基于改进CNN的化工设备故障分布式诊断技术研究  

Research on Distributed Fault Diagnosis Technology for Chemical Equipment Based on Improved CNN

作　　者：黄远 周文涛 Huang Yuan;Zhou Wentao(Yangzi Acetyl Chemical Co.,Ltd.,Chongqing,Sichuan 401254,China;School of Artificial intelligence and Automation,Huazhong University of Science and Technology,Wuhan,Hubei 430074,China)

机构地区：[1]扬子江乙酰化工有限公司,重庆401254 [2]华中科技大学人工智能与自动化学院,湖北武汉430074

出　　处：《化工设备与管道》2023年第6期8-17,共10页Process Equipment & Piping

基　　金：国家自然科学基金项目(编号:31571874)。

摘　　要：针对当前化工设备故障智能诊断方法准确率偏低和实时性较差等问题,提出一种基于神经网络故障诊断分类模型。首先引入分布式思想对1D-ResCNN进行改进,构建分布式特征提取卷积神经网络(DFECNN);然后引入了模型压缩技术对DFECNN模型进行轻量化处理,构建轻量型DFECNN模型。最后,基于上述内容,构建化工设备状态分布式诊断模型。结果显示,在研究构建的模型中,ReLU模块中的计算方式更具有细粒度,更适合整体标注模型,检测性能更好;DFECNN模型提升了全部评价指标的标注性能,保证了研究方法在设备故障检测中的有效性。轻量型DFECNN模型在经过训练后,在进行化工设备故障识别的实际效果上,对应的整体识别灵敏度、特异性、准确率依次为75.0%、83.3%、85.0%,优于其他模型。研究构建的轻量型DFECNN化工设备状态分布式诊断模型具有较高的精度和灵敏度,能够实现化工设备故障的智能诊断和检测,从而保障化工设备的正常运行,对化工产业发展有一定的促进作用。Aiming at problems of low accuracy and poor real-time performance of current intelligent fault diagnosis methods for chemical equipment,a classification model for fault diagnosis based on neural networks is proposed.Firstly,a distributed feature extraction convolutional neural network(DFECNN)was constructed by introducing a distributed idea to improve 1D-ResCNN;Then,model compression technology is introduced to lightweight the DFECNN model and construct a lightweight DFECNN model.Finally,based on the above content,a distributed diagnosis model for chemical equipment status is constructed.The experimental results show that among the models built in the study,the calculation methods in the ReLU module are more granular,more suitable for the overall annotation model,and have better detection performance;DFECNN model improves the labeling performance of all evaluation indicators,ensuring the effectiveness of research methods in equipment fault detection.After training,the lightweight DFECNN model has a corresponding overall recognition sensitivity,specificity,and accuracy of 75.0%,83.3%,and 85.0%in the actual effect of chemical equipment fault identification,which is superior to other models.The lightweight DFECNN distributed diagnosis model for chemical equipment status constructed through research has high accuracy and sensitivity,and can achieve intelligent diagnosis and detection of chemical equipment faults,thereby ensuring the normal operation of chemical equipment,and has a certain promoting effect on the development of the chemical industry.

关 键 词：人工神经网络 残差卷积思想 化工设备 分布式诊断 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]