Liquid Drip Detection in Power Plant based on Machine Vision

Abstract

In the power industry, equipment that has been running for a long time will have poor airtight performance over time, and the liquid in the equipment or pipeline will leak out, causing drip leakage.With the rapid development of computer vision technology, especially the development of convolutional neural networks and target detection models, target detection algorithms based on deep learning are increasingly being used in the power industry. In this paper, the YOLOV3 algorithm based on the PyTorch framework is used to detect the falling process of the dripping liquid. However, because the real power plant environment cannot be inspected on the spot, it can only simulate the falling process of the dripping liquid in the power plant by itself. The leakage of the water from the household faucet is used to simulate the leakage of the power plant equipment, and the pictures of the water droplets and the video of the falling process of the water droplets are taken with the mobile phone, which are used for the production, training and target detection of the subsequent data sets. By taking pictures of water droplets and performing data augmentation, a data set of droplet pictures is constructed, and the labelImg software is used to label the droplets in the pictures. The data set has only one droplet classification. Then, the data set of the augmented dripping pictures is made into the PASCL VOC data set, and the YOLOv3 model is used for training. The MAP of the training result is 0.996. When using the model obtained after training to detect the dripping video, it can be accurately identified. And locate the dripping target in the video frame.