Chassis domain controller test and calibration system based on machine learning

Abstract

The invention relates to the technical field of vehicle control and test, and discloses a chassis domain controller test and calibration system based on machine learning. The system is based on a machine learning technology and comprises a data acquisition unit, a calculation unit, a driving simulation unit, a test vehicle and a user interaction unit. The whole process of data acquisition, test scene design, test condition simulation, controller product test and vehicle calibration is intelligentized, the test efficiency can be greatly improved, the test cost can be reduced, the test consistency can be guaranteed, and the reliability and safety of the product can be improved.

Description

Technical Field

[0001] The present invention relates to the field of vehicle control and testing technology, and in particular to a chassis domain controller testing and calibration system based on machine learning. Background Art

[0002] With the development of autonomous driving and intelligent safety technologies, the number of on-board electronic control systems is increasing and their functions are becoming more and more complex. The traditional electronic and electrical architecture can no longer meet the increasingly complex needs of intelligent vehicles. The electronic control system has gradually developed towards domain control and central control. As an integrated system of vehicle driving, steering and braking, the performance of the chassis domain controller is directly related to the safety of the entire vehicle and the driving comfort. Therefore, the adequacy of testing and the consistency of calibration during its development process are of great significance to its performance. Traditional automotive electronic product development testing is mainly based on limited test scenarios, and the calibration process is heavily dependent on the subjective experience and feelings of the calibrator. The proportion of manual intervention factors is too large, which limits the test efficiency and the lack of product consistency and low applicability. Therefore, a new test method and calibration method is urgently needed for the development of chassis domain controllers to improve test efficiency and consistency of product performance.

[0003] The characteristics involved in the chassis domain control system include braking, handling stability, smoothness, passability and many other characteristics. The evaluation method mostly adopts a combination of subjective evaluation and objective evaluation. For the subjective evaluation part, due to the great differences in individual physiology and psychology, different drivers may have very different evaluations on the same characteristic of the same vehicle, resulting in great discreteness in the subjective evaluation results.

[0004] Therefore, the current testing and calibration methods based on limited test scenarios and highly dependent on driver evaluations have the following shortcomings: (1) The test scenarios are limited and the coverage is insufficient, resulting in the risk of insufficient product testing; (2) The test cycle is long and the efficiency is low, resulting in high product development costs; (3) The differences in drivers' subjective evaluations may lead to insufficient product consistency; (4) Extreme operating condition testing has high requirements on the driver's quality and poses a greater safety risk. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a chassis domain controller testing and calibration system based on machine learning, which intelligentizes the entire process from data acquisition, test scenario design, test condition simulation, controller product testing to vehicle calibration. In addition, through machine learning methods, the test scenarios can be quickly updated and expanded based on historical data to the greatest extent, thereby achieving efficient and sufficient product testing, and calibrating product performance based on highly standardized evaluation indicators, thereby ensuring product reliability and performance consistency.

[0006] The technical solution adopted by the present invention is: a chassis domain controller test and calibration system based on machine learning, including a data acquisition unit, a computing unit, a driving simulation unit, a test vehicle, a performance evaluation module, and a user interaction unit; the computing unit includes a data preprocessing module, a model training module, and a parameter optimization and adjustment module; the data acquisition unit and the data preprocessing module communicate directly, and the data acquisition unit transmits the data to the data preprocessing module; after receiving the data, the data preprocessing module classifies, cleans, and marks the data, and then transmits it to the model training module as a sample for learning, and adjusts and optimizes the calibration parameters based on the reward and punishment function rules; The test vehicle includes a chassis domain controller and a sensor component, and the chassis domain controller is installed on the test vehicle; the parameter optimization and adjustment module and the chassis domain controller are connected via CAN communication, and the adjusted parameters are calibrated to the chassis domain controller. After the calibration is completed, the test vehicle communicates with the driving simulation unit to confirm that the calibration is completed and start the driving scenario simulation; the test vehicle collects test data and uploads it to the performance evaluation module, which automatically evaluates the pros and cons of the test results and feeds the test results back to the model training module to provide an adjustment basis for model optimization. At the same time, the results are input into the user interaction unit for testers or third parties to conduct acceptance evaluation and target result screening.

