AUTHOR=Belim Marco , Meireles Tiago , Gonçalves Gil , Pinto Rui 

TITLE=Forecasting models analysis for predictive maintenance

JOURNAL=Frontiers in Manufacturing Technology

VOLUME=Volume 4 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/manufacturing-technology/articles/10.3389/fmtec.2024.1475078

DOI=10.3389/fmtec.2024.1475078

ISSN=2813-0359

ABSTRACT=This study examines the transition to predictive maintenance through real-time data analytics to reduce machine downtime and enhance insights into plant machinery. The research focuses on three primary objectives: sensorizing BA Glass equipment using OPC Router and PowerStudio SCADA for real-time data extraction, developing a predictive maintenance algorithm in Python to predict failures and trigger alarms, and evaluating various prediction models for integration into the algorithm. A comparative analysis of wired and wireless sensors determined that wireless sensors, despite their higher cost, are more practical and interchangeable. The study also compared prediction models such as Linear and Polynomial Regression, Simple and Double Exponential Smoothing, ARIMA, and Prophet models. An algorithm combining blocked cross-validation and rolling window methods was used to assess these models, calculating MSE, RMSE, and MAE for different training sizes and parameters. The findings suggest that for datasets with daily seasonality and gradual oscillations, the Double Exponential Smoothing model with an additive damped trend and linear models are optimal, while ARIMA and Prophet models are better suited for datasets with stable trends and higher frequency oscillations.