Pairwise comparison learning based bearing health quantitative modeling and its application in service life prediction

Cognitive computing is expected to meet the challenges posed by the avalanche problem of data being produced by experimental instruments and sensors in academia and industry. How to systematically, purposefully and reasonably interact with human beings and make-decision accordingly is one of the key factors for exerting the potential of cognitive computing and providing services for human beings. As one of the crucial supporting technologies for industrial equipment health management, bearing health analysis has increasingly become an important research field that is promising to improve the reliability and efficiency of modern industrial systems. One of the main challenges in condition-based maintenance and management of bearing is the health quantitative modeling and assessment. Therefore, a learning-based health modeling method, on the basis of newly defined multidimensional frequency-domain health feature, is proposed to realize quantitative assessment of bearing health state. First, a multilayer neural network with a special structure is designed. Then, a novel algorithm, namely PAirwiSe CompArison Learning (PASCAL) is proposed for network parameters learning. In addition, experiments are designed and carried out on a real industrial bearing testing dataset to verify the feasibility and efficiency of the proposed health modeling method. Experimental results are compared with those of two others recent research works, and the performance is measured with a percentage error metric.