Abstract

Underground coal mine workers who operate continuous mining machines rely on many cues to determine tool wear. This skill is difficult to train and proximity to the mining interface is a hazard to the machine operators. To create safer conditions for machine operators, an acoustic classification method for determining tool wear is proposed. To demonstrate this technique, a concrete sample is cut with conical picks of different wear levels using a linear cutting machine and the acoustic data is recorded for classification experiments. The differences in acoustic frequency spectra are highlighted and classification of short segments of the recorded acoustic data, less than 200 milliseconds in duration, is demonstrated using three popular classification techniques: the K-nearest neighbors classifier, the support-vector machine classifier, and the multi-layer perceptron classifier. The performance of these techniques is compared, and the effects of segment size and down sampling are examined. Of the tested methods, the support-vector machine gives good performance with little complexity. This technology could aid operators in performing their roles from a safer distance, alerting them to worn tool conditions in real time.

Keywords:Acoustic classification; Tool wear; Conical picks; Underground mining; Support-vector machine; K-nearest neighbours; Multi-layer perceptron classifier; Fourier transform; Vibration

Abbreviations: FTE: Full Time Equivalent; NIOSH: National Institute of Occupational Safety and Health; kHz: kilohertz; dBA: A-Weighted Decibel; KNN: K-Nearest Neighbours; SVM: Support-Vector Machine; MLP: Multi-Layer Perceptron; RBF: Radial Basis Function