Enabling Opportunistic Maintenance in the Sawmill Industry: LSTM-Based Prediction of Maintenance Opportunity Windows Using Discrete Event Simulation

Abstract

The sawmill industry has traditionally fell behind other manufacturing sectors in the adoption of modern maintenance and asset management practices, particularly regarding predictive and opportunistic maintenance strategies. This challenge is compounded by the limited amount of scientific literature and industrial case studies focusing specifically on sawmill environments. At the same time, this gap presents an opportunity to improve productivity and create competitive advantage through the implementation of advanced maintenance approaches. This study is conducted in collaboration with Södra Skogsägarna and investigates the feasibility of introducing opportunistic maintenance within the Swedish sawmill industry through the prediction of Maintenance Opportunity Windows (MOWs). The project builds upon recent research on MOW prediction and aims to adapt these concepts to the sawmill production context using Long Short-Term Memory (LSTM) networks. Since sufficient production data and data collection infrastructure are currently unavailable, a Discrete-Event Simulation model of a generic sawmill will be developed to generate synthetic production data. The simulation model will support an iterative process for identifying relevant input variables and defining future data collection requirements for real-world implementation. In addition to enabling the development of the prediction model, the simulation environment will also be used to evaluate the impact of different production policies and external factors on system performance. The study aims to contribute both to the academic understanding of opportunistic maintenance in sawmill systems and to the practical implementation of data-driven maintenance strategies in industry.