A multi-source data fusion fault diagnosis method and system based on DS evidence theory

Abstract

This invention belongs to the field of signal processing technology, specifically relating to a multi-source data fusion fault diagnosis method and system based on DS evidence theory. The invention first collects multi-source information from mechanical equipment, including vibration, image, and sound data, calculates corresponding feature indicators for each, then uses the PCA algorithm to reduce the dimensionality of these indicators, employs a Bayesian fault diagnosis model for fault diagnosis, and after obtaining the diagnostic results from each sensor, uses DS evidence theory for fault fusion diagnosis. This invention can comprehensively consider the characteristics of various types of data, extract more effective fault indicators, achieve more comprehensive fault identification, and fuse the diagnostic results of multiple sensors to ensure the accuracy of the final diagnosis. It is applicable to fault diagnosis of various complex mechanical systems.

Description

Technical Field

[0001] The invention belongs to the technical field of signal processing, and in particular relates to a multi-source data fusion fault diagnosis method and system based on DS evidence theory. Background Art

[0002] Mechanical fault diagnosis algorithms are crucial in modern industry. They can detect potential hazards in a timely manner through real-time monitoring to ensure the safety of employees and assets. They can also extend the service life of equipment, reduce equipment wear, identify factors that affect equipment performance by analyzing operating data, and make maintenance recommendations.

[0003] The operating data of mechanical equipment comes from a wide range of sources, including vibration data, sound data, image data, etc. The diversity of these data makes fault diagnosis face challenges such as inconsistent information and missing information. Through multi-source data fusion, we can fully utilize the advantages of different data sources to improve the accuracy and robustness of fault diagnosis. The fusion algorithm can comprehensively consider the characteristics of various types of data to achieve more comprehensive fault identification. The multi-source data fusion alarm diagnosis algorithm based on the Dempster-Shafer (DS) theory is gradually becoming an important research direction in the field of mechanical fault diagnosis due to its advantages in dealing with uncertainty and information fusion.

[0004] Existing methods or studies have one or more of the following defects and deficiencies:

[0005] 1. Many existing algorithms are still based on traditional methods, such as Kalman filtering and fuzzy logic, and lack innovative algorithms for new failure modes.

[0006] 2. Although the research on multi-sensor data fusion has increased, the data processing technology for the fusion of different types of sensors (such as vibration, temperature, acoustics, etc.) is still insufficient.

[0007] 3. Although deep learning has achieved remarkable results in image processing and natural language processing, its application in fault diagnosis is still in its early stages and there are few related patents.

[0008] 4. Existing patents are mostly focused on a single algorithm or technology, lack comprehensive solutions, and cannot effectively deal with complex mechanical system failures.

[0009] 5. In a high-speed, dynamically changing environment, existing technologies still have limitations in data processing speed and real-time performance, and it is difficult to meet the needs of industrial sites.

[0010] 6. The data collected by the sensor has many characteristic indicators, and it is difficult to extract effective indicators, resulting in deviations in alarm and diagnosis. Summary of the invention

[0011] The purpose of the present invention is to provide a multi-source data fusion fault diagnosis method and system based on DS evidence theory to solve the technical problems of existing mechanical fault diagnosis algorithms when processing multi-source data, such as information inconsistency, insufficient fusion technology, lack of comprehensive solutions, poor real-time performance, and difficulty in extracting effective feature indicators.

[0012] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0013] In the first aspect, the present invention proposes a multi-source data fusion fault diagnosis method based on DS evidence theory, the method comprising:

[0014] Collect multi-source information of the target device, including vibration data, image data and sound data;

[0015] Determine and extract characteristic indicators corresponding to the vibration data, the image data and the sound data;

[0016] Using principal component analysis algorithm to perform dimension reduction processing on the characteristic index;

[0017] According to the feature index after dimensionality reduction, a pre-built fault diagnosis model based on time-frequency domain features and naive Bayes is trained, and a first diagnosis result of the fault of the device to be tested is determined based on the fault diagnosis model;

[0018] The first diagnosis result is integrated by using DS evidence theory to determine the second diagnosis result. Further, the determining and extracting the characteristic indicators corresponding to the vibration data, the image data and the sound data includes:

[0019] Determine and extract characteristic indicators of vibration data of multiple target devices, including acceleration effective value, acceleration peak-to-peak value, acceleration pulse, acceleration margin, acceleration kurtosis, acceleration skewness, velocity effective value, velocity peak-to-peak value, velocity pulse, velocity margin, velocity kurtosis and velocity skewness, to form a first characteristic set;

[0020] Determine and extract characteristic indicators of sound data of multiple target devices, including effective value, loudness, soundprint roughness, frequency center of gravity, sharpness, high-low frequency band energy ratio, peak-to-peak value, impulse, margin, kurtosis and skewness, to form a second feature set;

[0021] Determine and extract feature indicators of multiple target device image data, including texture features, edge features, temperature distribution features and crack distribution features, to form a third feature set.

