Development of Deep Convoluted Neural Networks (DCNNs) and Change Detection Technology for Improved Railway Track Inspection

Federal Railroad Administration (FRA)-mandated railroad track inspections result in significant labor costs and occupy valuable network capacity. These factors, combined with advancements in the field of machine vision, have encouraged transition from human visual inspections to machine-based alternatives. While commercial machine vision technologies for railway inspection currently exist, many depend on human interpretation of captured information which suffers similar limitations to fully manual inspections. Automated analysis approaches which deliver objective results are also available in the industry. However, they are limited to a “pass/fail” approach through the detection of components which fail to meet maintenance or safety thresholds, as opposed to being able to detect subtle changes in track conditions to identify evolving problems. To overcome the limitations of human interpretation and simple “pass/fail” defect-finding, this paper describes field deployment and validation of a system that pairs three-dimensional (3D) machine vision with automated change detection technology. The change detection approach uses a combination of traditional image processing and a deep convolutional neural network (DCNN) to accurately characterize network conditions between repeat runs before analyzing differences. This paper first provides context for the current automated track inspection technology and discusses the applicability of change detection. Then it discusses the process for 3D image capture and how DCNNs were trained with these 3D images. Finally, it compares the trained DCNNs to an expert human inspector. Results from this study suggest that this technology can successfully identify present, broken, and missing spikes and fasteners with percent accuracies in excess of 98%.

Language

  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References; Tables;
  • Pagination: 15p

Subject/Index Terms

Filing Info

  • Accession Number: 01764399
  • Record Type: Publication
  • Report/Paper Numbers: TRBAM-21-01613
  • Files: TRIS, TRB, ATRI
  • Created Date: Dec 23 2020 11:04AM