Abstract:

[Objective] At present, the impact echo method is often used to detect the grouting compactness, but the signal interpretation relies on manual frequency domain analysis of the shock elastic wave signal, which has the shortcomings of poor objectivity and low efficiency. To address this issue, we propose an intelligent interpretation method based on deep learning and shock elastic wave detection signal. [Methods] Based on the elastic wave detection signal of the real grouting sleeve and the corresponding cloud picture, the one-dimensional time series information and the two-dimensional spectrum space information feature are innovatively fused for multi-mode intelligent analysis, and the anti-interference training is conducted in the network framework to send the similarity of the regression principal axis and the reference line into the rule base to obtain accurate compactness detection results. On the basis of the Transformer network, multi-task network branches are added. The feature extraction of one-dimensional signal and image segmentation of the spectrum cloud is performed through two front-end branches, and the regression is performed after feature fusion. [Results] This method can effectively tackle the pain points of grouting compactness detection and has the advantages of faster speed and higher accuracy. [Conclusions] The test data and field engineering verification are used to verify the method. The experimental results show that the method has great engineering application value.

Key words: impact echo (IE), intelligent signal analysis, transformer, convolutional neural network (CNN), multi-modal