2020/01/17 by Wilhelm Kirchgässner, Wilhelm Kirchgassner, Oliver Wallscheid +2 · 116 citations
Computer Science · Engineering · Materials Science · Mathematics · #Artificial neural network #Automotive industry #Benchmark (surveying) #Control theory (sociology) #Electric Motor Design and Analysis #Hyperparameter #Magnet #Magnetic Properties and Applications #Overheating (electricity) #Robustness (evolution) #Sensorless Control of Electric Motors #Support vector machine #Test bench #cs.LG #cs.SY #eess.SY #stat.ML
paper · pdf · doi:10.1109/tec.2021.3052546
published in IEEE Transactions on Energy Conversion 36(3), 2059-2067 (Institute of Electrical and Electronics Engineers) · preprint for TII: SS on Applications of Artificial Intelligence in Industrial Power Electronics and Systems
arxiv created 2020/01/17 · openalex created_date 2020/01/23 · arxiv updated 2021/01/27 · openalex publication_date 2021/02/06 · openalex updated_date 2026/08/05
Monitoring the magnet temperature in permanent magnet synchronous motors (PMSMs) for automotive applications is a challenging task for several decades now, as signal injection or sensor-based methods still prove unfeasible in a commercial context. Overheating results in severe motor deterioration and is thus of high concern for the machine's control strategy and its design. Lack of precise temperature estimations leads to lesser device utilization and higher material cost. In this work, several machine learning (ML) models are empirically evaluated on their estimation accuracy for the task of predicting latent high-dynamic magnet temperature profiles, specifically, ordinary least squares, support vector regression, k-nearest neighbors, randomized trees, and neural networks. Having test bench data available, it is shown that ML approaches relying merely on collected data meet the estimation performance of classical thermal models built on thermodynamic theory. Through benchmarking, this work reveals the potential of simpler ML models in terms of regression accuracy, model size, and their data demand in comparison to parameter-heavy deep neural networks, which were investigated in the literature before. Especially linear regression and simple feed-forward neural networks with optimized hyperparameters mark strong predictive quality at low to moderate model sizes.