Abstract: Fuzzy neural networks are composed of fuzzy logic and neural networks. They have the advantages of fuzzy technology and neural network technology. They are applied to power system fault diagnosis and are a convenient artificial intelligence processing system. The fault diagnosis of the rotating rectifier of the brushless excitation synchronous generator is studied based on the fuzzy neural network. After training, the fuzzy neural network has high accuracy and effectiveness in diagnosing rotating rectifier faults. Keywords: brushless excitation; rectifier fault; fuzzy neural network 1 Introduction Brushless excitation technology has completely solved a series of problems caused by excessive excitation current of large-capacity or ultra-large-capacity synchronous generators, ensuring that large generator sets can operate continuously for a long time. At present, more and more large brushless excitation generator sets are operating in the power system. However, the rotor rotating rectifier in the brushless excitation synchronous generator set is the most important link in the brushless excitation system. It is necessary to ensure that it works in a normal state. For rectifier faults, the nature of the fault must be distinguished first and different treatments should be carried out: (1) The open circuit fault of the rectifier element is a slow-changing fault and can be used as the object of fault diagnosis. The open circuit fault can be cleared by the timely action of the protection device. (2) For short-circuit faults in rectifier components, the first consideration should be the protection action, requiring the protection device to act as soon as possible to clear the short-circuit fault. Because a short-circuit fault may burn out the winding within seconds or minutes, there is not enough time to complete the fault diagnosis and processing. Therefore, when the power diode on any arm of the rotating rectifier fails (open circuit), the monitoring and protection system should correctly locate and diagnose which arm has failed and quickly provide early warning and protection control to achieve predictive maintenance of the rotor rotating rectifier of a large brushless excitation synchronous generator set. In response to this problem, this paper adopts fuzzy neural network recognition technology to establish a diagnostic model for the fault of the rotor rotating rectifier of the brushless excitation synchronous generator set. The establishment of fuzzy rules is achieved through the neural network BP learning algorithm, thereby establishing a high-precision positioning and diagnosis system, which not only expands the range of diagnosing the malfunction of the rotating rectifier, but also improves the positioning accuracy of the fault diagnosis. 2 Fault Diagnosis Signal Extraction Online real-time monitoring of the working status of the rotor rotating rectifier of the brushless excitation synchronous generator requires relevant information of the rotor-side rotating rectifier. Since the synchronous generator rotor eliminates slip rings and brushes, the extraction of common fault waveform characteristics (fault features) of the rotor-side rotating rectifier cannot be directly completed. One method for obtaining fault features of the rotor rotating rectifier element (generally referring to open circuit faults) is to directly use the stator excitation coil of the AC exciter, which rotates coaxially with the synchronous generator, as the detection coil. Its advantages include eliminating the need for designing, installing, and maintaining dedicated detection coils, facilitating on-site signal data acquisition, and utilizing the strong rotor signal with a large amount of rotor fault information. The extraction process is shown in Figure 1. Capacitors C1 and C2 are connected in parallel to the AC exciter stator excitation coil. The armature magnetic field generated by the AC exciter rotor armature current inevitably cuts its stator excitation coil and induces an electromotive force. Then, through voltage division by capacitors C1 and C2, only the voltage waveform across capacitor C2 containing rotor fault information needs to be collected. Frequency domain analysis is then performed to decompose different harmonic components of varying amplitudes, which serve as artificial intelligence fault diagnosis signals. Let the period of the voltage signal wave Uc(t) across capacitor C2 be T. The Fourier series expansion is: Since Fourier series decomposition calculations require a large number of additions, subtractions, multiplications, and divisions, a complex trapezoidal integral formula with a certain accuracy is used. The Fourier series coefficient expression for equation (2) is: 3 Fault Diagnosis Model The stator magnetic pole coils of the AC exciter are affected by the static excitation rectifier and the saturation of the magnetic pole core, increasing the signal-to-noise ratio of the voltage waveform signal across capacitor C2. By establishing an artificial intelligence fault diagnosis mode based on a fuzzy neural network, the distortion of the fault signal of the rotor rotating rectifier of the brushless excitation synchronous generator is corrected, thereby improving the location accuracy of the fault diagnosis. Since neural network theory simulates the working mode and structure of biological neurons, it