[0022] Furthermore, the principal component analysis algorithm is used to perform dimension reduction processing on the characteristic index, including:

[0023] Calculate the covariance matrix of each feature set after standardization;

[0024] Calculate the eigenvalues ​​and eigenvectors of each covariance matrix based on a numerical algorithm;

[0025] Determine the target eigenvector according to the size of the eigenvalue, and construct the principal component space based on it;

[0026] The standardized covariance matrix is ​​projected into the principal component space for matrix multiplication to obtain the feature index after dimensionality reduction.

[0027] Furthermore, the method of training a pre-built fault diagnosis model based on time-frequency domain features and naive Bayes according to the feature index after dimensionality reduction, and determining a first diagnosis result of the fault of the device under test based on the fault diagnosis model includes:

[0028] Determine the target device fault type according to the target device fault mode or historical fault record, use the fault type as a category label, and form a training set including feature indicators after dimensionality reduction and corresponding fault category labels;

[0029] Calculate the prior probability of each fault type in the training set ;

[0030] Calculate the likelihood probability of each characteristic indicator under each fault type ;

[0031] Calculate the characteristic index of the device under test after dimensionality reduction through Bayesian theorem The a posteriori probability of belonging to each fault type is , select the fault type with the maximum a posteriori probability as the first diagnostic result, and the expression is:

[0032] ;

[0033] ;

[0034] ;

[0035] in, Indicates the fault type; Is the fault type The prior probability of express In category The conditional probability under ; Represents a given feature In case of failure type The posterior probability of represents the normalization constant for all fault types; arg max is the choice The fault class that maximizes the posterior probability.

[0036] Furthermore, the adopting of DS evidence theory to fuse the first diagnosis result to determine the second diagnosis result includes:

[0037] Converting the first diagnosis result into an evidence body in DS evidence theory, each of the evidence bodies represents the degree of trust of a signal source on the fault type, the signal sources including vibration sensors, image sensors and sound sensors;

[0038] Construct a trust function and assign a trust function to each fault type in each signal source;

[0039] Merging the credibility functions of the plurality of evidence bodies to obtain a merged credibility function;

[0040] According to the combined credibility function, the fault type with the maximum credibility value is selected as the second diagnosis result.

[0041] Further, selecting the fault type with the maximum reliability value as the second diagnosis result according to the combined reliability function includes:

[0042] updating the combined credibility function based on the set conflict management strategy, and selecting the fault type corresponding to the maximum value of the credibility function from the updated credibility function as the second diagnosis result;

[0043] The conflict management strategy set includes:

[0044] (1) quantifying the degree of conflict between the different signal sources based on the conflict metric, and when the degree of conflict exceeds a preset threshold, selecting to exclude the conflicting information from the evidence synthesis to update the combined credibility function; the conflicting information is one or more of inconsistent diagnosis results, inconsistent probability distribution, inconsistent feature indicators, and inconsistent time or space;

[0045] (2) Adjust the weight of conflicting evidence based on the reliability of the evidence source or historical data to update the merged credibility function.

[0046] In a second aspect, the present invention proposes a multi-source data fusion fault diagnosis system based on DS evidence theory, which is applied to execute any of the multi-source data fusion fault diagnosis methods based on DS evidence theory described above, and the system comprises:

[0047] A data acquisition module is used to collect multi-source information of the target device, the multi-source information includes vibration data, image data and sound data;

[0048] A feature extraction module, used to extract feature indicators corresponding to the vibration data, the image data and the sound data;

[0049] A principal component analysis module, used for performing dimensionality reduction processing on the characteristic index by using a principal component analysis algorithm;

[0050] A first diagnosis module is used to train a pre-built fault diagnosis model based on time-frequency domain features and naive Bayes according to the feature index after dimensionality reduction, and determine a first diagnosis result of the fault of the device under test based on the fault diagnosis model;

[0051] The second diagnosis module is used to fuse the first diagnosis results using DS evidence theory to determine a second diagnosis result.

