TTCI investigated using a neural network technique to predict bolt hole crack, vertical split head, and crushed head rail flaws. The neural network model was developed to capture existing non-linear relationships between input variables pertaining to rail flaw development and rail defect outputs. Training and validation data were combined into three distinct groups (autumn/spring, summer, and winter) and models were developed for each defect type and group.
R T & S: Railway Track and Structures, April 2019, pp.9-12