Equipment defect analysis method and system based on voiceprint system and transfer learning

Abstract

The invention discloses an equipment defect analysis method and system based on a voiceprint system and transfer learning, and relates to the field of power plant equipment fault diagnosis and health monitoring, and the method comprises the steps: collecting an audio signal of equipment operation in real time; performing data representation and feature mapping based on the preprocessed audio signal to generate a first equipment defect analysis model; training the first equipment defect analysis model based on a transfer learning convolutional neural network to obtain a second equipment defect analysis model; analyzing through a second equipment defect analysis model, and outputting an equipment defect analysis result; according to the method, the problems of insufficient data and distribution difference in the target field are solved, and the generalization ability of the model and the defect detection precision are improved. The method can be widely applied to fault monitoring of power plant key equipment such as a fan, a pump and a steam turbine, and has high efficiency, adaptability and reliability.

Description

Technical Field

[0001] The present invention relates to the field of power plant equipment fault diagnosis and health monitoring, and particularly to a method and system for equipment defect analysis based on a voiceprint system and transfer learning. Background Art

[0002] With the improvement of the automation level of power industry equipment, equipment health monitoring and fault diagnosis technologies have become important means to ensure the safe operation of power plants.

[0003] Traditional vibration and ultrasonic signal analysis methods require a lot of manual experience and are difficult to effectively handle diverse equipment status data. In recent years, equipment monitoring based on voiceprint technology has gradually become a research hotspot. It extracts features through the operating sound of equipment, but there are still the following problems: insufficient data: there are few defective audio samples of specific equipment, and the model generalization ability is limited; feature extraction: traditional acoustic features cannot fully express equipment status information; adaptability: the model has insufficient migration ability among various equipment and scenarios. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, the technical problem to be solved by the present invention is: how to optimize the generalization ability of the model by combining a convolutional neural network (CNN) and transfer learning, and improve the accuracy and efficiency of defect detection.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for equipment defect analysis based on a voiceprint system and transfer learning, including:

[0008] Real-time collecting audio signals of equipment operation;

[0009] Based on the preprocessed audio signals, performing data representation and feature mapping to generate a first equipment defect analysis model;

[0010] Training the first equipment defect analysis model based on a transfer learning convolutional neural network to obtain a second equipment defect analysis model;

[0011] Analyzing through the second equipment defect analysis model and outputting equipment defect analysis results.

[0012] As a preferred scheme of the method for equipment defect analysis based on a voiceprint system and transfer learning, wherein:

[0013] The analyzing through the second equipment defect analysis model and outputting equipment defect analysis results includes:

[0014] Input the audio features to be analyzed into the second device defect analysis model;

[0015] Infer the input audio features and output the device status classification.

[0016] As a preferred solution of the device defect analysis method based on the voiceprint system and transfer learning, where:

[0017] The analysis by the second device defect analysis model and output of the device defect analysis result further includes:

[0018] Classify the device status of the analysis model and generate a device status diagnosis report;

[0019] Connect to the real-time monitoring system, set a judgment threshold according to the device status diagnosis report, and trigger an alarm or intervention measure.

[0020] As a preferred solution of the device defect analysis method based on the voiceprint system and transfer learning, where:

[0021] The generation of the first device defect analysis model includes:

[0022] Load the pre-trained model, lock the convolutional kernels in the lower layers of the pre-trained model, and retain its ability to extract low-level features;

[0023] Add a fully connected layer at the end of the pre-trained model to map the features to the target classification categories;

[0024] Use the alignment method of transfer learning to adjust the feature distributions of the source data and the target data, reduce the differences between domains, so as to generate the first device defect analysis model.

[0025] As a preferred solution of the device defect analysis method based on the voiceprint system and transfer learning, where:

[0026] The data representation includes:

[0027] Represent the collected audio signal x(t) as:

[0028]

[0029] where F is the Fourier transform and X(f) is the frequency domain representation; the generated spectrogram S is calculated as:

[0030] S(t, f) = log(|X(t, f)| 2 )

[0031] where S(t, f) is the spectrogram, representing the logarithmic energy spectrum of the signal at each time t and frequency f, and X(t, f) is the result of the short-time Fourier transform, representing the complex-valued spectrum of the signal at different times t and frequencies f.

