A survey of contact wire wear parameters and the development of a model to predict wire wear by using the artifical neural network

Abstract

To provide energy for the running of electric trains the overhead conductor system (OCS) is one that is commonly used. The OCS is fixed over the railway tracks with the contact wire above for the collection of electricity by the pantographs which are mounted on the roof of trains. The interaction of the pantograph collector strips and the contact wires is maintained by the pressure applied by the pantographs. This friction inevitably leads to the wear and tear of both the interfacing materials. Replacement of the pantograph collector strips is performed in the train maintenance facility whereas the replacement of the contact wires can be done during the limited non train running periods normally at night or planned system shutdowns. The railway operator has to plan in advance the replacement of the contact wire, before the cross-sectional area is worn below the point where, it infringes on either the electrical or mechanical limits. If the contact wire is worn below its limit then the electrical resistance may result in a voltage drop becoming too high and train motors may stall and burn out. If the contact wire fails mechanically i.e. below the minimum tensile strength, then it may break due to tension on the wire. Without a creditable method of predicting the condition of the contact wire then it would eventually reach the threshold and place the system at risk of a failure. The railway operator needs to be proactive and prudent in the planning and executing a wire replacement program, well before the wire reaches its engineering limits or a failure that can drain resources unnecessarily. With the scarcity in resources, the railway operators needs tools to predict the 'when' contact wire replacement needs to take place prior to its useful life expiring. A statistical approach using historical data is therefore proposed. The success of this statistical approach relies on a large quantity of data collected. In the past decade or two, advances in on-line measurement to collect the primary data on wear and contact force has provided far more information than a railway operator can assimilate. In this research, a review of the analytical and other approaches to predict contact wire wear was conducted. A comprehensive review of the possible contributing factors to the rate of wear for the contact wire and its pantograph counterpart is summarized. This includes as far fetching as the effect of humidity on the rate of wear of pantograph collector strips. Then, a statistical approach together with an artificial neural network (ANN) model is proposed to make the wear projection. Based on the data obtained from the 33-km double tracked Airport Express Line (AEL) taken over a two-year period, an attempt is to establish a feed-forward neural network to correlate the wire wear (as output/target pairs) with the contributing factors (i.e. input parameters corresponding to the output/target). Thus the ANN model, if successfully created, can predict when the wires in segments will need just-in-time replacement. The potential input parameters are wire height, wire stagger, train speed, contact force, traction mode, span length and pantograph passes. Due to the measurement of these parameters with data records taken from different train runs and devices, the data records are combined as input-target pairs to explore the correlation between output and target by the ANN model. A reasonable degree of correlation between wire wear and track topology is found. Following this, an attempt is to check the wire wear for different measurement runs over the two-year period to include the time as an input. However a sanity check of the wear data across datasets found that the average cross-sectional area of contact wires in segments increases with time. The increase in wire cross-sectional area is physically improper and this anomaly is attributed to calibration inconsistency in the measurement equipment. The calibration accuracy and consistency of the measuring equipment between runs needs improvement to predict wear rate over time as proposed research works for future study. The proposed method using ANN model has demonstrated the methodology for wire wear projection and the correlation of selected parameters to the contact wire wear even without the time factor. The methodology establishes the need to use statistical mean and variance of every data in a section as a form of abstraction of wire wear states by segments. This has the advantage over the analysis point-by-point using the raw data and makes it suitable for the training of the feed-forward ANN model and helps in the visualization of contact wire deterioration due to wear as well as the improvement if there is any modification works such as maintenance and renewal activities. In conclusion, this research involves using data obtained from a real railway network to build an ANN model which is able to correlate the wire wear using known operating parameters such as height, stagger, train speed, contact force etc. This prediction method can be refined periodically when new data is collected and integrated as target/input pairs to improve the model every year until the contact wire is finally replaced.