Development of a Methodology Using Artificial Neural Networks in the Detection and Diagnosis of Fauts for Pneumatic Controle Valves
Control Systems, Industrial Plants, Fault Detection and Diagnosis, Emulation, Digital Signatures
To satisfy an increasingly demanding and complex market, competition in the industrial field estimates for greater productivity and safety in the plant control systems. Therefore, the appearance of failure may compromise the functioning of a plant system and may even lead to a hazardous situation. Therefore, the area of FDD fault detection and diagnosis contributes to avoiding any unwanted event, since several techniques and methods have been the purpose of study for fault detection, isolation, identification and consequently fault diagnostic. In this work, a methodology will be applied through fault emulation aiming to obtain parameters similar to a benchmark reference model for the digital signatures determination step of the faults. This whole process will be carried out incorporating techniques such as clustering, data resizing, waste generation, neural networks, signature acquisition, and the decision tree method, which are necessary to perform the detection steps and, finally, to diagnose abnormalities that are present. To validate the methodology understudy, the DAMADICS benchmark model will be used, which is widely consolidated in the literature mainly due to its set of flaws, which served as the basis for the emulation process in a didactic plant system.