Health assessment and fault classification for hydraulic pump based on LR and softmax regression

The real-time health monitoring and fault diagnosis system of a hydraulic pump is important for the role of the pump as the power source of the entire hydraulic system. Thus, this study proposes health assessment based on logistic regression (LR) and fault classification based on softmax regression. The real-time state of the system is obtained by processing the data of vibration signals collected from the pumps, and maintenance can be performed as long as the failure or malfunction occurs. The vibration signal is decomposed into several product functions by local mean decomposition, and the product functions that contain fault information form a feature vector by extracting energy values and corresponding time-domain statistical magnitudes. Multidimensional scaling is used for feature reduction. The LR model and softmax regression model are trained by the reduced features to obtain health conditions and classify possible fault modes, respectively. The maximum likelihood method is applied to determine the parameters of the LR model, and the gradient descent method is used to determine the parameters of the softmax regression model. This method has been applied to process the vibration signals of a real hydraulic pump to verify its effectiveness and feasibility.