Method of predicting performance and optimizing design variable of vehicular drive motor using ai

Abstract

To provide an AI-model method for optimizing a motor design variable allowing for providing a motor design variable to achieve a targeted performance.SOLUTION: An AI-based vehicular drive motor developing system 1 includes a predictive AI model generating section 30 that provides a prediction of motor performance improvement including NVH (Noise / Vibration / Harshness) with a predictive AI model obtained from a motor design variable 10a changed in dimensions out of motor design variables in a motor CAD drawing, by polynomial curve fitting of noise-peak prediction and a design variable-optimizing AI model generating section 40 that uses a predictive AI model as a feature extractor for a target amount of motor performance improvement so as to provide dimensions of a motor design variable optimized with a design variable-optimization proposing AI model obtained by any one of reinforcement learning, Q-Learning and PSO (Particles Swarm Optimization).SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the design of a vehicle drive motor, and more particularly to a method for predicting the performance of a vehicle drive motor using AI (Artificial Intelligence) and optimizing the motor design based on this prediction. [Background technology]

[0002] Generally, motor CAD (Computer Aided Design) is utilized in the development of vehicle motors (i.e., drive motors).

[0003] As an example, the motor CAD motor development method determines motor design variables (e.g., factors of the stator-rotor assembly) that are determined to be changeable in the motor CAD drawings, analyzes all conditions regarding the interrelationships between the determined design variables on a case-by-case basis, and predicts motor performance improvement items, including NVH (Noise / Vibration / Harshness) performance, through simulation of the analysis results.The results are then applied to the dimensions of the motor design variables to complete the motor development.

[0004] Therefore, the motor CAD development method requires the generation and analysis of a simulation analysis model related to the CAD motor drawing, and for this purpose, it is very important to analyze the motor development target performance prediction and determine the motor design variables, which are the design variables.

[0005] However, in the motor CAD development method, specific units that motor designers have judged to be changeable based on their experience are determined as motor design variables, and when generating an analysis model related to these, it is currently not possible to reflect all combinations of variables with respect to the interrelationships between the motor design variables.

[0006] The reason for this is that the motor CAD development method is limited in that it is time- and cost-inefficient to generate an analysis model that reflects all combinations of motor design variables. Due to this limitation, the motor designer must rely on their own experience to determine the impact of some design variables that can be changed in the design. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] US Patent Publication US2008-0270093 (October 30, 2008) Summary of the Invention [Problem to be solved by the invention]

[0008] In view of the above, the present invention provides an AI model method for optimizing motor design variables that can provide motor design variables to achieve target performance by using artificial intelligence learning to present an AI model for predicting motor performance from data related to motor design variables and motor performance in motor CAD drawings, extracting features of the AI ​​model for predicting motor performance, and applying reinforcement learning to set target amounts for improving motor performance. [Means for solving the problem]

[0009] The method for predicting the performance of a vehicle drive motor and optimizing its design variables using AI includes the steps of acquiring motor design variables and motor performance data for a drive motor installed in a vehicle using a data acquisition unit, generating a motor performance prediction AI model using the motor design variables and motor performance data using an automated machine learning unit (Auto ML), and applying an evolutionary algorithm to the motor performance prediction AI model to generate a motor design variable optimization AI model through reinforcement learning.

[0010] Furthermore, the target data of the motor performance prediction AI model is the motor performance, which is obtained through analysis and testing of the motor design variables.

[0011] Furthermore, the motor design variables are combined using DOE.

[0012] Furthermore, the input data for the motor performance prediction AI model are the motor design variables, which include one or more of slots, teeth, stator tauts, bridges, magnets, and center posts.

[0013] Furthermore, the output data of the motor performance prediction AI model is the motor performance, and the motor performance can include one or more of NVH (Noise / Vibration / Harshness), torque, torque ripple, and magnetic flux.

[0014] In addition, the motor design variable optimization AI model has the motor performance, which is the output data of the motor performance prediction AI model, as input data, and the motor design variables, which are the input data of the motor performance prediction AI model, as output data.

