Rotating machinery fault identification method based on multi-source information fusion

Abstract

The invention relates to the technical field of fault detection, in particular to a rotating machine fault identification method based on multi-source information fusion, which comprises the following steps: collecting noise and vibration data of a target rotating machine; identifying and separating interference noise data generated by other sound production parts outside the target rotating machine; correcting the noise data of the target rotating machine by using the dry noise data; performing fusion processing on the corrected noise data and the vibration data, generating a comprehensive fault feature, matching the comprehensive fault feature with a known fault mode, and identifying a specific fault type; and generating a fault diagnosis report based on the fault type. Therefore, the accuracy of fault detection can be improved, the maintenance cost can be effectively reduced, and the service life of equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to a rotating machinery fault identification method based on multi-source information fusion. Background Art

[0002] Rotating machinery is a very common type of equipment in industrial production, including but not limited to motors, pumps, fans, compressors, etc. During operation, these mechanical equipment may have various faults due to design defects, manufacturing errors, improper installation, insufficient maintenance or operating errors.

[0003] In order to effectively prevent and deal with the above faults, it is necessary to regularly inspect and maintain the equipment, use advanced monitoring technologies such as vibration analysis and temperature monitoring to promptly detect potential problems, and take corresponding maintenance measures. The existing use of noise for fault analysis cannot properly handle the interference of external noise on the noise of rotating parts, thereby reducing the accuracy of fault judgment. Summary of the invention

[0004] The purpose of the present invention is to provide a rotating machinery fault identification method based on multi-source information fusion, aiming to reduce the influence of external noise, improve the accuracy of fault detection, and effectively reduce maintenance costs and extend the service life of equipment.

[0005] To achieve the above object, the present invention provides a rotating machinery fault identification method based on multi-source information fusion, comprising collecting noise and vibration data of a target rotating machinery; Identify and separate interfering noise data generated by other sound-generating components outside the target rotating machinery; The dry noise data is used to correct the noise data of the target rotating machinery; The corrected noise data is fused with the vibration data to generate comprehensive fault features and match them with known fault modes to identify specific fault types; Generate a fault diagnosis report based on the fault type.

[0006] The vibration data is collected by a plurality of vibration sensors installed on the rotating machine, and the noise data is acquired by a plurality of microphones arranged around the rotating machine.

[0007] The specific steps of collecting the noise and vibration data of the target rotating machinery include: Send a synchronous pulse signal through the central controller to start all sensors to collect data synchronously; During the data collection process, add a timestamp to each piece of data; Perform preliminary noise reduction processing on the collected noise data; The noise and vibration data are transformed into the frequency domain to extract spectral features.

[0008] The specific steps of identifying and separating the interference noise data generated by other sound-generating components outside the target rotating machinery include: Build a voiceprint library to store voiceprint features of known external noise sources; The support vector machine is trained using data from the voiceprint library to enable the model to identify and classify different noise sources; Independent component analysis is used to separate the independent interference noise data from the noise data.

[0009] The specific steps of constructing a voiceprint library to store voiceprint features of known external noise sources include: Set the type of external noise source around the rotating machinery, including fan, pump, motor, and wind noise; Record noise samples for each noise source under different operating conditions; Noise features of the noise source are stored based on the noise samples, where the noise features include the type of noise source, the acquisition time, and the acquisition location.

[0010] The specific steps of separating each independent interference noise data from the noise data by using independent component analysis include: organizing the noise data collected by multiple channels into a mixed signal matrix; Centralize the mixed signal matrix; Whiten the centralized data so that the covariance matrix of the data is the unit matrix; Initialize the weight matrix to represent the mixing matrix; Iteratively update the weight matrix using an optimization algorithm; The independent components are calculated using the final weight matrix.

[0011] The specific steps of using the dry noise data to correct the noise data of the target rotating machinery include: aligning the separated external noise data and the original noise data in time; For each time point, the corresponding external noise data was subtracted from the original noise data.

