primary crushing station is a throughput-critical node: haul trucks discharge run-of-mine material into the primary crusher dump pocket (hopper), and any loss of dump pocket availability directly propagates to upstream hauling and downstream processing. A persistent operational challenge is ore accumulation, bridging, and blockage in and around dump pockets, which can force partial or full stoppages to restore material flow. Based on industry estimates, blocked-crusher events in large-scale operations can cause 50–200 hours of unplanned downtime annually, with individual clearing procedures requiring 2–6 hours per incident; at typical primary gyratory capacities on the order of 2,000– 6,000 t/hr [6], this corresponds to approximately 100,000–1,200,000 tonnes of unprocessed ore per year, equivalent to several days of full production foregone. The disruption extends beyond lost crushing capacity: when the primary hopper is unavailable, loaded haul trucks must divert to a run-of-mine stockpile and complete a second full loading-and-hauling cycle before material reaches the crusher. This diversion can also affect ore tracking and blending control, increasing the risk of dilution or grade variability before processing. In one open-pit copper mine case study, hauling represented 40.52% of mining-stage OPEX [16], so this double-handling can impose a substantial incremental cost on the affected tonnes. From an operations perspective, industry handbooks highlight that even a one-percentage-point increase in process availability yields a 4.3% improvement in profit [6]. Current dump pocket cleaning and blocked-crusher clearing procedures are fre- quently performed under constrained access and harsh sensing conditions (dust, poor visibility, irregular ore geometry), and are often only partially mechanized (e.g., breaker booms or excavator-assisted probing). However, authoritative safety guidance emphasizes that blockage clearance should be performed from a position of safety and should not involve anyone entering or being lowered into the crushing area due to the potential for sudden, uncontrolled release of stored energy and falling material [2]. Consistent guidance for crushing operations also stresses eliminating routine presence on crusher access plat- forms and emphasizes prevention and safer clearing practices [3]. Despite these directives, fatal accidents continue to occur when personnel enter hoppers or confined spaces to clear obstructions; Mine Safety and Health Administration (MSHA) reports include cases where a worker entered a hopper to clear blocked material and was engulfed by dumped material [4], and MSHA safety alerts document multiple fatalities from engulfment while clearing obstructions in hoppers/bins/crushers [5]. While partial solutions exist (e.g., rock-breaker booms, probing, or localized “soften- ing”/clearing strategies), they do not provide an integrated pipeline for autonomous detection and removal of heterogeneous ore build-up under mine-realistic variability. Related work. Kim & Choi [15] demonstrated laboratory-scale hopper unblocking using RGB-D vision and a modular detect-then-clear pipeline. ExACT [8] applied Action Chunking with Transformers in a 21.5-ton excavator setting using LiDAR, camera, and joint-position inputs, with validation in a simulator built from real excavator data. Our work differs by: (i) benchmarking four IL architectures under identical conditions rather than a single method, (ii) targeting dump pocket cleaning, a safety-critical confined-space task with documented fatalities, and (iii) providing a low-cost ($250), reproducible testbed enabling systematic IL research in mining contexts where full-scale excavators are costprohibitive
RkJQdWJsaXNoZXIy MTM0Mzk2