[0052] The beneficial effects of the present invention are:

[0053] The present invention can more comprehensively reflect the equipment status and improve the accuracy of fault identification by comprehensively collecting multi-source information such as vibration, image, sound, etc. of mechanical equipment and calculating the corresponding characteristic indicators. The PCA algorithm is used to reduce the dimension of the characteristic indicators, effectively extract the features with the highest contribution, and further improve the accuracy of alarm and diagnosis. At the same time, the Bayesian fault diagnosis model and DS evidence theory are combined for fault fusion diagnosis, which not only enhances the system's comprehensive analysis capabilities for various types of data, but also significantly improves the system's anti-interference capabilities against noise and external interference. Even in the face of partial sensor or model failures, the present invention can ensure that the overall system maintains a high diagnostic level, providing a strong guarantee for the stable operation of mechanical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flowchart of a multi-source data fusion fault diagnosis method based on DS evidence theory provided in an embodiment of the present application;

[0055] Figure 2 A schematic diagram of another process of a multi-source data fusion fault diagnosis method based on DS evidence theory provided in an embodiment of the present application;

[0056] Figure 3 A structural diagram of a multi-source data fusion fault diagnosis system based on DS evidence theory provided in an embodiment of the present application;

[0057] Figure 4 This is a schematic diagram of vibration data in the case section of the specific implementation method of this application;

[0058] Figure 5 This is a schematic diagram of the sound data in the case section of the specific implementation method of this application;

[0059] Figure 6 This is a schematic diagram of image data of the case part of the specific implementation method of this application;

[0060] Figure 7This is a scatter plot of some vibration data in the case of the specific implementation method of this application after PCA dimensionality reduction processing. DETAILED DESCRIPTION

[0061] The present application is further described in detail below in conjunction with the accompanying drawings. It is necessary to point out here that the following specific implementation methods are only used to further illustrate the present application and cannot be understood as limiting the scope of protection of the present application. Technical personnel in this field can make some non-essential improvements and adjustments to the present application based on the above application content. Example

[0062] like Figure 1-2 As shown, this embodiment proposes a multi-source data fusion fault diagnosis method based on DS evidence theory, characterized in that the method includes the following steps:

[0063] S1. Collect multi-source information of the target device, including vibration data, image data and sound data;

[0064] In specific implementation, sensors installed on mechanical equipment collect multi-source information. Vibration sensors are used to collect vibration data of equipment, sound sensors are used to collect sound data of equipment, and image sensors are used to collect image data of equipment. The data collected by these sensors in real time will be processed later.

[0065] S2. Determine and extract characteristic indicators corresponding to vibration data, image data and sound data;

[0066] Preferably, determining and extracting characteristic indicators corresponding to the vibration data, image data and sound data includes:

[0067] Determine and extract characteristic indicators of vibration data of multiple target devices, including acceleration effective value, acceleration peak-to-peak value, acceleration pulse, acceleration margin, acceleration kurtosis, acceleration skewness, velocity effective value, velocity peak-to-peak value, velocity pulse, velocity margin, velocity kurtosis, and velocity skewness, to form a first characteristic set;

[0068] Determine and extract characteristic indicators of sound data of multiple target devices, including effective value, loudness, soundprint roughness, frequency center of gravity, sharpness, high-low frequency band energy ratio, peak-to-peak value, impulse, margin, kurtosis and skewness, to form a second feature set;

[0069] Determine and extract feature indicators of multiple target device image data, including texture features, edge features, temperature distribution features and crack distribution features, to form a third feature set.

[0070] In step S2, the vibration sensor data feature extraction includes extracting the time domain and frequency domain feature data of the vibration data. The specific indicators are as follows:

[0071] (1) Vibration acceleration amplitude. The vibration amplitude refers to the maximum deviation value of the vibration signal and is usually used to characterize the intensity of vibration.

[0072] (2) Effective value of acceleration, which indicates the intensity of vibration and is often used to measure the energy or power of vibration.

[0073] (3) Margin index: Margin is the ratio of the absolute value of the peak value to the root square amplitude, which reflects the impact characteristics of the signal. If this value is too large, it means that the signal has a strong impact and no kurtosis sensitivity.

[0074] (4) Kurtosis index: Kurtosis reflects the impact characteristics of the vibration signal. Kurtosis is sensitive to impact. Generally, the kurtosis value should be around 3, because the kurtosis of the normal distribution is equal to 3. If it deviates too much from 3, it means that the mechanical equipment has a certain impact vibration and there may be some hidden dangers of failure. Kurtosis is more sensitive than margin, but the indicator stability is poor.

[0075] (5) Skewness index: Skewness reflects the asymmetry of the vibration signal. Normally, the vibration signal is symmetrical about the x-axis, and the skewness should be close to 0. If the friction or collision in one direction of the equipment is large, it will cause vibration asymmetry and increase the skewness.

