Domain Adaptation in Intelligent Ultrasonic Logging Tool: From Microseismic to Pulse-Echo

Abstract

With high temporal resolution demand under noise contamination, the travel time estimation for the pulse-echo is noticed in acoustic logging instrument design. To this end, an intelligent ultrasonic logging system is built to collect borehole information, and a framework with domain adaptation theory is proposed. The modified maximum mean discrepancy (MMD) minimization combined with Spatial Pyramid Pooling (SPP) is constructed on different deep neural networks (DNN), where the transfer from the micro-seismic P-wave picking to the estimation of echo travel time is achieved. Versus the conventional travel time extraction algorithms, the proposed scheme improves the picking accuracy to 83.55% of the 10 dB signal-to-noise ratio (SNR). Experiments over the ultrasonic logging tool demonstrate the feasibility of the confusion domain, the effectiveness of travel time estimation, and the versatility of algorithm application.