System for identifying and detecting abnormal sound in oil cylinder of transformer based on neural network

Abstract

The invention provides a system for identifying and detecting abnormal sound in a transformer oil cylinder based on a neural network. According to the system, causes and characteristics of abnormal sound in the oil cylinder of the transformer are firstly analyzed, and then audio data are acquired by using a microphone system and characteristics are extracted. Then, a hybrid model combining a convolutional neural network (CNN) and a recurrent neural network (RNN) is constructed for training; and finally, verifying the recognition accuracy and robustness of the system through experiments. The method aims to improve the accuracy and efficiency of abnormal sound recognition of the transformer, provides powerful support for state monitoring and fault early warning of the transformer, and has a wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, specifically a system for identifying and detecting abnormal noises inside transformer cylinders based on neural networks. Background Technology

[0002] As a core component of an electron accelerator system, the stable operation of the transformer is crucial for the continuity and reliability of the system. However, during long-term operation, various factors such as changes in magnetic induction force, wear of mechanical parts, and abnormal oil flow conditions may cause abnormal noises within the transformer cylinder. These abnormal noises are often precursors to internal transformer faults. Traditional methods for detecting transformer abnormal noises mainly rely on manual auscultation and experience-based judgment, which are inefficient and cannot achieve long-term monitoring, failing to meet the requirements for accurate and comprehensive fault early warning. Summary of the Invention

[0003] To address the aforementioned issues, this invention proposes a neural network-based system for identifying and detecting abnormal noises inside transformer cylinders. This system aims to improve the accuracy and efficiency of abnormal noise identification, providing strong support for transformer condition monitoring and fault early warning.

[0004] The technical solution of the present invention includes the following steps: S1. Constructing the training set: Using a microphone system, audio sampling is performed to collect and label the sounds under normal and various abnormal conditions of the transformer in advance, and to construct the training set. S2. Model Building: Design a neural network model to classify input audio data; S3. Model Training: Use the constructed training set to train the designed neural network model; Audio data acquisition: Audio sampling is performed using a microphone system to collect the sound during transformer operation; S4. Actual target sound acquisition: Using a microphone system, audio sampling is performed on the actual target transformer; S5. Model Execution: Run the trained neural network model on a microcontroller or SoC system to identify and classify the actual target sound, and output the results of whether it is normal or not, and the type of abnormal sound.

[0005] The specific steps of S1 described in this invention include: S11. Use a microphone system to sample audio, collecting data for 1 normal operation and 6 abnormal states to construct a training set.

[0006] The specific steps of S2 in this invention include: S21. Use Fast Fourier Analysis (FFT) to extract frequency domain features from the input audio signal to obtain spectrum data; S22. Use a convolutional neural network (CNN) layer to perform convolution operations on the spectrum of the audio signal to extract local features; S23. Use pooling layers to compress data scale; S24. Use CNN and pooling again to extract features; S25. Use the pooling layer again to compress the data scale; S26. Use a gated loop unit (GRU) to capture the temporal dependencies in the audio signal; S27. Finally, the abnormal noise type is output by classifying it through a fully connected layer.

[0007] The specific steps of S3 in this invention include: S31. Use the constructed training set to train the designed neural network model.

[0008] The specific steps of S4 in this invention include: S41. Use a microphone system to sample audio from the actual target transformer.

[0009] The specific steps of S5 in this invention include: S51. On a microcontroller or other system, run the trained neural network model to identify and classify the actual target sound, and output the result of whether it is normal or not, and the type of abnormal sound.

[0010] Combining these beneficial effects, this invention, by introducing neural network technology, achieves automated and intelligent identification and detection of abnormal noises within transformer cylinders, improving the accuracy and efficiency of noise identification. The detection system of this invention can monitor the transformer's operating status in real time, promptly detect and warn of potential faults, providing strong protection for the safe and stable operation of the transformer. The technical solution of this invention has broad application prospects and can be extended to the field of fault monitoring and early warning for other power equipment. Attached Figure Description

[0011] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 Flowchart of an embodiment of the present invention; Figure 2 : System structure block diagram of the present invention; Figure 3 : Neural network structure diagram of this invention. Detailed Implementation

