Heuristic Algorithms for Aggregating Rail‐Surface‐Defect Data

Abstract

An optical inspection system has been developed to detect the presence of defects on the surface of rails. The system classifies each 6 in. (15 cm) length of railhead as defective or nondefective and generates large quantities of disaggregate, sequential condition data. Defective rail surfaces can then be corrected by grinding the surface of the rail. However, this requires that condition data be aggregated to a level suitable for making maintenance decisions, and that prior recognition be given to practical constraints such as adjusting minimum grinding length to the configuration of the particular grinding machine. Data‐aggregation procedures range from rule‐based techniques to mathematical optimization methods. This paper reviews these aggregation techniques and, consequently, formulates the grinding problem as a set‐packing integer programming formulation. Two heuristic solution methods are proposed to solve a set‐packing problem of high dimension resulting from a large number of feasible packs for rail‐surface‐condition data. These methods effectively moderate the computational intensiveness and time complexity associated with using existing procedures.