figure 4 The present invention is a flowchart of a method for deriving a motor design optimization model after generating an AI model for predicting the performance of a vehicle drive motor according to an embodiment of the present disclosure.

[0029] Figure 5 2 is a conceptual diagram of a process for generating an AI model for predicting the performance of a vehicle drive motor according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0030] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. These embodiments are examples and can be implemented in various forms by a person skilled in the art to which the present disclosure pertains, and are therefore not limited to the embodiments disclosed herein.

[0031] When a controller, part, device, element, assembly, unit, module, etc. of the present disclosure is described as having a purpose or performing an operation, function, etc., the controller, part, device, element, assembly, unit, or module should be considered herein as being "configured to" satisfy the purpose or perform the operation or function. Each controller, part, device, element, assembly, unit, module, etc. may be separately embodied or included in a processor and memory (e.g., a non-transitory computer-readable medium) as part of the device.

[0032] Reference Figure 1 The data set used in this disclosure includes motor design parameters 10a and motor performance 10b. The motor design parameters 10a can be provided from a motor computer-aided design (CAD) drawing, and the motor performance can be provided from analysis (simulation) results calculated based on the motor design parameters.

[0033] Motor design parameters 10a are the primary design parameters for the stator / rotor assembly, which is the motor drive unit for electric vehicle motors. In other words, in this disclosure, motor design parameters 10a include the dimensions of the stator and rotor, serving as data for learning an AI model for predicting vehicle drive motor performance. Motor performance 10b includes analysis result values ​​(e.g., noise / vibration / harshness (NVH), torque, torque ripple, radial flux, and tangential flux) as target data.

[0034] More specifically, the motor design parameters 10a adopt slot length, slot radius, tooth tip thickness, tooth tip angle, tooth width, slot width, 1 / 2 layer magnetic bridge thickness, 1 / 2 layer magnet thickness, center column thickness between two magnets, center column thickness between two 1 / 2 layer magnets, magnet angle, 1 / 2 layer magnet angle, 1 / 2 layer magnet length, stator tooth width, etc.

[0035] Motor design parameters can be obtained as design parameter combinations using design of experiments (DOE) and can also be applied by using competitor motor specification data or adding new data.

[0036] For example, as motor design parameters, data for 352 DOE points on 11 motor design parameters can be obtained, including the magnetic bridge thickness around the rotor outer diameter (OD), the center column thickness between the two magnets, the magnet thickness / width / angle, the stator tooth tip width, the stator tooth thickness, the stator tooth tip angle, and the slot length / width / radius.

[0037] The motor performance 10b is an analysis result based on the dimensions assigned to the motor design parameters 10a. The target data represents the motor performance produced by the interaction between the rotor and the stator according to the motor design parameters 10a.

[0038] In this case, the motor performance 10b includes NVH, torque ripple, torque ripple harmonic order, radiated power order, stator maximum density, noise level, noise area, peak number, peak torque, shaft speed, etc.

[0039] The AI ​​model learning unit 20 generates a vehicle drive motor performance prediction AI model using the motor design parameters 10 a as input data and the motor performance 10 b as target data.

[0040] The vehicle drive motor performance prediction AI model 30 trained by the AI ​​model learning unit 20 predicts performance result data 31a based on changes in motor design parameters 10aa. When new dimensions are applied to the motor design, replacing the original motor design parameters 10a, the model generates updated performance predictions.

[0041] For example, the motor design parameter 10aa changed to a new size is a design parameter that represents the main performance of the motor and can be applied by changing the slot length / width, tooth tip thickness / angle, 1st / 2nd layer magnetic bridge thickness, 1st / 2nd layer magnet thickness / length, magnet angle, center column thickness between two magnets, stator tooth tip width, etc.

[0042] The vehicle drive motor performance prediction AI model 30 generated by the AI ​​model learning unit 20 is a motor performance prediction AI model in which the correlation, which represents performance changes due to changes in motor design parameters, is set to be greater than 0.95.

