Use of Artificial Intelligence for Predictive Maintenance and Management of Addis Ababa Light Rail Transit

Condition based monitoring is gaining much importance in the industry because of the need to increase machine reliability and reduce the potential loss of production due to breakdowns caused by different defects. In this thesis, we are interested in a condition based monitoring techniques using artificial neural network approach, especially Multiple Layer Perceptron (MLP). The multiple layer perceptron networks trained with backpropagation algorithm are very frequently used to solve a classification problems. In order to keep the machine performing at its best, one of the principal tools for the diagnosis of signaling equipment problems is the acoustic analysis and also vibration analysis which can be used to extract the fault features and then identify the fault patterns. In addition, there is a demand for techniques that can make decision on the running health of the machine automatically and reliably. Artificial intelligent techniques have been successfully applied to automated detection and diagnosis of railway signaling equipment conditions. They largely increase the reliability of fault detection and diagnosis systems. Accordingly, the aim of this paper is to apply a MLP to classify a large number of faulty signals acquired from turn out in different states: crack signal and fatigue signal. The extracted parameters is the peak ratio, one of the best indicators. The main impact of this neural network is to generate answers that give the combined state of crack and fatigue simultaneously whereas most of previous neural networks have focalized mainly on gears or on bearings alone. Information about the signaling equipment obtained in the form of time signal indicators is converted into a frequency signal indicator using an algorithm designed and coded using MATLAB. The frequency data obtained using the algorithm then is used as an input for continuous learning by an artificial neural network. Based on this learning outcome, the state of signaling equipment can easily be defined and classified. We chose the renowned Multi-layer perceptron (MLP) an artificial neural network for the classification phase. From simulation, we obtained a learning rate of 98% showing our algorithm and equipment state classification as per the signal generated during operation is acceptable.