Fault detection for industrial processes based on time-serial maximal deviation analysis

Abstract

The applications of feature extraction algorithms in data-driven fault detection have been widely investigated for industrial processes, and there are plenty of multivariate analytics available in the literature for fault detection. However, the mainstream methods all focus on exploiting the systematic variation in historical normal samples, and a fault is usually isolated as a deviation from the characterized normal signatures. With respect to the sole purpose of fault detection for dynamic processes, timely and adaptively uncover deviation inherent in time-serial variation for online sequential samples of current fault detection interest, could be much more appropriate than extracting latent representations for only normal samples. Therefore, a novel feature extraction algorithm titled as time-serial maximal deviation analysis (TSMDA) is proposed. The proposal of TSMDA is orientated to derive a projecting scheme with dynamically determined projecting vectors in a timely manner, so that possible time-serial deviation in newly measured sequential samples could be adaptively uncovered to maximal extent. Most importantly, TSMDA cannot be executed without the involvement of online monitored samples, and it is expected to guarantee consistently enhanced fault detectability. The effectiveness of the proposed fault detection method is validated through practical experiments on the two industrial processes, and the influence of the key model parameter on the fault detectability is also evaluated.