[0043] Furthermore, the design parameter optimization AI model 40 can be generated from the vehicle drive motor performance prediction AI model 30 generated by the AI ​​model learning unit 20. Hereinafter, the vehicle drive motor performance prediction AI model may be referred to as a motor performance prediction AI model.

[0044] The motor performance prediction AI model is generated by setting n points representing the operating area of ​​the electric vehicle motor; obtaining main performance result data from the n points; and then using them as input data and target data to perform learning to have a correlation of motor performance to motor design parameters with an accuracy of more than 95%.

[0045] The following process is performed in the motor performance prediction AI model: the current-driven power supply unit is operated according to the input and target data obtained through testing or calculation, and the performance of the electromagnetic system based on electromagnetic force, the performance of the mechanical system based on speed, the performance of the acoustic environment based on acoustic noise, etc. are calculated in sequence.

[0046] Specifically, the performance of the electromagnetic system is calculated by combining any one or more of air gap force calculation, Maxwell stress tensor method, and Fourier analysis as an FE-based model. The performance of the mechanical system is calculated by combining any one or more of an analytical-based model, free motion response, and force response. The performance of the acoustic noise is calculated by combining any one or more of an analytical-based model and sound power level.

[0047] The motor performance prediction AI model 30 may have improved reliability compared to conventional models that predict performance based on the domain knowledge results of motor developers.

[0048] The AI ​​model generated by the AI ​​model learning unit 20 may be referred to as the motor performance prediction AI model 30 for improving reliability. The place where the motor performance prediction AI model 30 is provided or stored may be referred to as the motor performance prediction AI model generation unit.

[0049] The AI ​​model generated by the motor performance prediction AI model 30 can be stored and referred to as a design parameter optimization AI model 40 for optimizing design parameters. The location where the design parameter optimization AI model 40 is provided or stored can be referred to as a design parameter optimization AI model generation unit.

[0050] The design parameter optimization AI model 40 provides a design parameter optimization AI model that can predict optimized motor design parameters and dimensions 41a for target performance input 10bb of motor performance improvement including NVH reduction based on the motor performance prediction AI model 30 for reliability improvement.

[0051] In other words, in the design parameter optimization AI model generation unit, the motor target performance including NVH is the input data, the motor design parameters are the output data, and as an AI recommendation algorithm, reinforcement learning, Q-learning and particle swarm optimization (PSO) are combined or selectively applied so that the input / output is set opposite to the motor performance prediction AI model.

[0052] In other words, through the AI ​​recommendation algorithm, a design parameter optimization proposal AI model with changes in design parameters and dimensional outputs that can achieve target performance including NVH motor performance improvements can be provided.

[0053] Reference Figure 2 , motor design parameters and performance data of the motor 200 are acquired from the vehicle 300 for each segment classified based on the overall length of the vehicle (the length from the front bumper to the rear bumper) and the price.

[0054] For example, the A-type motor 200a is an example of extracting motor design parameters and NVH performance characteristics suitable for segment A vehicles. The B-type motor 200b is an example of extracting motor design parameters and NVH performance characteristics suitable for segment B vehicles. The C-type motor 200c is an example of extracting motor design parameters and NVH performance characteristics suitable for segment C vehicles. The motor design parameters and motor performance including NVH of the A-, B-, and C-type motors can be used as input data and target data, respectively, as data acquired for modeling, or can be used as input data for established models.

[0055] Figure 1 and Figure 3 It is a detailed configuration of the AI ​​model learning unit 20 generating a performance prediction AI model and a design parameter optimization AI model. Figure 1 and Figure 3 This diagram is a conceptual diagram of a step-by-step method for predicting the performance of vehicle drive motors and optimizing their design.

[0056] Specifically, the motor performance prediction AI model 30 generated by the AI ​​model learning unit 20 can be provided by the data acquisition unit 32 , the data version management unit 33 , the model generation unit 34 , the model test and evaluation unit 35 , and the performance prediction model completion unit 36 ​​.

[0057] For example, the data acquisition unit 32 of the learning model uses motor design parameters as input data. The motor design parameters used as input data include motor design parameters and their sizes.

