Cloud-assisted high-Sulfur fuel monitoring for connected heavy-duty vehicles based on transformer neural network

High-sulfur diesel usage in heavy-duty vehicles (HDVs) worsens harmful emissions, harming public health, and hindering progress in clean energy and low-emission technologies. Toward this end, this study investigates the early-stage impact of high-sulfur diesel on HDV after-treatment systems and introduces HS-FuelFormer, a cloud-assisted framework for real-time diesel quality monitoring using connected HDV sensor data. To reduce reliance on data integrity, it employs a transformer neural network (TNN) with a sliding-window technique for instant diesel type estimation. Multiple instant results are then integrated to enhance the evolving likelihood estimation of high-sulfur diesel usage. A credibility assessment method is also introduced to enhance framework transparency by interpreting the TNN’s decision-making process, fostering trust in the framework by drivers and regulators. A case study with real vehicular data demonstrates HS-FuelFormer’s ability to fill existing gap in online diesel quality monitoring. Experimental results also highlight its effectiveness in early detection, low-frequency operation to reduce cloud load, and reliable performance in unstable communication environments.