Research On Electromagnetic Compatibility Fault Diagnosis Method Based On Probabilistic Neural Networks

With the improvement of the integration of electronic systems and the increasing complexity of the electromagnetic environment,the electromagnetic compatibility of electronic systems has become more and more prominent.For the electromagnetic emission items in the electromagnetic compatibility standard test,how to realize the rapid location and effective diagnosis of the electromagnetic compatibility fault object of the electronic system becomes the key.Electromagnetic compatibility testing has the characteristics of many coupling influencing factors,many working status changes,and relatively poor test data consistency.Even if the same system is tested in different laboratories and at different times,there are often differences in results.It brings great uncertainty to the electromagnetic compatibility interference location and diagnosis.The existing fault tree analysis or feature similarity comparison greatly reduces the dependence and requirements on personnel.However,there are limitations such as a single criterion,no uncertainty,and no effective use of a priori diagnostic data.The dissertation is aimed at this research.Based on the idea of pattern classification,the dissertation makes full use of the prior conclusions and test data of experts,and adopts probabilistic neural network to establish the many-to-one mapping relationship between system abnormal data and fault interference sources,and predicts the possibility through PNN network during fault diagnosis.which can provide a quantitative basis for testers to troubleshoot,reduce the requirements for test and diagnosis personnel,and improve the effectiveness of diagnosis.The main research content of this paper are as follows:Firstly,comparing the traditional electromagnetic compatibility fault diagnosis methods,combining the characteristics of electromagnetic compatibility fault diagnosis with various interference modes and the characteristics of multiple data characteristics and diagnosis requirements for specific fault modes,a PNN-based electromagnetic compatibility fault diagnosis method is proposed.Then take the electromagnetic compatibility fault diagnosis of the servo system as the research object,analyze its composition and electromagnetic radiation characteristics,combine the test experience to divide the servo system EMC failure mode and data frequency band,and determine the test data segmentation and failure mode during PNN network training Corresponding relationship;and taking the test of the photoelectric pod servo system as a typical object,the initial data source for establishing the PNN fault diagnosis network of the servo system is obtained.Then,pre-processing algorithms are proposed for all kinds of data needed in the PNN fault diagnosis network.Specifically,according to the original test data in the electromagnetic compatibility standard test,the energy norm characteristics of the data are extracted through wavelet packet decomposition,and the influence of the internal noise of the data on the validity of the data is analyzed,and the high-order noise part of the data is eliminated;the application test case verifies The PNN fault diagnosis method realizes the matching recognition of the data samples to be tested under different interference modes;in the verification,the PSO algorithm is used to optimize the smoothing factor of the PNN network,so that the fault recognition rate is improved under different sample numbers;comparative analysis The influence of different sample sizes on the recognition rate of fault diagnosis is verified,and the applicability of the algorithm is verified.Finally,use the Qt and MatLab platform to develop EMC fault diagnosis software based on the PNN network,establish and train the EMC fault diagnosis network of the servo system through the test abnormal data of the photoelectric pod servo system,and finally use the test data of a certain type of servo system to perform the software The test,through comparative analysis based on template matching method,preliminarily verified the applicability and feasibility of this method,and provided a new research idea for EMC fault location and diagnosis.