| The axial fan of expressway is an important part of the ventilation system of long and big tunnel.Preventive maintenance is great significance to prolong the life of fan and ensure the safe operation of tunnel.Adding fault diagnosis system to the stock fan,real-time detection and feedback of the function and performance of the equipment,can be helpful to realize preventive maintenance.However,for the scenario that it is not suitable to install fault diagnosis system based on wired transmission,the narrow bandwidth transmission condition of wireless mode is an important bottleneck to limit the sampling information.Therefore,the optimization method of fan status monitoring points and fault feature selection algorithm are studied in this dissertation,so that the sampling information can meet the objective conditions of narrow bandwidth transmission.The research content of this dissertation is as follows:1.The purpose of selecting condition monitoring points is to sample key monitoring points first and reduce the number of monitoring points when the bandwidth is limited.The current fault diagnosis system based on wired transmission only selects monitoring points based on validity factors,that is,only considers whether the installed monitoring points are helpful for fault diagnosis.Although the diagnosis accuracy is improved,it is difficult to transmit a large amount of sampled information in real time through wireless.Therefore,a monitoring point optimization method based on reliability analysis is proposed for wireless scenarios.Based on the reliability reverse fault analysis theory,this method determines the risk degree,probability importance degree,easy detection degree,easy condition monitoring degree and maintenance economic factors of fan faults.Then,the faults are ranked,and monitoring points are set for the faults with high ranking priority according to the bandwidth conditions.Finally,among the 33 types of fan faults,it is determined that the motor bearing fracture,rotor unbalance and other faults need to install monitoring points first.The monitoring points determined by this method provide a research object for the following research feature selection algorithm.2.The purpose of feature selection is to find features with small quantity,high quality and general purpose and transmit them to the background for fault diagnosis.When the bandwidth is limited,important fault features are sampled preferentially to reduce the sampling information of a single monitoring point.Currently,Wrapper methods commonly used in fault diagnosis systems based on wired transmission.The features selected by Wrapper method are less in quantity and better in quality,but poor universality.The features selected by the Filter method are few and universal,but the feature quality is poor.When these two methods are applied to wireless scenario separately,the updating of fault diagnosis model and diagnosis accuracy will be affected.In order to balance the contradiction between feature quality and universality,this dissertation designed a two-step feature selection method combining Filter and Wrapper.The Filter method was used for transmission of features with less selection and strong universality,and the Wrapper method was used for secondary optimization of features after receiving data.Experimental results show that the fault diagnosis accuracy of the proposed algorithm reaches 98.33% when the sampling data is reduced by 65.385%,which is 1.33%higher than that of the Filter method alone,proving that the selected features have the advantages of less quantity and better quality.In terms of universality,the accuracy error of this algorithm on K-Nearest Neighbor(KNN)and Support Vector Machine(SVM)fault diagnosis model is 2.98%,which is lower than 9.34% of Wrapper method.The proposed method preferentially samples important fault features such as small quantity,good quality and strong universality of key monitoring points when the bandwidth is limited,and reduces the sampling information of monitoring points. |
PDF Full Text Request