Project: #386 Deterioration Digital Twins of Commercial Trucks and Trailers for Targeted Inspection and Maintenance Progress Report - Reporting Period Ending: March 30, 2023 Principal Investigator: Pingbo Tang Status: Active Start Date: July 1, 2022 End Date: June 30, 2023 Research Type: Applied Grant Type: Research Grant Program: FAST Act - Mobility National (2016 - 2022) Grant Cycle: 2022 Mobility21 UTC Progress Report (Last Updated: March 30, 2023, 9 p.m.) % Project Completed to Date: 60 % Grant Award Expended: 60 % Match Expended & Document: 60 USDOT Requirements Accomplishments What are the major goals and objectives of the project? The proposed project aims to enable targeted inspection and maintenance of commercial tractors and trailer fleets by a vehicle deterioration digital twin that integrates historical inspection records and real-time sensor data for predicting high-risk vehicles and components. Such a vehicle deterioration digital twin should support the prioritization of vehicles and vehicle components for inspection and maintenance to balance fleet safety, mobility, and maintenance costs. Specific objectives include: - Generating Process Digital Twins from Inspection Data and Sensor Logs of Telematics System. - Generating Inspection and Maintenance Plans for Truck and Trailer Fleets. - Discussions with State and Federal Stakeholders about the Use of Inspection Reports and Telematics Data for Guiding the Safety Inspection Policy What was accomplished under these goals? Since September 2022, the project team has focused on analyzing multiple vehicle inspection data and has developed a vehicle data analytics and visualization dashboard for analyzing deterioration rates of vehicles based on historical inspection data, identifying the high-risk vehicles of a given commercial vehicle fleet based on historical deterioration rates of various vehicles and components, and providing inspection suggestions that enable the cost-safety balance. - Identified deterioration pattern of heavy-duty trucks and trailers and developed a general Markov deterioration model based on the whole population of heavy-duty vehicles for predicting a certain vehicle’s state in the future. The prediction of vehicles’ future states can contribute to fleet management. - Clustered the heavy-duty trucks and trailers according to the deterioration patterns and identified four groups of heavy-duty vehicles corresponding to four deterioration patterns. - Developed specific Markov deterioration models for different groups of vehicles with different deterioration patterns, which provides a more specific and fittable deterioration model for a certain vehicle. - Developed an inspection strategy that identifies risky vehicles periodically using the Markov deterioration model, which can balance the operation safety and costs. - Examined different inspection strategies’ (e.g., annual inspection for all vehicles, monthly inspection for all vehicles, monthly inspection for high-risk vehicles) performances in balancing operation safety, losses of uptime, and inspection costs. - Developed a data augmentation method to augment the limited historical inspection data and refine the Markov model. The general Markov model with data augmentation and the specific models with data augmentation both have been validated to have good performances in vehicle condition prediction. - Developed a transfer-learning-based state prediction method to predict the vehicle component’s state in the future according to the vehicle’s specific deterioration pattern. The researchers developed a transfer learning algorithm to transfer the information in the general Markov deterioration model to several specific models for different deterioration patterns. Heavy-duty trucks and trailers have different deterioration patterns. For example, some vehicles’ brakes deteriorate faster due to different driving behaviors and vehicle properties. The general deterioration model can hardly accurately reflect the deterioration patterns of all heavy-duty vehicles. In contrast, the specific deterioration models can fit different deterioration patterns and achieve more accurate predictions. Such an accurate prediction can help to generate more safety-cost-balanced inspection decisions that can detect all unsafe vehicles and not waste inspections on safe vehicles. - Quantified the impacts of the critical vehicle attributes (e.g., vehicle characteristics, driving behaviors, driving environments) on the deterioration patterns of various vehicle components and explained the reasonability of clustering results. - PI Tang has reached out to Amazon (October, 2022, through submitting a proposal in response to Amazon Faculty Award program) and Re:Build Manufacturing (March 2023, through a meeting on CMU campus) for collaboration on using telematics for automated performance management of vehicles and batteries of electrical vehicles - Developed an explainable model that creates inspection plans and provides explanations as to why different vehicles have different rates of deterioration, as well as why a specific vehicle has a certain rate of deterioration. This explanation can greatly enhance managers’ trust in the model’s inspection plan. By making the model’s process transparent and clear, it becomes easier for managers to understand and trust its outputs. - Developed an interface that enables fleet managers to create inspection plans for their vehicles. This interface allows managers to upload information about their fleet of vehicles, and then generates suggestions for inspection and maintenance plans based on that information. In addition, the interface provides explanations for the suggested plans, making it easier for managers to understand why certain plans are being suggested. With this information, managers can then make decisions about the final inspection plan. The interface thus provides a valuable tool for managers to obtain inspection plans from the algorithm while also understanding the suggested plans and making final decisions. - Designing and implementing an experiment to evaluate how the explainable decision-making algorithm improves the manager’s trust. The experiment contains three tests for fleet managers. In test 1, the managers will make inspection decisions relying on their experience and a database with historical recordings of vehicle deterioration. In test 2, the managers will make inspection decisions relying on the inspection plans suggested by the algorithm, database, and experience. In test 3, the managers can access the suggested inspection plans and the database and obtain the reasonability under the suggested inspection plan. Then, the experiment quantities the managers’ trust by calculating the percentage of inspection suggestions accepted by the managers. - Developing an algorithm to explore how safety and energy efficiency relate to vehicle violation. Specifically, the algorithm explores crash rates and component violation rates among commercial fleets with different attributes. By examining these factors, the algorithm is able to estimate crash rates for a specific commercial fleet. - Developing an algorithm to generate inspection and maintenance plans for