Automatic material classification via proprioceptive sensing and wavelet analysis during excavation

Abstract

This paper presents an excavation material classification methodology that uses wavelet analysis and unsupervised learning on acceleration measurements. The technique was validated by using acceleration data that were acquired from three inertial measurement units (IMUs) on an instrumented 1-tonne capacity wheel loader. One IMU was installed on the loader’s boom and two on the bucket. The acceleration signals were logged for 32 manual excavation trials in three excavation materials with different rock size distributions, and the data were processed offline. The continuous wavelet transform was applied to the acceleration signals to extract features from the acceleration signals. The results show that classifying the wavelet feature set using an unsupervised k-means algorithm provides an average material classification accuracy of 81 %, when attempting to simultaneously classify all three materials.