The results of TCRP Project E-2, Artificial Intelligence for Transit Railcar Diagnostics, were published as TCRP Report No. 1 in 1994. The objectives of this project were to assess the potential application of artificial intelligence (AI) techniques in diagnostic practices in the railcar maintenance environment and, where appropriate, to recommend steps to introduce such practices. The researchers (1) identified AI techniques that are applicable to the diagnosis or prediction of railcar failures; (2) identified the AI techniques with high probabilities of success; (3) estimated the magnitude of potential benefits from using these techniques; (4) identified in order of priority the railcar subsystems (e.g., propulsion, brakes, doors) that benefit most from application of each of these techniques; and (5) developed a research program for systematically evaluating and implementing these techniques.

The final report for this project recommended that a follow-up demonstration of AI technology be considered. The report recommended that a demonstration of the technology focusing on the railcar propulsion system (including traction motors, gearboxes, power switchgear, and control logic units) be performed to evaluate its benefit to the diagnostic process. The report concluded that the use of AI technology in diagnosing railcar propulsion system problems could pay for itself in 1 year if it could provide a 7.2 percent reduction in the propulsion system mean time to repair.

As a result of this recommendation, funding was obtained to undertake a demonstration of this AI technology. The demonstration used a combination of model-based reasoning and expert systems AI techniques to develop a "mechanics' assistant" tool that can be used to diagnose problems with the railcar propulsion system. The Washington Metropolitan Area Transit Authority was selected as the site for the demonstration.

The demonstration project included the following elements: (1) selection of a demonstration site based on appropriate selection criteria; (2) specification and procurement of AI software shell and workstation hardware; (3) establishment of a diagnostic baseline for evaluation of the AI demonstration results; (4) development of functional and causal models of the propulsion system and generation of diagnostic rules for input into the AI software; (5) pre-test of the AI system, with debugging as necessary; (6) test of the AI system in actual field operation; (7) evaluation of the AI system diagnostic capability; and (8) preparation of a report documenting the demonstration effort and results.

Status: The final report for the project has been published as TCRP Report 44, "Demonstration of Artificial Intelligence Technology for Transit Railcar Diagnostics," and is available in portable document format (PDF). Double-click on the files below to access the report. (A free copy of the Adobe Acrobat Reader is available at https://www.adobe.com.) PLEASE NOTE: Because of the very large size of these files, it may take a long time--possibly more than 1 hour collectively--to download. We regret the inconvenience.