[0076] (6) Peak-to-peak value: The meaning of peak-to-peak value is similar to peak value, which reflects the variation range of the maximum and minimum values ​​of the vibration waveform.

[0077] (7) The pulse reflects the overall fluctuation of the vibration data, and the calculation formula is the maximum value of the data divided by the absolute average value.

[0078] In step S2, the sound sensor data features are extracted, and the specific indicators are as follows:

[0079] (1) Short-time zero-crossing rate (ST-ZCR): It can reflect the fluctuation degree of the sound signal and simply reflect the frequency domain information from the time domain. The original sound signal needs to pass through an anti-aliasing bandpass filter first, with a low-end cutoff frequency higher than 50 Hz to filter out environmental noise and power supply interference.

[0080] (2) Short-time energy (STE): used to distinguish the boundaries of sounds.

[0081] (3) Linear Prediction Coefficient (LPC): The basic idea is that the human vocal system is a linear time-invariant system, and the sound sampling data with similar time have a strong correlation.

[0082] (4) Linear Prediction Cepstral Coefficients (LPCC): Calculated by LPC, the last C represents the cepstral domain. Like LPC, LPCC models linear systems. Compared with LPC, LPCC uses less data to express more effective information and is easy to implement.

[0083] (5) Mel spectrum coefficient (MFSC): Mel spectrum converts non-uniformly distributed frequencies into uniformly distributed Mel frequencies (the effect is better when the signal-to-noise ratio is low). The main process of the sound MFSC extraction algorithm is: framing, windowing, fast Fourier transform, modulus calculation, Mel filtering, and logarithm calculation (mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=40)).

[0084] (6) Mel Cepstral Coefficient (MFCC): After discrete cosine transformation, MFCC outputs less data, has lower dimensions, weaker correlation, and carries less information than MFSC, so it is generally necessary to calculate the differential coefficients of different orders of MFCC in combination. (Preprocessing, adding Hamming window, FFT transformation, passing through Mel triangular filter bank, calculating mean and variance, calculating transposition, performing FFT transformation, performing logarithmic operation through filter bank, and performing DCT transformation.)

[0085] (7) GFCC coefficient: Use the Gamma-tone filter group to filter the sound.

[0086] (8) Peak value and corresponding frequency of the power spectrum: According to the size of the spectrum energy, the frequency and corresponding peak value (energy) corresponding to the first 8 peaks are taken as the feature vectors.

[0087] (9) Loudness: The loudness of a sound is a subjective indicator that describes the intensity of the sound perceived by the human ear. Loudness is used to reflect the properties of the auditory judgment of the strength of a sound. It is related not only to the sound pressure, but also to factors such as frequency and waveform.

[0088] (10) Sharpness: The Sharpness index describes the timbre characteristics in the evaluation of sound quality. The sharpness of sound describes the feeling that is related to the frequency component of the sound but not to the loudness of the sound. It reflects the degree of sharpness, high frequency, and discomfort of the sound, and is the comparison between the high-frequency energy and the total energy in the sound.

[0089] In step S2, the features of the image data are extracted, and the specific indicators include:

[0090] (1) Texture features Texture features can characterize the state of the equipment surface through the grayscale value, edge, directionality, roughness, etc. of the image. When a device fails, different texture features often appear on the surface, which can reflect wear, cracks, corrosion, etc.

[0091] (2) Edge Features: Edge features are used to identify mutations, defects, and changes in the surface or internal structure of a device. Faults often result in significant changes in the edge, such as cracks, wear, or the presence of foreign matter.

[0092] (3) Thermal Imaging Features In thermal imaging images, faults are usually manifested as uneven heat distribution. For example, faults such as motor overheating and bearing wear may cause local temperature rise, and these abnormalities can be detected through temperature distribution maps.

[0093] (4) Crack Detection Image processing technology can effectively detect cracks on the surface or inside of equipment, especially in structural parts, welding points, pipelines, etc. The occurrence of cracks is usually a precursor to failure.

[0094] S3. Use principal component analysis algorithm to reduce the dimension of feature indicators.

[0095] It should be noted that before applying the principal component analysis algorithm (PCA), the data needs to be preprocessed to ensure data quality and applicability.

[0096] The steps include:

[0097] (1) Data cleaning: Delete or correct missing values, outliers, and erroneous data.

