Enhanced Particle Filter and Cyclic Spectral Coherence based Prognostics of Rolling Element Bearings

Abstract

In the era of Internet of Things and Industry 4.0, the demand for condition monitoring of rotating machinery in industry becomes ever more significant. In addition to the steps of fault detection and diagnosis, the accurate estimation of the Remaining Useful Life (RUL) of machinery components may provide significant economic merits, optimizing the maintenance and avoiding potential human casualties and environmental pollution. During the last decades, a number of methodologies have been developed in the area of Prognostics and Health Management (PHM), categorized mainly in three groups, the physics based, the data-driven and the hybrid approach groups. Physical models are related to the load, the speed, the material, the geometry, etc. of a specific component and are able to make accurate RUL predictions but at an expensive computation cost and high complexity. In order to facilitate the applicability, a number of data-driven methodologies have been proposed including various versions of Kalman and Particle Filters. Among others, the Bayesian inference based Particle Filter provides high prediction accuracy for complex nonlinear systems utilizing little amount of data compared to machine learning techniques. The Monte Carlo step of the method is optimal to deal with the stochastic degradation process of bearings and appears to be able to handle any form of noise distribution. However, the traditional resampling methods frequently present the problem of particle leanness which heavily influences the performance of PF. The existing prediction methodologies are mainly based on classical diagnostic features, e.g. RMS, reaching a limit of efficacy. In order to overcome the abovementioned bottlenecks, an advanced prognostic methodology is proposed based on PF, the “systematic” resampling method and the Cyclic Spectral Coherence (CSCoh). The “systematic” resampling is proposed in order to address the problem of impoverishment. Moreover the CSCoh has been recently proposed as a powerful tool revealing weak modulations masked in the signals. The integration of the CSCoh over the frequency leads to the Enhanced Envelope Spectrum and a diagnostic feature is estimated based on the sum of the amplitudes of three harmonics of the characteristic fault frequencies of rolling element bearings. The methodology is tested and evaluated on experimental vibration signals, while the performance is quantitatively evaluated using prognostic metrics in the consideration of accuracy, precision and convergence.