tires in the commercial fleet, which optimize the plans in the safety, productivity, and energy consumption level. What opportunities for training and professional development has the project provided? - Three Ph.D. students (Ruoxin Xiong, Ying Shi, and Ava Jahan Biglari) in the Department of Civil and Environmental Engineering have got the opportunity of learning the practice of commercial vehicle fleet inspection and maintenance planning, accumulating data analysis results for future presentations and research publications. - Three Ph.D. students (Ruoxin Xiong, Ying Shi, and Ava Jahan Biglari) have weekly meetings with industry collaborators to develop their skills of presenting the work to industry professionals and identifying scientific problems from practical problems, they also get feedback from the industry professionals about various data quality issues in the inspection data and have started developing new scientific methods for data imputation and new data analysis methods that enable vehicle data analytics and predictive fleet management based on partial and faulty data. One Ph.D. student (Ying Shi) presented the research work on Deterioration Digital Twins of Bridges and Commercial Trucks and Trailers for Targeted Inspection and Maintenance at a Digital Twin Research Collaboration Meeting with ANSYS in Pittsburgh, PA, on July 15, 2022 - One Ph.D. student (Ying Shi) presented the research work on “Safety-Cost Aware Inspection Strategy for Commercial Vehicle Fleets” at Mobility21 Deployment Partner Consortium Symposium on November 3, 2022 - One Ph.D. student (Ying Shi) presented the research work on “Identifying Safety-Critical Heavy-Duty Vehicles in Fleets with Complementary Vehicle Inspection Datasets Through Cross-Database Clustering Analysis” at TRB 102nd Annual Meeting in Washington, DC on January 10, 2023 How have the results been disseminated? If so, in what way/s? - The most frequent distribution of the results is through bi-weekly meetings with industry professionals for showing them the results of data collection and analysis for supporting predictive commercial vehicle fleet management. - One M.S. student (Jiayi Li) presented at the Summer Research Symposium of the Civil and Environmental Engineering Department. The short presentation is on the truck fleet management dashboard for visualizing truck fleet deterioration rates based on historical inspection reports. - The project team presented use cases of a user interface design to CompuSpections and Truck-Lite (Clarience Technologies) for simplifying the uses of the developed algorithms in guiding the inspectors in selecting critical vehicles and components and prioritizing the maintenance plans of a given fleet. - One Ph.D. student (Ying Shi) presented the research work on “Identifying Safety-Critical Heavy-Duty Vehicles in Fleets with Complementary Vehicle Inspection Datasets Through Cross-Database Clustering Analysis” at TRB 102nd Annual Meeting in Washington, DC on January 10, 2023 - The project team is in the process of developing a journal manuscript for submission around April 2023 What do you plan to do during the next reporting period to accomplish the goals and objectives? - Plan to reach out to industry and government and dig more into the policy perspective. - Develop and test the new dashboard that allows inspectors and fleet managers to view clusters of vehicles for an interactive explanation of the inspection and maintenance plans generated by the algorithms - Communicate with Turnpike for potential collaborations on using the developed data-driven models and dashboard in helping policemen at the entrance of highway 76 in identifying critical trucks or trailers for preventing high-risk vehicles from entering the highway. - Submit the inspection and maintenance strategy generation journal manuscript Impacts During this reporting period, the project team continued the following activities that have some impacts: - The two industry collaborators (Compuspections and Clarience Technologies) provided more truck/tractor inspection data and helped the project team clean and organize their data for supporting integrated analysis of historical inspection reports and real-time data. The two industry collaborators (Compuspections and Clarience Technologies) have their engineers and professionals participate in the development of data analytics techniques and started improving their software and hardware platform based on the findings of the project (e.g., critical components that deserve automatic inspection). The project team has a new user interface design approved by professionals from CompuSpections and Clarience Technologies) because of its simplified design. This new user interface design integrates clustering and machine learning algorithms with interactive data visualization features for an interactive explanation of inspection and maintenance plans generated by the algorithms. The new data augmentation and transfer learning algorithms for deterioration prediction of vehicles have scientific impacts. These new deterioration models and algorithms contribute scientific knowledge about vehicle deterioration models for supporting proactive fleet management and the interfaces for queries and suggestions of inspection and maintenance plans of certain components of certain vehicles with specific properties. - The project team continued investigating more detailed reasonings for the inspection plan generation model to provide more detailed explanations to improve the managers’ trust. Other - Website: TrSafety - Towards Data-Driven and Continuous Safety Inspection of Commercial Trucks and Trailers: https://sites.google.com/andrew.cmu.edu/trsafety/home - Database - A database that integrates the crash data from MCMIS with inspection records from MCMIS and CompuSpections for supporting the analysis of "Which types of vehicles tend to involve into crashes"? - Algorithms Algorithms for 1) predicting vehicle conditions and identifying risky vehicles periodically to ensure the operation safety with the minimizing number of inspections; 2) clustering similar vehicles with similar brake deterioration patterns; 3) augmenting limited historical data for obtaining more reliable deterioration models; 4) explaining the clustering results by quantifying the influences of vehicle characteristics, driving behaviors, and driving environments on deterioration patterns given a certain component; 5) explaining the reasons for the variant deterioration rates of variant vehicles. - Models 1) Deterioration models of commercial trucks and tractors for supporting the simulation of different inspection and maintenance policies for managing commercial vehicle fleets; 2) explainable models for interpreting the reasons for variant deterioration patterns for variant vehicles. - Educational aids or curricula Education and outreach materials for training industrial professionals in the effective use of historical inspection records of commercial vehicles for preventive commercial vehicle fleet inspection and maintenance planning - Software or NetWare Two interfaces for inspectors and fleet managers, respectively, are in development. Outcomes New Partners None Issues None