Industrial equipment fault prediction method and device

Abstract

The invention provides an industrial equipment fault prediction method and device, relates to the technical field of fault analysis, and solves the technical problem of insufficient fault prediction accuracy in the prior art. The method specifically comprises the following steps: acquiring equipment data; the equipment data comprises first data and second data; performing noise filtering on the third data through a moving average filtering algorithm; performing feature extraction on the third data after noise filtering to obtain equipment feature data; calculating a correlation coefficient of the equipment feature data and the equipment fault tag, and selecting and removing non-correlation features based on a threshold value; principal component analysis is carried out on the equipment feature data after the non-relevant features are removed, and low-dimensional equipment feature data are obtained; and inputting the low-dimensional equipment feature data into the dual-channel model for fault classification and fault prediction. The method is used for industrial equipment fault prediction.

Description

Technical Field

[0001] This application relates to the field of fault analysis technology, and in particular to a method and apparatus for predicting faults in industrial equipment. Background Technology

[0002] Industrial equipment fault prediction is a crucial link in ensuring continuous and stable equipment operation, reducing unplanned downtime losses, and improving production and maintenance efficiency in industrial production, and it is of great significance to the high-quality development of the industrial sector. In existing technologies, industrial equipment fault prediction largely relies on the subjective judgment of maintenance personnel based on their on-site practical experience, or uses basic prediction methods such as setting fixed alarm thresholds and simple data statistical comparisons. This approach initially identifies fault risks by directly comparing real-time equipment operating data with preset standards. However, this method is not only greatly affected by the experience level and subjective judgment differences of maintenance personnel, making it prone to judgment bias, but also, simple prediction methods fail to capture the complex correlation characteristics and potential fault evolution patterns hidden in equipment operating data. It often fails to predict potential faults in advance, leading to delayed fault warnings and frequent omissions and misjudgments. Therefore, existing technologies suffer from insufficient accuracy in fault prediction. Summary of the Invention

[0003] This application provides a method and apparatus for predicting industrial equipment failures, which solves the technical problem of insufficient accuracy in failure prediction in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, a method for predicting industrial equipment faults is provided, comprising: acquiring equipment data; the equipment data includes first data and second data; the first data satisfies a normal distribution; the second data does not satisfy a normal distribution; removing outliers from the equipment data to obtain third data; performing noise filtering on the third data using a moving average filtering algorithm; extracting features from the noise-filtered third data to obtain equipment feature data; the equipment feature data includes time-domain feature data and frequency-domain feature data; calculating the correlation coefficient between the equipment feature data and the equipment fault label, and selecting to remove irrelevant features based on a threshold; performing principal component analysis on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data; and inputting the low-dimensional equipment feature data into a dual-channel model for fault classification and fault prediction; the dual-channel model includes a first model and a second model; the first model is used for fault classification based on the low-dimensional equipment feature data; and the second model is used for fault prediction based on the low-dimensional equipment feature data.

[0005] In conjunction with the first aspect mentioned above, in one possible implementation, outliers in the device data are removed to obtain third data, including: based on 3 The criteria are used to remove outliers from the first set of data; outliers in the second set of data are then removed based on the interquartile range to obtain the third set of data.

