Optimal Strategy in Predicting Equipment Sensor Failure Using Genetic Programming and Histogram of Residual

Abstract

Detecting sensor abnormality is challenging because the data are normally acquired using IoT approach and stored offline in a dedicated server (data logs). The objectives of this research are to device an approach to detect sensor abnormality and perform this in a ”white box” approach. In the proposed approach, the compressor sensor output is modelled as a function of other sensors using static approach, comparing regression results of Genetic Programming (GP) with Multiple Linear Regression (MLR) and Neural Network Regression (ANN). Subsequently, the model output is used for detecting abnormality by observing the residuals.