Preprocessing of the Observed Data and the Recognition of the Hidden Signs of Fault

Abstract

This paper describes methods and algorithms for preprocessing of the observed data to fault prognostics. The main focus is on identifying the hidden signs of fault (nonamplitude failure predictor) and the method, defined their evolution. The signal from the vibration sensor, installed on the engine case, is taken as the initial observed signal. Further analysis is based on the representation of the signal in the form of coefficients of its wavelet decomposition. Each coefficient of wavelet decomposition is represented as a discrete sequence of data. Further pre-processing signal is a representation of the signal in the form of finite segments of wavelet coefficients. Thus, the transition is made to the high-dimensional vector processes with discrete time. The last stage of preprocessing is the decomposition of vector processes on "amplitude" and nonamplitude "phase" components (K-decompositions). If the physics of process is known, methods of a nonlinear stochastic filtration are used for allocation of the hidden signs. In the opposite case, particularly interesting from a practical point of view, K- decompositions of the wavelet coefficients segments are analysed. Pseudo-dynamics of phase vector component is analysed to detect of non-amplitude failure predictors. Algorithm and software in automatic mode analyses the evolution of the phase component on the basis of continuous or periodic monitoring. Prognosis of evolution of the phase variable and life time estimate is based on the definition of evolution equations, or by monitoring the entropy characteristics of the pseudo-phase component of K- decomposition. Pseudo-phase component is physically interpreted as characteristics of process regularities. The described algorithm is implemented in PHM computing cluster. For full analysis of life time estimates the cloud computing service is used.