An ultrasonic gas leak localization method based on mode switching

Abstract

With the increasing demands for industrial safety, environmental protection, and energy efficiency, ultrasonic gas leak localization has attracted increasing attention. However, conventional single-model localization algorithms often exhibit degraded accuracy when the source-to-sensor distance varies, owing to the different propagation characteristics of near-field and far-field acoustic signals. To address this problem, this study proposes an ultrasonic gas leak localization method that combines an improved phase-transform (PHAT)-N-ρr weighted cross-correlation algorithm with adaptive near-/far-field mode switching. In the far-field mode, the improved PHAT-N-ρr algorithm introduces a power-law normalization factor ρ and a perturbation term r to enhance the stability of time-delay estimation under weak-signal and low-SNR conditions. In the near-field mode, energy decay (ED) information is combined with the time difference of arrival (TDOA) to improve localization accuracy under spherical-wave propagation conditions. A five-element acoustic array was designed for 40 kHz ultrasonic leakage signals, and an amplitude-difference-based decision mechanism was introduced to select the appropriate localization model according to the distance-dependent acoustic characteristics of the leakage signal. Experimental results obtained under controlled laboratory conditions at different source-to-array distances show that the proposed method achieved a mean localization error of 23.33 mm and a maximum error of 39.10 mm. Compared with standalone far-field PHAT-N-ρr and standalone near-field ED-TDOA methods, the mean error was reduced by approximately 26.8% and 42.2%, respectively. These results indicate that the proposed method improves localization stability by combining far-field weak-signal time-delay enhancement with near-/far-field adaptive localization model selection, rather than applying a single localization model over the entire measurement range.