[0006] In conjunction with the first aspect mentioned above, in one possible implementation, noise filtering of the third data is performed using a moving average filtering algorithm, including: setting the size of the filtering window; performing traversal filtering on the third data based on the filtering window; the traversal filtering is to calculate the average value of the data within the filtering window range with the data point as the center for each data point, and replace the original data point with the average value.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, feature extraction is performed on the noise-filtered third data to obtain device feature data, including: performing time-domain feature extraction on the noise-filtered third data to obtain time-domain feature data; the time-domain feature data includes mean, variance, root mean square value, and peak factor; performing frequency-domain feature extraction on the noise-filtered third data to obtain frequency-domain feature data; the frequency-domain feature data includes power spectral density, dominant frequency, and amplitude spectrum.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the correlation coefficient between equipment feature data and equipment fault labels is calculated, and irrelevant features are removed based on a threshold. This includes: setting a correlation threshold; calculating the correlation coefficient between equipment feature data and equipment fault labels; and removing the corresponding feature from the equipment feature data if the correlation coefficient is less than the correlation threshold.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, principal component analysis is performed on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data. This includes: standardizing the equipment feature data after removing irrelevant features; calculating the covariance matrix and performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; selecting a set number of eigenvectors to form a transformation matrix based on a set variance contribution rate threshold; and projecting the equipment feature data after removing irrelevant features onto a low-dimensional space based on the transformation matrix to obtain low-dimensional equipment feature data.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, the first model is the SVM model; the second model is the LSTM model.

[0011] In conjunction with the first aspect mentioned above, one possible implementation involves fault classification based on low-dimensional equipment feature data, including: normalizing the low-dimensional equipment feature data to determine whether it is linearly separable or nonlinearly separable; for linearly separable low-dimensional equipment feature data, constructing an objective function and finding the optimal hyperplane through Lagrange duality to perform fault classification; for nonlinearly separable low-dimensional equipment feature data, introducing a kernel function and solving the dual problem through quadratic programming to perform fault classification.

[0012] In conjunction with the first aspect mentioned above, one possible implementation involves fault prediction based on low-dimensional device feature data, including: converting the low-dimensional device feature data into a format suitable for LSTM input; inputting the converted low-dimensional device feature data into a second model and outputting device fault prediction results; the second model is a model trained using historical fault data; the historical fault data includes device data and corresponding device fault labels.

[0013] Secondly, an industrial equipment fault prediction device is provided, comprising: a communication unit and a processing unit; the communication unit is used to acquire equipment data; the equipment data includes first data and second data; the first data satisfies a normal distribution; the second data does not satisfy a normal distribution; the processing unit is used to remove outliers from the equipment data to obtain third data; to perform noise filtering on the third data using a moving average filtering algorithm; to perform feature extraction on the noise-filtered third data to obtain equipment feature data; the equipment feature data includes time-domain feature data and frequency-domain feature data; to calculate the correlation coefficient between the equipment feature data and the equipment fault label, and to select and remove irrelevant features based on a threshold; to perform principal component analysis on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data; and to input the low-dimensional equipment feature data into a dual-channel model for fault classification and fault prediction; the dual-channel model includes a first model and a second model; the first model is used for fault classification based on the low-dimensional equipment feature data; and the second model is used for fault prediction based on the low-dimensional equipment feature data.

[0014] This application provides a method and apparatus for predicting industrial equipment faults. It can differentiate and remove outliers from equipment data with different distribution characteristics, reduce noise interference through filtering, comprehensively extract time-domain and frequency-domain features and filter core information, and achieve accurate fault classification and early prediction with the help of a dual-channel model. It effectively avoids the limitations of reliance on human experience and simple prediction methods, and solves the technical problem of insufficient fault prediction accuracy in the prior art.

[0015] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0016] Figure 1 A flowchart illustrating an industrial equipment fault prediction method provided in this application embodiment; Figure 2 A flowchart illustrating another industrial equipment fault prediction method provided in this application embodiment; Figure 3 A flowchart illustrating another industrial equipment fault prediction method provided in this application embodiment; Figure 4 A flowchart illustrating another industrial equipment fault prediction method provided in this application embodiment; Figure 5 A flowchart illustrating another industrial equipment fault prediction method provided in this application embodiment; Figure 6 A flowchart illustrating another industrial equipment fault prediction method provided in this application embodiment; Figure 7 A flowchart illustrating another industrial equipment fault prediction method provided in this application embodiment; Figure 8 A flowchart illustrating another industrial equipment fault prediction method provided in this application embodiment; Figure 9 This is a schematic diagram of the structure of an industrial equipment fault prediction device provided in an embodiment of this application. Detailed Implementation