[0012] The specific steps of fusing the corrected noise data with the vibration data, generating a comprehensive fault feature and matching it with a known fault mode, and identifying a specific fault type include: Combining the feature vectors of the noise data and the vibration data into a comprehensive feature vector; Assigning weights to noise signatures and vibration signatures; The comprehensive feature vector is classified using a multimodal support vector machine to obtain a fault feature database; The fused fault signatures are matched with known fault patterns in the fault signature database to identify the specific fault type.

[0013] The rotating machinery fault identification method based on multi-source information fusion of the present invention first needs to install appropriate sensors, such as acoustic sensors and vibration sensors, on the target rotating machinery to collect noise and vibration data during the operation of the equipment. Since there are often multiple sound sources in the actual environment, these sound sources may generate noise that is unrelated to the target rotating machinery, namely the so-called interference noise. The task of this stage is to identify and separate these interference noises from the collected noise data in order to more accurately analyze the real state of the target machinery. This usually involves complex signal processing techniques, such as spectrum analysis, waveform analysis, etc.

[0014] Using dry noise (i.e. pure noise without interference) data as a reference, the noise data of the separated target rotating machinery is corrected. This process helps to eliminate or reduce the impact of environmental factors on the measurement results and improve the purity and reliability of the data. The corrected noise data is combined with the vibration data and fused through a specific algorithm or model (such as a machine learning algorithm, a deep learning model, etc.). The purpose is to generate a comprehensive fault feature that can fully reflect the operating status of the mechanical equipment.

[0015] Compare the comprehensive fault characteristics obtained in the above steps with the pre-established known fault mode database. Through the matching process, it is possible to determine whether the current mechanical equipment has a fault and what type of fault it is. Based on the results of fault identification, a detailed fault diagnosis report is generated. The report should include the type of fault, possible causes, recommended maintenance measures, etc., to provide clear operating instructions for maintenance personnel. In addition, preventive maintenance suggestions can be made based on actual conditions to avoid similar problems that may occur in the future. This can improve the accuracy of fault detection, effectively reduce maintenance costs, and extend the service life of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a flow chart of the rotating machinery fault identification method based on multi-source information fusion of the present invention.

[0018] Figure 2 It is a flow chart of collecting noise and vibration data of a target rotating machine according to the present invention.

[0019] Figure 3 It is a flow chart of the present invention for identifying and separating the interfering noise data generated by other sound-generating components outside the target rotating machinery.

[0020] Figure 4 It is a flow chart of constructing a voiceprint library to store voiceprint features of known external noise sources according to the present invention.

[0021] Figure 5 It is a flowchart of using independent component analysis to separate each independent interference noise data from the noise data of the present invention.

[0022] Figure 6 The present invention is a flow chart of using dry noise data to correct the noise data of a target rotating machine.

[0023] Figure 7 The present invention fuses the corrected noise data with the vibration data to generate a comprehensive fault feature and matches it with a known fault mode to identify a specific fault type. DETAILED DESCRIPTION

[0024] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0025] See also Figure 1 to Figure 7 The present invention provides a rotating machinery fault identification method based on multi-source information fusion, comprising: S101 collects noise and vibration data of target rotating machinery; In the process of mechanical parts rotating, many different types of data can actually be generated, such as temperature data, power data of driving components, etc. These data have certain reference value for a comprehensive understanding of the operating status of the equipment. Temperature data can reflect the changes in the thermal state of mechanical equipment during operation, which is very important for preventing failures caused by overheating; the power data of driving components can reveal the working efficiency and load conditions of power components such as motors, which helps to evaluate energy efficiency and discover potential power system problems.

[0026] However, despite the value of the above data, it is often difficult to obtain these data in practical applications. First, the installation location, quantity and accuracy of temperature sensors and power monitoring devices will affect the quality of the data. Improper installation will cause data distortion, which in turn affects the accuracy of the analysis results. Secondly, the correlation between these data and equipment failures is relatively weak, that is, their changes do not directly point to specific failure modes, which makes the effectiveness and reliability of fault diagnosis based on these data low. Therefore, when selecting a data source for fault judgment, priority is usually given to data types that are strongly correlated with the fault, easy to obtain and can provide clear diagnostic information, such as vibration data and noise data.

