PROGNOSTICS OF ROLLING ELEMENT BEARINGS USING SOFT COMPUTING TECHNIQUES

Abstract

Prognostics aims at extending the life cycle of engineering assets with reducing exploitation and maintenance costs. For this reason, prognostics is considered as a key process with future capabilities. Indeed, accurate estimation of the Remaining Useful Life (RUL) of bearings enable defining a further plan of actions to increase safety and minimize downtime of rotary machines. Effective prognostics are essential aspects for maintenance engineers to schedule a repair and to acquire replacement of bearing before it fails. Although, variety of prognostics methodologies have been reported, but their applications in industry are still relatively new and mostly focused on the prediction of specific degradations. Only in few studies, the damage detection results are extended to the estimation of the RUL and these are achieved by modeling the underlying degradation process using a surrogate measure of degradation. Furthermore, the parameters of degradation model are utilized for prognosis and these are estimated using the degradation data. Besides this, significant and sufficient number of fault indicators are required to accurately prognoses the components faults. Hence, an adequate usage of health indicators in prognostics for the effective interpretation of bearing degradation process is still required. Recent advances show that data-driven approaches (mainly based on machine learning methods) are increasingly applied for fault prognostics. These can be seen as black-box models that learn system behavior directly from Condition Monitoring (CM) data, and use that knowledge to infer its current state and predict future progression of failure. Such issues constitute the problems are addressed in this study and have led to develop a novel approach beyond conventional methods of data-driven prognostics. Soft computing techniques are developed and presented for rolling element bearings used for the assessment of performance degradation and estimation of the remaining useful life (RUL) of bearings. The primary objectives are using to measure the degradation, which is developed to undertake the damage detection and prognosis in a single framework. The methodology assumes that the degradation consists of a series of degraded states (health states), which are effectively represent the dynamic and stochastic process of bearing failure. The main contributions are as follows:  The data-processing step is improved by introducing a new approach for features extraction using signal processing techniques, where features selection is based on three characteristics, i.e., monotonicity, trend ability and predictability. The main idea of this development is to transform raw data into features that improve accuracy of long-term predictions. ii  To account these issues, a new prediction algorithm of supervised and unsupervised machine learnings are proposed. An ensemble approach is also proposed to quantify uncertainty and improve accuracy of estimates.  Performance of bearing degradation assessment is enhanced for the proposition of new health indicators and classifier. The degradation index are obtained using self-organizing maps as the dissimilarity values between the normal and faulty datasets. It is representing as an error as shown the evolution of bearing. The higher order cumulants as health indicators from the stationary wavelet decomposition (SWT) and Extra Trees Regression (ETR) are used. The prognosis of bearing degradation is estimated using k-Nearest Neighbours (kNN), Support Vector Machine (SVM) and Decision Tree (DTs). The bearing fault features are extracted from vibration data using an intrinsic mode functions (IMfs). The fault features are applied to the k-medoids clustering for the evaluation of the performance of bearing degradation.  The prognostics model is developed for the estimation of bearings RUL. To overwhelm, a measurement series of Weibull distribution based new health indicator is proposed. Finally, the modified artificial neural networks (ANN) model is a one-step ahead of the usual ANN is performed for the prediction of bearing RUL. In another approach, the combination with sum of limiting value of control chart by mahalanobis distance is used to develop the health degradation indicator (HDI). Simplified Fuzzy Adaptive Resonance Theory Map (SFAM) with the proposed health indicator is taken for the prediction of bearing RUL. The proposed prognostic methodology is successfully tested and validated with an experimentation. It is conducted in the vibration laboratory to real industry application. The results of experiments indicate that accurate estimation of health states is achievable and the proposed method provided an accurate long-term prediction of bearing remaining useful life. Also, the results of experimental tests showed that the proposed model has the capability of providing early warning of abnormal bearing operating conditions by identifying the transitional states of bearings fault conditions. The results are very encouraging and showed that the proposed prognostic model has the potential to be used as a generic and scalable asset health estimation tools in industrial machinery.