Transforming motor calibration
Learn why a Japanese motor manufacturer believes Secondmind will help it achieve 85% faster calibration
Challenge
For the motor manufacturers, the motor performance measurements are a major obstacle to efficient calibration. These methods require extensive manual testing across various operating conditions, such as different speeds, voltages, and currents. Each motor calibration involves engineers spending significant time manually measuring and adjusting motor performance. This time-consuming approach drains valuable human resources, incurs considerable costs while also delaying time-to-market.
Testing and data inefficiency
To calibrate the motor, they rely on full factorial Design of Experiments (DoE) methodology, which can be extremely data-intensive. Often, the number of data points collected far exceeds what was practically necessary, leading to testing inefficiencies. For example, up to two-thirds of collected data is often unused in performance optimization, which indicates there is significant potential for data reductions. However, reducing data volume using conventional DoE methods risks compromising the precision of the resulting calibration maps and poses a major challenge.
Challenges with control parameters
Compounding this are the four distinct control parameters and one constraint condition motor manufacturers have to manage: motor speed, DC voltage, phase current, lead angle, and modulation index. Each parameter can take many different values, resulting in an exponential growth of combinations. This complex interplay between parameters requires rigorous data collection and constant manual supervision, which ties senior engineering experts in a cycle of repetitive, low-value tasks.
The iterative nature of the traditional DoE approach also adds to the difficulty. The labor-intensive process involves engineers repeatedly and manually adjusting motor settings, running new tests, and creating a calibration map based on the collected data.
Operational constraints with physical testing
Engineers running physical tests face further constraints, such as the availability of motor prototypes, testing equipment, setup durations, and the inherent wear and tear on testing hardware. Ensuring consistent calibration means running the motor at a precise speed and current without fluctuations, a task often complicated by real-world conditions and equipment limitations. Data points where the motor fails to reach target speed or shows instability in electric current vectors have to be excluded, requiring further iterations and adjustments.
The traditional calibration methods require a lengthy and resource-intensive process, which might threaten the ability to bring new motor models to market swiftly, risking operational inefficiencies and potential loss of market share. The urgent need for a more efficient, data-driven calibration method is evident — one that could achieve high accuracy with reduced data and time investment, enabling faster and more streamlined optimization processes.
Results
Lorem ipsum dolor sit amet consectetur. Mi quis vitae ornare elit eget. Habitant fringilla pretium id praesent. Neque eget morbi malesuada tortor laoreet. Vitae aliquam tincidunt aliquam at libero cursus quam eu augue. Scelerisque lorem maecenas sed vitae tempus aliquam facilisis. Tristique magna id nullam diam massa. Pharetra lobortis volutpat sit tortor libero. Egestas mus diam quis congue feugiat leo. Nec sed fusce morbi et metus nisi mollis. Cras sed volutpat proin quam leo amet sed sit viverra. Odio scelerisque nec augue purus sem in suscipit lectus non. Nunc ultrices dolor fames fusce quam. Turpis leo non lectus scelerisque habitant. Enim egestas sapien amet risus amet adipiscing aliquam facilisis.
Process
Secondmind collaborated with a motor manufacturer to prove we can overcome the substantial challenges posed by traditional calibration methods. The collaboration achieved a significant reduction of time and data requirements while maintaining high precision.
Secondmind fully intelligent, automated DoE
Intelligent automation of the DoE process to create the virtual motor
Secondmind Active Learning helps engineers enhance Design of Experiments (DoE) methodologies with AI, by offering intelligent, automated experimental design, data acquisition, model creation and optimization. The figure above illustrates how Secondmind for Calibration connects with the existing test bench.
Engineers can automatically adjust parameters, homing in on the optimal settings with each iteration. This meant calibration engineers only ran the most informative experiments, to effectively reduce the overall number of tests required and swiftly improve model accuracy. As a result, calibration engineers will achieve high precision models in significantly fewer steps, directly addressing the inefficiencies of prior methods.
The plot below illustrates the amount of data reduction achieved by Secondmind compared with the original calibration method. Secondmind for Calibration enables engineers to build high-precision models with the same level of accuracy to the original method but using 80% less data.
Comparison of the amount of data required to achieve the target difference between the grand truth and the model torque value. Scale removed from axes to protect confidentiality.
Secondmind used the acquired data to create a virtual motor model, to enhance the calibration process. Using Secondmind’s advanced machine learning (ML) models, the virtual motor could accurately predict motor performance as it closely represented real-world conditions.
Handling complex parameter interactions with ease
Secondmind can also empower calibration engineers to more effectively manage the complex interplay of the motor’s control parameters: motor speed, DC voltage, phase current, lead angle, and modulation index. Secondmind uses machine learning algorithms to dynamically respond to control variables and adapt to motor characteristics under varying operating conditions, ensuring precise calibration and optimized motor performance.
Conclusion
Secondmind succeeded in proving that motor calibration processes could quickly transition from cumbersome, data-hungry and labor-intensive to streamlined, efficient, and highly accurate. This transformation not only reduces the burden on human resources but also enhances the company's ability to quickly innovate within the fast-paced EV market.
80% data savings: Secondmind achieved remarkable data efficiency, reducing the need for motor measurements by 80%. This reduction in data requirements streamlines the calibration process, easing data collection burdens and expediting the workflow.
Reduction in calibration time: By automating and optimizing the most cumbersome parts of the calibration process with Secondmind Active Learning, total calibration time is expected to reduce by 85%.
Optimized resource utilization: The virtual motor is expected to reduce test bench usage to approximately 20% of previous methods thereby allowing for more efficient and precise calibrations.
This case study demonstrates the impact of Secondmind for Calibration in the electric vehicle industry. By reducing data requirements and speeding up the calibration process, Secondmind provides a scalable, efficient solution that can be integrated into existing workflows. This approach accelerates product development and enhances overall efficiency, paving the way for more sustainable and cost-effective electric vehicle production.
Looking ahead, the motor manufacturer expects to achieve even more impressive results, potentially reducing data usage by more than 80%. This vision is not just an aspiration; it is grounded in the understanding that much of the traditional data previously collected was not critical to achieving optimal calibration. By utilizing Secondmind data-efficient products to only focus on the most informative data points, motor manufacturers can continue refining its processes and optimizing performance without requiring extensive, redundant data collection.
Secondmind’s solution is more than just a tool; it's a partner in the journey towards smarter, greener electric vehicles. As this experience shows, embracing advanced ML can bring substantial benefits, making it a critical component in modern automotive engineering.