[0007] Furthermore, the data acquisition unit is connected to the sensor component to obtain the input and output data of the test vehicle in real time. The data transmitted by the data acquisition unit includes remote data, historical data and real-time test data, providing a data source for machine learning training.

[0008] Furthermore, the computing unit preprocesses, analyzes, and learns the data acquired by the data acquisition unit based on a machine learning algorithm. The parameter optimization and adjustment module autonomously adjusts and optimizes the parameters that need to be calibrated. By comparing with the historical standard driving database, it autonomously calculates and evaluates in real time whether the performance characteristics of the current test vehicle or the preset control parameters meet expectations.

[0009] Furthermore, the driving simulation unit includes a driving simulation robot and configuration software for different driving conditions. The driving simulation robot is installed on the test vehicle. The driving simulation unit simulates different preset driving conditions by simultaneously controlling the steering, throttle and braking, providing the computing unit with a large amount of preset condition test data and quickly evaluating the quality of the machine learning system parameter matching.

[0010] Furthermore, the user interaction unit records and displays test data, providing an interface for user interaction.

[0011] Furthermore, the parameter optimization and adjustment module calibrates the adjusted parameters to the chassis domain controller based on the CCP protocol.

[0012] A chassis domain controller testing and calibration method based on machine learning is applied to a chassis domain controller testing and calibration system based on machine learning, including: vehicle test data acquisition, data cleaning and key parameter extraction; setting of penalty function and reward function in machine learning, establishment of learning model; quantification and boundary setting of evaluation method; driving condition simulation and extraction of key control parameters.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. Improved testing and calibration efficiency: Using historical data and experienced drivers as training samples and machine learning technology to automatically perform the testing and calibration process of the chassis domain controller, this system can significantly reduce the time required to perform these tasks, greatly improving efficiency compared to traditional manual or semi-automatic methods.

[0014] 2. Improved calibration accuracy: Machine learning models are used to learn and identify complex test patterns and key parameters in test scenarios and their relationships from a large amount of test data, so that test patterns and test scenarios can be optimized more accurately, and the chassis domain controller can be calibrated in a refined manner, which is beneficial to improving the overall performance and stability of the chassis domain controller.

[0015] 3. Reduce labor costs: The present invention relies on the reuse of historical test data and test scenarios, which greatly improves test automation, thereby reducing dependence on the technical level and experience of testers. Therefore, it can significantly reduce the manpower requirements in the testing and calibration process, thereby reducing testing costs.

[0016] 4. Improve chassis domain controller performance and stability: Using machine learning calibration methods, the system can more intelligently import more complex and targeted test scenarios, improve test conditions, optimize controller control parameters, and improve the adaptability of chassis domain controllers to complex conditions, thereby improving system stability and ensuring vehicle driving safety and comfort.

[0017] 5. Adaptability and flexibility: Through the classification of historical data and test scenarios, the system can test and calibrate chassis domain controllers of different types of vehicles, making the test system more adaptable and flexible. In addition, with the further training and optimization of machine learning models, their effectiveness and accuracy will continue to improve over time, and the applicability of test scenarios will also be further optimized and expanded. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the structure of the testing and calibration system of the chassis domain controller based on machine learning described in the present invention. DETAILED DESCRIPTION

[0019] The purpose of the present invention is to provide an intelligent testing and calibration system and method for a chassis domain controller based on machine learning, which is used for testing and calibration during the development of the chassis domain controller, aiming to improve the development efficiency of the chassis domain controller and ensure the adequacy of the testing and the consistency of the calibration during the development of the domain controller. The present invention is further described below in conjunction with the accompanying drawings.

