Generating Mathematical Model of Equipment and Its Applications in PHM

Abstract

We developed a method for generating a mathematical model of equipment. The model can be used in many model-based applications of prognostics and health management. The method processes sensor data obtained from target equipment to generate a model that contains sensors, latent variables, and approximate equations. First, latent variables are generated by analyzing correlation coefficients. Next, the method divides the variables (latent variables and sensors) into several groups by applying a hierarchical clustering method. Finally, it generates approximate equations of variables within each group. The generated equations can work as features to help users detect potential failures or estimate remaining useful life. The results of experiments using data obtained from electric generators shows the effectiveness of the features. We also discuss the differences between generating features by using a neural network and the proposed method.