When implementing new vehicle functions, engineers have to do a lot of work, because with the increasing range of functions and the increased system connectivity of modern vehicles, the number of tests and the number of relevant parameters required for calibration also increases significantly. grows on. To cope with higher complexities in the future as well as increase development efficiency, Porsche Engineering uses artificial intelligence (AI) to calibrate control units and input data. “With Porsche Engineering Reinforcement Learning — PERL for short — we're turning the calibration of driving functions into an intelligent decision-making process,” says Matteo Scull, lead engineer at Porsche Engineering.
PERL is based on deep reinforcement learning, a self-learning AI process. The basic idea is that instead of optimizing individual parameters, the AI develops a strategy that leads to the best possible calibration result for the entire function. The advantages are the high degree of process efficiency of the method, as it is self-learning, and the universal applicability to many areas of vehicle development (see also Porsche Engineering Magazine 1/2021 on Pearl's principle). “Porsche Engineering recognized at an early stage the potential of deep reinforcement learning for automatic calibration of control units. Since 2017, we have been working on PERL together with AI experts from the Porsche Engineering sites in Cluj and Timisoara, Romania. and have been continuously developing the method ever since,” says Dr. Matthias Bach, Senior Manager HV Battery Calibration and Diagnostics. and evaluation at Porsche Engineering.
Porsche Engineering has registered more than 50 patents for PERL.
Since then more than 50 patents have been registered for PERL. Thanks to this expertise, Porsche Engineering is one of the first companies worldwide to integrate it into the development process of new vehicle systems. Meanwhile, PERL has already been used for the calibration process in two customer projects. In the first, Porsche Engineering has been collaborating with the FZI Research Center for Information Technology in Karlsruhe and Porsche for almost three years. In this project, PERL is used to coordinate the formulation of fuel blends for a new gasoline engine for hybrid vehicles. In another project, PERL is used to calibrate vibration damping in the powertrain of a Porsche electric vehicle.
Calibration of controllers
In both cases, AI is used to calibrate the controllers, which presents a particular challenge. Thomas Rudolph, Porsche Engineering and PhD student at FZI, explains: “The calibration of the control functions is a complex task, as we need to precisely control highly dynamic processes. The engine mixture In designing, for example, the injection quantity must be precisely adjusted for each speed and torque combination using parameter maps to obtain the lambda value—the optimum operation of the exhaust gas aftertreatment. For the control variable—match the target.
The biggest challenge involved is the dead time resulting from the spatial distance between the engine and the sensor system at the end of the exhaust track, coupled with the high speed at which the control system must operate, say, under load. During the conversion. If the control system reacts too slowly, the emission control system will malfunction and emissions will increase. If the regulation is too aggressive, the system can see. So it is very important to find a balance between the two extremes. “Custom parameterization of complex and diverse control systems remains a major challenge in the automotive industry. Prof. Søren Hohmann, head of the Institute of Control Systems at the Karlsruhe Institute of Technology (KIT) and director of the FZI, says that advanced learning The methods can make extensive, costly, and sometimes manual calibration significantly faster.Summary of the benefits of using AI.
“Modern learning methods can significantly expand, lower cost, and sometimes speed up manual calibration and make it more efficient.”
Prof. Dr.-Ing. Sören Hohmann, head of the Institute for Control Systems at KIT and director at FZI
PERL is capable of configuring a wide range of control parameters in a manner that determines the optimum mixture for highly dynamic engine operation. “This makes PERL an indispensable development tool—especially for future, stricter emissions standards that will require more precise control of lambda in all operating areas,” says Bach. Dr. Galabina Aleksieva-Rausch, who is responsible for Process, Quality, and Methodology Development at Porsche, shares her assessment: “As part of the capability study offered by PERL, we used it as a conventional calibration. The results were much better than expected: the calibration using the AI approach was already as good as the series calibration, even without fine adjustments. .
The maturity of calibration data in a computer using AI is highly dependent on the task in question, but is typically between 80 and 90 percent. The crude calibration is then fine-tuned and validated through bench tests and test drives, which remain an integral part of future calibration due to quality control and assurance requirements. PERL also assists the applicant in these tasks, as the program can run in the background during the test and make suggestions to further improve the calibration using the data obtained.
Currently, the performance gains from using PERL for series development cannot yet be quantified. However, it is already clear that AI-supported calibration methods such as PERL can significantly speed up the overall development process: “Thanks to PERL, we can run calibrations at a much earlier stage and correct more quickly. can get results. We then take these results and approach the later stages of development in a much more focused and therefore more efficient way,” says Stefano Cheney from the Alekseeva-Rausch team.