[0015] Furthermore, the motor design variable optimization AI model can be applied to one or more of reinforcement learning, Q-Learning, and PSO (Particles Swarm Optimization).

[0016] Furthermore, the motor design variable optimization AI model calculates multiple combinations of optimized motor design variables, and prioritizes the combinations of the multiple motor design variables under constraints on the motor performance.

[0017] Furthermore, if the goal of improving motor performance is to reduce NVH, the motor design variable optimization AI model sets a minimum torque change amount as a power performance constraint.

[0018] In addition, the noise level is predicted from a machine learning model for noise level during the optimization process of the motor design variable optimization AI model. [Effects of the Invention]

[0019] The optimization method for vehicle drive motor design using the AI-based motor development system of the present invention proposes an optimized motor design variable dimension group based on the target amount of motor performance improvement of the design variable optimization proposal AI model that utilizes the performance prediction AI model based on changes in the design variables of a vehicle motor, particularly an electric vehicle motor, and achieves the following actions and effects.

[0020] It is possible to construct a performance prediction AI model for performance prediction by labeling NVH performance, magnetic force, torque ripple, and motor torque performance data by changing the design variables of analytical simulations using experimental radiated noise data of the motor and CAD drawing data.

[0021] By using the performance prediction model based on changes in design variables as a new feature extractor, it is possible to propose multiple combinations of design variables that have a high probability of influencing the desired target performance.

[0022] In the process of applying the performance prediction AI model as a feature extractor for the design variable model, the SHAP function, which is a basic descending sorting based on the highest importance of the feature importance plot, can be applied to enable the motor developer's domain knowledge and physical explanation of the prediction model, and multiple combinations of optimized design variables to achieve the target performance are provided.

[0023] The experimental and analytical results of the input data are labeled through a two-stage process. In particular, in the first stage, the performance prediction model is able to accurately predict performance for a variety of changing combinations of design variables. In the second stage, the performance prediction AI model from the first stage is used as one of the feature extractors, and reinforcement learning can be used to propose multiple combinations of optimized design variables for the target performance.

[0024] It is possible to verify the performance of the combinations of all motor design variables, and it is possible to present feasible design variables. [Brief explanation of the drawings]

[0025] [Figure 1] 1 is a block diagram of a method for predicting the performance and optimizing the design of a vehicle traction motor according to the present invention; [Figure 2] 10 is an example showing the difference in prediction result data of a performance prediction AI model for each vehicle segment according to the present invention. [Figure 3] 1 is a conceptual diagram illustrating a step-by-step method for predicting the performance and optimizing the design of a vehicle traction motor according to the present invention; [Figure 4] 1 is a flowchart of a method for deriving a motor design optimization model after an AI model for predicting vehicle traction motor performance is created according to the present invention. [Figure 5] FIG. 1 is a conceptual diagram illustrating how a design optimization model is created from an AI model for predicting vehicle drive motor performance according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0026] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, these embodiments are merely examples and those skilled in the art will be able to realize the present invention in various different forms, and therefore the present invention is not limited to the embodiments described herein.

[0027] 1, the datasets used in the present invention are motor design variables 10a and motor performance 10b. The motor design variables 10a are provided from motor CAD (Computer Aided Design) drawings, and the motor performance is provided from simulation results calculated based on the motor design variables.

[0028] The motor design variables 10a are the main design variables of the stator / rotor assembly, which is the drive unit of the electric vehicle motor. That is, in the present invention, as data for training the vehicle drive motor performance prediction AI model, the motor design variables 10a include the dimensions of the stator and rotor, and the motor performance 10b includes analysis result values ​​for each performance, such as NVH, torque, torque ripple, radial magnetic flux, and tangential magnetic flux, as target data.