[0020] like Figure 1 As shown in the figure, the test and calibration system of the chassis domain controller based on machine learning mainly includes data acquisition, data preprocessing, model training, parameter calibration, driving condition simulation and visual result feedback, etc. The main structure of the system includes a data acquisition unit, a computing unit, a driving simulation unit, a test vehicle, and a user interaction unit; among them, the data acquisition unit: the data source is not limited to real-time data, and existing historical data and remote cloud data or industry shared data can also be used; real-time data mainly obtains the input and output data of the chassis domain controller in real time through sensor components. The sensor components can include lidar, cameras, inertial measurement units (IMUs), etc., which are used to capture the chassis motion trajectory and environmental information.

[0021] Computing unit: The computing unit includes data preprocessing, model training and parameter optimization and adjustment. Based on the machine learning algorithm, the computing unit preprocesses, analyzes and learns the data obtained by the data acquisition unit and transmits it to the model training unit for training, so as to autonomously adjust and optimize the parameters that need to be calibrated. The computing unit can identify the behavior pattern of the chassis domain controller and evaluate its performance. Through continuous learning and optimization, the system can adapt to different chassis domain controllers and environmental conditions, and autonomously calculate and evaluate whether the current driving or test parameters meet expectations in real time by comparing with the historical standard driving database.

[0022] Driving simulation unit: The driving simulation unit mainly includes a driving simulation robot and configuration software for different driving conditions. It can simulate different preset driving conditions by controlling the steering, throttle and brake at the same time. The driving simulation unit built based on hardware-in-the-loop or vehicle-in-the-loop can realize repeated testing of multiple scenarios and multiple conditions. On the one hand, it can provide a large amount of test data for the computing unit to train the algorithm. On the other hand, it can quickly evaluate the matching quality of machine learning system parameters, thereby realizing a closed loop of data collection, algorithm training, system calibration and algorithm verification, improving development efficiency while improving the effectiveness of the learning algorithm.

[0023] Chassis domain controller: As an actual development product, the chassis domain controller optimizes the internal parameters through machine learning algorithms, verifies the matching effect of the calibration parameters, and provides feedback for further optimization of the algorithm.

[0024] Test vehicle: The test vehicle is the target vehicle to be calibrated. A chassis domain controller, a driving simulation robot and a sensor component are installed on the vehicle. The driving simulation robot actively manipulates the vehicle to simulate different driving conditions, and transmits the vehicle status to the data acquisition unit in real time through the sensor component for processing and evaluation, so as to continuously optimize the calibration parameters.

[0025] User interaction unit: The user interaction unit is used to display test results and provide configuration options for user interaction. Users can use the interface to grasp the performance indicators and calibration parameter tuning trends of the chassis domain controller in real time, and can record historical calibration parameters, provide an interface for user callback confirmation, and allow users to select the appropriate parameter group for acceptance confirmation according to the driving scenario.

[0026] The key technologies of the system of the present invention include: (1) Vehicle test data acquisition, data cleaning and key parameter extraction: In order to achieve automated testing and calibration, the present invention first trains, tests and iteratively optimizes the model through a large number of test data samples, where the vehicle test data mainly comes from massive real vehicle test data of calibration testers and simulation test data based on hardware-in-the-loop (HIL) test platform; then the data is preprocessed. Since there is a large amount of random interference information and invalid data irrelevant to the test in the data, it is denoised and outliers are removed through manual labeling and automated filtering algorithms. In addition, normalization, missing value processing and feature value selection are used to extract valid data and apply it to model training. During the training, key parameters that affect the subjective evaluation of the vehicle, such as longitudinal acceleration, yaw angular velocity, lateral acceleration, etc., are gradually extracted, providing a basis for the training optimization objectives.

[0027] 2) Setting of penalty function and reward function, and establishment of learning model in machine learning: The present invention aims at the needs of automated testing and calibration of chassis domain controllers. For different test scenarios, the rationality boundary of the vehicle's subjective evaluation parameters is used as the basis for setting the penalty function and reward function. A learning model is established by combining the minimum error penalty, performance threshold penalty, and stability and safety rewards, thereby promoting the optimization of test behaviors under different working conditions and providing a basis for parameter calibration.