retains the learned information through training, thereby describing the process characteristics. Fuzzy logic systems can be represented as the expansion of fuzzy basis functions, and the essence of neural networks can be regarded as a linear combination of some functions. As a logic system with learning algorithms, fuzzy neural networks are widely used in power systems. Studies have shown that the spectral characteristics of the voltage waveform signal across the external capacitor C2 of the AC exciter stator have a certain correspondence with the operating state of the rotor rotating rectifier of the brushless excitation synchronous generator. When the rotor rotating rectifier fails, it will cause a change in the spectral characteristics of the voltage waveform signal across capacitor C2. By utilizing the learning and memory functions of neural networks, the nonlinear relationship between the spectral characteristics of the magnetic pole capacitor voltage Uc(t) waveform signal under different operating conditions and the rotor rotating rectifier fault can be extracted, and a fuzzy logic rule can be established. Since the rule considers the influence of the AC exciter stator magnetic pole coil on the static excitation rectifier and the saturation of the magnetic pole core, the fuzzy diagnostic system has high reliability. Therefore, the fuzzy diagnostic system can be trained using the learning algorithm of the neural network to realize the parameter identification of the fuzzy neural network. The fuzzy neural network consists of an input layer, a fuzzy inference layer, and an output layer. The input fuzzification function is taken as a Gaussian membership function. The number of nodes in the fuzzy inference layer can be obtained by clustering samples, or it can be adjusted according to actual requirements to meet accuracy requirements. θ[sub]ij[/sub], αi[sub]j[/sub], and λ[sub]ij[/sub] correspond to the nodes of the fuzzy inference layer, respectively. The output function can be expressed as: For the output function (6), the problem of determining the diagnostic model can be transformed into the error equation: To obtain the minimum value, the learning algorithms for each parameter can be obtained from the BP algorithm as follows: Based on the neural network BP algorithm, the parameter correction of the learning process is performed according to equations (8), (9), and (10) to train the parameters of the system network, and then the fault diagnosis system based on the fuzzy neural network can be determined, as shown in Figure 2. In Figure 2, the input layer has 6 neurons, and its input signal is {, which are the fundamental, second, third, fourth, fifth, and sixth harmonic values ​​of the voltage Uc(t) waveform signal across the external capacitor C2 of the AC exciter stator; the hidden layer is the fuzzy inference layer with 5 neurons; the output layer has 3 neurons, and the output result is, which can accurately diagnose the 8 operating states of the rotary rectifier rotor of the brushless excitation synchronous generator. Based on the characteristics of the sigmoid function of the fuzzy neural network, the diagnostic output of the fuzzy neural network for rotary rectifier faults is set to the expected diagnostic values ​​of (0.1, 0.1, 0.1), (0.1, 0.1, 0.9), (0.1, 0.9, 0.1), (0.1, 0.9, 0.9), (0.9, 0.1, 0.1), ..., (0.9, 0.9, 0.9). 4. Fault Diagnosis Example Eight operating conditions of the rotating rectifier were collected from the two ends of the external capacitor C2 of the AC exciter stator of a 360 kW brushless synchronous generator, as shown in Table 1. Let Q be the set of operating conditions reflecting the rotating rectifier of the synchronous generator, where: P1 represents normal operation of the rotating rectifier; P2 represents an open circuit fault in the A-phase positive group (A+); P3 represents an open circuit fault in both the A-phase negative group (A-) and the B-phase positive group (B+); P4 represents an open circuit fault in the B-phase positive group (B+); P5 represents an open circuit fault in both the A-phase negative group (A-) and the B-phase positive group (B+). P1+ is an open-circuit fault in both arms; P6 is an open-circuit fault in both arms of the positive and negative groups of phase A (A+, A-); P7 is an open-circuit fault in all three arms of the negative group of phase A (A-), the positive group of phase B (B+), and the negative group of phase C (C-); P8 is an open-circuit fault in all three arms of the negative group of phase A (A-), the positive group of phase B (B+), and the negative group of phase C (C-). Fault diagnosis was performed on the 360 ​​kW brushless synchronous generator rotating rectifier under eight operating conditions, and the diagnostic results are shown in Table 2. 5. Conclusion When a fault occurs in the rotating rectifier of a large synchronous generator, it will cause a change in the armature magnetic field of the AC exciter, inducing a series of harmonic potentials in the stator pole coils of the AC exciter. These harmonic potentials reflect the operating conditions of the rotor rotating rectifier. Therefore, this paper introduces the application of a fuzzy neural network system in the fault diagnosis of the rotating rectifier of a large synchronous generator. The results show that the established diagnostic model has good diagnostic performance.