[0032] As a preferred solution of the device defect analysis method based on the voiceprint system and transfer learning, wherein:

[0033] The feature mapping includes:

[0034] Convert the spectrogram into the Mel spectrogram M through the Mel filter:

[0035]

[0036] where H k (f) is the weight function of the k-th Mel filter.

[0037] As a preferred solution of the device defect analysis method based on the voiceprint system and transfer learning, wherein:

[0038] The training of the first device defect analysis model by the transfer learning convolutional neural network includes:

[0039] Adopt a convolutional neural network based on transfer learning, and its loss function is:

[0040]

[0041] where CE is the cross-entropy loss, N represents the number of samples in the target domain, y i is the true label of the i-th sample, is the predicted probability distribution of the model for the i-th sample; λ is a weight parameter used to balance the classification loss and the domain adaptation loss; MMD is the maximum mean discrepancy used for feature alignment between the source domain and the target domain; φ is the CNN feature extraction mapping, φ(S t ) is the feature representation of the target domain data, φ(S s ) is the feature representation of the source domain data.

[0042] In a second aspect, an embodiment of the present invention provides a device defect analysis system based on a voiceprint system and transfer learning, including:

[0043] An audio acquisition and preprocessing module for real-time acquisition of audio signals during device operation;

[0044] A feature extraction and model generation module for performing data representation and feature mapping based on the preprocessed audio signals to generate a first device defect analysis model;

[0045] A training module for training the first device defect analysis model based on a transfer learning convolutional neural network to obtain a second device defect analysis model;

[0046] An analysis module for analyzing through the second device defect analysis model and outputting the device defect analysis result.

[0047] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0048] a memory and a processor;

[0049] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the device defect analysis method based on the voiceprint system and transfer learning as described in any embodiment of the present invention.

[0050] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a processor, the device defect analysis method based on the voiceprint system and transfer learning is implemented.

[0051] Advantages of the present invention: The present invention combines transfer learning technology, uses a pre-trained model to extract general features, and aligns the feature distributions of the source domain and the target domain through the maximum mean discrepancy (MMD), solving the problems of insufficient data in the target domain and distribution differences. The loss function optimizes both the classification accuracy and the feature distribution consistency, thereby improving the generalization ability of the model and the accuracy of defect detection. The present invention can be widely applied to the fault monitoring of key equipment in power plants such as fans, pumps, and steam turbines, and has high efficiency, adaptability, and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0053] Figure 1 is the overall flowchart of the device defect analysis method based on the voiceprint system and transfer learning described in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] In the following description, many specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways than those specifically described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0056] Secondly, as used herein, "one embodiment" or "an embodiment" refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other.

[0057] Embodiment 1

[0058] Referring to Figure 1 , which is the first embodiment of the present invention, this embodiment provides a method for analyzing device defects based on a voiceprint system and transfer learning, including:

[0059] S1: Real-time collect the audio signals of the device during operation;

[0060] S2: Based on the preprocessed audio signals, perform data representation and feature mapping to generate a first device defect analysis model;

[0061] S3: Train the first device defect analysis model based on a transfer learning convolutional neural network to obtain a second device defect analysis model;

[0062] S4: Analyze through the second device defect analysis model and output the device defect analysis result.

[0063] It should be noted that through steps S1 - S4, a comprehensive monitoring and defect analysis of the device operation state can be achieved; the audio signals of the device operation can be collected in real time, and the signal quality can be improved through preprocessing. Subsequently, through feature extraction and mapping, the audio signals are converted into a data form that can be processed by a machine learning model. Finally, a device defect analysis model is trained using a transfer learning-based convolutional neural network, and this model is applied to accurately analyze device defects.

