Univariate Sensor Data Prediction Using Conventional and Machine Learning Based Time Series Techniques

Abstract

Availability of data from sensors is becoming easy and in abundance due to the era of industrial revolution 4.0. These data carry rich information about the health condition of the process and equipments in industries along with the current status of the process from which they are acquired. Analysis of this data reveals the interaction and impact among variables involving the control loop. Forecasting and prediction of sensor data is important for the effective functioning of the predictive maintenance stream. Time stamped data can be predicted using time series forecasting techniques. In this paper, the temperature data from a temperature sensor installed to a hydraulic rig is considered for the analysis. The univariate data is predicted for future cycles using times series forecasting techniques. Comparison study between conventional and machine learning algorithms is well defined. These techniques are evaluated using different accuracy metrics like MAE, MSE and RMSE.