Another application case using PERL is vibration damping in an electric vehicle powertrain. In this example, engineers are trying to reduce a disruptive vibration in the powertrain to a specific feedback measurement input—the same principle used in headphones, say, and called noise cancellation. known as. This calibration is primarily concerned with excitation in the range of 1 to 15 Hz, which is often perceived by vehicle occupants as vibration and, in the worst case, can damage the powertrain. As with the composition of the mixture in the engine, optimal control is crucial for the system to function, because for effective damping, the vehicle's electric motor must be properly timed and properly synchronized. Also, the countervibration should not be too strong, since the motor torque necessary to generate the counterpulse is not available for propulsion. This means that, when this concept is applied, rest and driving dynamics interact directly with each other.
Finding the perfect balance
So applicants must find an optimal balance between comfort and play when designing vibration damping. To get the best possible result—a typical result for Porsche—three maps per electrically driven axle are provided with simultaneous data to obtain vibration damping. Due to the high level of complexity, a robust initial input of data to the control units requires a lot of effort if done manually. Only when the data is received by the control units can the applicators start fine calibration. “One of the main goals of PERL was to shorten this period. From the beginning, the focus was on a universal application for AI for very different vehicle platforms and derivatives,” says Maurice Haus, software engineer at Porsche. .
To be able to use PERL, the Porsche engineering and Porsche development teams first used individual test bench data to model the chassis physics of real-life vibration profiles. This model was extended using neural networks. “The hardware model simulates physics while the artificial neuronal network bridges the accuracy gap between the real world and the simulation,” explains Skull. The PERL basic procedure was then applied. The AI agent was trained on a hybrid model using a large dataset and randomly generated initial maps. This ensures good generalization of the strategy.
After the end of the extensive training, target deployment took only a few seconds for the agent to complete the vehicle-specific calibration adaptation. Furthermore, it was possible to transfer the AI strategy from the original vehicle to other derivatives—including with different powertrains—without modification and in some cases without the need for vehicle-specific measurements far exceeding the expectations associated with using PERL. . exceeded. “With the new variants, thanks to PERL, we can start calibration with pre-calculated data. Other test car users can work in parallel on vehicle calibration during this time and we don't need to wait. No,” explains Tobias Rowlett, Manager e-Vehicle Calibration Porsche. “Overall, we can save one to two weeks of calibration time. Furthermore, PERL pre-enters the correct data, so when the powertrain is started for the first time due to incorrect calibration data. There is no risk of harm.”
“PERL has demonstrated that the approach is universally applicable to all types of data entry and that AI can significantly reduce the work and time required in calibration.”
Dr. Matthias Bach, Senior Manager of Battery Calibration and Evaluation at Porsche Engineering
Following the successful completion of both customer projects, the customer evaluation has been entirely positive. “PERL has demonstrated that the approach is universally applicable to all types of data entry and that AI can significantly reduce the work and time required in calibration. This means that newly developed The PERL methodology has made a major contribution to Porsche Engineering's strategic goal: to ensure short delivery times of high-quality solutions for complex tasks, to the benefit of our customers,” according to Bach. Expansion of the use of PERL at Porsche is already in the planning stages. The program is to be used, for example, to optimize camshaft bearing control calibration.
Vibration damping calibration
The neural network improves the accuracy of the simulation.
For effective vibration damping, the motor in an electric vehicle must be activated at the right time and properly synchronized. Also, the countervibration should not be too strong, since the motor torque necessary to generate the counterpulse is not available for propulsion. To be able to use PERL, developers first used testbench data to model the chassis physics of actual vibration profiles. The model was then extended using neuronal networks. The hardware model simulates physics while the artificial neural network bridges the accuracy gap between the real world and the simulation.
PERL policy training
Cloud computing speeds up calculations.
The agent was trained using a large dataset and randomly initialized maps to ensure good generalization of the strategy. 64 CPU cores in the cloud processed the simulations in parallel, resulting in significant acceleration of calculations compared to using a local machine. The strategy is optimized with the graphics processor in the cloud.
Abstract
Porsche Engineering uses reinforcement learning (RL) to calibrate driving functions. The new approach means that a large part of the data input is performed by AI. Only at the end is manual fine-tuning done on the test track. In addition, a single trained RL agent can handle the calibration of multiple vehicle derivatives. As a method, RL significantly reduces the time and costs involved in calibration.
Information
The text first appeared in Porsche Engineering Magazine, Issue 1/2024.
Text: Richard Backhouse
Copyright: All images, videos and audio files published in this article are subject to copyright. Reproduction in whole or in part is not permitted without the written permission of Dr. Ang. HCF Porsche AG. Please contact newsroom@porsche.com for more information.