[0027] The vibration data is collected through multiple vibration sensors installed on rotating mechanical equipment. These sensors can capture subtle vibration changes generated during the operation of the equipment, providing key information for monitoring the health status of the equipment and predicting failures. The noise data is obtained through multiple microphones set up around the rotating mechanical equipment. These microphones can record the sounds generated when the equipment is running. By analyzing these sounds, we can further understand the working status of the equipment and possible problems.

[0028] It is worth noting that different types of rotating machinery have their own unique structural characteristics and working methods, so there are also differences in the layout of sensors and microphones. In order to ensure that data with diagnostic value can be effectively collected and to avoid data redundancy or invalidity due to improper sensor layout, it is usually necessary to determine the most appropriate layout through experiments based on the specific equipment type and application scenario. This process may involve multiple adjustments and optimizations to ensure that the collected data can fully and accurately reflect the actual operating conditions of the equipment, thereby providing a reliable basis for subsequent analysis and decision-making.

[0029] The specific steps include: S201 sends a synchronous pulse signal through the central controller to start all sensors to collect data synchronously; Configure the central controller to ensure that it can send synchronization pulse signals. The central controller can be a high-performance microprocessor or a dedicated data acquisition device. Connect the vibration sensor and microphone to the central controller. Ensure that the communication interface and protocol of all sensors are consistent so that the central controller can manage them uniformly. Send a synchronization pulse signal through the central controller, which is transmitted to all sensors through a hardware interface (such as a TTL signal). After all sensors receive the synchronization pulse signal, they immediately start to collect data synchronously.

[0030] S202 adds a timestamp to each piece of data during the data collection process; Generate accurate timestamps in the central controller to ensure that each data point has a unique time identifier. During the data collection process, record the timestamps together with the collected noise and vibration data. The timestamp can be an absolute time (such as UTC time) or a relative time (such as the time from the start of the collection). Ensure that the data format is consistent to facilitate subsequent data processing and analysis. For example, the data can be saved as a CSV file with each line containing a timestamp, noise data, and vibration data.

[0031] S203 performs preliminary noise reduction processing on the collected noise data; Select appropriate noise reduction algorithms, such as spectral subtraction, wavelet transform, adaptive filtering, etc. These algorithms can effectively remove background noise and retain the main noise source characteristics.

[0032] According to the characteristics of the noise data, set the parameters of the denoising algorithm. For example, for spectral subtraction, you can set the threshold; for wavelet transform, you can choose the appropriate wavelet basis function and the number of decomposition levels.

[0033] The collected noise data is subjected to noise reduction processing to generate noise data after noise reduction.

[0034] S204 converts the noise data and the vibration data into the frequency domain to extract frequency spectrum features.

[0035] Use Fast Fourier Transform (FFT) to convert noise data and vibration data from the time domain to the frequency domain. FFT can decompose the signal into components of different frequencies, which is convenient for subsequent feature extraction. Spectral features such as band energy and harmonic components can be extracted from frequency domain data. For example, the energy distribution of different frequency bands can be calculated, and the harmonic frequency and amplitude can be extracted. The extracted spectral features are combined into feature vectors to facilitate subsequent fault diagnosis.

[0036] S102 identifies and separates interference noise data generated by other sound-generating components outside the target rotating machinery; The specific steps include: S301 builds a voiceprint library to store voiceprint features of known external noise sources; The specific steps include: S401 sets the type of external noise source existing around the rotating machinery, wherein the noise source type includes fan, pump, motor, and wind noise; Identify the types of external noise sources that may be present around rotating machinery. Common types of noise sources include: Fan: Fan for cooling system and ventilation system.