[0012] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. Where there is no conflict, the embodiments and technical features in the embodiments can be combined with each other. The term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0013] Example 1 Figure 1 and Figure 2 The flowchart and schematic diagram of the internal inverter voltage regulation method of the X-ray machine power supply disclosed in this embodiment are shown respectively. The steps are as follows: S1. Constructing the Training Set: A training set is constructed using a microphone system for sampling. This includes: S11. Uses the Panasonic WM-61A microphone system to sample audio at a sampling rate of 44100Hz; S12. Pre-collect the sounds of the transformer under one normal operating condition and six abnormal conditions and label them. 200 samples of each condition are collected to construct a training set with 1400 samples in total. S2. Model Construction: Design a neural network model algorithm to classify input audio data; the algorithm structure includes: S21, 1024-point Fast Fourier Transform (FFT); S22, 32x1024 convolutional layer 1; S23, 32x512 pooling layer 1; S24, 64x512 convolutional layer 2; S25, 64x256 pooling layer 2; S26, 256x128 GRU layer; Hidden layers of S27 and 128; S28, 64 fully connected layer 1; S29, 6 fully connected layer 2; S3. Train the model 200 times using the training set, using early stopping, and take the best result as the training result. S4. Use a Panasonic WM-61A microphone to sample the audio of the actual target transformer being measured. S5. On the Kendryte K230 microcontroller system, run this neural network algorithm, input the audio signal collected by the actual target transformer, and determine whether the classification output is normal or not, and the type of abnormality.

Claims

1. A system for identifying and detecting abnormal noises inside a transformer cylinder based on a neural network, characterized in that, include: a) Training set construction module, used to collect sound under normal and various abnormal conditions of the transformer using a microphone system, and to label it in order to construct a training set; b) A model building module for designing a neural network model that includes Fast Fourier Transform (FFT) units, Convolutional Neural Network (CNN) layers, pooling layers, Gated Recurrent Units (GRUs), and fully connected layers to classify input audio data; c) Model training module, used to train the designed neural network model using the constructed training set; d) Audio data acquisition module, used to acquire sound during transformer operation using a microphone system; e) Actual target sound acquisition module, used to sample audio from the actual target transformer using a microphone system; f) Model running module, used to run the trained neural network model on a microcontroller or SoC system, to identify and classify the actual target sound, and output whether it is normal or not and the type of abnormal sound.

2. The system according to claim 1, characterized in that, The training set construction module further includes: using a microphone system to sample audio, collecting sounds under one normal operating condition and at least six abnormal conditions, and labeling them to construct a training set.

3. The system according to claim 1, characterized in that, The model building module further includes: using Fast Fourier Analysis (FFT) to extract frequency domain features from the input audio signal to obtain spectral data; using at least one Convolutional Neural Network (CNN) layer to perform convolution operations on the spectrum of the audio signal to extract local features; using at least one pooling layer to compress the data scale; using CNN and pooling again to further extract features; using a gated recurrent unit (GRU) to capture the temporal dependencies in the audio signal; and finally, using a fully connected layer to classify and output the type of abnormal noise.

4. The system according to claim 1, characterized in that, The model training module further includes: training the designed neural network model using the constructed training set and obtaining the best training results.

5. The system according to claim 1, characterized in that, Both the audio data acquisition module and the actual target sound acquisition module use the same microphone system for audio sampling.

6. The system according to claim 1, characterized in that, The model running module runs a trained neural network model on a microcontroller or SoC system to identify and classify the actual target sound, and outputs the result of whether it is normal or not and the type of abnormal sound. The microcontroller system includes, but is not limited to, the Kendryte K230 microcontroller.

7. The system according to claim 1, characterized in that, The neural network model further includes: a 32x1024 convolutional layer 1, a 32x512 pooling layer 1, a 64x512 convolutional layer 2, a 64x256 pooling layer 2, a 256x128 GRU layer, a 128 hidden layer, a 64-layer fully connected layer 1, and a 6-layer fully connected layer 2.

8. The system according to claim 1, characterized in that, The microphone system includes, but is not limited to, the Panasonic WM-61A microphone system.

Images