Research Projects

Analyzing Human Learning and Decision-Making in Heavy-Duty Fleet Management to Improve Inspection Practices

Inconsistencies in inspection outcomes, whether by human inspectors or AI systems, pose significant safety risks for heavy-duty vehicles (HDVs). Human inspectors struggle with vast vehicle states and fault modes, while AI systems face challenges due to insufficient or sparse historical data. These challenges contribute to critical safety issues, as brake failures account for 29% of all HDV crashes, and in 2021, 15.6% of HDVs in crashes had brake inspection violations. Human inspectors, with their tacit knowledge, uniquely determine where to focus, when to stop collecting information, and how to use the collected information to make decisions. Understanding human decision-making is key to improving predictive inspections and prioritizing high-risk components. To study how humans learned from inspection trials and errors, the authors surveyed 20 participants to capture human learning processes during five rounds of vehicle inspections, evaluating eight vehicles per round with brake condition ground truths. Then the authors applied an advanced statistical framework for analyzing humans’ feature selection and answer correction behaviors to recover and explain the learning processes. Results showed participants prioritized key features like brake pad thickness (77%), vehicle age (72%), and mileage driven (62%), aligning with expert-identified factors. Participants also reduced their inspection attempts by over 33% from the first to the final round. These findings highlight the value of human heuristics in overcoming data limitations and improving predictive inspections, supporting a collaborative human-AI framework for more reliable HDV assessments. 

Link to Inspection Game