Intelligent Fault Diagnosis And Life Prediction Method For Typical Hydraulic Pump Based On Deep Learning

The development of big data,deep learning,cloud platform,and other technologies provide new theories and technical support for intelligent operation and maintenance of equipment.“Condition-based maintenance” and “predictive maintenance” have become the development direction of modern equipment operation and maintenance systems.As the cross and combination of intelligent operational theory and hydraulic technology,and combined with the characteristics of the mobile mechanical hydraulic system,such as widely distributed operation sites,mobile operation,and harsh operating environment,how to realize the condition monitoring and intelligent operation and maintenance of hydraulic systems on the premise that the equipment meets the operation requirements has become a common scientific problem in the equipment manufacturing industry.Therefore,this paper takes typical hydraulic pump as the research object and establishes a intelligent operation and maintenance system of typical hydraulic pump based on cloud platform as the research objective.Focus on edge side data intelligent signal preprocessing algorithm,health assessment,Remaining Useful Life(RUL)prediction and fault diagnosis algorithm based on cloud platform and deep learning,and develop intelligent operation and maintenance system of constant pressure variable displacement pump.It provides new theory,new technology,and new method for the intelligence operation and maintenance of modern equipment.Firstly,aiming at the shortcomings of traditional feature engineering methods for extracting original signals,the adaptive signal decomposition denoising algorithms were studied and the adaptive decomposition and elimination algorithms for vibration signals based on adaptive local iterative filtering and variational mode decomposition were proposed respectively.The algorithms were applied to the vibration signal decomposition and denoising of constant pressure variable displacement pump,and the high frequency noise in the vibration signal of constant pressure variable displacement pump was effectively filtered out.The recognition accuracy of the fault diagnosis result of constant pressure variable displacement pump is 99% by using the algorithms,which verifies the effectiveness and advancement of the method.Given the serious unbalance between normal state data and fault state data in the process of equipment operation,the data-driven deep learning model has a strong dependence on the quality and quantity of data,the time-conditional generative adversarial network algorithm was studied.A time-conditional generative adversarial network based vibration data enhancement algorithm was proposed.The algorithm was used to enhance the unbalanced data of constant pressure variable displacement pump big bearing fault.The algorithm not only significantly improves the balance rate of data samples,but also improves the fault identification accuracy of the model from 80% to 98% after data enhancement.Aiming at the problems of limited installation space or difficult installation and disassembly of vibration sensors in some devices,the extraction algorithm of voice print features based on sound signals was studied,a fault diagnosis method for constant pressure variable displacement pump based on the combination of voicing feature and ivector recognition model was proposed.By collecting the non-contact sound signals around the constant pressure variable displacement pump and extracting the voiceprint features,the fault diagnosis of constant pressure variable displacement pump based on the voiceprint features is realized,and the recognition accuracy reaches 92.63%,which provides a new idea and new method for solving the fault diagnosis problem of the noncontact sensor of the hydraulic pump.Aiming at the problem that the traditional single sensor fault diagnosis algorithm of hydraulic pump is not comprehensive and complete in obtaining the state information of hydraulic pump,the feature fusion algorithm based on Residual Network(Res Net)was studied,a fault diagnosis algorithm based on smooth pseudo Wigner-Ville distribution and Res Net was proposed.Under the condition of strong noise interference,the algorithm realizes the original feature extraction,multi-dimensional feature fusion and feature dimension reduction of the constant pressure variable displacement pump vibration signal.The fault diagnosis result has strong stability,and the recognition accuracy can reach 90%,which establishes a theoretical foundation for the realization of the intelligent fault diagnosis system of the constant pressure variable displacement pump.Aiming at the problems that the slow performance degradation process of hydraulic pump could not be accurately evaluated and the equipment RUL could not be accurately predicted,the equipment health evaluation and RUL prediction algorithm were studied,a health evaluation and RUL prediction algorithm based on Res Net and long short-term memory neural network was proposed.Based on this algorithm,a gear pump health assessment model was established and the vibration signal of the gear pump was analyzed to obtain the gear pump health indicator.The health indicator correction coefficient was proposed,and the model correction of the differences within the product class is realized.The mean absolute error of the gear pump RUL prediction is reduced by 50% before and after the correction.Finally,the system architecture and data management architecture of intelligent fault diagnosis system of hydraulic system are proposed.The intelligent fault diagnosis system of constant pressure variable displacement pump was developed on Advantech WISEPaa S cloud platform,and the edge side and cloud platform deployment of intelligent fault diagnosis algorithm proposed in the previous paper were realized.Subsequently,the system was tested online by using the fault simulation test bed of constant pressure variable displacement pump.The real-time condition monitoring,fault diagnosis and fault degree evaluation of the constant pressure variable displacement pump were realized,and the accuracy of diagnosis and evaluation is 93.36%.The research results can not only provide key technologies for big data mining of hydraulic pump operation but also lay a theoretical foundation for complex equipment hydraulic system intelligent operation and maintenance.It can also provide an example for China’s “Cloud Platform + Artificial Intelligence for IT Operations”,which has important guiding significance for promoting the intelligent transformation and upgrading of hydraulic equipment in China.