[0029] More specifically, the motor design variables 10a include a slot length, a slot radius, a tooth tip thickness, a tooth tip angle, a tooth tip width, a slot width, a first / second layer bridge thickness, and a second layer bridge width. st / 2 nd Layer), 1st / 2nd layer magnet thickness (Magnet Thickness 1 st / 2 nd Layer), Center Post Thickness between Two Magnets, Center Post Thickness between Two Magnets 1 st / 2 nd Layer), Magnet Angle, 1st / 2nd Layer Magnet Angle st / 2 nd Angle), 1st / 2nd layer magnet length (Magnet Length 1st / 2 nd Apply settings such as Layer, and Stator Tooth Tip Width.

[0030] Meanwhile, motor design variables can be obtained in a form in which design variables are combined using DOE (Design of Experiment), and data on motor specifications of competitors can be used, or new data can be added and applied.

[0031] As an example, data for 352 DOE points is obtained for 11 motor design variables, including bridge thickness around the rotor (Bridge Thickness Around Rotor OD), center post thickness between two magnets (Center Post Thickness between Two Magnets), magnet thickness / width / angle, stator tooth tip width, stator tooth thickness, stator tip angle, and slot length / width / radius.

[0032] The motor performance 10b is target data that indicates the motor performance generated by the interaction between the rotor and the stator according to the motor design variables 10a, and is the result of an analysis of the dimensions given to the motor design variables 10a.

[0033] In this case, the motor performance 10b includes NVH (Noise / Vibration / Harshness), torque ripple percentage (Torque Ripple in%), torque ripple harmonic order (Torque Ripple Harmonics Order006 / 012 / 018 / 024), radiated power order (Radiated Power Order6 / 12 / 18 / 24), stator maximum density (Fr / Ft_Density_Stator_Max), noise level (OA Noise Level), noise area (Noise Area), number of peaks (Peak1 / 2 / 3), peak torque (Peak Torque), and shaft speed (Shaft Speed).

[0034] An AI model for predicting vehicle drive motor performance is created in an AI model learning unit 20 using motor design variables 10a as input data and motor performance 10b as target data.

[0035] When a motor design variable 10aa that has been changed to a new dimension from the motor design variable 10a is input, the vehicle drive motor performance prediction AI model 30 trained by the AI ​​model learning unit 20 predicts modified performance prediction result data 31A based on the modified motor design 10aa.

[0036] As an example, the motor design variables 10aa changed to the new dimensions are design variables that indicate the main performance of the motor, and can be applied by changing the dimensions of the slot length / width, tooth tip thickness / angle, first / second layer bridge thickness, first / second layer magnet thickness / length, magnet angle, center post thickness between two magnets, and stator tooth tip width.

[0037] The vehicle drive motor performance prediction AI model 30 created by the AI ​​model learning unit 20 is provided as a motor performance prediction AI model set so that the correlation, which is the performance change due to the change in motor design variables, is higher than 0.95.

[0038] Furthermore, a design variable optimization AI model 40 is generated from the vehicle drive motor performance prediction AI model 30 created 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.

[0039] The motor performance prediction AI model is generated by setting n points that represent the operating range of the electric vehicle motor, obtaining key performance result data at the n points, and then using this as input data and target data.The model is then trained so that the accuracy of motor performance relative to the motor design variables has a correlation of 95% or more.

[0040] The electric power supply is driven by current from the input and target data obtained through testing and calculations, and the motor performance prediction AI model sequentially calculates performance such as the electromagnetic system based on electromagnetic force, the mechanical system based on velocity, and the acoustic environment based on acoustic noise.

[0041] In particular, the performance of electromagnetic systems is calculated using FE based models, which are any one or more combinations of Air Gap Force Calculation, Maxwell Stress Tensor Method, and Fourier Analysis; the performance of mechanical systems is calculated using analytical based models, free motion response, and force response, which are any one or more combinations; and the acoustic noise performance is calculated using analytical based models and sound power levels, which are any one or more combinations.

[0042] The motor performance prediction AI model 30 can be more reliable than conventional models in which motor developers predict performance based on their own domain knowledge. To improve reliability, the AI ​​model generated, stored, and provided by the model learning unit 20 may be referred to as the motor performance prediction AI model 30. The location where the motor performance prediction AI model 30 is provided or stored may be referred to as a motor performance prediction AI model generation unit.