[0028] 3) Quantification and boundary setting of evaluation methods: The system needs to evaluate the test results and then adaptively adjust the key parameters. Therefore, the system includes quantification and boundary setting of evaluation methods. The evaluation methods include but are not limited to mean square error, mean absolute error, stability (test consistency), response time, safety and other indicators; and the evaluation criteria or quantitative boundaries mainly include vehicle performance indicators, such as braking distance threshold, braking deceleration threshold, acceleration threshold, lateral acceleration threshold, etc. Through these evaluation methods and boundary settings, it is possible to quantify and ensure that the machine learning model achieves the expected performance and meets safety standards in the actual application of chassis domain controller testing and calibration. These measures help promote the industrial application of the model and ensure its reliability and safety in actual vehicle operations.

[0029] 4) A chassis domain controller intelligent testing and calibration system based on machine learning includes driving condition simulation and extraction of key control parameters. Specifically, driving condition simulation provides test scenarios and training data for machine learning, and the parameter settings of the test scenarios or conditions must meet the functional and performance requirements of the vehicle, based on chassis control industry standards such as national standards and ISO standards, and taking into account different operating environments, such as different road conditions, weather conditions, temperature changes, etc., and can be combined with historical data, expert experience, etc. to enrich and optimize the scenarios. Through the process of condition simulation testing, key test parameters that affect vehicle performance can be extracted. Using sensitivity analysis, experimental design methods and data-driven methods, key control parameters such as the correlation between vehicle smoothness indicators and solenoid valve pressure increase and decrease time, the correlation between oversteering or roll trend and steering wheel steering speed, etc., are extracted, so as to optimize the parameters in continuous testing and learning. By extracting and confirming key parameters, the behavior of the model in actual operation can be better understood, ensuring that the testing and calibration activities of the chassis domain controller are more accurate and effective, which helps to improve vehicle performance and ensure driving safety.

[0030] In this embodiment, the data acquisition unit provides the system with the raw data required for chassis domain controller testing. Through various sensors installed on the test vehicle, such as acceleration sensors, wheel speed sensors, steering angle sensors, and tire pressure sensors, the data acquisition unit continuously collects the vehicle's operating data, including but not limited to the vehicle's speed, acceleration, steering angle, and pressure value of each tire.

[0031] To ensure data accuracy and consistency, the collection process takes into account different driving conditions and environmental changes, including different road conditions (such as dry, wet, icy) and different driving modes (such as normal, intense, idle) and different driving styles (cautious, general, aggressive). The data collection unit can also automatically record environmental conditions (such as temperature, humidity, visibility, wind speed, etc.), which are important factors affecting the performance of the chassis domain controller.

[0032] The data preprocessing module performs a series of preprocessing operations on the collected raw data, including filtering, denoising, normalization, and data missing value processing. Advanced signal processing techniques, such as Kalman filters, are used to optimize data quality, remove outliers and noise, and obtain more accurate and reliable data for subsequent processing. In addition, feature scaling techniques (such as normalization) are used to adjust all data to the same level to improve the training effect of the machine learning model.

[0033] The preprocessed data is fed into the model training module for feature extraction. In this module, features that are critical to calibrating and testing chassis domain controller performance are extracted from the processed data. These features may include acceleration change patterns, steering response time, braking distance, etc., which are key indicators for evaluating the performance of vehicle chassis systems. By applying principal component analysis (PCA) and other feature selection techniques, the most informative features can be identified and extracted from large amounts of data.

[0034] After feature extraction, these features are used to train the machine learning model. In this embodiment, machine learning algorithms such as support vector machine (SVM) and deep neural network (DNN) are selected for model training. Through a large amount of real vehicle test data, the machine learning model can learn the best response mode of the chassis domain controller under different working conditions.

[0035] During the model training process, cross-validation and other techniques are used to avoid overfitting and ensure that the model has good generalization ability. At the same time, the model is optimized by adjusting the parameters of the algorithm to achieve the best testing and calibration performance.

