Health indicator construction and remaining useful life estimation for mechanical systems using vibration signal prognostics

Health indicator (HI) construction and remaining useful life (RUL) estimation are the most critical factors in prognostic health management of industrial machinery. A computationally less complex and economic prognostic approach could attract industrialists to implement a predictive maintenance strategy for their mechanical systems. In this paper, a prognostic approach combining the principal component analysis (PCA) and statistical exponential degradation model is employed for the HI construction and remaining useful life estimation of mechanical systems. Vibration signals acquired from the machinery, from the initial healthy state to the final faulty state are utilized for the prognostic analysis. Statistical vibration signature features in the time–frequency domain are extracted using discrete wavelet transform. The features well representing the health degradation of the mechanical system are selected by analyzing the monotonicity, trendability, and prognosability factors and the fused employing PCA to construct a HI. Finally, an exponential degradation model is used to fit into HI for RUL estimations. The algorithm is validated using FEMTO bearing vibration data set for accelerated failure tests. It is observed that the constructed HI gives a better representation of the degradation phenomenon and perfectly fits the exponential degradation model for RUL estimation. The error of predicted RUL from the actual experimental value is 21 min. This reliable RUL estimation model can be utilized to develop a predictive maintenance and maintenance decision-making module.