1537 Trucks (haulers) and excavators (loaders) are the main entity in overburden removal for open pit coal mine. The excavator digs the broken material of overburden, which previously has been drilled and blasted, at the front area. Each excavator is supplied with several trucks which constantly travel to front and ready to receive the overburden material to be filled in its bucket. Once the truck’s bucket is full, the truck will haul the material to designed disposal area. Pair of a loader with several trucks in a same objective to load and haul overburden from front to disposal area is known as fleet. Operational inefficiency particularly suboptimal speed selection of the trucks can contribute to increased cycle times, higher fuel consumption, brake and tire wear, vehicle overheating, and production delays. Current systems often use fixed speed limits or rule-based recommendations without adapting to real-time vehicle conditions (payload, grade, weather, operator behavior). As mining environments change daily, static models fail to deliver optimal guidance. Adaptive Machine Learning (Adaptive ML) integrates new data in near-real-time, retraining predictive models to reflect the latest road conditions and operator performance. This enables dynamic operator coaching that improves productivity and safety. Truck operators require optimal speed guidelines tailored to road grade and curvature, payload, engine RPM and power capability, rolling resistance, weather and road deterioration, operator behavior which the traditional guidance systems cannot adapt to this complexity. This research aims to design and evaluate a system that: 1. Segments the haul road automatically using GPS. 2. Computes correlations between speed, grade, acceleration, payload, curvature, fuel rate, and other factors. 3. Builds a physics-informed, adaptive machine learning model that recommends optimal speeds and predicts cycle times. 4. Produces real-time operator coaching messages via a rule engine. 5. Allows continuous retraining using recent high-quality operational data. The existing studies focus on stochastic cycle-time modeling, terrain classification, fuel consumption optimization, dynamic dispatching. However, few integrate with GPS-based road segmentation, telematics fusion, incremental/adaptive learning, physics-informed ML, real-time operator feedback. This research addresses that gap by fusing domain knowledge with ML.
RkJQdWJsaXNoZXIy MTM0Mzk2