[0064] Embodiment 2

[0065] Referring to Figure 1 , which is an embodiment of the present invention. Based on the previous embodiment, a method for analyzing device defects based on a voiceprint system and transfer learning is provided, including:

[0066] In the embodiment of the present application, the real-time collection of the audio signals of the device during operation in the above step S1 includes:

[0067] Arrange multiple high - sensitivity microphone arrays around the key equipment in the power plant (such as steam turbines, pumps, fans, etc.) to ensure comprehensive coverage of audio signals.

[0068] It should be noted that high - sensitivity microphone devices are used to collect the sound signals of the equipment under different operating states. The collected samples should include various working conditions (such as normal operation, minor faults, serious faults), and cover different load conditions as much as possible.

[0069] In another possible implementation, the position and number of microphones can also be dynamically adjusted according to the equipment operating state and environmental changes to better cover the complex equipment operating environment.

[0070] Intelligent sensors with computing capabilities can also be used, which can perform preliminary signal processing and compression at the acquisition end to reduce the amount of transmitted data.

[0071] Denoise, down - sample and normalize the collected original audio signals to eliminate environmental noise and equipment background interference and ensure data quality.

[0072] Specifically, the pre - processing includes annotating the audio samples as "normal" and "fault" categories according to the equipment operation records.

[0073] Denoising processing: Use filtering algorithms (such as low - pass filtering) to remove background noise and electromagnetic interference.

[0074] Signal segmentation: Segment the continuous audio into segments with a fixed duration (such as 2 seconds) for subsequent processing.

[0075] Normalization: Standardize the signal amplitude to a fixed range to eliminate the influence of volume differences.

[0076] When processing the equipment sound signals, normalization is used to scale the audio feature values (such as spectrogram data, MFCC features, etc.) to a fixed range (usually [0, 1]) to eliminate the dimensional differences between different feature values. The specific formula is as follows:

[0077]

[0078] where \(x\) i，j represents the \(j\) - th feature value of the \(i\) - th audio sample. For example, the amplitude of a certain frequency component in the spectrogram or a certain dimension value of the MFCC feature. \(x\) i,j norm represents the result after normalizing the \(j\) - th feature value of the \(i\) - th audio sample. After normalization, all feature values will be scaled to the range of [0, 1]. \(x\) min,j represents the minimum value of the \(j\) - th feature value in the entire audio sample dataset. For example, the global minimum value of a certain frequency component in the spectrogram data. \(x\) max,jDenote the maximum value of the j-th eigenvalue in the entire audio sample dataset. For example, the maximum value of the 1st dimension feature of MFCC in all samples.

[0079] In another possible implementation, an adaptive filtering algorithm (such as LMS, RLS) can also be used to dynamically adjust the filtering parameters to more effectively remove background noise and interference.

[0080] Multi-microphone synchronization technology can also be used to ensure that the signals collected by different microphones are aligned in time, improving the integrity of the signals.

[0081] The preprocessed signal can also be encoded and compressed to reduce data transmission and storage costs while retaining key information.

[0082] In the embodiment of the present application, based on the preprocessed audio signal in the above step S2, data representation and feature mapping are performed to generate the first device defect analysis model, including:

[0083] Feature extraction is performed on the audio signal. The audio signal is converted into a spectrogram through the short-time Fourier transform (STFT) to show the change of frequency over time. The mel-frequency cepstral coefficients method is used to extract the frequency-domain features of the sound, which is suitable for capturing features within the human ear's audible range.

[0084] Specifically, the collected audio signal x(t) is expressed as:

[0085]

[0086] where F is the Fourier transform and X(f) is the frequency-domain representation. The generated spectrogram S is calculated as:

[0087] S(t, f) = log(|X(t, f)| 2 )

[0088] where S(t, f) is the spectrogram, representing the logarithmic energy spectrum of the signal at each time t and frequency f, and X(t, f) is the result of the short-time Fourier transform, representing the complex-valued spectrum of the signal at different times t and frequencies f.

