Automatic Classification of Fragmented Rock using Proprioceptive Sensing

Thumbnail Image
Artan, Unal
Fragmentation , Mining , Classification
Acquiring meaningful information about excavation material characteristics, such as material type, density, cohesion, and size distribution, can help to optimize upstream and downstream operations in applications such as mining, construction, aggregate material handling and preparation, and space exploration and development. Furthermore, information about the material characteristics might also be used to adapt machine controllers used for autonomous excavation. Though some properties, including rock size distribution, can be estimated visually by using exteroceptive sensors (e.g., cameras and laser scanners), issues such as occlusion and poor lighting conditions (e.g., underground) and dust can significantly limit the performance of vision-based methods. The automatic classification of fragmented rock research presented in this thesis focuses on the development and study of approaches that use proprioceptive sensing and wavelet analysis. Proprioceptive sensing is applicable in dark, dusty and harsh environments that limit the use of exteroceptive sensor data. Signals generated from proprioceptive sensors are transient and non-periodic, making it well suited for wavelet analysis as compared to Fourier analysis. Preliminary dumping experiments using small scale equipment were conducted that showed the potential of using custom wavelet features, created using wavelet analysis, and the connection to the size of fragmented rocks. Further experiments conducted using larger equipment that interacts with material through excavation reinforced the connection between size distribution and the wavelet features. An analytical solution was developed and tested connecting the wavelet features to the mean particle size of fragmented rock. The solution requires the use of a reference or ground truth mean particle size that can be generated using any method of size distribution estimation and the ratio of wavelet features to estimate the mean size of an unknown fragmented rock pile. Through extensive experiments performed using small, medium and full sized equipment and numerous fragmented rock piles, the developed method was compared with the ground truth. All experiments were performed with manually operated equipment, which introduced variability in the acquired data. The results show that the proposed methodology has promise as a new technique for providing mean rock size estimates by using only proprioceptive information.
External DOI