Digging for Data: Experiments in Rock Pile Characterization Using Only Proprioceptive Sensing in Excavation

Characterization of fragmented rock piles is a fundamental task in the mining and quarrying industries, where rock is fragmented by blasting, transported using wheel loaders, and then sent for further processing. This field report studies a novel method for estimating the relative particle size of fragmented rock piles from only proprioceptive data collected while digging with a wheel loader. Rather than employ exteroceptive sensors (e.g., cameras or LiDAR sensors) to estimate rock particle sizes, the studied method infers rock fragmentation from an excavator’s inertial response during excavation. This paper expands on research that postulated the use of wavelet analysis to construct a unique feature that is proportional to the level of rock fragmentation. We demonstrate through extensive field experiments that the ratio of wavelet features, constructed from data obtained by excavating in different rock piles with different size distributions, approximates the ratio of the mean particle size of the two rock piles. Full-scale excavation experiments were performed with a battery electric, 18-tonne capacity, load-haul-dump (LHD) machine in representative conditions in an operating quarry. The relative particle size estimates generated with the proposed sensing methodology are compared with those obtained from both a vision-based fragmentation analysis tool and from sieving of sampled materials.

💡 Research Summary

The paper presents a novel method for estimating the relative particle size of fragmented rock piles using only proprioceptive data collected by a wheel loader during excavation. Traditional fragmentation assessment relies on exteroceptive sensors such as cameras, LiDAR, or on labor‑intensive sieve analysis, both of which have limitations: visual methods only capture surface information and can be biased by lighting or occlusion, while sieve analysis is time‑consuming, requires manual sampling, and may not be representative of the whole pile.

The authors propose to exploit the inertial and hydraulic signals that naturally arise when the loader’s bucket collides with rock fragments. These collisions generate vibration, acceleration, and pressure signatures whose frequency and energy content depend on the size and mass of the impacted particles. By applying a continuous wavelet transform to the raw time‑series (denoted g(t)), they compute a “wavelet feature” β defined as the maximum over scales of the convolution between g(t) and a scaled mother wavelet Ψs (Equation 8). The scale s is related to frequency by f = fc · s, where fc is the mother wavelet’s centre frequency. Physically, larger particles produce longer contact times and stronger low‑frequency components, leading to higher β values.

Field experiments were conducted at an operating quarry using an 18‑tonne battery‑electric load‑haul‑dump (LHD). Five distinct rock piles with differing fragmentation levels were excavated multiple times. For each excavation run, the start and end times (α1, α2) were recorded, and both accelerometer data (mounted on the bucket) and hydraulic pressure data were processed independently to extract β. The ratio of β between any two piles was then compared to the ratio of their mean particle sizes (¯x) obtained from a Rosin‑Rammler model fitted to sieve analysis data. The results show a strong linear correlation: when the particle‑size difference exceeds 30 %, the β‑ratio predicts the ¯x‑ratio within 5 % error.

The proposed proprioceptive approach was benchmarked against a commercial vision‑based fragmentation tool (WipFrag™) and conventional sieve analysis. The vision system, which relies on image segmentation and deep‑learning edge detection, provides rapid estimates but only of the surface layer, leading to systematic deviations from the true bulk distribution. Sieve analysis remains the most accurate ground truth but is impractical for real‑time operations. In contrast, the wavelet‑based proprioceptive method delivers continuous, on‑the‑fly estimates without additional hardware, leveraging sensors already present on most modern excavators.

Limitations identified include (1) sensitivity of the signal to extraneous factors such as soil moisture, ambient vibrations, and equipment wear, which were not fully compensated in the current model, and (2) the method currently yields only a relative mean‑size metric rather than a full particle‑size distribution. Future work will explore multi‑sensor fusion (combining vibration, pressure, motor current, and possibly acoustic data) and machine‑learning regression techniques to infer absolute size distributions. The authors also envision integrating the fragmentation estimate into closed‑loop autonomous excavation controllers, enabling real‑time adaptation of digging trajectories, bucket trajectories, and hydraulic settings based on the inferred material hardness and granularity.

Overall, the study demonstrates that proprioceptive sensing, coupled with wavelet analysis, can serve as a practical, low‑cost alternative for rock‑pile characterization in mining and quarrying, opening pathways for more intelligent, energy‑efficient, and autonomous material‑handling operations.