A Data-Driven Fault Detection and Diagnosis by NSGAII-t-SNE and Clustering Methods in the Chemical Process Industry

Abstract

Abstract This study aims to illustrate a novel unsupervised learning method for fault detection and diagnosis of chemical processes. The data-driven fault detection and diagnosis contains two main steps. a) data preparation and feature selections as preprocessing step and b) fault detection and diagnosis for fault indication. In this study, a non-dominated sorting genetic algorithm (NSGAII) was utilized for selecting the most relevant variables from the measured variables for each fault. The t-distributed stochastic neighbor embedding (t- SNE) algorithm was used for information extraction from the selected variables which will lead to visualization of extracted features. Meanwhile, fault detection was performed by k-means and density-based spatial clustering of applications with noise (DBSCAN) clustering algorithms. The Tennessee Eastman benchmark process and faults of process variation were applied to validate the proposed method. Results show that the combination of NSGAII, t-SNE, and clustering methods is an efficient method for Tennessee Eastman process fault detection and diagnosis. And the proposed method could be used in chemical processes for early fault detection.