An enhanced kernel learning data-driven method for multiple fault detection and identification in industrial systems

Abstract

Traditional fault detection methods focus mainly on a single abnormal condition of the system. However, successive multiple faults are more common than a single fault in industrial systems. Hence, this paper proposes a novel algorithm for detecting and identifying multiple faults associated with the quality indicators of the process. Considering the dynamic feature and measurement noise in the system, an enhanced kernel learning data-driven (EKLDD) algorithm is designed to improve the performance of modeling and multiple fault detection. In addition, a monitoring scheme is proposed to evaluate the quality status under every fault based on the fault line and the angle statistics. Lastly, a simulation case and a real-world case are presented to illustrate the feasibility and effectiveness of the proposed EKLDD method.