Intermittent Fault Diagnosis in Sensors using Artificial Neural Network: A Comparative Study

Abstract

Processing algorithms and processing hardware have developed at a very rapid pace in the last three decades. Sensors play a vital role in the condition monitoring of various components of a manufacturing system and to measure different variables that may affect the performance of these systems. Real-time collection of data, processing of the collected data, and decision-making are challenges. Intermittent fault diagnosis is difficult as it occurs randomly and does not follow a pattern. Non-availability of a suitably labeled dataset for testing the algorithms is also a challenge. Keeping these things in mind in the present research fault-injected sensor signals are developed from the publicly available Intel Lab dataset for temperature and light sensor signals. With the help of the proposed algorithm, a Matlab code for intermittent fault injection is developed. The performances of the various machine-learning algorithms extensively used in literature are also compared with the help of accuracy, precision, recall, and F1 score values. The performances of the artificial neural networks are compared with SVM, Ensemble, and k-NN in classifying various intermittent fault modes using the classification learner application of Matlab. The trained bilayer neural network has achieved an F1 score of 0.89 which is the highest among the other tested machine learning algorithms.