Implementation of Data-Driven Fault Detection Approach for a Three-Tank System: An Educational Perspective

Abstract

Precise and efficient fault management are essential in today's industrial environment to avoid downtime and monetary losses. Any faulty events could affect factory process flow as well as product quality. Recently, due to the advancement in signal monitoring with cloud data storage, fault detection approach has changed from a model-based into to a data-driven approach. The main goal of this study is to provide an initial insight of the data-driven fault detection approach focusing on undergraduate or postgraduate students, including practitioners that are just new in this area. At the first stage, the data generation and collection approach are described with one process plant example, namely; the three-tank system. Later, detailed principal component analysis (PCA) and threshold determination are explained using normal datasets. After that, the faulty dataset is used to determine the performance of the PCA approach. It is found that the PCA approach is only able to detect six out of eleven faulty conditions, which results in 54.4% of the fault detection performance.