[0058] Experimental signature values ​​or simulation (analysis) signature values ​​regarding motor design parameters are acquired as target data, and learning model data from which features are extracted is selected using a data selection process by performing off-line data analysis on the signature values.

[0059] The learning model data acquisition section 32 includes a data collection section 32a, which contains a large number of experimental and analytical value signatures representing the performance of each motor design parameter. The learning model data acquisition section 32 also includes a data exploration section 32b, which queries and analyzes the acquired data, including the large number of experimental and analytical value signatures, and derives useful information; a data cleaning section 32c, which identifies, modifies, and filters out errors, missing, and inaccurate values ​​in the dataset; and a feature engineering section 32d, which optimizes the features of the external data between the experimental value signatures and simulation (analysis). Furthermore, the feature engineering section 32d is configured to extract features as useful information to ultimately obtain the learning model data to be used.

[0060] The data version management section 33 manages data versions of previously acquired data and newly acquired data sets with respect to motor design parameters and performance that have been used for learning, and performs systematic model management through index configuration of data amplification.

[0061] The data version management unit 33 provides training data and verification data from the acquired data to the model generation unit 34 , and provides test data for model evaluation in the model test evaluation unit 35 .

[0062] The model engineering unit 34a of the model generation unit 34 generates a model using training data, and establishes a machine learning model by automating the process of developing the machine learning model through the automated machine learning (AutoML) unit 34c.

[0063] The established machine learning model is verified by the model evaluation section 34 b using the verification data provided from the data version management section 33 .

[0064] In addition to the model engineering section 34a, the model evaluation section 34b, and the AutoML section 34c, the model generation section 34 is also executed based on the ML lifecycle platform 34d. The ML lifecycle platform 34d is a holistic solution that supports the development, distribution, monitoring, and maintenance of machine learning models.

[0065] The model generation unit 34 uses the motor design parameters as input data of the motor performance prediction AI model and the motor performance as output data to establish a model combined with a deep neural network structure to improve the accuracy of the model output data relative to the actual performance.

[0066] Specifically, the model engineering unit 34a and the model evaluation unit 34b increase the correlation between the performance of the training model and the verification model through mutual data exchange and verification. In addition, the AutoML unit 34c increases the correlation between the performance of the performance prediction model through data exchange and verification between the model engineering unit 34a and the model evaluation unit 34b.

[0067] The model testing and evaluation unit 35 evaluates the performance prediction model of the AutoML unit 34c as test data by performing a sensitivity analysis of performance changes based on changes in model design parameters (i.e., size) and visualizing the results. Furthermore, the model testing and evaluation unit 35 compares the test data with the developer's domain knowledge. As a result, the model testing and evaluation unit 35 improves reliability compared to the performance prediction AI model.

[0068] To this end, the model testing section 35 a of the model test evaluation section 35 applies test data to the performance prediction model and tests the accuracy of the model.

[0069] The Shapley Additive Explanation (SHAP) 35b of the Model Testing and Evaluation 35 is a statistical technique and framework for interpreting and explaining the predictions of machine learning models. SHAP 35b is used to assess the contribution of certain features or elements to the predicted value to help analyze the model's predictions.

[0070] The performance prediction model completion section 36 completes the performance prediction model 36a and receives the interpretable interface from the SHAP section 35b in the interpretable interface section 36b.

[0071] The design parameter optimization AI model generator 40 extracts features of the design parameter optimization AI model 40 from the motor performance prediction AI model 30 completed by the final NVH prediction model 36a. The final design parameter optimization AI model 40 uses the motor's target performance as input data and outputs the motor design parameters of the target performance.

[0072] Reinforcement learning technology is applied to the design parameter optimization AI model, and optimization is used to extract features from the motor performance prediction AI model. The design parameter optimization AI model generation unit 40 is divided into an evolutionary algorithm unit 42a for design optimizer development and a deep reinforcement learning unit 42b for design engine development. The performance prediction model is ultimately used as a feature extractor in the optimization method evaluation unit 43.

