28-layer Multilayer PCB board, Base Material: FR4 TG 180. inner/outer copper thickness 1 oz. Immersion Gold surface treatment with green color soldermask. White color Legend.board thickness 3.6 mm. Min.line width/Min.line spacing 0.18/0.18 mm. E-test: 100%
Most of multilayer PCB boards are designed with impedance. There are two impedance tester in our factory.General single end impedance at 50 ohm, differential 100 ohm. Tolerance 10%.
Multilayer PCB Multilayer PCB,Electronics Multilayer PCB,Custom-Made Multilayer PCB,Custom Multilayer PCB Orilind Limited Company , https://www.orilind.com
The new fusion transformer network explains it from the following points:
1 FUZZY NEURAL NETWORK 2 FUZZY INTEGRATION DATA FUSION The transformer in this paper has 9 kinds of fault states: no fault; low energy density partial discharge; high energy density partial discharge; 0 low energy discharge; 1 high energy discharge; 2< Thermal fault of 1503C; Low temperature overheat fault of 41503003C; Overtemperature fault of 53007003C; High temperature overtemperature fault of 6>7003C.
According to the trained neural network, different fault samples are identified separately, and the correct recognition rate of each neural network classifier for each fault is recorded as the importance of the information of the classifier for each fault classifier. When the neural network classifier performs fault diagnosis, the max operator is used to complete the local decision. When the fuzzy fusion decision is completed, the max operator is also used to determine the maximum value as the final fault diagnosis result. The diagnostic steps are as follows: The fuzzy density value gi is determined according to the diagnostic accuracy of each neural network for different faults, and the local decision is calculated by the max operation according to the actual output of each neural network, and the fk is calculated; it is calculated by formula (2); (5) Calculate the g fuzzy measure; 0 Calculate the final fuzzy integral result according to formula (4), and use the fusion result as the diagnostic output.
3 Simulation study The three inputs are C2H2/C2H4, CH4/H2 and C2H4/C2H6. Take 60 samples of each fault and divide it into 4 groups. Each group has 15 samples to identify and test 4 RBF networks to determine each nerve. The correct recognition rate for each fault in the network, that is, the confidence level of each neural network, is the fuzzy density value used for fuzzy integral fusion. This paper selects 18 sets of fault samples from the non-training samples for simulation analysis.
From the simulation analysis, it can be seen that regardless of whether the local decision-making of each neural network is consistent, all kinds of transformer faults can be correctly identified. The fusion result is closely related to the output of the local decision-making of each fuzzy neural network. The confidence of each neural network is related to the higher the confidence, the better the fusion effect. Samples 7 and 12 show that there are 3 kinds of faults on the output of 4 neural networks, and the fault corresponding to the actual situation is obtained after the fusion processing by fuzzy integral.
4 Conclusions Transformer fault identification method based on fuzzy integral fusion technology is proposed. The fuzzy radial basis function neural network is used to identify the faults. The fuzzy integral fusion is based on fully considering the importance degree of the output information of each fuzzy neural network classifier. The structural system of the method determines that it has certain fault tolerance, reduces the risk of misdiagnosis of faults, and has broad application prospects. 