CNN-LSTM network-based damage detection approach for copper pipeline using laser ultrasonic scanning

Abstract

Copper pipeline is a commonly used industrial transmission pipeline. Nondestructive testing of copper pipeline early damage is very important. Laser scanning has attracted extensive attention because it can realize the visualization of guided wave propagation and non-contact on-line detection. However, the damage points detection in laser scanning imaging method rely on the difference between the damage points signals and surrounding normal points signals. This limits the applicability of laser scanning and may lead to inaccurate in large-area detection. Facing with such challenges, a damage detection method based on CNN-LSTM network is proposed for laser ultrasonic guided wave scanning detection in this paper, which can detect each scanning point signal without relying on the surrounding detection points signals. Firstly, the proposed data conversion algorithm is used to preprocess the laser scanning signals. Next, CNN-LSTM network is used to train the damage detection model. Four 1D Conv channels with different convolution kernel sizes and depths are designed in Convolutional Neural Network (CNN) module. The module can extract the signal time domain features. Then the features are input into the Long Short-Term Memory Network (LSTM) for feature extraction and classification. Finally, the CNN-LSTM is trained using the laser scanning detection data collected on the copper pipeline with crack and corrosion damages, and applied to detect the copper pipeline damage signal. At the same time, the state-of-the-art methods is compared with proposed method. The experimental results show that the detection accuracy of the method is 99.9%, 99.9%, 99.8% and 99.8% for copper pipeline 0.5 mm deep crack damage, penetrating crack damage, corrosion damage and inside crack damage, respectively. The damage location and size can be accurately detected by the proposed method.