A Machine Learning Approach on Data-Driven Thermal Monitoring of Induction Motors in Electrified Powertrains

The research is aimed to determine a data-driven approach towards the thermal monitoring of Induction Motors in the context of electrified powertrains. We apply the examination of the temperature data with the help of the appropriate set of sensors, including Thermocouples, Resistance Temperature Detectors , as well as Infrared sensors in combination with the mixture of the Machine Learning algorithms to predict the service requirements and operational needs of the examined system . The data, provided by the thermal sensors, makes it possible to predict the demand for the maintenance of the part or optimize the work of the coolant pumps. Thus, we conduct the performance analysis of several ML models, including Artificial Neural Networks , Support Vector Machines , Decision Trees , and Random Forests across the range of different epochs of the training. Our analysis has shown that ML-based approaches can be highly efficient in accurately predicting the forthcoming maintenance requirements using the sensor-based information. Precisely, the most efficient was the ANN model, producing the result of 97.65%. It was followed by SVM and DT, producing the results of 94.5% and 92.3%, respectively. Finally, RF was only able to produce the result of 90.25%, falling behind the other approaches. Our comparison has provided multiple implications to the relative strengths and weaknesses of the employed models, and the choice of whether they are appropriate must be conducted through the prism of these comparisons. In addition, the analysis of the performance across different epochs has shown that the subsequent ones have produced the better results, which makes the training process crucial.