Detection of Mechanical Defects by Neural Networks “Experimental Analysis”
Various methods are implemented to identify the nature of a defect on a rotating machine, by using vibratory measures; they differ in their precision, simplicity of implementation and their sensitivity to errors measurement. The identification of several defects combination is still difficult to implement by conventional signal processing, as the vibration signal that emerges is disturbed, thus making any identification so hard. In this study, we proposed a method based on the neural networks to identify one defect or several combinations of mechanical defects. Thus we propose the neuronal method: the Radial Basis Function (RBF). We highlight their capacity to detect the defect and their sensibility with regard to a signal noise characterizing the other independent sources to the defects. This evaluation will be done with measurement will be carried out on a housing bearing and test bench made up of a toothed gearing on two floors, and without lubrication. Some provoked defects will be analyzed in this study.
B. Bélaid and N. Hamzaoui, "Detection of Mechanical Defects by Neural Networks “Experimental Analysis”", Applied Mechanics and Materials, Vol. 232, pp. 674-681, 2012