A Proposal of Robust Fault Diagnosis System in Presence of Missing Data and Noise in Mechanical Systems

Abstract

Currently, the modern industry requires to achieve high economic returns with a continuous increase in the quality of the final products, to have high levels of industrial safety and to minimize possible effects on the environment. For accomplishing these requirements, it is necessary a fast detection and identification of faults that occur in the industrial systems. The evolution of the Internet of things and technological advances in automation devices, industrial networks and wireless communications, among other elements, have allowed a significant growth in the number of tools to be used for the treatment and management of the information obtained from the industrial processes by the supervision, control and data acquisition systems (SCADA). However, the performance of these tools, and especially of the fault diagnosis systems, are affected by two specific problems: the presence of noise and missing information on the measured variables. In this paper, a novel methodology for fault diagnosis in mechanical industrial systems is proposed by using computational intelligence tools. The proposal presents a robust behavior in presence of missing data and noise in the measurements by achieving high levels of performance. The proposed methodology is applied to the DAMADICS actuator FDI benchmark which is an electro-pneumatic valve widely used in modern industrial systems. The satisfactory results obtained demonstrate the effectiveness and validity of the proposal.