Stacked Sparse Auto Encoder Network Based Multimode Process Monitoring

Abstract

It is difficult for conventional multivariate statistical process control (MSPC) methods to choose the optimal control limits in multimode online monitoring. This may not only ignore some potential relationships within raw mode data, but also cover the impact of incipient faults. Thus a stacked sparse auto encoder (SSAE) based multimode monitoring method is proposed in this paper to pursuit deeper architecture information and essential characteristics of raw measurements. It is put forward by incorporating real-time mode identification with fault discrimination online within a global model. For incipient faults that cannot be detected by conventional statistical methods, the proposed framework shows a remarkable high efficiency. Experiments on Tennessee Eastman Process (TEP) have shown its efficiency for not only multimode identification, but also fault detection.