[0036] Using the trained machine learning model, the chassis domain controller is automatically calibrated in the parameter optimization module. During the calibration process, the model will guide how to adjust the controller parameters to achieve the best vehicle performance. The results of the model evaluation will indicate which parameters need to be adjusted and how to adjust them, so as to accurately control the vehicle's driving performance and ensure the stability and safety of the vehicle under different conditions.

[0037] The calibrated chassis domain controller will be repeatedly tested for performance on the target calibration vehicle, including but not limited to simulated environment tests and actual road tests. The test results will be fed back to the machine learning model training module and the human-computer interaction module, and the results before and after adjustment will be fed back to the designer in a graphical way for further optimization and adjustment of the model. At the same time, the results will be presented to users for evaluation to achieve continuous performance improvement.

Claims

1. A chassis domain controller testing and calibration system based on machine learning, characterized in that: The system includes a data acquisition unit, a computing unit, a driving simulation unit, a test vehicle, a performance evaluation module, and a user interaction unit; the computing unit includes a data preprocessing module, a model training module, and a parameter optimization and adjustment module; the data acquisition unit communicates directly with the data preprocessing module, and the data acquisition unit transmits data to the data preprocessing module; after receiving the data, the data preprocessing module classifies, cleans, and marks the data, and then transmits the data to the model training module as a sample for learning, and adjusts and optimizes the calibration parameters based on the reward and punishment function rules; the test vehicle includes a chassis domain controller and a sensor component, and the chassis domain controller is installed on the test vehicle; The parameter optimization and adjustment module and the chassis domain controller are connected via CAN communication, and the adjusted parameters are calibrated to the chassis domain controller. After the calibration is completed, the test vehicle communicates with the driving simulation unit to confirm the completion of the calibration and start the driving scene simulation; the test vehicle collects test data and uploads it to the performance evaluation module, and the performance evaluation module automatically evaluates the pros and cons of the test results, and feeds the test results back to the model training module to provide an adjustment basis for model optimization, and at the same time inputs the results into the user interaction unit for testers or third parties to conduct acceptance evaluation and target result screening.

2. The chassis domain controller testing and calibration system based on machine learning according to claim 1, characterized in that: The data acquisition unit is connected to the sensor component to obtain the input and output data of the test vehicle in real time. The data transmitted by the data acquisition unit includes remote data, historical data and real-time test data, providing a data source for machine learning training.

3. The chassis domain controller testing and calibration system based on machine learning according to claim 1, characterized in that: The computing unit preprocesses, analyzes, and learns the data acquired by the data acquisition unit based on a machine learning algorithm. The parameter optimization and adjustment module autonomously adjusts and optimizes the parameters that need to be calibrated, and autonomously calculates and evaluates in real time whether the performance characteristics of the current test vehicle or the preset control parameters meet expectations by comparing with a historical standard driving database.

4. The chassis domain controller testing and calibration system based on machine learning according to claim 1, characterized in that: The driving simulation unit includes a driving simulation robot and configuration software for different driving conditions. The driving simulation robot is installed on a test vehicle. The driving simulation unit simulates different preset driving conditions by simultaneously controlling steering, throttle and braking, providing the computing unit with a large amount of preset condition test data and quickly evaluating the quality of machine learning system parameter matching.

5. The chassis domain controller testing and calibration system based on machine learning according to claim 1, characterized in that: The user interaction unit records and displays test data and provides an interface for user interaction.

6. The chassis domain controller testing and calibration system based on machine learning according to claim 1, characterized in that: The parameter optimization and adjustment module calibrates the adjusted parameters to the chassis domain controller based on the CCP protocol.

7. A chassis domain controller testing and calibration method based on machine learning, applied to a chassis domain controller testing and calibration system based on machine learning as claimed in any one of claims 1 to 6, characterized in that: include: Vehicle test data acquisition, data cleaning and key parameter extraction; Setting penalty functions and reward functions in machine learning, establishing learning models; quantification and boundary setting of evaluation methods; driving condition simulation and extraction of key control parameters.

Images