一种无接触式压缩机运行故障检测技术研究

Research on a Fault Detection Technology for Non-contact Compressor Operation

在传统机械设备故障诊断技术中,数据采集的方式主要是使用振动传感器采集振动信号,然而这种采集方式非常不便于大规模部署,效率低下,灵敏度不足。利用麦克风采集空气压缩机运行时声音,实现了非接触式数据采集;然后利用VGGish模型和小波包分别提取128维Embedding特征和小波能量特征,并且将两种方式提取的特征进行了特征融合;最后借助遗传算法将SVM、RandomForests、Adaboost三个强基学习器进行了模型集成,提高了故障模型的识别准确率。

In the traditional fault diagnosis technology of mechanical equipment,the method of data acquisition is mainly to use vibration sensors to collect vibration signals.However,this acquisition method is very inconvenient for large-scale deployment and has low efficiency.In this paper,the microphone is used to collect the sound of the air compressor during operation,and the non-contact data acquisition is realized.Then the VGGish model and the wavelet packet are used to extract the128-dimensional embedding feature and the wavelet energy feature respectively,and the features extracted by the two methods are combined.Finally,the three strong base learners of SVM,RandomForests and Adaboost are integrated with the genetic algorithm,which improves the recognition accuracy of the model.