Relating track geometry to vehicle performance using neural network approach

Abstract

As part of a research that led to the development of the performance-based track geometry (PBTG) inspection technology, the Transportation Technology Center, Inc., Pueblo, CO, USA, a wholly owned subsidiary of the Association of American Railroads, has developed a technique relating track geometry to vehicle performance, real-time. This technique is based on the neural network (NN) approach, an emerging powerful tool in recognizing complex patterns and non-linear relationships between many inputs and an output, such as the relationship between track geometry and vehicle response. On the basis of this technique, many NNs have been developed (trained) from actual vehicle/track interaction test results. For a given vehicle type, the trained NNs directly relate three-dimensional track geometry and vehicle operating speed to vehicle performance. The effects of other track conditions such as lubrication, rail profile, and track stiffness are indirectly considered on the basis of their statistical distributions from test results. Because the PBTG inspection is intended to optimize the track geometry maintenance, the vehicle types selected for testing to date were the ones most sensitive to track geometry (in North America, these are the tank car, covered hopper car, and coal gondola car). However, more NNs for other vehicle types can be easily trained on the basis of the NN technique developed using actual vehicle performance and track geometry test results.