| Pulse bag filter is widely used in the industrial field with its high-efficiency filtering advantages.A large number of dust removal data are obtained by monitoring the operation status of the equipment.The utilization rate of these data is low in the industry,resulting in a great waste of data resources.Therefore,how to mine unknown patterns and information from these data is the key to realize intelligent in the field of dust collector.In this context,data mining research is carried out for the data set obtained from the actual production of pulse bag filter.Combined with the operation mechanism of the dust collector,machine learning and statistical analysis methods are used to analyze the motion characteristics of pulse bag filter.First of all,for the problem of insufficient cleaning capacity,bag rupture and equipment wear of pulse jet cleaning method for pulse bag filter,by analyzing the filtering mechanism of filter bag and the shortcomings of static injection method based on fixed time pressure difference,a dynamic cleaning method of double pressure difference threshold control high and low pressure injection based on pressure difference time sequence prediction method is proposed,Compared with static injection,dynamic injection has better adaptability,and can reconcile the contradiction between cleaning efficiency and filtration efficiency.Three models are selected to predict the pressure difference data.The experimental results show that the three models can predict the change trend of pressure difference data,and the MSE of BP neural network is the lowest,which is 0.00994.Secondly,to study the common faults of the pulse bag filter,choose to analyze the characteristics of various faults from two aspects of the filter system and the ash cleaning system.In order to analyze the correlation between various types of faults,the fault feature data set is extracted from the database,and the FP-growth correlation algorithm is used to mine the frequent itemsets and association rules of the fault data set,and the strong association rules generated from the mining results are filtered.Through correlation analysis,the correlation between different faults is obtained,the correlation information is summarized,and the analysis results are used to provide references for technicians in fault diagnosis and equipment maintenance.Finally,in order to solve the problem that the overflow of fault alarm information of pulse bag filter interferes with the fault diagnosis and analysis of technicians,the unsupervised learning method in machine learning is used to analyze the fault type data set.Firstly,the k-means clustering analysis is carried out on the fault data set,and the K is determined by a variety of parameter indexes.It is found that the clustering effect is the best when the K is 11.The random forest and support vector classification are trained on the labeled fault data set after clustering,and the three methods are applied to the new unlabeled fault data set for fault classification and prediction.By comparing the unity of the classification results of the three algorithms,it is found that the unity of the three algorithms is up to 67%when they have nine K values,At the same time,the unified rate of the three algorithms under different K values is greater than 0.998,which proves the feasibility of using this classification algorithm and has good prediction accuracy,which can help technicians to accurately locate the fault location and improve the efficiency of troubleshooting. |