[0043] The design variable optimization AI model 40 provides a design variable optimization AI model in which the dimensions of the motor design variables are optimized in response to input of target performance for improving motor performance, including reducing NVH. The location where the design variable optimization AI model 40 is provided or stored may be referred to as a design variable optimization AI model generation unit.

[0044] The design variable optimization AI model 40 can provide a design optimization model that is capable of predicting motor design variables and dimensions 41a that can achieve the target NVH performance changes and power performance 10bb, based on the motor performance prediction AI model 30 with improved reliability.

[0045] That is, the design variable optimization AI model generation unit 40 uses the target performance of the motor, including NVH, as input data and the motor design variables as output data, and applies reinforcement learning, Q-Learning, and PSO (Particle Swarm Optimization) in combination or selectively as AI recommendation algorithms, with the input / output set in reverse to that of the motor performance prediction AI model.

[0046] In other words, the AI ​​recommendation algorithm provides an AI model that proposes design variable optimization and has design variable changes and dimensional outputs that can achieve the target performance for improving motor performance, including NVH.

[0047] Referring to FIG. 2, motor design variables and performance data of the motor 200 are acquired for vehicles 300 classified by segment based on the vehicle's electrical equipment (length from the front bumper to the rear bumper) and price.

[0048] As an example, the type A motor 200a is an example of extracted motor design variables and NVH performance characteristics suitable for an A-segment vehicle, the type B motor 200b is an example of extracted motor design variables and NVH performance characteristics suitable for a B-segment vehicle, and the type C motor 200c is an example of extracted motor design variables and NVH performance characteristics suitable for a C-segment vehicle. These motor design variables and motor performance including NVH for each of A, B, and C are used as input data and target data, respectively, to be acquired for modeling, or as input data for a constructed model.

[0049] Referring to Figures 1 and 3, a detailed configuration for generating a performance prediction AI model and a design variable optimization AI model in the AI ​​model learning unit 20 is shown, which is a conceptual diagram of a step-by-step method for predicting the performance and optimizing the design of a vehicle traction motor.

[0050] Specifically, the motor performance prediction AI model 30 generated by the AI ​​model learning unit 20 is provided by a data acquisition unit 32, a data versioning unit 33, a model generation unit 34, a model test evaluation unit 35, and a performance prediction model completion unit 36.

[0051] As an example, the data acquisition unit 32 of the learning model takes motor design variables as input data 10a. Here, the motor design variables used as input data include the motor design variables and their dimensions.

[0052] Experimental labeling values ​​or simulation (analysis) labeling values ​​for motor design variables are obtained from the target data 10b, and learning model data with extracted features is selected through a data selection process based on outer-line data analysis of these labeling values.

[0053] The learning model data acquisition unit 32 is composed of a data indigestion part 32a containing a huge number of experimental and analytical labeling values ​​that indicate performance for each motor design variable, a data exploration part 32b that investigates and analyzes the acquired data containing a huge number of experimental and analytical labeling values ​​to derive only useful information, a data cleaning part 32c that identifies, corrects, and filters errors, omissions, and inaccurate values ​​in the data set, and a data feature engineering part 32d that optimizes the characteristics of the outer-line data between the experimental labeling values ​​and simulation (analysis) and extracts features as useful information to acquire the learning model data that will ultimately be used.

[0054] The data version management unit 33 manages data versions relating to already acquired data and newly acquired data sets for motor design variables and performance used in learning, and performs systematic model management with an index structure for data amplification.

[0055] The data version management unit 33 provides training data and validation data from the acquired data for the model generation unit 34, and provides test data for model evaluation in the model test evaluation unit 35.

[0056] The model engineering unit 34a of the model generation unit 34 uses training data for model generation, and constructs a machine learning model by automating the process of developing the machine learning model using an automated machine learning unit (AutoML Part) 34c.

