Fault Diagnosis and Analysis of Marine Filter Based on SOM Network

Marine filters are widely used in various marine auxiliary equipment and power module equipment. The quality of filter directly affects the performance of the ship's power system and decides whether the ship can operate normally or not. Therefore, the quality detection of the filters plays a key role in the whole system. However, the structure of marine filter is very complex, the input and output of the system are inconspicuous, so it is difficult to describe the filter effectively with an accurate model. But with the development of pattern recognition and neural network theory, the new methodologies provide a new way for fault diagnosis. In this paper, we use the non-linear mapping properties of SOM network, and improve the inadequacy of initialization of network weights, use "probability normal distribution" to distribute the initial weights reasonably, and by balancing the difference between weights and input vectors to determine the neighborhood range. The fault is effectively diagnosed and analyzed combined with the detection of flow and pressure signals filtered by filters, and the filters with different faults in internal structure can be distinguished, so as to achieve the purpose of analyzing the fault grade and category of filters.