Reliable Sensor Diagnosis for Fuel and Emissions Monitoring in Heavy-Duty Diesel Vehicles

Heavy-duty vehicles remain a major source of transport emissions, and diesel vehicles already in service will continue operating during the transition to lower-carbon transport. Faults in engine sensing can alter fuel use, emissions, and maintenance decisions. Detecting an abnormal reading is therefore not enough. A diagnostic system must determine which measurement is unreliable and avoid acting when the evidence does not support that conclusion. We develop a selective sensor-diagnosis system that detects faults, identifies the affected sensor, estimates the fault type, and reports the complete diagnosis only when it is confident that the diagnosis is correct. A separate experiment evaluates whether a corrupted fuel-flow measurement can be reconstructed from the remaining measurements. We evaluate the diagnostic system on 217 steady-state operating points from a diesel engine using five types of injected sensor faults. When three complete engine-speed levels are excluded from model training and threshold selection, the system issues a diagnosis for 72.5 percent of test batches. Of these diagnoses, 94.6 percent correctly detect the fault, identify the faulty sensor, and identify the fault type across the generated test distribution. In a separate experiment with separately measured fuel mass flow, the system reports a source decision for 63.2 percent of corrupted fuel-flow cases, identifies fuel mass flow correctly in 99.9 percent of reported cases, and removes 83.8 percent of the injected measurement error. By improving the reliability of sensor diagnosis under operating conditions absent from training, this work addresses a practical bottleneck in maintaining trustworthy fuel-use and emissions monitoring for heavy-duty vehicles that will remain in service during the transition to lower-carbon transport.