APIS - Acoustic process control in the sawmill

Purpose and goal

** Denna text är maskinöversatt ** During the time that good data collection could have been done, the number of errors / stops of the same type has been too few to safely build a reliable AI model that also tells what kind of error it is. With the data we have collected so far, an AI model for analysis of normal / abnormal operation has been developed. The model shows that there is potential to detect when process errors requiring production shutdown soon occur. We can say with better certainty than 60% that a sound sequence indicates normal or abnormal operation.

Expected results and effects

By having an AI system acoustically monitor the sawmill process, it is expected that disturbances and breakdowns can be predicted and avoided. A normal sawmill has an availability of 60-80%. The low availability is due to, among other things, engine failures and saw blades that cut obliquely or go off. With available audio data, a model for normal / abnormal operation has been developed, which shows that there is potential to detect when faults leading to a production shutdown soon occur. We can say with better certainty than 60% that a sound sequence indicates normal or abnormal operation.

Planned approach and implementation

When the installation of microphones was ready everything worked. However, the sound was not good enough. This was because the microphones did not make contact with the storage computer, but occasionally flashed, and that a button on the microphones was incorrectly set. All this took time to detect and fix. Audio data is continuously collected and saved as 1-minute FLAC files. The audio files were correlated with the stop times. The two stop codes we used were stops due to the saw band tearing, which requires replacement of the saw band. The second reason for stopping is band breakage.