Unbalance Detection and Prediction in Induction Machine Using Recursive PCA and Wiener Process

This paper focuses on the detection of unbalance in induction machines and the prediction of its dynamic evolution based on the analysis of the current signals. The proposed method is based on the combination of recursive Principal Component Analysis (PCA) and a Wiener process. The recursive PCA is processed on the different features computed from the current signals in order to choose the most relevant ones. This way the fault detection is processed with the only knowledge of the healthy state of the machine. A linear Wiener process is then used to model the behavior of PCA components and predict their evolution over time in order to estimate the dynamic of the tracked fault along with the remaining useful life (RUL). The proposed method is applied on real data from a 5.5 kW induction machine with three different levels of unbalance and obtains very promising results.