Research On Real-time Fault Diagnosis System Of Subway Fan Based On Machine Learning

As a special tool for exhausting smoke and supplying air,the subway axial flow fan is the key to ensuring the normal operation of the subway and the safety of passengers.Its health status is directly related to the safe operation of the subway.Metro axial fans,which are basically in full-time working conditions,have many safety hazards when operating in subways in underground confined spaces.If the fan is working in a faulty state,the operation of the subway environmental control system will cause the whole system to run out of paralysis,which will seriously stop the car control room.Ensuring the safe operation of these wind turbine equipment is the premise and basis for maintaining the stability of the subway environmental control system,and it is also related to the safety of each citizen's personal life and property.With the development of artificial intelligence technology,a timely and rapid fault monitoring and diagnosis platform for subway axial fans can be effectively reduced.However,after field visits and investigations,it was found that the top wind turbine companies in China have not yet applied the complete set of fan monitoring and diagnosis systems to engineering practice or are still in the laboratory research stage.Therefore,it is necessary to establish a sound turbine condition monitoring and fault diagnosis system that can be initially applied in engineering practice.In this paper,for the DTF-3.55 subway axial flow fan,the experimental device of this type of fan is established for the first time.The fault simulation experiment of this type of fan is carried out and the corresponding fault library is established.Furthermore,a suitable improved machine learning method was established,and a wind turbine fault diagnosis system including signal analysis,signal feature extraction,real-time fault diagnosis and training mode was established with LabVIEW as the software platform.The main work and achievements of this thesis are as follows:(1)For the DTF-3.55 subway axial flow fan,the fault experimental device was built for the first time.The selection and installation of various components of the experimental device were completed,three common types of fan failures were simulated,and the obtained raw experimental data was subjected to signal processing and feature extraction to support the research of the subsequent fault diagnosis core.The corresponding fault library.(2)Based on fault data,the basic classifier and AdaBoost multi-classification algorithm are studied.The different basic classifiers are compared with AdaBoost multi-classification algorithm.The improved machine learning method AdaBoost.M2-DS is established.The fault recognition rate is better than BP.The neural network multi-classification algorithm has an improvement of more than 30%.(3)The DTF-3.55 subway axial flow fan fault real-time diagnosis system was established and the overall architecture and system of the system were introduced.The modules of this system were introduced in detail.Among them,the training mode of the system can enrich the fault library by continuously collecting relevant parameters,and upgrade the diagnostic core of the system based on this.Finally,a real-time fault monitoring platform for DTF-3.55 subway fans based on subway fans and diagnostic systems was established.The fault diagnosis accuracy and stability of the real-time fault monitoring platform of the fan established in this paper has been significantly improved compared with the traditional diagnostic system,and has certain practical value.