[0089] In another possible implementation, wavelet transform (CWT), time-frequency distribution

[0090] (such as Wigner-Ville distribution) and other methods can also be used for time-frequency analysis; combined with multi-scale analysis methods (such as multi-resolution analysis), features of the audio signal are extracted from different scales to improve the richness and robustness of the features.

[0091] The spectrogram is converted into a mel spectrogram M through a mel filter:

[0092]

[0093] Among them, H k (f) is the weight function of the k-th Mel filter.

[0094] In another possible implementation, feature extraction methods such as linear predictive coding (LPC), zero crossing rate (ZCR), fundamental frequency (F0), etc. can also be used.

[0095] More advanced deep learning models (such as deep convolutional generative adversarial network DCGAN, autoencoder AE) can also be used for feature extraction to automatically learn complex audio features.

[0096] Take the audio features (spectrogram or MFCC matrix) obtained by feature extraction as the input of the model. These features represent the time-frequency information of the audio and can reflect the operating state of the device.

[0097] The device defect analysis model includes:

[0098] Load a pre-trained model: Select a deep learning model (such as ResNet, MobileNet) trained on a large-scale dataset (such as ImageNet or LibriSpeech).

[0099] Freeze the lower convolutional layers of the model (used for extracting general features), and only retain the parameters of these layers to ensure that the basic feature extraction ability is not affected.

[0100] Unfreeze the higher-level network so that it can adapt to the target task through transfer learning.

[0101] Add a classification layer: Add a fully connected layer at the end of the model to map the features to the target classification categories (such as normal, fault 1, fault 2, etc.).

[0102] Feature mapping: Map the input audio features to a high-dimensional feature space through the convolutional and fully connected layers of the pre-trained model.

[0103] Exemplarily, the input of the spectrogram may be mapped to a 512-dimensional feature vector, representing the high-level abstract features of the audio.

[0104] Target classification: Add a custom classifier (such as a fully connected layer + Softmax activation function) on the mapped high-dimensional feature space to further map the features to the target classification categories (such as "normal", "bearing fault", "gear fault", etc.).

[0105] Output the classification result: The classifier outputs the probability distribution of each class, and finally predicts the class to which the current state of the device belongs.

[0106] Feature alignment: Use alignment methods based on transfer learning (such as Maximum Mean Discrepancy, MMD) to adjust the feature distributions of source data (such as laboratory equipment data) and target data (actual power plant data), reducing the differences between domains.

[0107] Specifically, feature alignment is carried out during the model training process and is part of the feature mapping, rather than an adjustment after classification is completed.

[0108] Through alignment methods such as Maximum Mean Discrepancy (MMD), adjust the feature distributions between the source domain (pre-trained data) and the target domain (equipment audio data), making their distributions in the high-dimensional feature space as consistent as possible.

[0109] Calculate the difference in feature distributions between the source domain and the target domain on the intermediate feature layer of the model, and optimize this part of the error during training.

[0110] The process of feature alignment can be regarded as a constraint condition for model training, aiming to improve the generalization ability of the classifier in the target task.

[0111] In another possible implementation, other pre-trained models (such as MobileNet, DenseNet, EfficientNet) can also be used as the base model.

[0112] Combine multiple models (such as the fusion of CNN and RNN) to improve the robustness and generalization ability of the model.

[0113] In the embodiments of the present application, training the first device defect analysis model based on the transfer learning convolutional neural network in step S3 to obtain the second device defect analysis model includes:

[0114] The model training uses the data of the target domain to optimize the classifier and the unfrozen high-level parameters, while minimizing the following two losses:

[0115] Classification loss (Cross-Entropy): Measures the difference between the predicted class of the model and the true label.

[0116] Transfer loss (MMD or other methods): Measures the difference in feature distributions between the source domain and the target domain.

[0117] After training is completed, the model can not only accurately distinguish the device states of the target domain, but also has strong generalization ability.