[0098] (2) Standardization: Since PCA relies on the covariance matrix, and the size of the covariance is affected by the scale of each feature, the data needs to be standardized or normalized to ensure that all features have the same scale. The mean of the standardized data is 0 and the standard deviation is 1.

[0099] Preferably, a principal component analysis algorithm is used to reduce the dimension of the characteristic index, including:

[0100] Calculate the covariance matrix of each feature set after standardization;

[0101] Calculate the eigenvalues ​​and eigenvectors of each covariance matrix based on numerical algorithms (such as QR algorithm, Jacobi method, etc.);

[0102] Determine the target eigenvector according to the size of the eigenvalue, and use it to construct the principal component space; specifically, select the first few principal components whose cumulative explained variance reaches a predetermined threshold (such as 90% or 95%) according to the size of the eigenvalue to form a new feature space;

[0103] The standardized covariance matrix is ​​projected into the principal component space for matrix multiplication to obtain the feature index after dimensionality reduction.

[0104] S4. According to the feature indicators after dimensionality reduction, a pre-built fault diagnosis model based on time-frequency domain features and naive Bayes is trained, and a first diagnostic result of the fault of the device under test is determined based on the fault diagnosis model;

[0105] Preferably, according to the feature index after dimensionality reduction, a pre-built fault diagnosis model based on time-frequency domain features and naive Bayes is trained, and a first diagnosis result of the fault of the device under test is determined based on the fault diagnosis model, including:

[0106] Determine the target device fault type according to the target device fault mode or historical fault record, use the fault type as a category label, and form a training set including feature indicators after dimensionality reduction and corresponding fault category labels;

[0107] Calculate the prior probability of each fault type in the training set ;

[0108] Calculate the likelihood probability of each characteristic indicator under each fault type ;

[0109] Calculate the characteristic index of the device under test after dimensionality reduction through Bayesian theorem The posterior probability of belonging to each fault type , select the fault type with the maximum posterior probability as the first diagnosis result, the expression is:

[0110] ;

[0111] ;

[0112] ;

[0113] in, Indicates the fault type; Is the fault type The prior probability of express In category The conditional probability under ; Represents a given feature In case of failure type The posterior probability of represents the normalization constant for all fault types; arg max is the choice The fault class that maximizes the posterior probability.

[0114] S5. Use DS evidence theory to integrate the first diagnosis result to determine the second diagnosis result.

[0115] Preferably, the DS evidence theory is used to fuse the first diagnosis result to determine the second diagnosis result, including:

[0116] The first diagnosis result is converted into an evidence body in the DS evidence theory, each evidence body represents the degree of trust of a signal source on the fault type, and the signal sources include vibration sensors, image sensors and sound sensors;

[0117] Construct a trust function and assign a trust function to each fault type in each signal source;

[0118] The credibility functions of multiple evidence bodies are merged to obtain a combined credibility function;

[0119] According to the combined credibility function, the fault type with the maximum credibility value is selected as the second diagnosis result.

[0120] In some optional embodiments, according to the combined credibility function, selecting the fault type with the maximum credibility value as the second diagnosis result includes:

[0121] updating the merged credibility function based on the set conflict management strategy, and selecting the fault type corresponding to the maximum value of the credibility function from the updated credibility function as the second diagnosis result;

[0122] The conflict management strategies set include:

[0123] (1) Based on the conflict metric, the conflict degree between different signal sources is quantified. When the conflict degree exceeds a preset threshold, the conflict information is excluded from the evidence synthesis to update the combined credibility function. The conflict information is one or more of inconsistent diagnosis results, inconsistent probability distribution, inconsistent feature indicators, and inconsistent time or space.

[0124] (2) Adjust the weight of conflicting evidence based on the reliability of the evidence source or historical data to update the merged credibility function.

[0125] More specifically, regarding the definition of the above conflict information:

[0126] Inconsistent diagnostic results: Different signal sources give different diagnostic results for the same fault type.

[0127] Inconsistent probability assignment: In the Bayesian fault diagnosis model, there are significant differences in the probability assignment of different signal sources to the same fault type.

[0128] Inconsistent feature indicators: After PCA dimensionality reduction processing, the feature indicators extracted from each signal source have obvious differences in values ​​or change trends.

[0129] Inconsistency in time or space: There is a mismatch between different signal sources in the time point or spatial location of the fault. For example, the vibration sensor detects the fault earlier than the image sensor.

[0130] In the specific implementation, after obtaining the first diagnosis result of each sensor, DS evidence theory is used to perform fault fusion diagnosis. The specific steps are as follows:

[0131] (1) The first diagnosis result is converted into an evidence body in the DS evidence theory. Each evidence body represents the degree of trust of a signal source (such as vibration sensor, image sensor, and sound sensor) in the fault type.

