AUTOMATED HEALTH STATE ASSESSMENT OF ROLLING ELEMENT BEARINGS

Abstract

The aim of vibration assessment and prognostics is to extend the engineering life cycle while reducing operating and sustaining costs. Therefore, prognosis is regarded as a key process for the future. Still, accurate estimates of the equipment's real state condition and fault characters allow a further plan of measures that improves the performance, reduce downtime, guarantee the completion of missions and make production effective. As a key component, rolling element bearings are taken as the part of consideration in this research. Recent progress has demonstrated the use of data-driven approaches for fault prognostics of rotatory elements in machinery. Automated models that learn the behaviour of the system directly from the vibration data, also able to forecast the potential progress of failure and categorize the conditions over which the element passes. Approximating and conventional interpolation over the indicators of essential machinery behaviour, however, is a bad practice which can lead to poor predictions. There are some points to ponder over data driven approaches for condition evaluation: (1) How to efficiently process raw vibration data to obtain appropriate characteristics that clearly reflect development in deprivation? (2) How to differentiate deprivation conditions and describe catastrophe criteria? (3) How to be sure that learned models will be robust enough to show steady performance over uncertain inputs that deviate from learned experiences, and to be reliable enough to encounter unknown data (i.e., operating conditions, engineering variations, etc.)? (4) How to diagnose the faults? Such issues constitute the problems addressed in this thesis and have led to develop a novel approach beyond conventional methods of data-driven prognostics. The main contributions are as follows. A novel approach of bearing condition assessment utilizing Local mean decomposition (LMD) and spectral clustering (SC) is proposed. To enrich the efficiency of SC, a novel decision criterion, PI (Parting Index) aiming the optimal number of clusters based on a new similarity indicator and disorder indicator coined as Synergic Association Index (SAI) and Separation Index (SI) respectively, has been proposed. The work aims to achieve better performance degradation assessment (PDA). In addition to this, a novel self-adaptive signal decomposition technique: Concealed component decomposition (CCD) is proposed. The proposed CCD technique is utilized to develop a precise bearing fault diagnosis model and a practical mode selection criterion. The proposed model has been validated over different simulated and experimental datasets of varying fault types. To make the assessment model more effective, a novel texture-based image processing technique for feature extraction is proposed and its proficiencies are utilized to develop a health assessment model. Various grayscale images are i prepared and normalized with a proposed synergic affinity index (SAI) that brings the temporal intrinsic evidence presents in different vibration modes. Finally, dimensionality, a decent term utilized for the dimension related issues of the feature vectors for condition assessment of bearings is addressed. Two models using linear and nonlinear approach is proposed to reduce the effect of dimensionality. All the proposed approaches are validated over experimental and simulated data sets. The proposed methods are also found superior over existing methods on various grounds.