Explainability: Relevance based dynamic deep learning algorithm for fault detection and diagnosis in chemical processes

• Published in: Computers & chemical engineering Vol. 154; p. 107467
• Main Authors: Agarwal, Piyush, Tamer, Melih, Budman, Hector
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2021
• Subjects: Autoencoders; Deep learning; Explainability; Fault detection and diagnosis; Tennessee eastman process; Deep learning; Tennessee eastman process; Autoencoders; Explainability; Fault detection and diagnosis
• ISSN: 0098-1354; 1873-4375
• DOI: 10.1016/j.compchemeng.2021.107467

•An Explainability based methodology is employed for deriving contribution plots for fault diagnosis.•Pruning of irrelevant input variables for the supervised classification task results in enhanced fault detection and fault diagnosis accuracy.•The proposed relevance based method shows superior accuracy compared to other methods. [Display omitted] The focus of this work is on Statistical Process Control (SPC) of a manufacturing process based on available measurements. Two important applications of SPC in industrial settings are fault detection and diagnosis (FDD). In this work, a deep learning (DL) based methodology is proposed for FDD. We investigate the application of an explainability concept (explainable artificial intelligence (XAI)) to enhance the FDD accuracy of a deep neural network model trained with a dataset of relatively small number of samples. The explainability is quantified by a novel relevance measure of input variables that is calculated from a Layerwise Relevance Propagation (LRP) algorithm. It is shown that the relevances can be used to discard redundant input feature vectors/ variables iteratively thus resulting in reduced over-fitting of noisy data, increasing distinguishability between output classes and superior FDD test accuracy. The efficacy of the proposed method is demonstrated on the benchmark Tennessee Eastman Process.