Operational cycle detection in underground haul trucks via data-driven hydraulic pressure reconstruction based on on-board SCADA data

Operational cycle detection in underground haul trucks relies on the hydraulic oil pressure (HYDOILP) pulse, yet HYDOILP sensors are often unavailable. We propose a soft-sensing framework that reconstructs HYDOILP from seven CAN bus signals to detect unloading events without direct pressure measurement. A hybrid Conv1D+LSTM network achieves R 2 = 0.779 and RMSE = 0.698 MPa on a chronologically separated test set, preserving peak prominence for event extraction. Peak-based detection yields F1 = 1.00 across 14 test cycles; adaptive boundary estimation is the next step. The framework degrades gracefully under noise, engine and vehicle speed are the most critical inputs, and multi-machine validation is required before deployment.