[0017] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0018] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0019] To address the technical problem of insufficient accuracy in fault prediction in existing technologies, this application provides an industrial equipment fault prediction method. The method includes: acquiring equipment data containing both normally and non-normally distributed data; sequentially performing outlier removal, moving average filtering, time-domain and frequency-domain feature extraction, irrelevant feature removal, and principal component analysis for dimensionality reduction; and then inputting the low-dimensional feature data into a dual-channel model for fault classification and prediction. Based on this, the effective information in the equipment operation data can be fully extracted, enabling accurate fault identification and early prediction, significantly improving the accuracy of industrial equipment fault prediction.

[0020] like Figure 1 As shown in the embodiments of this application, the industrial equipment fault prediction method includes: S101. Obtain device data.

[0021] The equipment data includes first data that follows a normal distribution and second data that does not follow a normal distribution, covering various types of characterizing data during equipment operation.

[0022] In this embodiment of the application, various types of data generated during the operation of the equipment are acquired through industrial sensors or Internet of Things (IoT) acquisition methods.

[0023] As an example, the temperature of an industrial motor is collected using an industrial temperature sensor.

[0024] Based on the above steps, equipment operation data is obtained, providing basic data support for subsequent fault analysis.

[0025] S102. Remove outliers from the device data to obtain the third data.

[0026] Outliers refer to invalid data that deviates from the normal operating data range of the equipment.

[0027] In this embodiment of the application, outliers in the first data and the second data are removed based on the different distribution characteristics of the device data to obtain the third data.

[0028] Based on the above steps, invalid interference in the data is eliminated, improving the reliability of subsequent data processing.

[0029] S103. Noise filtering is performed on the third data using a moving average filtering algorithm.

[0030] Among them, the moving average filtering algorithm is a filtering method that smooths the original data by using the mean of the data within the window.

[0031] In this embodiment of the application, noise filtering is performed on each data point of the third data based on the filtering window.

[0032] Based on the above steps, noise interference in the data is suppressed, making the data more consistent with the actual operating status of the equipment.

[0033] S104. Perform feature extraction on the third data after noise filtering to obtain equipment feature data.

[0034] Among them, equipment characteristic data includes time-domain characteristic data that reflects the overall level and dispersion of the data, and frequency-domain characteristic data that reflects the frequency distribution.

[0035] In this embodiment of the application, time-domain and frequency-domain related features are extracted from the noise-filtered third data to form device feature data.

[0036] As an example, the mean and peak factor time-domain features and the dominant frequency and amplitude spectrum frequency-domain features of the data are extracted.

[0037] Based on the above steps, core data representation information is extracted to provide key basis for fault diagnosis.

[0038] S105. Calculate the correlation coefficient between equipment feature data and equipment fault labels, and select and remove irrelevant features based on the threshold.

[0039] Among them, the equipment fault label is used to identify the equipment operating status (normal or faulty) corresponding to the data.

[0040] In this embodiment of the application, the correlation coefficient between each feature and the fault label is calculated, and irrelevant features are eliminated based on a set threshold.

[0041] It should be noted that the correlation threshold can be adjusted according to different equipment types and fault scenarios.

[0042] As an example, a threshold of 0.3 is set to remove features whose absolute correlation coefficient is lower than this value.

[0043] Based on the above steps, redundant feature interference is reduced, and the efficiency of subsequent data processing is improved.

[0044] S106. Perform principal component analysis on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data.

[0045] Principal component analysis is a dimensionality reduction method that preserves the main information of the data through linear transformation.

[0046] In this embodiment of the application, principal component analysis is performed on the filtered feature data to transform it into low-dimensional device feature data.

[0047] Based on the above steps, the data dimensions are simplified, reducing the computational complexity of subsequent models.

[0048] S107. Input the low-dimensional equipment feature data into the dual-channel model for fault classification and fault prediction.

