Machine Failure Prediction Using Sensor Data

Abstract

This study proposes a framework to predict machine failures using sensor data and optimize the predictive/corrective maintenance schedule. Using historical data, machine learning (ML) models are trained to predict the failure probabilities for the next five days. Multiple algorithms, including Feature Engineering (FE), feature extraction techniques, selections, and ML models (both regression and classification based) are compared. The machine learning models' output is fed to optimization models to propose an optimized maintenance policy. These models include both a linear optimization model and a Markov Decision Processes (MDP) model. An MDP is employed as a discrete-time stochastic control process for decision making. Therefore, demonstrating how prediction models can help increase system reliability at lower costs.

Date of Award	May 2021
Original language	American English