Study On Fault Diagnosis Technique For Marine Main Engine System Based On Community Detection Algorithm

Fault monitoring and diagnosis of marine engine system is of great significance to improve safety and reliability of marine diesel engine work, reduce equipment maintenance costs and reduce economic losses, and avoid the occurrence of major accidents. The most commonly used traditional methods of fault diagnosis for marine diesel are thermal parameters of analysis, oil analysis and vibration analysis. Some of the ideas and methods of other disciplines, such as time domain analysis, wavelet analysis, wavelet packet method, the local wave method, rough set theory are also transplanted to the study of marine engine fault diagnosis. At the same time, the community detection as one of the hotpots in complex networks has penetrated into many engineering fields, including fault monitoring and diagnosis of marine engine system.The community structure analysis of complex network is a very important issue in the analysis of complex network structure. The community is a node subset of network nodes, the connection edges between whose internal nodes are very densely populated, and the connection edges among subsets are sparse. Community detection algorithm objective is to use as little information as accurately as possible the network community structure. Graph segmentation method such as Kernighan-Lin algorithm and the spectral bisection method, splitting methods such as GN algorithm, consolidation method whose benchmark is modularity, methods based on network dynamics and statistical reasoning are introduced. Each sample of the marine engine system is as a network node, and the value of the various thermodynamic parameters is as node attributes. The degree of similarity between the fault samples is defined, and is as the edge weight between nodes. The fault node classification is community detection. Clauset, Newman and Moore computed and updated the network modularity using the heap data structure, so CNM algorithm is gotten. Calculate the similarity between income classes and class, and fault diagnosis is converted to solve the detection problem to meet within-class similarity and between-class similarity of sub-networks. The final conversion is the modularity gain function optimization problem. In order to reduce the computational complexity, the initial partitioning process is set, the nodes whose similarity is greater than the threshold are merged, and then the CNM algorithm is executed. The MATALB software is used for the simulation of the CNM algorithm for friendly platform programming environment, easy-to-use programming languages, scientific computing, data processing capabilities and full graphics processing capabilities. The fault data are collected from STX Dalian Shipbuilding Co., Ltd.7boat-related information and records. Accuracy and running time with threshold from0.01to0.99are counted to test the pros and cons of the algorithm. The results show, using the appropriate similarity function, CNM algorithm has a very high accuracy rate, and stability on the threshold, which can get the correct results within a very short time.The application of Complex network community detection algorithm on the marine engine system fault diagnosis is feasible shown by simulation results, however, the interface program between actual fault samples taken and theoretical test data due to the difference of the actual situation, needs further research. In the entire marine system fault diagnosis applications, the applicability of this method need to be further validation.