Increasing Reliability of Fault Detection Systems for Industrial Applications

There is an increasing need of advanced fault detection (FD) systems that can provide early warning of developing abnormalities in industrial systems. Plant measurements are often corrupted by noise, errors, and other disturbances. The equipment in the plant is subject to wear and minor changes that lead to drift in plant measurements. This causes false alarms and misdiagnoses, which reduce confidence in the FD systems. In this work, various steps are proposed to enhance the reliability of fault detection systems for industrial applications. The steps discussed in this paper include data pre-processing for improving data quality, adaptive thresholds for better decision making, and adaptive learning for responding to slowly evolving drift. The proposed algorithm is applied for detection of faults in the coal mill system of thermal power plants. The historic data collected from an actual 500 MW plant is employed for validation. The effectiveness of the algorithm with and without the suggested steps is compared using appropriate performance measures.