[0132] (2) Construct a basic probability assignment (BPA) to assign a credibility function value to each fault type in each signal source, indicating the credibility of the fault type.

[0133] (3) The credibility functions of multiple evidence bodies are fused to obtain a combined credibility function. During the fusion process, the conflict management strategy set above is used to deal with the conflicts and uncertainties between the evidences.

[0134] (4) According to the combined reliability function, the fault type with the maximum reliability value is selected as the second diagnosis result. This fault type is the most likely fault type and is used to guide subsequent maintenance and fault handling.

[0135] It should be further explained that in the actual application of updating the merged credibility function based on the set conflict management strategy in this embodiment, different evidence sources may give conflicting information. The DS theory is used to measure these conflicts through the conflict degree (K value). The greater the conflict degree, the lower the credibility of the final diagnosis result.

[0136] Specific steps:

[0137] During the synthesis process, when a conflict occurs, it needs to be handled. There are usually two ways to handle it:

[0138] (1) Ignore conflicting information: exclude conflicting information from the synthesis.

[0139] (2) Adjusting the BPA: Adjust the final BPA by reducing the weight of conflicting evidence.

[0140] After completing the update of the fusion belief function, subsequent diagnostic reasoning is performed based on the synthesis results.

[0141] In this embodiment, according to the combined credibility function, the fault type with the maximum credibility value is selected as the second diagnosis result. In specific implementation, the following steps are included:

[0142] From the fused BPA, the fault type with the largest BPA value is selected as the final diagnosis result. This fault type is the most likely fault type.

[0143] If the BPA values ​​of certain fault types are very close, further diagnosis can be performed to confirm the final result.

[0144] According to the above-mentioned embodiments of the present invention, the present invention can comprehensively consider the vibration, image and sound data of mechanical equipment, extract more effective fault indicators, and realize more comprehensive fault identification. At the same time, through PCA dimensionality reduction processing and the construction and training of Bayesian fault diagnosis model, the complexity and computational cost of data processing are reduced. Finally, fault fusion diagnosis is performed through DS evidence theory, which improves the accuracy and reliability of fault diagnosis. This method can significantly improve the operating efficiency and safety of mechanical equipment in practical applications, and reduce maintenance costs and production losses.

[0145] Combination Figure 3 In another embodiment of the present invention, a multi-source data fusion fault diagnosis system based on DS evidence theory is proposed, which is applied to execute the multi-source data fusion fault diagnosis method based on DS evidence theory. The system includes:

[0146] A data acquisition module is used to collect multi-source information of the target device, the multi-source information includes vibration data, image data and sound data; accordingly, the data acquisition module includes a vibration sensor, an image sensor and a sound sensor;

[0147] A feature extraction module is used to extract feature indicators corresponding to vibration data, image data and sound data;

[0148] The principal component analysis module is used to perform dimension reduction processing on feature indicators using the principal component analysis algorithm;

[0149] A first diagnosis module is used to train a pre-built fault diagnosis model based on time-frequency domain features and naive Bayes according to the feature index after dimensionality reduction, and determine a first diagnosis result of the fault of the device under test based on the fault diagnosis model;

[0150] The second diagnosis module is used to fuse the first diagnosis result by using DS evidence theory to determine the second diagnosis result.

[0151] It should be noted here that each module in the above-mentioned fault diagnosis system corresponds to steps S1 to S5 in implementing the above-mentioned fault diagnosis method, and the instances and application scenarios implemented by multiple modules and corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned embodiment 1.

[0152] In order to more clearly illustrate the present invention and its advantages, the fault diagnosis method provided by the present invention will be further explained below in combination with specific cases and related partial figures.

[0153] 1. Data Collection

[0154] The sensor collects the vibration data, sound and image data of the actual bearing on site and inputs them into the server. The server may include the feature extraction module, the principal component analysis module, the first diagnosis module and the second diagnosis module of the fault diagnosis system in the above embodiment; the corresponding vibration data is as follows: Figure 4 As shown ( Figure 4 The horizontal axis represents the corresponding time, and the vertical axis represents the vibration amplitude). The sound data is as follows Figure 5 As shown ( Figure 5 The horizontal axis represents the corresponding time, and the vertical axis represents the sound waveform amplitude). The device image data is as follows Figure 6 As shown, the PCA diagram after dimensionality reduction is as follows Figure 7 As shown ( Figure 7 Each axis of the coordinate system corresponds to a principal component. The first principal component (PC1) is the direction with the largest variance in the data, and the second principal component (PC2) is the direction orthogonal to the first principal component).

