Quantification of Forecasting and Change-Point Detection Methods for Predictive Maintenance

Abstract : In order to evaluate the advantages and disadvantages of change detection techniques using Singular Spectral Transform (SST) and Autoregressive Integrated Moving Average (ARIMA) applied to equipment diagnosis, these two techniques are applied to signal data sets and their performance is evaluated. Synthesized signals, periodic and non-periodic, are used to evaluate the capability of detection of both methods for several types of changes. SST was applied to change detection in rotating machines by quantitative evaluation of misalignment in a turbopump assembly. It was shown that the SST method is suitable for detecting change in periodicity, and that it can even be applied to data acquired intermittently. On the other hand, the ARIMA method was effective in detecting change points in continuous data. When comparing the RMS of vibration signals in the case of misalignment to the case of a properly lined pump, no significant difference is detected, but a statistically significant change is present when using the SST Score for change detection. Structural abnormality in rotating machines is difficult to detect using the magnitude of vibration but since the SST detects changes in the shape of the signal, it is much more sensitive to changes related to abnormality.