RANCANG BANGUN MODEL ESTIMASI TEMPERATUR MAGNET PERMANEN MOTOR SINKRON BERBASIS PEMBELAJARAN MESIN

• Main Author: Valentin, Renaldo
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/52742
• Institution: Institut Teknologi Bandung

Permanent magnet synchronous motor (PMSM) has experienced a significant increase in production by various automotive industries (Bocker, 2020). In order to optimize the capacity, reliability, and lifetime of PMSM, overheat in PMSM’s magnet permanent must be avoided. At present, the majority of the automotive industry cannot monitor permanent magnet’s temperature accurately. This problem has the potential to cause high thermal stress on PMSM components, and leads to damage and decreased lifetime of the MSMP which results in increased production cost or operational costs in the industry. One alternative solution to solve this problem is to design a temperature estimation model based on machine learning. The temperature estimation model for the PMSM component was designed using the Xtreme Gradient Boosting (XGBoost), Decision Tree, and Random Forest Algorithm. The design of this model is carried out in two stages, namely the design of the model using default hyperparameters and hyperparameters that are optimized by the Bayesian Optimization method. In the evaluation phase of the model, three different metrics are used, namely mean absolute error, mean squared error, and R2. Designing the best model using the default hyperparameter produces performance metrics MAE 2.55 +/- 0.0083 (oC), MSE 11.96 +/- 0.0567 (oC), and R2 0.79 +/- 0.0009, while designing the best model after parameter optimization results in MAE 1.01 performance metrics. +/- 0.0083 (oC), MSE 2.58 +/- 0.0146 (oC), and R2 0.97 +/- 0.539. The final temperature estimation model that has been designed is able to detect permanent magnet temperatures with absolute maximum prediction error in the range of 1.49oC - 1.59oC with a confidence level of 95%.