[0037] Pump: A pump in a liquid delivery system.

[0038] Motor: An electric machine that drives rotating machinery.

[0039] Wind noise: The natural sound of wind or air flow in the environment.

[0040] Record these noise source types in the voiceprint library to ensure that each noise source is clearly identified.

[0041] S402 records noise samples of each noise source under different working conditions; Identify different operating conditions, such as different speeds, loads, temperatures, etc., to ensure that the noise samples collected are representative. Place highly sensitive microphones near each noise source to ensure that noise data can be collected from different positions and angles. Ensure that all microphones start collecting data at the same time to maintain time synchronization. Use additional sensors to record environmental conditions during collection, such as temperature, humidity, background noise, etc., so that the impact of these factors can be considered in subsequent analysis. Label each noise sample with the corresponding noise source type, operating conditions, and collection time.

[0042] S403 stores noise characteristics of the noise source based on the noise sample, where the noise characteristics include the type of noise source, the collection time, and the collection location.

[0043] Extract features from the collected noise samples. Common features include: Noise source type: records the noise source type corresponding to each noise sample.

[0044] Acquisition time: Record the acquisition time of each noise sample to ensure the accuracy of the timestamp.

[0045] Collection location: Record the location where each noise sample was collected, including the specific microphone position.

[0046] Frequency domain features: Perform fast Fourier transform (FFT) on the noise data to extract spectrum features, such as frequency band energy, harmonic components, etc.

[0047] Time domain features: Extract the time domain features of noise data, such as root mean square value (RMS), peak value, kurtosis, etc.

[0048] Time-frequency features: Use short-time Fourier transform (STFT) or wavelet transform to extract the time-frequency features of the noise signal.

[0049] The extracted noise features are stored in the voiceprint library to form a structured database. The database table may include the following fields: Type of noise source: e.g. fan, pump, motor, wind noise, etc.

[0050] Acquisition Time: The acquisition time of the noise samples.

[0051] Collection location: The location where the noise samples were collected.

[0052] Frequency domain characteristics: frequency domain characteristics of noise samples, such as frequency band energy, harmonic components, etc.

[0053] Time domain characteristics: time domain characteristics of noise samples, such as RMS value, peak value, kurtosis, etc.

[0054] Time-frequency characteristics: time-frequency characteristics of noise samples, such as STFT or wavelet transform results.

[0055] Data annotation: Add labels to each noise sample, including the noise source type, collection time, collection location, etc., for subsequent analysis and identification.

[0056] S302 uses data from the voiceprint library to train a support vector machine so that the model can identify and classify different noise sources; Divide the data into training and test sets, usually with a ratio of 70% training set and 30% test set. Ensure that the types of noise sources in the training and test sets are evenly distributed.

[0057] Use feature selection methods (such as recursive feature elimination, principal component analysis, etc.) to select the most representative features, reduce feature dimensions, and improve model training efficiency and accuracy.

[0058] Select support vector machine as the classification model. SVM performs well in processing high-dimensional data and multi-classification problems, and is suitable for noise source identification tasks. Then select a suitable kernel function, such as linear kernel, polynomial kernel, radial basis function (RBF) kernel, etc. Select the most appropriate kernel function according to the characteristics of the data.

[0059] Use the training set data to train the support vector machine model. During the training process, you can use optimization algorithms such as batch gradient descent and stochastic gradient descent to ensure model convergence. Save the trained model for noise source identification in practical applications. Use the test set data to evaluate the performance of the model, calculate indicators such as classification accuracy, recall rate, F1 score, etc., to ensure that the model has good generalization ability.

[0060] S303 uses independent component analysis to separate each independent interference noise data from the noise data.

[0061] The specific steps include: S501 organizing the noise data collected by multiple channels into a mixed signal matrix; Multiple high-sensitivity microphones are arranged around the rotating machinery to ensure that noise data can be collected from different positions and angles. The noise data collected by multiple channels are sorted into a mixed signal matrix X, where each column represents the noise data collected by a microphone and each row represents the noise data at a time point.

