| Popis: |
Energy demand from global industries only grows, and is affected by the fact that installed bearings in industrial electric motors are constituted as the primary mode of failure affecting energy consumption. Therefore, the demand for efficient maintenance in electric motors is critical. As a solution, preventive maintenance has typically been employed as a philosophy for asset management that aims to maximize operation through routine inspections with increasing frequency when abnormalities are exhibited, but this leads to an increase in the probability of failure due to the repetitive intervention and the inherent human error. This document presents an integrated diagnostic and prognostic framework for remaining useful life prognosis in bearings, based on the estimation of probability of degradation subject to a defined faults modes and severities induced. The methodological approaches presented incorporate vibration analysis, which actively support the nondestructive and noninvasive fault diagnostics of rotating machines at early stages, but they pose a challenge regarding the signal properties, e.g., highly dynamic and non-stationary. The methodology assumes that bearing degradation consists of a series of discrete degraded states which effectively represent the dynamic and stochastic process of failure process. Prior empirical knowledge is also embedded in the integrated system for classification of detecting faults and severities. In short, the proposed methodology characterizes different failure signatures from bearings using vibration signals and several domains of signal representation with the purpose of deal with stochastic nature and complex relationships in the data concerning failures and severities. For the characteristics selection stage, a study on the fusion and selection of domains and characteristics for signal representation is carried out, in order to discriminate the relevant information. Specifically, here is introduced a fusion and selection schemes based on forward and backward procedures, as well as a stochastic feature selection approach. These techniques are intended to highlight relevant multi-domain features from vibration signals in bearing fault diagnosis and severity evaluation tasks, at the same time that the dimensionality of the data for machine training is reduced. For the training stage, approaches are based on stochastic systems based on the estimation of the probability of a set of discrete states, such as: Hidden Markov Models with discrete observation, Hidden Markov Models with continuous observation and Hierarchical Hidden Markov Models. |
