Research On Integrated Condition Monitoring Methods For Key Parts Of Rotating Machinery

In modern industry,as the key equipment,rotary machinery plays an important role in metallurgy,wind power,oil,gas and other industries.Bearings,gears and other key parts of the rotating machinery are the main sources of failure.With the upgrade of machinery,the demand for its intelligent maintenance is increasing.Theoretically,the quantification of the machine degradation assessment is achieved by defining a distance metric that measures the deviation from the health baseline.As the degradation of rotating machinery is dynamic and nonlinear,when the selection of the health baseline cannot correctly represent the optimum healthy state of the machinery,the accuracy and reliability of the degradation assessment in condition monitoring are with uncertainty,which brings difficulties and challenges to the whole process.In addition,few existing condition monitoring methods can achieve health information extraction,degradation assessment and fault diagnosis under the same technical scheme,which is inefficient in practical engineering application.In this paper,an integrated condition monitoring method is developed under the same technical scheme,based on perceptual vibration hashing and self-organizing maps neural network.It includes feature extraction,degradation assessment and intelligent fault diagnosis.(1)For feature extraction,an improved perceptive vibration hashing algorithm is proposed,which represents the raw vibration signal by using machine condition hashes with several symbols.Not only the effectiveness of information extraction can be guaranteed,but also the data dimension is compressed and reduced.(2)For degradation assessment,the distance metric quantifying the similarity between machine condition hashes is used as a health index to perform the degradation of the machinery.Different from the traditional method that using the initial working state as the health baseline,this study proposed an optimum healthy state selection method based on the clustering center of self-organizing maps neural network,and then the distance metric deviating from the optimum healthy state is defined as a health index.(3)For fault diagnosis,when the value of the health index exceeds the preset threshold,a visualized fault diagnosis method based on the self-organizing maps neural network is proposed for fault type recognition.Finally,the accuracy and robustness of the proposed method are validated by a public benchmark dataset and a rotary torsional fatigue test bench respectively.The results show that the proposed method can achieve a good performance on early degradation detection and fault type recognition.