A comprehensive OBD data analysis framework: Identification and factor analysis of high-emission heavy-duty vehicles

On-Board Diagnostic (OBD) systems enable real-time monitoring of NOx emissions from heavy-duty diesel vehicles (HDDVs). However, few studies have focused on the root cause analysis of these emissions using OBD data. To address this gap, this study proposes an integrated analysis framework for HDDV NOx emissions that combines data processing, high-emission vehicle identification, and emission cause analysis. The framework employs a fuel-based window method to identify high-emission vehicles, while binning and machine learning techniques trace the causes of NOx emissions. A case study is conducted using data from 32 vehicles sourced from Tianjin On-Board Diagnostic Platform. Of these, five vehicles were identified as high emitters. A machine learning model was trained for each vehicle, with a detailed analysis conducted on three of them. The analysis involves a preliminary investigation of vehicle emissions status, followed by bin analysis to initially identify the causes of emissions. Finally, machine learning analysis is conducted, including the generation of individual conditional expectation (ICE) plots and multivariable partial dependence plots (PDPs), serving as a supplement to bin analysis when it cannot effectively pinpoint the causes of high emissions. This approach effectively uncovers the underlying factors within OBD big data. Using the analysis framework, we discover the identified causes of high NOx emissions were uneven heating of the Selective Catalytic Reduction (SCR) system and prolonged idling and high-power operation, catalyst degradation at 200-250 °C, and SCR system failure before 425 °C. The proposed framework offers a clear approach for identifying the causes of NOx emissions, aiding policymakers in implementing effective NOx control strategies for HDDVs.