Fault Detection of Production Equipment by Using Spectral Characteristics and Autoencoder

Abstract

In this paper, we discuss on fault detection method for pressure washers. In our previous study, we proposed a fault detection method using an AutoEncoder, focusing on the amplitude value of each frequency spectrum in the FFT analysis of time series pressure data. We also verified the detection accuracy of artificially generated anomaly data. As a result, we found that the detection performance deteriorated when the amplitude of the spectrum near the resonance frequency changed. Therefore, in addition to the conventional AutoEncoder, we propose a method to improve the performance by treating the spectrum as a lumped spectrum in a set frequency range.