[0049] The dual-channel model includes a first model for fault classification and a second model for fault prediction.

[0050] In this embodiment, low-dimensional device feature data is input into a dual-channel model to complete fault category identification and fault trend prediction respectively. The first model outputs whether the device is faulty and the fault type, and the second model outputs the probability of fault occurrence.

[0051] It should be noted that the two models work together to improve the comprehensiveness and reliability of the results.

[0052] Based on the above steps, accurate fault classification and early prediction can be achieved, improving the accuracy of prediction results.

[0053] Based on the above technical solution, invalid interference data was eliminated, core features were mined, and the calculation was simplified by dimensionality reduction. Finally, the dual-channel model was used to achieve accurate identification and early prediction of faults, which effectively improved the accuracy and practicality of industrial equipment fault prediction.

[0054] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 2 As shown, the above S102 removes outliers from the device data to obtain the third data. This can be achieved through the following S201 and S202, which are explained in detail below: S201, based on 3 The criterion is to remove outliers from the first set of data.

[0055] Among them, the 3σ criterion is an outlier judgment method based on the characteristics of normal distribution, and the first data is the equipment operation data that meets the normal distribution.

[0056] In this embodiment of the application, the mean and standard deviation of the first data are first calculated, and then outlier data points are screened and removed according to the 3σ criterion.

[0057] Optionally, outlier data points are filtered and removed according to the 3σ criterion, satisfying the following formula:

[0058] in, The mean of the first data point. For the data points to be judged, is the standard deviation of the first data point.

[0059] As an example, the mean of the motor winding temperature data (normal distribution) is calculated to be 38℃ and the standard deviation is 3℃, and data below 29℃ or above 47℃ are removed.

[0060] Based on the above steps, outliers in the first set of data are accurately removed, ensuring data validity.

[0061] S202. Remove outliers from the second data based on the interquartile range to obtain the third data.

[0062] Among them, the interquartile range (IQR) is the difference between the third quartile and the first quartile, and the second data is the equipment operation data that does not meet the normal distribution.

[0063] In this embodiment, the second data is first sorted, the first quartile Q1, the third quartile Q3, and the IQR are calculated, and then outliers are filtered and removed according to the boundaries. For outlier boundaries: lower limit = Q1 - 1.5 × IQR, upper limit = Q3 + 1.5 × IQR.

[0064] It should be noted that this step is adapted to non-normally distributed data to avoid processing bias caused by misjudgment of distribution.

[0065] As an example, for the pump body vibration amplitude data (non-normal distribution), Q1=1.2mm and Q3=2.1mm are calculated, and data below -0.15mm (the negative sign indicates the direction) or above 3.45mm are removed.

[0066] Based on the above steps, the characteristics of non-normal data are adapted to effectively remove outliers from the second set of data.

[0067] Based on the above technical solution, a differentiated outlier removal method is adopted for equipment data with different distribution characteristics. This method not only ensures the processing accuracy of normally distributed data, but also adapts to the characteristics of non-normally distributed data, significantly improving the rationality and effectiveness of outlier removal and providing high-quality data support for subsequent data processing.

[0068] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 3 As shown, S103 above performs noise filtering on the third data using a moving average filtering algorithm, which can be specifically implemented through S301 and S302, as explained below: S301. Set the size of the filter window.

[0069] The filter window size refers to the number of data points involved in the moving average calculation. In this embodiment, the filter window size is reasonably set based on the acquisition frequency and fluctuation characteristics of the third data, with odd-numbered windows preferred to facilitate data point centering. It should be noted that the window size needs to balance the filtering effect with the timeliness of the data, avoiding being too large or too small. As an example, for equipment vibration data acquired at 10Hz, the filter window size is set to 5 data points. Based on the above steps, a filter window suitable for the data characteristics is determined to ensure the subsequent filtering effect.

