An ultrasonic echo characterization approach based on particle swarm optimization

Abstract

Presented is a hands-free approach for the extraction and characterization of ultrasonic echoes embedded in noise. By means of model-based nondestructive evaluation approaches, echoes can be represented parametrically by arrival time, amplitude, frequency, etc. Inverting for such parameters is a non-linear task, usually employing gradient-based, least-squared minimization such as Gauss-Newton (GN). To improve inversion stability, suitable initial echo parameter guesses are required which may not be possible under the presence of noise. To mitigate this requirement, particle swarm optimization (PSO) is employed in lieu of GN. PSO is a population-based optimization technique wherein a swarm of particles explores a multidimensional search space of candidate solutions. Particles seek out the global optimum by iteratively moving to improve their position by evaluating their individual performance as well as that of the collective. Since the inversion problem is non-linear, multiple suboptimal solutions exist, and in this regard PSO has a much lower propensity of becoming trapped in a local minima compared to gradient-based approaches. Due to this, it is possible to omit initial guesses and utilize a broad search range instead, which becomes far more trivial. Real pulse-echoes were used to evaluate the efficacy of the PSO approach under varying noise severity. In all cases, PSO characterized the echo correctly while GN required an initial guess within 30% of the true value to converge.