[0155] 2. Data feature extraction

[0156] 2.1. Extract characteristic indicators of vibration data, according to Figure 4 The corresponding vibration index is calculated from the collected acceleration vibration waveform data, and the extracted indexes are 12 indexes: acceleration effective value, acceleration peak-to-peak value, acceleration pulse, acceleration margin, acceleration kurtosis, acceleration skewness, velocity effective value, velocity peak-to-peak value, velocity pulse, velocity margin, velocity kurtosis, and velocity skewness. The vibration characteristic index array is calculated as follows:

[0157] a = [12, 106, 5.47, 10.48, 3.05, -0.048, 21, 72.89, 1.52, 0.01,19.34, 1.99]

[0158] Select multiple devices, including healthy and faulty device data, and measure multiple sets of data to establish vibration data set A:

[0159] A= [a1,a2,a3,...an]

[0160] 2.2. Extract the characteristic indicators of sound data, according to Figure 5 The collected sound waveform data is used to calculate the corresponding sound indicators. The extracted indicators are effective value, loudness, voiceprint roughness, frequency center of gravity, sharpness, high-low frequency energy ratio, peak-to-peak value, pulse, margin, kurtosis, and skewness. The calculated indicators are as follows:

[0161] s = [435, 51, 976, 890, 0.264, 0.359, 3670, 8.042, 6.459, 3.69, -0.018]

[0162] Select multiple devices, including healthy and faulty device data, and measure multiple sets of data to establish a vibration data set S:

[0163] S = [s1,s2,s3,...sn]

[0164] 2.3. Extract image data feature indicators, according to Figure 6 The collected equipment images calculate the corresponding image data feature indicators. The extracted indicators are gray-level co-occurrence matrix (GLCM), which describes the spatial relationship between pixel grayscales in the image. It is usually used to detect defects such as cracks and wear. Common GLCM features include contrast, homogeneity, roughness, and complexity. The calculated indicators are as follows:

[0165] P = [65.15, 0.4594, 0.0284, 7.96144]

[0166] 3. PCA dimensionality reduction

[0167] Taking vibration data as an example, after dimensionality reduction, the following data is obtained:

[0168] Variance ratio of each principal component: [0.67284757, 0.10768493];

[0169] Eigenvalues ​​of each principal component: [8.65089732, 1.3845205];

[0170] Cumulative variance ratio: [0.67284 0.7805325];

[0171] The scatter plot after dimensionality reduction is as follows Figure 7 As shown, similarly, the sound and image are also processed using the above method to obtain the corresponding main components.

[0172] 4. Construction of fault diagnosis model based on time-frequency domain features and naive Bayes

[0173] 4.1. Organize the reduced-dimensional dataset, including the corresponding features and labels.

[0174] 4.2. Calculate the prior probability of each category , that is, the frequency of occurrence of each category in all samples:

[0175] ;

[0176] 4.3. For each feature, calculate the probability of the feature under the condition of a given category .

[0177] If the feature is continuous data, it is usually assumed to follow a Gaussian distribution and the probability is calculated using the mean and variance. If it is discrete data, the conditional probability is calculated using the frequency.

[0178] 4.4. For new samples , calculate the posterior probability of belonging to each category according to Bayes' theorem:

[0179] ;

[0180] in is a normalization constant for all categories and can be ignored since it is calculated the same for each category.

[0181] 4.5. Select the maximum posterior probability: Calculate the posterior probabilities of all categories and select the category corresponding to the maximum value as the prediction result:

[0182] ;

[0183] 5. Fault diagnosis fusion

[0184] Based on the DS evidence theory in the above embodiment, the diagnostic results of the three sensor data, namely, sound, image and vibration, are diagnostically fused to obtain the final conclusion.

[0185] It should be noted that the algorithm mentioned in the present invention can be implemented in the upper computer software in the form of signal processing, or can be implemented in other ways, such as digital chips, hardware circuits, etc., to realize the calculation process.

