Flexible log cutting in sawmill: an approach based on Value Maps and Derivative-free optimization

The log-to-board cutting process within sawmills represents a significant opportunity for optimization, especially with the advent of advanced measurement technologies such as computed tomography (CT). Despite the availability of these cutting-edge technologies, their full potential remains underutilized within flexible cutting lines. In fact, a very accurate model of the log and its internal defects can be derived from the tomographic data. This model can be used to obtain an estimate of the value of each possible board that can be cut from the log. An accurate estimate can be beneficial to the solution of the problem of defining an optimal cutting pattern. However, the evaluation process is too computationally intensive to be directly applied in optimization procedures for sawmills with flexible saw lines, where we have several degrees of freedom in determining feasible cutting patterns, and a huge number of estimations are required. This study proposes a realistic and innovative solution to fasten the board valorization process, using value maps generated by convolutional neural networks to dramatically reduce the time required to valorize a board. Furthermore, the study introduces a comprehensive methodology for addressing the flexible log optimization problem, leveraging a dynamic programming algorithm integrated with black-box optimization techniques for effective exploration of the cutting pattern solution space.