[0057] The constructed machine learning model is subjected to model verification by the model evaluation unit 34b using verification data provided by the data version management unit 33.

[0058] The model engineering part 34a, model evaluation part 34b, and automated machine learning part (AutoML Part) 34c are all included and are based on the machine learning lifecycle platform 34d. The machine learning lifecycle platform 34d is a comprehensive solution that supports the development, distribution, monitoring, and maintenance of machine learning models.

[0059] The model generation unit 34 constructs a model that incorporates a deep neural network structure, with the motor design variables as input data for the motor performance prediction AI model and the motor performance as output data, to improve the accuracy of the model's output data compared to the actual performance.

[0060] In particular, the model engineering unit 34a and the model evaluation unit 34b improve the correlation of performance between the training model and the validation model through mutual data exchange and verification, and the automated machine learning unit 34c improves the correlation of performance between the performance prediction model through data exchange and verification between the model engineering unit 34a and the model evaluation unit 34b.

[0061] The model test evaluation unit 35 evaluates the performance prediction model of the automated machine learning unit 34c using test data by analyzing the sensitivity of performance changes due to changes in model design variables (i.e., dimensions) and visualizing the results, allowing developers to compare the results with their own domain knowledge results and improve the reliability of the performance prediction AI model.

[0062] To this end, the model testing unit 35a of the model testing and evaluation unit 35 applies test data to the performance prediction model to test the accuracy of the model.

[0063] On the other hand, the Shapley Additive Explanation Part (SHAP) 35b of the model testing and evaluation unit 35 is a statistical technique and framework for analyzing and explaining the predictions of machine learning models, and is used to evaluate how much a predicted value contributes to which characteristics or factors, thereby assisting in the model's predictive analysis.

[0064] The performance prediction model completion unit 36 ​​finalizes (36a) the performance prediction model, and receives an explainable interface from the SHAP unit 35b at the result visualization processing unit 36b.

[0065] The design variable optimization AI model generation unit 40 extracts the features of the design variable optimization AI model 40 from the motor performance prediction AI model 30 finally determined by the motor performance prediction model determination unit 36a, and the final design variable optimization AI model 40 has the target performance of the motor as input data and the motor design variables for the target performance as output data.

[0066] A reinforcement learning method is applied to the design variable optimization AI model, and the features of the motor performance prediction AI model are extracted through optimization. The design variable optimization AI model generation unit 40 is divided into an evolutionary algorithm unit 42a for design optimization development and a reinforcement learning unit 42b for design engine development, and both are finally used as feature extractors for the performance prediction model in an evaluation unit 43 for the optimization method.

[0067] 4 shows an example in which NVH is applied as motor performance, where a training dataset including peak values ​​and peak positions of the overall noise level is acquired as motor performance for the motor design variables to which motor design (DOE) is applied in the data acquisition unit 32, and an overall noise level model is generated by the motor performance prediction AI model. The noise peak prediction is output from the overall noise level model using polynomial curve fitting.

[0068] Figure 5 shows the process of deriving an AI model for optimizing motor design variables for NVH as a motor performance factor. If the goal of improving motor performance is to reduce NVH, then the AI ​​model for optimizing motor design variables sets a minimum torque change as a power performance constraint.

[0069] Therefore, if NVH improvement performance is input as a target quantity under the condition of minimizing torque change as NVH performance improvement progresses, the optimal motor design variables for achieving the motor performance improvement goal will be output. To derive an optimization model that can achieve this, one or more of reinforcement learning, Q-Learning, and PSO (Particle Swarm Optimization) are applied, and the NVH design machine learning model 44 is optimized to show motor performance with a correlation of 95% or more with the motor design variables. When the motor performance improvement goal, including NVH, is achieved, the NVH design machine learning model is confirmed as an AI model proposing design variable optimization, and motor design variables are output.

[0070] The motor design variable optimization AI model calculates n combinations of design variables 46a. Furthermore, under the constraints on the motor performance, priorities are set for the combinations of the n motor design variables.

