Advanced vibration analysis for the diagnosis and prognosis of rotating machinery components within condition-based maintenance programs

Machines used in the industrial field may deteriorate with usage and age. Thus it is important to maintain them so as to avoid failure during actual operation which may be dangerous or even disastrous.The literature has focused its attention on the development of optimal maintenance strategies, such as condition-based maintenance (CBM), in order to improve system reliability, to avoid system failures, and to decrease maintenance costs. CBM aims to detect the early occurrence and seriousness of a fault, to estimate the time interval during which the equipment can still operate before failure, and to identify the components which are deteriorating. CBM has been widely and effectively applied to rotating machines, which usually operate by means of bearings. The reliable and continuous work of bearings is important as the break of one of them can compromise the work of the system. Thus the monitoring, prognosis and diagnosis of bearings represent crucial and important tasks to support real-time maintenance programs. This research has carried out a complete analysis of advanced soft computing techniques ranging from the multi-class classification to one-class classification, and of combination strategies based on classifier fusion and selection. The purpose of this analysis was to design and develop high accurate and high robust methodologies to perform the detection, diagnosis and prognosis of defects on rolling elements bearings. We used vibration signals recorded by four accelerometers on a mechanical device including rolling element bearings: the signals were collected both with all faultless bearings and after substituting one faultless bearing with an artificially damaged one. Four defects and three severity levels were considered. This research has brought to the design and development of new classifiers which have proved to be very accurate and thus to represent a valuable alternative to the traditional classifiers. Besides, the high accuracy and the high robustness to noise, shown by the obtained results, prove the effectiveness of the proposed methodologies, which can be thus profitably used to perform automatic prognosis and diagnosis of rotating machinery components within real-time condition-based maintenance programs.