[0062] S502 performs centralization processing on the mixed signal matrix; Calculate Mean: Calculate the mean μ of each column of the mixed signal matrix X.

[0063] Centering subtracts the mean of each column of the mixed signal matrix X so that the mean of the data is zero.

[0064] Xcentered=X−1μ T Among them, 1 is the full 1 vector, and μ is the mean vector of each column of data.

[0065] S503 performs whitening processing on the centralized data so that the covariance matrix of the data is a unit matrix; Calculate the covariance matrix: Calculate the covariance matrix C of the centered data.

[0066] C=1 / nX centered X centered T Where n is the number of data points.

[0067] Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalue λi and eigenvector vi.

[0068] Construct a whitening matrix W such that WX centered The covariance matrix of is the identity matrix.

[0069] W=VΛ −1 / 2 V T Where V is the eigenvector matrix and Λ is a diagonal matrix whose diagonal elements are the square roots of the eigenvalues.

[0070] Whitening: Multiply the centralized data by the whitening matrix W to obtain the whitened data X whitened =WX centered .

[0071] S504 initializes a weight matrix to represent a mixing matrix; A weight matrix A is randomly initialized, representing the mixing matrix. The size of the weight matrix is ​​m×m, where m is the number of noise sources.

[0072] The initial weight matrix can be randomly generated using a uniform distribution or a normal distribution.

[0073] S505 iteratively updates the weight matrix using an optimization algorithm; Select an appropriate optimization algorithm, such as the FastICA algorithm, to separate independent components by maximizing the non-Gaussianity of the signal. Use the gradient ascent method or the fixed point algorithm to iteratively update the weight matrix AA until convergence. By calculating the gradient and updating the weight matrix, the optimal solution is gradually approached. The weight matrix is ​​directly updated through the fixed point iteration method to speed up convergence.

[0074] S506 calculates independent components using the final weight matrix.

[0075] Calculate independent components: Use the final weight matrix A to calculate the independent components S.

[0076] S=A −1 X whitened Where S is the separated independent component matrix, and each column represents an independent noise source.

[0077] Result verification: Compare the separated independent components with known noise sources to verify the separation effect. The separation effect can be evaluated by visualizing waveform and spectral characteristics, and calculating indicators such as signal-to-noise ratio (SNR).

[0078] S103 uses the dry noise data to correct the noise data of the target rotating machinery; The specific steps include: S601 aligns the separated external noise data and the original noise data in time; Ensure that the separated external noise data and the original noise data have the same timestamp. If the timestamps are inconsistent, timestamp correction is required. Use the timestamp alignment method to align the timestamps of the separated external noise data and the original noise data. Interpolation methods (such as linear interpolation and spline interpolation) can be used to generate missing time point data.

[0079] Ensure that the two data sets are completely synchronized in time. This can be achieved by using the synchronization pulse signal sent by the central controller to ensure that all sensors start collecting data at the same time. During the data processing stage, select the same time window for analysis to ensure that the time range of the data is consistent.

[0080] S602: For each time point, the corresponding external noise data is subtracted from the original noise data.

[0081] Point-by-point subtraction: For each time point t, from the original noise data N original (t) minus the corresponding external noise data N external (t).

[0082] N clean (t) = N original (t)−Nexternal (t) Among them, N clean (t) is the pure target mechanism noise data.

[0083] In some cases, negative values ​​may appear after external noise is subtracted. Negative values ​​can be set to zero or handled using other methods.

[0084] N clean (t)=max(N clean (t),0) Verify the corrected noise data to ensure that the data after external noise removal is in line with expectations. Visualize the original noise data, the separated external noise data, and the pure target mechanism noise data, compare the waveform and spectrum characteristics, and verify the removal effect. Calculate the spectrum of the corrected noise data, observe the changes in the spectrum, and ensure that the external noise has been effectively removed.

