基于深度学习的往复压缩机故障检测方法研究

Fault Detection Method of Reciprocating Compressor Based on Deep Learning

利用AlexNet卷积神经网络和深度学习方法,对往复压缩机故障检测开展了研究。具体以往复压缩机示功图为研究对象,提出了基于AlexNet卷积神经网络的压缩机运行工况分类模型。在分类模型全连接层做一维特征向量的处理,提出了故障检测算法,搭建了往复空气压缩机实验测试平台。测取了4个工况下的示功图,组成了训练图集进行网络训练并进行了网络训练参数寻优。对进、排气阀泄漏故障进行的模拟测试结果表明,当故障评判阈值在80%~90%之间时,模型的故障识别准确率可达90%以上,可在仅有压缩机正常运行数据的情况下,很好地完成往复压缩机热力性能故障的检测,并进一步对故障气缸进行定位,为故障早期快速报警奠定了基础。

A study on reciprocating compressor fault detection method using AlexNet convolutional neural network deep learning method is carried out.Taking the reciprocating compressor dynamometer diagram as the research object,a compressor operating condition classification model based on AlexNet convolutional neural network is proposed.After processing the one-dimensional feature vector of the fully connected layer of the classification model and proposing a fault detection algorithm,an experimental testbed is built for reciprocating air compressors,and the indicator diagrams are measured under four operating conditions,and a training atlas is composed for network training and network training parameter search.Simulation test results on intake and exhaust valve leakage faults show that when the fault evaluation threshold is between 80%and 90%,the fault identification accuracy of the model can reach more than 90%.It can well complete the detection of thermal performance faults in reciprocating compressors with only normal compressor operation data,and further locate the faulty cylinder.It lays a foundation for early and rapid fault alarm.