High-Sensitivity Eddy Current Probe Design via Multipath ResNet and Bayesian Optimization

Eddy current testing (ECT) is a vital technique for pipeline defect detection, where the sensitivity of detection is heavily influenced by probe design parameters. However, traditional optimization methods for probe parameters often suffer from limitations such as neglecting interactions among parameters, ignoring potential optimal combinations within the step size, and being quite time-consuming. To address these challenges, an advanced optimization framework is proposed, which combines a neural network with Bayesian optimization (BO). A probe configuration consisting of two coaxially arranged coils connected via a bridge circuit is investigated. A multipath residual neural network is developed as a surrogate model to evaluate the design parameters, including coil inner diameter, number of turns, height, and spacing. Bayesian optimization then uses this model as the objective function to identify optimal parameter combinations. Simulation and experimental results validate that the surrogate model demonstrates enhanced prediction accuracy, and the optimization process achieves superior performance with fewer iterations. Compared with the comparison groups, the optimized probes exhibit higher sensitivity for defects in the 1–4-mm depth range. These prove the effectiveness of the proposed method for efficient and high-performance ECT probe design, indicating its significant application potential.