基于卷积神经网络的乙烯球罐泄漏检测研究

Study on Leakage Detection of Ethylene Spherical Tank Based on Convolutional Neural Network

乙烯球罐泄漏会对安全生产造成严重威胁。为了解决泄漏初期不易检测到的难题,研究应用卷积神经网络进行自动泄漏识别的方法。以不同角度下的正常工况和泄漏工况的场景图像作为训练对象,搭建基于卷积神经网络的识别模型,研究确定采用最大池化法和ReLU激活函数的网络设置,可使网络性能达到最优。测试结果表明,本文提出的方法和模型能有效实现乙烯球罐泄漏的自动检测。

The leakage of ethylene spherical tank will pose a serious threat to safety in production.Considering the difficulty of leakage detection at early stage,the convolution neural network method for automatic leak identification was studied.The recognition model based on convolution neural network was built with scene images of normal and leakage conditions from different angles as training objects.The network settings of maximum pooling method and ReLU activation function were studied and selected,which optimized the network performance.The test results showed that the proposed method and model could effectively realize the automatic leak detection of ethylene spherical tank.