Design of experiment for diagnosis of the temperatures in AC motor drive

Abstract

The paper describes statistical methods used to diagnose power semiconductors in DTC drive system. Simulation was used to train the data-mining model and then trained model was used to diagnose temperatures. To select the most significant training points the design of experiment was applied. From the measurement point of view the challenge is in predicting the maximum temperature reached by the semiconductors (a momentary effect) based on signals that vary on much larger time scales, e.g., the cooling water temperature. In this paper we present some requirements regarding collection and analysis of drive signals for the purpose of diagnosing temperatures.