[0071] In other words, the optimization model presents n combinations of motor design variables and dimensions as output data, and prioritizes the output data based on the design variables that have the greatest impact on improving NVH performance and that minimize torque change in terms of power performance.

[0072] A machine learning model 45 for the noise levels determined from the optimized NVH design machine learning model 44 can predict noise levels 46B for the n combinations.

[0073] That is, an optimized NVH design machine learning model 44 and a machine learning model 45 for noise levels are derived from the design variable optimization AI model generation unit 40, and the NVH design machine learning model and noise levels are optimized so that the NVH performance improvement target can be achieved by repeated execution of the NVH design and noise level machine learning model. [Explanation of symbols]

[0074] 1: AI-based vehicle drive motor development system 10: Motor design variable provider 10a: Motor design variables 10a: Dimension change input data 10b: Motor performance improvement input data 10b: Motor performance 20: AI model learning section 30: Motor performance prediction AI model (generation part) 31A: Performance prediction result data 32: Data acquisition section 32a: Data Indigestion Part 32b: Data Exploration Part 32c: Data Cleaning Part 32d: Feature Engineering Part 33: Data Versioning 34: Model generation unit 34a: Model Engineering Part 34b: Model Evaluation Part 34c: Automated Machine Learning Part (AutoML Part) 34d: Machine Learning Lifecycle Platform (ML Lifecycle Platform Part) 35: Model Test and Evaluation Department 35a: Model Testing Part 35b: Result analysis model part (SHAP part) 36: Performance prediction model completion section 36a: Final NVH Prediction Model Part 36b: Result visualization processing part (Explainable Interface Part) 40: Motor design variable optimization AI model (generation part) 41a: Design variable optimization result data 42a: Evolutionary Algorithms for Design Optimization Development 42b: Reinforcement learning section for design engine development 43: Evaluation section for optimization method 44: NVH Design Machine Learning Model 45: Machine learning model for noise levels 46a: n combinations of design variables 46B: Noise levels for n combinations 100: Motor CAD drawing 200: Motor 201: Stator / rotor assembly 200a: A type motor 200b: B type motor 200c: C type motor 300: Vehicle

Claims

1. acquiring motor design variables and motor performance data for a traction motor mounted on a vehicle; generating a motor performance prediction AI model in an automated machine learning unit (AutoML) using the motor design variables and the motor performance data; applying an evolutionary algorithm to the motor performance prediction AI model to generate a motor design variable optimization AI model through reinforcement learning; A method for predicting vehicle drive motor performance and optimizing design variables using AI.

2. The target data (Target Data) of the motor performance prediction AI model is the motor performance (10b) and is obtained by analysis of the motor design variables. A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 1.

3. The motor design variables are characterized in that they can be combined using DOE. A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 1.

4. Input data of the motor performance prediction AI model are the motor design variables, and the motor design variables include one or more of slots, teeth, stator tauts, bridges, magnets, and center posts. A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 1.

5. The output data of the motor performance prediction AI model is the motor performance, and the motor performance includes one or more of NVH (Noise / Vibration / Harshness), torque, torque ripple, and magnetic flux. A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 1.

6. The motor design variable optimization AI model is characterized in that the motor performance, which is output data of the motor performance prediction AI model, is input data, and the motor design variables, which are input data of the motor performance prediction AI model, are output data. A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 1.

7. The motor design variable optimization AI model is characterized in that one or more of reinforcement learning, Q-learning, and PSO (Particle Swarm Optimization) is applied. A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 6.

8. The motor design variable optimization AI model calculates the optimized motor design variables using a plurality of combinations, a priority order is set for combinations of the plurality of motor design variables under constraints on the motor performance; A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 6.

9. The motor design variable optimization AI model is characterized in that, if a target of the motor performance improvement is reduction of NVH, a minimum torque change amount is set as a power performance constraint condition. A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 6.

10. The noise level is predicted from a machine learning model for the noise level during the optimization process of the motor design variable optimization AI model. A method for predicting performance and optimizing design variables of a vehicle drive motor using AI according to claim 6.

Images