[0085] The signal-to-noise ratio (SNR) of the corrected noise data was calculated to evaluate the effect of noise reduction.

[0086] S104 fuses the corrected noise data with the vibration data to generate a comprehensive fault feature and matches it with a known fault mode to identify a specific fault type; The specific steps include: S701 combines the feature vectors of the noise data and the vibration data into a comprehensive feature vector; The feature vectors of the noise data and the vibration data are combined into a comprehensive feature vector.

[0087] F=[F noise ,F vibration ] Among them, F noise is the feature vector of the noise data, F vibration is the eigenvector of the vibration data.

[0088] S702 assigns weights to noise features and vibration features; Different weights are assigned to noise features and vibration features according to different fault types and application scenarios. The weight selection can be determined based on experience or through experiments.

[0089] F weighted =w noise F noise +w vibration F vibration Among them, w noise and w vibration are the weights of noise and vibration features.

[0090] In practical applications, the weights can be dynamically adjusted according to the specific requirements of the fault type to optimize the diagnosis results.

[0091] S703 uses a multimodal support vector machine to classify the comprehensive feature vector to obtain a fault feature database; The comprehensive feature vector and the corresponding fault label form a training set. A part of the data is divided as a test set to evaluate the performance of the model. In order to reduce the training cost of the model, the present invention uses small sample data to train the model.

[0092] Multimodal support vector machine (SVM) was selected as the classification model.

[0093] Adjust model parameters (such as kernel function, penalty parameter C, etc.) to obtain the best classification effect. Use the training set data to train the multimodal SVM model.

[0094] Use the test set data to evaluate the performance of the model and calculate indicators such as classification accuracy, recall rate, and F1 score.

[0095] The trained multimodal SVM model is stored in the fault feature database. The comprehensive feature vectors and fault labels in the training set are stored in the fault feature database. The fault types in the fault feature database include rotation imbalance, bearing failure, seal failure, material deformation, impurity ingress, etc.

[0096] Rotational imbalance: When the mass of a rotating part is unevenly distributed, the axis of rotation will deviate from the center, causing abnormal vibration. This fault is common in high-speed rotating equipment, such as fans and pumps, and in severe cases it will affect the stability and service life of the equipment.

[0097] Bearing failure: Bearing is one of the indispensable key components in rotating machinery, and its performance directly affects the operating status of the entire system. Bearing failure may be caused by poor lubrication, overload, wear and tear, and manifests as abnormal vibration, increased noise and other phenomena.

[0098] Seal failure: The seal in mechanical equipment is used to prevent external pollutants from entering or internal media from leaking. Once the seal fails, it may not only cause environmental pollution, but also cause damage to other components.

[0099] Material deformation: After long-term operation or under extreme conditions, some parts of mechanical equipment may undergo plastic deformation or fatigue damage. This change will affect the normal working performance of the equipment and even cause equipment failure.

[0100] Ingress of impurities: If dust, particles and other impurities in the external environment enter the interior of mechanical equipment, they may cause increased wear on the friction pair surface, reduce transmission efficiency, or even clog the pipeline, affecting the normal operation of the system.

[0101] By storing the above fault types and their corresponding feature vectors in the fault feature database, rich reference information can be provided for subsequent fault diagnosis. When new fault data is input into the system, the fault type can be quickly identified by comparing it with the existing data in the database, and corresponding maintenance measures can be taken to effectively ensure the safe and stable operation of mechanical equipment.

[0102] S704 matches the fused fault signature with known fault patterns in the fault signature database to identify the specific fault type.

[0103] Collect noise data and vibration data from the rotating machinery to be diagnosed and extract comprehensive feature vectors. Input the extracted comprehensive feature vectors into the trained multimodal SVM model for classification prediction. According to the prediction results of the model, identify the specific fault type. For example, the model may output "bearing fault", "gear fault", etc.

[0104] Compare the identification results with historical fault data to verify the consistency of the identification results.

[0105] S105 generates a fault diagnosis report based on the fault type.

