Fault Diagnosis For Industry System Based On Similarity Measure And Analysis

Fault diagnosis is very important for the stability and the security of the industry system.This research studies the fault location for industry system based on vector similarity measure.By using topology structure,data-driven machine learning methods and similarity measure for a variety of variable data matrix,the fault location can be found by the monitoring system.The main contents are as follows:Distributed fault diagnosis based on partial model information.A new fault monitoring framework based on the combination of model-based fault detection architecture and data-based learning method to achieve the fault detection and the location of the series connected process.The serially connected system is analyzed to obtain the partition method of the whole system,and then this information of model is used to combine with data-base method.This framework can make up loss of model structure information which is the insufficiency of the data-driven method,and improve the accuracy and interpretability of fault diagnosis.The location of fault components based on optimized ReliefF feature selection algorithm.The optimal ReliefF feature selection algorithm is used to select features which are most related to the fault category.And then combined with the location of the feature collection in the model,the faults can be located.During the operation process,the industrial system will generate a lot of information,which reflect the change of the system state.After extracting,this information can be used for system status monitoring and fault diagnosis,and get the most original variables that cause the system failure.Fault recognition and classification strategy based on the K nearestneighbor of diffusion distance.Using the idea of Diffusion Maps(DM),the nonlinear high dimensional data are mapped to the diffusion space.This approach keeps the low dimensional manifold of the data and explores the potential geometric characteristics and inherent laws of high dimensional data.The diffusion distance is used to measure the similarity between the test data samples and the training data samples.Finally the fault identification and classification of industrial systems are realized by using the classification algorithm based on the combination of k nearest neighbor and the diffusion distance.A fault diagnosis system software based on expert system is designed.The software combines the result of data analysis mentioned above and the original expert knowledge of the equipment.Its main work process is as follows: in the process of operation,the actual data set which is collected is entered into the fault diagnosis model of the above text,and the type of fault is obtained.According to the rules in the expert knowledge base,the inference engine is used to deduce the fault symptom corresponding to the fault type,and the fault coping strategy is put forward,and the degree of the fault can be controlled in time.During failure repair,referring to the fault tree constituted by the rules of expert system,the specific cause of failure can be diagnosed,and maintenance suggestions will be given.