Optimal design of a guided wave rail web transducor using numerical modelling

Abstract

Ultrasonic Guided Waves can propagate over long distances, and are thus suitable for the interrogation of long structural members such as rails. A recently developed Ultrasonic Broken Rail Detection (UBRD) system for monitoring continuously welded train rail tracks, primarily detects complete breaks. This system uses a guided wave mode with energy concentrated in the head of the rail, which propagates large distances and which is suitable for detecting defects in the rail head. Exploiting a second mode, with energy concentrated in the web section, would allow us to e ectively detect defects in the web of the rail. The objective of this study is to develop an ultrasonic piezoelectric transducer that can excite a guided wave mode with energy concentrated in the web of the rail. It is required that the transducer must strongly excite such a mode at the operational frequency of the UBRD system. The objective is thus to obtain a design with optimal performance. A recently developed numerical modelling technique is used to model the interaction of the transducer with the rail structure. The technique employs a 2D Semi-Analytical Finite Element (SAFE) mesh of the rail cross-section and a 3D nite element mesh of the transducer; and is thus referred to as SAFE-3D. The accuracy of the SAFE-3D method was validated though experimental measurements performed on a previously developed transducer. A design objective function representative of the energy transmitted by the transducer to the web mode was selected. The identi ed design variables were the dimensions of the transducer components. The performance of the transducer was optimized using a response surface-based optimization approach with a Latin Hypercube sampled design of experiments (DoE) that required SAFE-3D analyses at the sampled points. A Nelder- Mead optimization algorithm was then used to nd an optimal transducer design on the response surface. The performance of the optimal transducer predicted by the response surface was found to be in good agreement with that computed from SAFE-3D. The optimum transducer was manufactured and experimental measurements veri ed that the transducer model was exceptionally good. The design method adopted in this study could be used to automate the design of transducers for other sections of the rail or other frequencies of operation.