A novel fault detection and diagnosis approach based on orthogonal autoencoders

• An unsupervised fault detection and diagnosis scheme based on orthogonal autoencoders is proposed for the monitoring of industrial chemical processes. The use integrated gradients allows exploration bottleneck autoencoder, providing candidate variables root cause analysis. method performs well compared to traditional methods in offering compelling interpretable results. analysis shows how difference way faults are introduced affects performances principal component analysis-based control charts. In recent years, there have been studies focusing different types (AEs) complex nonlinear data coming from However, many cases focus was placed detection. As a result, practitioners encountering problems trying interpret such models obtaining once an alarm raised. This paper proposes novel statistical process (SPC) framework (OAEs). OAEs regularize loss function ensure no correlation among features latent variables. extremely beneficial SPC tasks, as it invertibility covariance matrix when computing Hotelling T 2 statistic, significantly improving performance highly correlated. To support identification analysis, we propose adaptation (IG) method. Numerical simulations benchmark Tennessee Eastman Process used evaluate approach by comparing approaches (PCA) kernel PCA (KPCA). explore information useful stored intermediate layers encoder network. We also investigate structure faulty results show combination IG represents ready-to-use solution, improved over methods.