[0073] Figure 4 In this example, NVH is used as a motor performance indicator. Data acquisition unit 32 acquires a training dataset containing peak values ​​and peak positions of overall noise levels as motor performance indicators for motor design parameters using a motor design DOE. This dataset then generates an overall noise level model as a motor performance prediction AI model. The overall noise level model outputs a noise peak prediction using a curve fitted with four polynomials.

[0074] Figure 5 The process of deriving a motor design parameter optimization AI model for NVH as motor performance is shown. In the motor design parameter optimization AI model, when the goal of motor performance improvement is to reduce NVH, minimizing torque variation is set as the power performance constraint.

[0075] Therefore, when NVH performance improvement is input as a target, with minimal torque change when improving NVH performance, the optimal motor design parameters for achieving the target motor performance improvement are output. To derive an optimization model capable of achieving this target, one or more of Q-learning and particle swarm optimization (PSO) are applied, and the NVH design ML model 44 is optimized to achieve motor performance with a correlation of at least 95% with the motor design parameters. When the motor performance improvement target, including NVH, is achieved, the NVH design ML model is confirmed as a design parameter optimization proposal AI model, and the motor design parameters are output.

[0076] The motor design parameter optimization AI model calculates n design parameter combinations 46a. In addition, under the constraints of motor performance, the priorities of the n motor design parameter combinations are set.

[0077] In other words, the optimization model recommends n combinations of motor design parameters and sizes that significantly improve NVH performance. The design parameters that minimize torque variation, which is a power performance, are prioritized in the output data.

[0078] With respect to the noise levels determined from the optimized NVH design ML model 44 , the noise levels of the n combinations 46 b may be predicted from the noise level ML model 45 .

[0079] In other words, by repeatedly executing the NVH design and noise level ML models when deriving the optimized NVH design ML model 44 and noise level ML model 45 from the optimized design parameter AI model generation unit 40, optimization of the NVH design ML model and noise level is performed to achieve the target NVH performance improvement.

Claims

1. A method for predicting the performance of a vehicle drive motor and optimizing design parameters using artificial intelligence (AI), the method comprising the following steps: Acquiring data on motor design parameters and motor performance based on a drive motor installed on a vehicle; Based on the acquired motor design parameters and motor performance data, an automated machine learning (AutoML) unit generates a motor performance prediction AI model. Applying an evolutionary algorithm to the motor performance prediction AI model; and Through reinforcement learning, an AI model for optimizing motor design parameters is generated.

2. The method according to claim 1, wherein The target data of the motor performance prediction AI model is motor performance, and The target data is obtained by analyzing motor design parameters.

3. The method according to claim 1, wherein A combination of motor design parameters is received through a design of experiments (DOE).

4. The method according to claim 1, wherein The input data of the motor performance prediction AI model are motor design parameters, and The motor design parameters include any one or more of slots, teeth, stator teeth, magnetic bridges, magnets and center columns.

5. The method according to claim 1, wherein The output data of the motor performance prediction AI model is the motor performance, and The motor performance includes any one or more of noise / vibration / harshness (NVH), torque, torque ripple and magnetic flux.

6. The method according to claim 1, wherein The motor design parameter optimization AI model uses the output data of the motor performance prediction AI model, namely the motor performance, as input data, and uses the input data of the motor performance prediction AI model, namely the motor design parameters, as output data.

7. The method according to claim 6, wherein: The motor design parameter optimization AI model adopts any one or more of reinforcement learning, Q-learning and particle swarm optimization (PSO).

8. The method according to claim 6, wherein: The motor design parameter optimization AI model is calculated by optimizing multiple combinations of motor design parameters, and The priorities of the multiple combinations of motor design parameters are set under the limitation condition of motor performance.

9. The method according to claim 6, wherein: In the motor design parameter optimization AI model, when the goal of motor performance improvement is to reduce noise / vibration / harshness (NVH), the minimum torque change is set as the power performance constraint.

10. The method according to claim 6, wherein: In the process of optimizing the motor design parameter optimization AI model, the noise level is predicted from a machine learning (ML) model for the noise level.

Images