[0118] Specifically, adopt a convolutional neural network based on transfer learning, and its loss function is:

[0119]

[0120] Among them, CE is the cross-entropy loss, N represents the number of samples in the target domain, yi is the true label of the i-th sample, is the predicted probability distribution of the model for the i-th sample; λ is a weight parameter used to balance the classification loss and the domain adaptation loss; MMD is the maximum mean discrepancy used for feature alignment between the source domain and the target domain; φ is the CNN feature extraction mapping, and φ(S t ) is the feature representation of the target domain data, and φ(S s ) is the feature representation of the source domain data.

[0121] In another possible implementation, for the hybrid loss function: in addition to the cross-entropy loss (CE) and the maximum mean discrepancy (MMD), other loss functions (such as KL divergence, Huber loss) can be introduced to improve the robustness of the model.

[0122] According to the importance of different classes (for example, the weight of a serious fault is greater), the loss function is weighted to optimize the model's recognition ability for important classes.

[0123] Use data augmentation techniques (such as noise injection, time stretching, frequency masking) to increase the diversity of training samples and improve the generalization ability of the model.

[0124] Introduce the batch normalization technique during training to accelerate the convergence speed of the model and improve the stability of training.

[0125] In addition to the maximum mean discrepancy (MMD), other transfer learning techniques (such as domain adversarial training, adaptive domain normalization) can be used for feature alignment.

[0126] After training, prune the model to reduce the number of model parameters and improve the running efficiency of the model.

[0127] Quantize the model to convert floating-point parameters into fixed-point parameters, reducing the storage and computing resource requirements of the model.

[0128] Optimize the model deployment method, such as using lightweight frameworks (such as TensorFlow Lite, ONNX Runtime) for model deployment to improve the running efficiency of the model in the actual environment.

[0129] In the embodiment of the present application, in the above step S4, analysis is performed through the second device defect analysis model, and the output device defect analysis results include:

[0130] The device status classification of the analysis model is generated to produce a device status diagnosis report;

[0131] Connect to the real-time monitoring system, set a judgment threshold according to the device status diagnosis report, and trigger an alarm or an intervention measure.

[0132] Specifically, preprocess and extract features from the audio signal to be analyzed, and input them into the trained device defect analysis model. The model infers the input audio features and outputs device status classifications including normal operation, minor faults, and severe faults; analyze the output results of the analysis model to generate a device status diagnosis report; interface with the real-time monitoring system, and trigger alarms or intervention measures including device shutdown and maintenance scheduling according to the model analysis results.

[0133] Exemplarily, during real-time operation, input new audio features into the trained model, and output target classification results, such as: "The device is operating normally (probability 0.85)"; "Bearing fault (probability 0.10)"; "Other faults (probability 0.05)".

[0134] Embodiment 3

[0135] The above is a schematic solution of the device defect analysis method based on the voiceprint system and transfer learning in this embodiment. It should be noted that the technical solutions of the device defect analysis system based on the voiceprint system and transfer learning belong to the same concept as the technical solutions of the device defect analysis method based on the voiceprint system and transfer learning. For the details not described in the technical solutions of the device defect analysis system based on the voiceprint system and transfer learning in this embodiment, reference can be made to the description of the technical solutions of the device defect analysis method based on the voiceprint system and transfer learning.

[0136] This embodiment also provides a system for the device defect analysis method based on the voiceprint system and transfer learning, including:

[0137] An audio acquisition and preprocessing module for real-time acquisition of the audio signal of the device operation;

[0138] A feature extraction and model generation module for performing data representation and feature mapping based on the preprocessed audio signal to generate a first device defect analysis model;

[0139] A training module for training the first device defect analysis model based on the transfer learning convolutional neural network to obtain a second device defect analysis model;

[0140] An analysis module for analyzing through the second device defect analysis model and outputting the device defect analysis result.

[0141] This embodiment also provides a computing device applicable to the situation of the device defect analysis method based on the voiceprint system and transfer learning, including:

[0142] A memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the device defect analysis method based on the voiceprint system and transfer learning as proposed in the above embodiments.