[0106] Generates detailed fault diagnosis reports including fault type, fault location, severity and repair suggestions.

[0107] Through the above steps, the corrected noise data and vibration data can be effectively fused and processed to generate comprehensive fault features, and matched with known fault patterns to identify specific fault types, thereby improving the accuracy and reliability of rotating machinery fault diagnosis. The rotating machinery fault identification method of the present invention has an accuracy rate of about 80% in identifying the type of faulty parts. It can quickly identify the fault type by comparing the existing data in the database, and take corresponding maintenance measures to effectively ensure the safe and stable operation of mechanical equipment.

[0108] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. Rotating machinery fault identification method based on multi-source information fusion, It is characterized in that Includes: Collecting noise and vibration data of target rotating machinery; Identify and separate interfering noise data generated by other sound-generating components outside the target rotating machinery; The dry noise data is used to correct the noise data of the target rotating machinery; The corrected noise data is fused with the vibration data to generate comprehensive fault features and match them with known fault modes to identify specific fault types; Generate a fault diagnosis report based on the fault type.

2. The rotating machinery fault identification method based on multi-source information fusion according to claim 1, characterized in that: The vibration data is collected by a plurality of vibration sensors installed on the rotating machine, and the noise data is acquired by a plurality of microphones arranged around the rotating machine.

3. The rotating machinery fault identification method based on multi-source information fusion according to claim 2, characterized in that: The specific steps of collecting the noise and vibration data of the target rotating machinery include: Send a synchronous pulse signal through the central controller to start all sensors to collect data synchronously; During the data collection process, add a timestamp to each piece of data; Perform preliminary noise reduction processing on the collected noise data; The noise and vibration data are transformed into the frequency domain to extract spectral features.

4. The rotating machinery fault identification method based on multi-source information fusion according to claim 3 is characterized in that: The specific steps of identifying and separating the interference noise data generated by other sound-generating components outside the target rotating machinery include: Build a voiceprint library to store voiceprint features of known external noise sources; The data in the voiceprint library is used to train the support vector machine so that the model can identify and classify different noise sources; Independent component analysis is used to separate the independent interference noise data from the noise data.

5. The rotating machinery fault identification method based on multi-source information fusion according to claim 4, characterized in that: The specific steps of constructing a voiceprint library to store voiceprint features of known external noise sources include: Set the type of external noise source around the rotating machinery, including fan, pump, motor, and wind noise; Record noise samples for each noise source under different operating conditions; Noise features of the noise source are stored based on the noise samples, where the noise features include the type of noise source, the acquisition time, and the acquisition location.

6. The rotating machinery fault identification method based on multi-source information fusion according to claim 5, characterized in that: The specific steps of separating each independent interference noise data from the noise data by using independent component analysis include: organizing the noise data collected by multiple channels into a mixed signal matrix; Centralize the mixed signal matrix; Whiten the centralized data so that the covariance matrix of the data is the unit matrix; Initialize the weight matrix to represent the mixing matrix; Iteratively update the weight matrix using an optimization algorithm; The independent components are calculated using the final weight matrix.

7. The rotating machinery fault identification method based on multi-source information fusion according to claim 6, characterized in that: The specific steps of using the dry noise data to correct the noise data of the target rotating machinery include: aligning the separated external noise data and the original noise data in time; For each time point, the corresponding external noise data was subtracted from the original noise data.

8. The rotating machinery fault identification method based on multi-source information fusion according to claim 7, characterized in that: The specific steps of fusing the corrected noise data with the vibration data, generating a comprehensive fault feature and matching it with a known fault mode, and identifying a specific fault type include: Combining the feature vectors of the noise data and the vibration data into a comprehensive feature vector; Assigning weights to noise signatures and vibration signatures; The comprehensive feature vector is classified using a multimodal support vector machine to obtain a fault feature database; The fused fault signatures are matched with known fault patterns in the fault signature database to identify the specific fault type.

Images