[0070] S302. Perform traversal filtering on the third data based on the filtering window.

[0071] Among them, traversal filtering is a processing method that replaces the original data with the mean within a window, taking each data point as the center, according to the data time sequence.

[0072] In this embodiment, data points are selected one by one according to the data acquisition order, a filtering window is defined with the selected data point as the center, the average value of all data within the window is calculated, and the average value is used to replace the original data points. When the window exceeds the data range, boundary data is used to supplement it.

[0073] Alternatively, the filtering calculation satisfies the following formula:

[0074] in, Let m be the value of the i-th data point after filtering, and m be the size of the filtering window. This represents the value of the kth original data point of the third data point within the window, where [·] indicates rounding down.

[0075] It should be noted that the traversal process must maintain the temporal order of the data and not change the temporal correlation characteristics of the original data.

[0076] As an example, we iterate through the temperature data with a window size of 5, and using the 3rd data point as the center, we calculate the mean of the 1st to 5th data points to replace the original 3rd data point.

[0077] Based on the above steps, data fluctuations are smoothed out, and noise interference in the third data is effectively suppressed.

[0078] Based on the above technical solution, by first setting a filtering window that adapts to the characteristics of the data, and then calculating the mean by traversing the time sequence, the filtering is achieved. This not only takes into account the timeliness and completeness of the data, but also efficiently removes noise, significantly improves the smoothness and reliability of the data, and provides high-quality data for subsequent feature extraction.

[0079] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 4 As shown, S104 above performs feature extraction on the third data after noise filtering to obtain device feature data. Specifically, this can be achieved through the following S401 and S402, which are explained in detail below: S401. Extract time-domain features from the third data after noise filtering to obtain time-domain feature data.

[0080] Among them, the time-domain feature data includes the mean, variance, root mean square value and peak factor, which reflect the overall level, dispersion and signal strength of the data.

[0081] In this embodiment of the application, based on the third data after noise filtering, various time-domain features are extracted through statistical calculation to form time-domain feature data.

[0082] Optionally, the peak factor satisfies the following formula:

[0083] in, The root mean square value, For a single data value, is the maximum absolute value of the data, and CF is the peak factor.

[0084] It should be noted that time-domain feature extraction does not require changing the time attributes of the data; it is calculated directly based on the original time-series data.

[0085] Based on the above steps, the core characteristics of the time domain dimension of the data are captured, providing basic feature support for fault diagnosis.

[0086] S402. Extract frequency domain features from the third data after noise filtering to obtain frequency domain feature data.

[0087] Among them, the frequency domain characteristic data includes the power spectral density reflecting the frequency distribution of the signal, the dominant frequency of the energy concentration frequency, and the amplitude spectrum of each frequency amplitude.

[0088] In this embodiment, the filtered third data is first transformed to the frequency domain by Fourier transform, and then the power spectral density, main frequency and amplitude spectrum are extracted from the frequency domain data.

[0089] Based on the above steps, frequency domain features can capture frequency-related fault information that is difficult to reflect in the time domain, mine key information in the frequency domain dimension of data, and supplement the deficiencies of time domain features.

[0090] Based on the above technical solution, by extracting time domain and frequency domain features respectively, the time distribution characteristics and frequency distribution characteristics of the data are fully covered, and the fault correlation information in the equipment operation data is fully mined, providing comprehensive and diverse feature support for subsequent feature selection and model input, thereby improving the accuracy of fault judgment.

[0091] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 5 As shown, the above S105 calculates the correlation coefficient between equipment feature data and equipment fault labels, and removes irrelevant features based on a threshold. This can be achieved through the following S501, S502, and S503, which are explained in detail below: S501, Set the correlation threshold.

[0092] Among them, the correlation threshold is the critical value for judging the strength of the correlation between equipment characteristics and fault labels.

[0093] In this embodiment of the application, the correlation threshold is flexibly set according to the type of industrial equipment, the fault scenario, and the distribution of characteristic data.

