A hybrid ANN-GA back-analysis technique for local anomaly detection in railway track substructure

Abstract

The UK's ageing railway transportation network is increasingly at risk of substructure failure, often caused by malfunctioning buried drainage systems. These drainage issues lead to localised soil weaknesses in the substructure layers, which, if undetected, can result in costly maintenance interventions or, worse, catastrophic system failure. Regular non-destructive testing (NDT) assessments are essential for monitoring the condition of the substructure, yet current interpretation techniques for NDT data provide limited insight into the size, location, and even presence of weakened zones. This results in an incomplete understanding of the substructure's condition, impeding effective maintenance planning. This study proposes a novel hybrid back analysis technique to detect weakened zones in railway substructures caused by drainage malfunctions, addressing a critical gap in existing solutions. The method employs an artificial neural network (ANN) surrogate model, trained on virtual experimental data generated through finite element (FE) simulations, and couples it with a genetic algorithm (GA) to optimise the match between modelled and measured deflections. This novel method is computationally efficient, independent of seed modulus values, and thoroughly validated for accuracy. It delivers a precise understanding of soil weaknesses in railway substructures, transforming maintenance strategies by improving safety, reducing costs, and promoting infrastructure sustainability.