Using Noise Dosimeter Data and Machine Learning For Working Operation Detection

Since data collection on site is very time-consuming and costly, the aim of the study was to determine how accurately parts of the working time and working operations of chainsaw felling can be determined using noise dosimeter data and machine learning. For the study, a noise dosimeter and a video camera were used to record the felling of 53 spruce trees with a DBH of 8 to 72 cm. The time study and noise data were divided into a training set (27 trees) and a test set (26 trees) and analysed in the software Orange using the three machine learning methods Random Forest, Neural Network and Decision Tree. The results showed that the Random Forest method was most successful in classifying parts of the working time and operations. The effective and operating time of the chainsaw were determined with 96.0 % and 99.4 % accuracy, the main and auxiliary productive time with 92.0 % and 65.7 % accuracy and the sub-phases of felling and timber production with 84.1 % and 94.9 % accuracy. The highest accuracy in determining the work operations was observed in the case of notch- cutting and delimbing (80.3% and 89.5%, respectively), while the lowest accuracy was found for stump debarking, butt trimming and crosscutting (7.6% to 14.0%). The results show great potential for predicting time study data using machine learning.