Independent and Related Variable Fault Detection Based on Information Concentrated Variational Auto-encoder

Abstract With the rapid development of deep learning methods, variational auto-encoder (VAE) has been utilized for nonlinear process monitoring. However, most VAE-based methods hardly consider inner independent and related relationship each variable. To overcome this problem, a novel VAE named variable information concentrated (IRVIC-VAE) is proposed. concentrate information, loading weight matrix regularization based on mutual variables with gaussian distribution introduced so that can separate into two subspaces contain in latent variables. The original data space decomposed via IRVIC-VAE orthogonal approximate to normal distribution. For monitoring, are combined establish statistics according Kullback-Leibler divergence 2-norm. Finally, performance effectiveness testified by Tennessee Eastman (TE) process.