| Popis: |
Rolling bearings are widely used in domestic and industrial applications. Most of the industrial machinery often includes this kind of elements. Because of that, it is important to devise predictive maintenance tools in order to detect defects in the bearings and to avoid failure of the machinery. The number of rolling elements and their positions in the load zone change with bearing rotation, producing a periodical variation of the total stiffness of the bearing assembly. Hence, the bearing generates vibrations. When a bearing has a defect, these vibrations are increased. In this paper, a detection and classification system of fault bearings is presented. This system is based on the frequency domain response and the application of a Neuro-Fuzzy method. In the literature, different approaches in the frequency domain have been proposed. In this paper, a particular Neuro-Fuzzy approach has been chosen, given its learning properties and its capability of expressing the resultant detection and classification system by rules. A certain amount of trials have been carried out and it has been concluded that several Neuro-Fuzzy systems in a cascade configuration is a better option to solve the classification problem than other classification methods. In addition to, the results of these initial trials have determined the modifications of the learning phase suggested in this paper. The Neuro-Fuzzy systems with the proposed modifications and only two outputs improve the system performance and the rule association capability. Different results of the proposed approach are shown, where satisfactory results have been achieved.Copyright © 2009 by ASME |
