The operational problem
Mining LHD engines operate under changing load, restricted ventilation and stop-start duty. These conditions make thermal stress, fuel efficiency and loss of performance important maintenance signals, but many relevant quantities are not measured by the standard ECU. The paper presents a hybrid model that estimates a wider group of engine KPIs from three accessible inputs: engine speed, load and fuel consumption.
Model architecture
Data-driven mappings first infer the intermediate variables needed by calibrated physical sub-models. Those sub-models represent gas exchange, combustion, power production and turbocharger behaviour. The architecture includes a non-adiabatic treatment of turbine exhaust temperature and uses turbocharger geometry in place of proprietary performance maps, allowing the model to operate over multiple rotational speeds.
This combination preserves physical interpretation while avoiding a sensor requirement for every target KPI. It is particularly suited to a digital twin that must work with data already available on a machine.
Experimental and calibration approach
Controlled engine tests combined CANedge2 ECU logging with QuantumX pressure and temperature measurements. The measurements supplied calibration and validation evidence across different operating points, including conditions selected to represent the variable duty of underground equipment.
The model’s thermal predictions achieved a mean absolute percentage error of 6.21%. Pressure quantities achieved 5.08%, while power and brake-specific fuel consumption were predicted to approximately 4.1%. Across the complete KPI set, the average error was 5.36%.
Engineering value
The model makes hidden engine state more visible without demanding a permanently instrumented test cell. A mine could use the estimates to establish expected behaviour at a given speed and load, detect persistent departures and focus maintenance attention on systems responsible for temperature, pressure or efficiency changes.
The brake-specific fuel-consumption relationship was centred near the engine’s maximum-torque region. That observation is directly relevant to LHD duty, where usable torque matters more than operating at maximum rated speed.
Limits and next steps
The paper recommends further work on the assumed combustion-cycle split, intercooler parameters, transient coolant behaviour and validation at the edges of the operating envelope. Higher sampling rates and robust handling of missing intermediate data would improve diagnostic use. The possibility of multiple parameter combinations producing similar outputs also needs to be managed when the model is recalibrated for other engines.