A RTMAP-Fuzzy Artificial Neural Network Applied to the Monitoring and Fault Identification in Structural

Abstract. This paper presents an ARTMAP-Fuzzy artificial neural network for monitoring and fault diagnosis in mechanical structures. The goal is to use the ARTMAP-Fuzzy neural network in the identification and characterization of structural failure. This methodology can help professionals in the inspection of mechanical structures, identifying and characterizing faults, in order to perform preventative maintenance and decision-making. In order to validate the methodology we propose the modeling and simulation of signals from a numerical model using an aluminum beam. The results show the efficiency, robustness and accuracy of the methodology adopted. 1. Introduction Increasingly preventive maintenance has been used in industries, businesses, buildings, machinery monitoring, among others. Its use is justified by the need to reduce costs and increase reliability and safety of structures and equipment, preventing disasters from structural failures [5]. Structural failure can occur due to several factors, such as wear of a component, loosening of bolted joints, cracks or even the combination of these elements. Regardless of the origin and intensity, in most cases, structural failure causes an appreciable variation in the spatial parameters of structure, such as reduction in structural rigidity, slight reduction in mass and an increase in damping, which leads to a change of the dynamic behavior structure. Thus the spatial variation of the parameters affects the main dynamic parameters, response functions, resonance frequencies, damping ratio and eigenmodes of the structure [10]. There are some preventive maintenance techniques based on non-destructive testing (NDT) that are applied in oil analysis, magnetic particle, liquid penetrant, methods based on vibration analysis, etc.