A fault diagnosis methodology for an external gear pump with the use of Machine Learning classification algorithms: Support Vector Machine and Multilayer Perceptron

This paper presents a fault diagnosis Machine Learning (ML) computational strategy for an external gear pump. The method uses supervised learning of descriptive features. The focus is on two types of ML nonlinear multi-class classification algorithms: Support Vector Machine (SVM) and Multilayer Perceptron (MLP) algorithms. Although significant work has been reported by previous authors, it is still difficult to optimise ab initio the choice of the hyper parameters (ML method dependent) for each specific application. For instance, the type of SVM kernel function or the selection of the MLP activation function and the optimum number of hidden layers (and neurons). As it is well known, reliability of ML algorithms is strongly dependent upon the existence of a sufficiently large quantity of high-quality training data. In our case, and in the absence of experimental data, high-fidelity in-silico data (generated via the software PumpLinx) have been used for the training of the underlying ML metamodel. A variety of working conditions are recreated, ranging from healthy to various kinds of faulty scenarios (i.e., clogging, radial gap variations, viscosity variations). In addition, noise perturbation has been considered in order to increase the sample data available for ML training. This paper explores and compares the use of SVM and MLP algorithms for predictive maintenance. To reduce the high computational cost during the training stage in the MLP algorithm, some predefined network architectures, like 2n neurons per hidden layer, are used to speed up the identification of the precise number of neurons (shown to be useful when the sample data set is sufficiently large). A series of benchmark tests are presented, enabling to conclude that the use of wavelet features and SVM or MLP algorithms can provide the best accuracy for classification.