Reinforcement Learning algorithm for wood log cutting pattern optimization

This thesis focuses on the development and implementation of a Reinforcement Learn- ing algorithm for optimizing log cutting patterns in the wood processing industry. The proposed algorithm is a Deuling Double Deep Q-Network (DDQN) that autonomously learns optimal cutting strategies, maximizing material utilization and minimizing waste. This thesis is done as part of the research and development activities of Microtec Srl, a leading company in scanning and optimization technologies for the wood industry. Conventional systems used nowadays in industrial log cutting rely on predefined cutting patterns and are limited by computational constraints, the proposed algorithm attempts to overcome these limitations by using a Reinforcement Learning Framework. The model training and validation was carried out using real data provided by Microtec Srl, ensuring the robustness and reliability of the proposed solutions. The obtained results shows much Reinfocement Learning is useful for approximating solutions to computationally intensive tasks that are otherwise too complex to be solved using traditional methods.