[0186] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0187] In addition, each functional module in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0188] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-source data fusion fault diagnosis method based on DS evidence theory, characterized in that: The method comprises: Collect multi-source information of the target device, including vibration data, image data and sound data; Determine and extract characteristic indicators corresponding to vibration data, image data and sound data; Using principal component analysis algorithm to perform dimension reduction processing on the characteristic index; According to the feature index after dimensionality reduction, a pre-built fault diagnosis model based on time-frequency domain features and naive Bayes is trained, and a first diagnosis result of the fault of the equipment under test is determined based on the fault diagnosis model, including the following: Determine the target device fault type according to the target device fault mode or historical fault record, use the fault type as a category label, and form a training set including feature indicators after dimensionality reduction and corresponding fault category labels; Calculate the prior probability of each fault type in the training set ; Calculate the likelihood probability of each characteristic indicator under each fault type ; Calculate the characteristic index of the device under test after dimensionality reduction through Bayesian theorem The posterior probability of belonging to each fault type , select the fault type with the maximum a posteriori probability as the first diagnostic result, and the expression is: ； in, Indicates the fault type; Is the fault type The prior probability of express In category The conditional probability under ; Represents a given feature In case of failure type The posterior probability of represents the normalization constant for all fault types; arg max is the choice The fault category that maximizes the posterior probability; The first diagnosis result is integrated by using DS evidence theory to determine the second diagnosis result, including the following: The first diagnosis result is converted into an evidence body in the DS evidence theory, each evidence body represents the degree of trust of a signal source on the fault type, and the signal sources include vibration sensors, image sensors and sound sensors; Construct a trust function and assign a trust function to each fault type in each signal source; The credibility functions of multiple evidence bodies are merged to obtain a combined credibility function; According to the combined credibility function, the fault type with the maximum credibility value is selected as the second diagnosis result.

2. According to the multi-source data fusion fault diagnosis method based on DS evidence theory in claim 1, it is characterized in that: The determining and extracting characteristic indicators corresponding to the vibration data, the image data, and the sound data includes: Determine and extract characteristic indicators of vibration data of multiple target devices, including acceleration effective value, acceleration peak-to-peak value, acceleration pulse, acceleration margin, acceleration kurtosis, acceleration skewness, velocity effective value, velocity peak-to-peak value, velocity pulse, velocity margin, velocity kurtosis and velocity skewness, to form a first characteristic set; Determine and extract characteristic indicators of sound data of multiple target devices, including effective value, loudness, soundprint roughness, frequency center of gravity, sharpness, high-low frequency band energy ratio, peak-to-peak value, impulse, margin, kurtosis and skewness, to form a second feature set; Determine and extract feature indicators of multiple target device image data, including texture features, edge features, temperature distribution features and crack distribution features, to form a third feature set.

3. The multi-source data fusion fault diagnosis method based on DS evidence theory according to claim 2 is characterized in that: The adopting of the principal component analysis algorithm to perform dimension reduction processing on the characteristic index includes: Calculate the covariance matrix of each feature set after standardization; Calculate the eigenvalues ​​and eigenvectors of each covariance matrix based on a numerical algorithm; Determine the target eigenvector according to the size of the eigenvalue, and construct the principal component space based on it; The standardized covariance matrix is ​​projected into the principal component space for matrix multiplication to obtain the feature index after dimensionality reduction.

4. The multi-source data fusion fault diagnosis method based on DS evidence theory according to claim 1 is characterized in that: The step of selecting the fault type with the maximum reliability value as the second diagnosis result according to the combined reliability function includes: updating the combined credibility function based on the set conflict management strategy, and selecting the fault type corresponding to the maximum value of the credibility function from the updated credibility function as the second diagnosis result; The conflict management strategy set includes: (1) quantifying the degree of conflict between the different signal sources based on the conflict metric, and when the degree of conflict exceeds a preset threshold, selecting to exclude the conflicting information from the evidence synthesis to update the combined credibility function; the conflicting information is one or more of inconsistent diagnosis results, inconsistent probability distribution, inconsistent feature indicators, and inconsistent time or space; (2) Adjust the weight of conflicting evidence based on the reliability of the evidence source or historical data to update the merged credibility function.

5. A multi-source data fusion fault diagnosis system based on DS evidence theory, applied to execute the multi-source data fusion fault diagnosis method based on DS evidence theory as described in any one of claims 1 to 4, characterized in that: The system comprises: A data acquisition module is used to collect multi-source information of the target device, the multi-source information includes vibration data, image data and sound data; A feature extraction module, used to extract feature indicators corresponding to the vibration data, the image data and the sound data; A principal component analysis module, used for performing dimensionality reduction processing on the characteristic index by using a principal component analysis algorithm; A first diagnosis module is used to train a pre-built fault diagnosis model based on time-frequency domain features and naive Bayes according to the feature index after dimensionality reduction, and determine a first diagnosis result of the fault of the device under test based on the fault diagnosis model; The second diagnosis module is used to fuse the first diagnosis results using DS evidence theory to determine a second diagnosis result.

Images