[0143] This embodiment also provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the device defect analysis method based on the voiceprint system and transfer learning as proposed in the above embodiments.

[0144] The storage medium proposed in this embodiment and the device defect analysis method based on the voiceprint system and transfer learning proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be referred to the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0145] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A device defect analysis method based on voiceprint system and transfer learning, characterized in that: include: Collect audio signals of equipment operation in real time; Based on the preprocessed audio signal, perform data representation and feature mapping to generate a first equipment defect analysis model; The first equipment defect analysis model is trained based on a transfer learning convolutional neural network to obtain a second equipment defect analysis model; The second equipment defect analysis model is used to perform analysis and output equipment defect analysis results.

2. The device defect analysis method based on voiceprint system and transfer learning according to claim 1, characterized in that: The analyzing by the second equipment defect analysis model and outputting the equipment defect analysis result comprises: Inputting the audio features to be analyzed into the second device defect analysis model; Infer the input audio features and output device status classification.

3. The device defect analysis method based on voiceprint system and transfer learning as claimed in claim 2, characterized in that: The performing analysis by the second device defect analysis model and outputting the device defect analysis result further comprises: Analyze the equipment status classification of the model and generate equipment status diagnosis report; Connect to the real-time monitoring system, set judgment thresholds based on equipment status diagnostic reports, and trigger alarms or intervention measures.

4. The device defect analysis method based on voiceprint system and transfer learning as claimed in claim 3, characterized in that: The generating of the first equipment defect analysis model comprises: Load the pre-trained model and lock the convolution kernels at the lower layers of the pre-trained model to retain its ability to extract low-level features. Add a fully connected layer at the end of the pre-trained model to map features to target classification categories; Using the alignment method of transfer learning, the feature distributions of the source data and the target data are adjusted to reduce the differences between domains to generate the first equipment defect analysis model.

5. The device defect analysis method based on voiceprint system and transfer learning as claimed in claim 4, characterized in that: The data representation includes: The collected audio signal x(t) is expressed as: Where F is the Fourier transform and X(f) is the frequency domain representation; the generated spectrum S is calculated as: S(t,f)=log(|X(t,f)| 2 ) Where S(t,f) is the spectrogram, which represents the logarithmic energy spectrum of the signal at each time t and frequency f, and X(t,f) is the result of the short-time Fourier transform, which represents the complex-valued spectrum of the signal at different times t and frequencies f.

6. The device defect analysis method based on voiceprint system and transfer learning according to claim 5, characterized in that: The feature map includes: Convert the spectrogram to a Mel spectrum M through a Mel filter: Among them, H k (f) is the weight function of the kth Mel filter.

7. The device defect analysis method based on voiceprint system and transfer learning according to claim 6, characterized in that: The training of the first equipment defect analysis model based on the transfer learning convolutional neural network includes: A convolutional neural network based on transfer learning is used, and its loss function is: Where CE is the cross entropy loss, N is the number of samples in the target domain, and y i is the true label of the i-th sample, is the predicted probability distribution of the model for the i-th sample; λ is a weight parameter used to balance the classification loss and domain adaptation loss; MMD is the maximum mean difference, which is used to align the features of the source domain and the target domain; φ is the CNN feature extraction mapping, φ(S t ) is the feature representation of the target domain data, φ(S s ) is the feature representation of the source domain data.

8. A system using the device defect analysis method based on voiceprint system and transfer learning as described in any one of claims 1 to 7, characterized in that: include: Audio acquisition and preprocessing module, used to collect audio signals of equipment operation in real time; A feature extraction and model generation module, used for performing data representation and feature mapping based on the preprocessed audio signal to generate a first device defect analysis model; A training module, used for training the first device defect analysis model based on a transfer learning convolutional neural network to obtain a second device defect analysis model; The analysis module is used to perform analysis using a second device defect analysis model and output device defect analysis results.

9. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the device defect analysis method based on the voiceprint system and transfer learning as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the device defect analysis method based on a voiceprint system and transfer learning as described in any one of claims 1 to 7.

Images