This study aims to improve maintenance management in a port company by forecasting machine downtimes, thereby enhancing equipment reliability, reducing corrective maintenance costs, and supporting strategic decisions.
The research implemented and compared forecasting models using real data on equipment downtime from the maintenance sector of a port company located on the coast of Paraná, Brazil. Advanced artificial intelligence (AI) methods – multilayer perceptron (MLP) and long short-term memory (LSTM) neural networks – were applied and compared to the classical ARIMA statistical model. Performance was evaluated based on prediction accuracy.
The results demonstrated that AI-based models outperformed the traditional ARIMA model, showing substantially lower error values. Among the tested approaches, the LSTM neural network achieved the highest predictive accuracy, proving to be the most suitable for forecasting maintenance stoppages in the studied context. The model's outputs also supported a recommendation to increase the size of the company's maintenance team to better manage operational demands.
The findings provide valuable insights into how AI forecasting can enhance maintenance strategies, leading to cost reductions and improved resource allocation in industrial operations.
This work contributes to the field by applying and validating state-of-the-art machine learning models for predictive maintenance in a real industrial setting. It offers a data-driven approach that aligns with Industry 4.0 practices and supports decision-making processes aimed at improving operational efficiency.
