A Novel Condition Assessment Technique For Railway Track Substructure Using Soft Computing

Abstract

The railway infrastructure in the UK is one of the oldest transportation systems in the world. Substructure is a key component of railway track, and similar to other surface transportation systems, track substructures are subjected to ageing and deterioration. Additionally, drainage malfunction in railway track substructure causes local soil weakness that, subsequently, can lead to railway failure. Although there are different destructive and non-destructive tests (NDTs) used in railway substructure condition assessment, there is limited knowledge about how to interpret surface deflection data for the purpose of substructure condition assessment. Limited knowledge about the current conditions of substructure layers in the presence of any local structural weakness can lead to the employment of inefficient and time- and cost-consuming maintenance actions. Therefore, this research proposes the use of a novel back-analysis technique to interpret and estimate the stiffness properties of substructure components using a falling weight deflectometer test data (a well-established and widely used NDT in the UK) and to detect any existing local anomalies in the ground layers. The proposed technique is an integration of an artificial neural network (ANN) with metaheuristic optimisation algorithms. In this regard, where the ANN surrogate forward model is trained based on a database generated by the validated finite element (FE) models. The results indicate that the proposed hybrid technique is a reliable approach to estimating substructure’s layer moduli, as well as identifying a weakness zone’s modulus and its geometrical properties. The corresponding limitations of the proposed technique are then discussed, and further avenues of research are suggested.