Improving the efficiency and effectiveness of railcar safety appliance inspection using machine vision technology

Abstract

Before a train departs a yard, the cars and locomotives undergo inspection, including safety appliance inspection. Safety appliances are handholds, ladders and other objects that serve as the interface between humans and rail cars during transportation. Currently, inspections are carried out by carmen, railroad personnel who are trained in detecting defects in railcars. These inspections are primarily visual and most take place while the inspectors either walk or travel alongside the train in some type of vehicle. Current regulations require that cars be inspected each time a train departs even if they have recently passed previous inspections. A cost model for current safety appliance inspection methods is developed and discussed in this paper. The model considers failure costs, which result from defective safety appliances, and the cost of ensuring defective appliances are caught by inspections, known as improvement costs. Regarding improvement costs, there exists a potential to increase both the effectiveness and efficiency of safety appliance inspections by utilizing machine vision technology to partially automate the car inspection process. Machine vision consists of capturing digital video and using algorithms capable of detecting and analyzing the particular objects or patterns of interest. These systems can objectively inspect railcars without tiring or becoming distracted and can also focus on certain parts of the railcar not easily seen by an inspector on the ground. Benefits of the addition of machine vision to the inspection process are evident in the inspection cost model. Machine vision is being developed for several inspection tasks in the railroad industry and the Association of American Railroads is sponsoring research at the University of Illinois to develop a system for safety appliance inspection. The use of machine vision algorithms makes it possible to recognize the safety appliances on railcars and to identify and report defective appliances. With nearly 1.3 million railroad freight cars in circulation, the development of an algorithm robust enough to detect safety appliance violations on all car types under a variety of environmental conditions is nontrivial. A machine vision system consists of the image acquisition system, algorithms, and the preliminary portable field setup, all of which are discussed in this paper.