Predicting the failure of railway point machines by using Autoregressive
Integrated Moving Average and Autoregressive-Kalman methods
Sahand Abbasnejad and Ahmad Mirabadi

In this paper, forecasting methods that use autoregressive integrated moving average (ARIMA) and autoregressiveKalman (AR-Kalman) are presented for the prediction of the failure state of S700K railway point machines. Using signal processing methods such as wavelet transform and statistical analysis and the stator current signal, the authors have acquired the time series data of the point machine behavior using a near-failure test point machine. Prediction methods are implemented by utilizing the acquired time series data, and the results are compared with the specified failure margin. Furthermore, the proposed ARIMA method used in this study is compared with the AR-Kalman prediction method, and prediction errors are analyzed.