A Multivariate Model to Predicting Vibration Features for Equipment Prognosis

Abstract

Vibration analysis, a vital tool in the scheduling of equipment for maintenance is used to assess the useful life of equipment for allocation of resources to mitigate downtime. Compared to previous approaches of univariate prediction, this study presents a more practical model by employing vibration analysis data as a multivariate problem in predicting the remaining useful life (RUL) of an equipment. Applying the model, Multiple Linear Regression (MLR) and Linear Programming (LP) were explored to determine the deterioration rate and the RUL of the equipment. The results showed that the MLR had a high predictive accuracy on the data sets. Furthermore, a p-value of 1.546e-06 and Multiple R-squared value of 0.8215 were obtained showing that the MLR appears to be a good prediction model. From the solution of the LP formulation, the RUL of the equipment was 181 days. These results closely matched the historical data of the equipment which implied this model could be used for planning of maintenance activity for this equipment and any similar equipment.