Research On Fault Early Warning And Diagnosis Method Of Centrifugal Blowers Based On Transfer Learning

As typical rotating fluid conveying equipments,centrifugal blowers play an indispensable role in energy,environment,manufacturing and other fields.With the continuous development of the high-speed,automation and large-scale,the various components of the equipment and the correlation between different equipment are getting closer and closer.A large number of cases have proved that once a centrifugal blower fails,the failure can easily expand rapidly,causing equipment to stall,and even causing disorder in the production line and threatening life safety.In order to ensure the smooth and safe operation of equipment,networked monitoring systems have been widely used.The equipment operation data collected by the monitoring system is analyzed,and the hidden equipment health information is excavated to realize fault warning and diagnosis.This is an effective method to reduce the faults,and important measures to promote the intelligent transformation of the industry.Funded by the special funding of Chongqing technology innovation and application project-"Development and Application Demonstration of High-end Turbine Equipment Remote Operation and Maintenance Service Platform Based on Big Data and Intelligence"(cstc2018jszx-cyzd X0146).This paper studies the faults early warning and diagnosis method of centrifugal blower,realizes the early warning in time at the beginning of the fault,and accurately locates the fault type and fault location after the fault occurs,so as to ensure the safe and stable operation of the equipment,reduce the use cost,and provide reference for other equipment operation and maintenance research,so it has important academic significance and practical value.First,the composition and common faults of the centrifugal blowers are analyzed,and the common faults are summarized.A more complete fault tree is established through top-down,layer-by-layer deductive reasoning.Secondly,in view of the problem that the equipment operating data collected by the factory cannot support the construction of high-precision early warning models,a transfer learning method is proposed based on the autoencoding network.The data collected in the laboratory is used as the source domain,and the monitoring data of the equipment in actual use is the target domain.By minimizing the maximum mean discrepancy and finetuning of the characteristics of the source domain and target domain encoders,the fault warning for centrifugal blowers in actual use is constructed,and the accuracy is effectively improved.Then,in order to solve the problem of low fault diagnosis accuracy due to insufficient fault samples of equipments in actual use,a transfer learning method based on stacked autoencoding is proposed.The autoencoding fault diagnosis model established based on rich samples is retrained and reused in the actual use of the rotor fault diagnosis,the diagnosis accuracy rate is improved.Finally,the proposed fault tree,fault early warning and fault diagnosis methods of centrifugal blowers are embedded in the remote operation and maintenance system of turbine equipments,which makes the methods proposed in this paper applied to engineering practice.