AI-based assessment of log quality

Wood quality is a critical factor for value creation, resource efficiency, and revenue in the forestry and sawmill industries. It plays a decisive role in determining how logs are priced, processed, and utilized. In many small and medium-sized sawmills, however, wood quality assessment and subsequent sorting are still predominantly carried out manually through visual inspection. This process is time-consuming, highly labor-intensive, and is increasingly becoming a bottleneck in light of the growing shortage of skilled workers. While high-precision systems such as X-ray or CT scanners provide detailed information about the internal structure of the log, they are often too expensive, space-intensive, and complex to integrate for this target group.

Against this backdrop, the project “DatMuSSS – Data-Driven Multimodal Sorting of Logs in Sawmills” is developing an AI-based solution for the automated assessment of roundwood quality, specifically tailored to the needs of small and medium-sized sawmills. The aim is to reliably assign logs to the sorting classes relevant for operations and to objectively support wood sorting—with significantly lower system complexity and investment costs than established high-end solutions.

The focus is on developing a cost-effective retrofit solution, i.e., a system that can be retrofitted to existing facilities. This solution is intended to be integrable into existing sawmill lines without fundamental modifications and, in the long term, to represent an economically attractive alternative to CT or X-ray systems.

 

Comprehensive Assessment of Wood Quality

For the sorting of roundwood, characteristics such as knots, internal rot, and cavities are of central importance. While external characteristics can be visually assessed, internal defects often remain hidden during a purely surface analysis. DatMuSSS addresses this limitation by specifically supplementing the evaluation of wood quality with acoustic information, thereby going beyond traditional visual inspection. This creates a solid foundation for more precise and consistent wood sorting.

The system combines multiple sensor sources: Cameras capture end views and top views of the logs and provide information on externally visible quality characteristics. Additionally, each log on the conveyor belt is subjected to an impulse. The resulting airborne and structure-borne sound signals are recorded and evaluated using data-driven methods to incorporate quality-relevant internal structural characteristics such as voids or internal rot into the classification. Furthermore, existing laser scanner data on the shape and dimensions of the logs is utilized and incorporated into the assessment of wood quality.

 

AI-supported sorting during ongoing operations

The evaluation of the sensor data is entirely AI-based. Manual sorting by experienced sawmill employees serves as a reference for labeling the data and forms the basis for training the models. The goal is to directly classify wood quality into the relevant sorting classes to objectively support wood sorting during ongoing operations without having to interpret individual physical parameters. The analysis is designed to be compatible with standard cycle times in real-world sawmill operations and does not interrupt the ongoing process. 

The goal is a misclassification rate of less than one percent, comparable to manual sorting by qualified personnel.

 

Responsibilities of Fraunhofer IDMT

Fraunhofer IDMT develops the acoustic-based sensor technology and AI methods for the automated assessment of roundwood quality. To this end, the institute designs the acoustic measurement system—including excitation, sensor technology, and noise suppression—creates multimodal datasets from acoustic, image, and laser data, and trains AI models for the reliable classification of relevant sorting classes. The developed methods are integrated into a demonstrator and validated under real-world conditions in sawmill operations.