Design of Algorithm for Identification of Locomotive Electrical Machine Unit During Repair

Abstract

The purpose of this article is the design of an algorithm for identification of locomotive electrical machine unit by robotic cleaning facility during repair. The recognition and identification of objects is affected by several random factors and is the probabilistic process. the Johnson criterion was used as the criterion that allows making decision on fulfilment of the identification task with the specified degree of reliability. The diameter of locomotive fuel and oil pump motor fittings was adopted as the critical minimal dimension of the object to be identified. It was calculated that for identification of locomotive electrical machine unit the digital image should have dimensions of 80Ñ…80 pixels. The identification of 80Ñ…80 two-dimensional vector requires larger memory space and longer system learning time. The decrease of input data amount is possible by image additional processing. In order to decrease the unit identification system input data amount the algorithm foreseeing input vector of values generation using summarization of pixel binary codes over lines was developed. The unit identification system was built on the basis of the multilayer perceptron and was modeled in Neural Network Toolbox MATLAB package. The best learning error magnitude result was shown by the network with 40 hidden layers.