[0094] It should be noted that the threshold should not be too large or too small to prevent excessive retention of redundant features or accidental deletion of valid features.

[0095] As an example, in the context of motor fault prediction, the correlation threshold is set to 0.3.

[0096] Based on the above steps, clear criteria for feature selection are established, providing a basis for subsequent removal of invalid features.

[0097] S502. Calculate the correlation coefficient between equipment characteristic data and equipment fault labels.

[0098] The correlation coefficient is used to quantify the degree of linear correlation between equipment feature data and fault labels.

[0099] In this embodiment of the application, the Pearson correlation coefficient calculation method is used to traverse all equipment feature data and corresponding equipment fault labels, and calculate the correlation coefficient between the two one by one.

[0100] Optionally, the Pearson correlation coefficient is calculated according to the following formula:

[0101] in, Let be the correlation coefficient between the i-th feature and the fault label. For the j-th sample value of the i-th feature, Let be the mean of the i-th feature. Let j be the fault label value of the j-th sample. denoted as the mean of the fault labels, and n is the number of samples.

[0102] As an example, the correlation coefficient between the peak factor feature and the motor bearing fault label is calculated to be 0.42.

[0103] Based on the above steps, the correlation between features and faults can be quantified.

[0104] S503. If the correlation coefficient is less than the correlation threshold, the corresponding feature will be removed from the device feature data.

[0105] Among them, irrelevant features refer to feature data with a correlation coefficient lower than a set threshold and a weak correlation with the fault.

[0106] In this embodiment of the application, the correlation coefficient of each feature is compared with a preset threshold. If the correlation coefficient is less than the correlation threshold, the feature is removed.

[0107] It should be noted that only features that do not meet the threshold requirements are removed, while features with correlation coefficients greater than or equal to the threshold are retained, without changing the data structure of the effective features.

[0108] As an example, if the threshold is 0.3, variance features with a correlation coefficient of 0.21 will be removed from the device feature data.

[0109] Based on the above steps, redundant and invalid features can be eliminated, thereby improving the relevance of feature data.

[0110] Based on the above technical solution, by setting reasonable thresholds, quantifying the correlation between features and faults, and eliminating low-correlation features, redundant information interference can be effectively reduced, feature dimensions can be simplified, and subsequent data processing efficiency can be improved. At the same time, the effectiveness and relevance of the retained features can be guaranteed, providing high-quality feature input for fault prediction.

[0111] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 6 As shown, S106 above performs principal component analysis on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data. Specifically, this can be achieved through S601, S602, S603, and S604, which are explained in detail below: S601. Standardize the device feature data after removing irrelevant features.

[0112] Standardization is a preprocessing method that eliminates differences in the dimensions of different features and unifies the data scale.

[0113] In this embodiment of the application, the device feature data after removing irrelevant features is standardized so that the mean of each feature is 0 and the variance is 1.

[0114] Alternatively, standardization satisfies the following formula:

[0115] in, The standardized value of the i-th feature in the j-th sample. These are the original sample values. Let be the mean of the i-th feature. Let be the standard deviation of the i-th feature.

[0116] It should be noted that standardization only processes the data within a feature and does not change the relative relationships between features.

[0117] Based on the above steps, the differences in dimensions are eliminated, so that different types of features have equal weight and comparability.

[0118] S602. Calculate the covariance matrix, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors.

[0119] The covariance matrix reflects the degree of linear correlation between features, the eigenvalues ​​characterize the information importance of the corresponding eigenvectors, and the eigenvectors are orthogonal vectors of the covariance matrix.

[0120] In this embodiment, the covariance between each pair of features is calculated based on the standardized feature data to construct a covariance matrix. Then, all eigenvalues ​​and corresponding eigenvectors of the covariance matrix are obtained through an eigenvalue decomposition algorithm. The covariance matrix satisfies the following: The eigenvalue decomposition satisfies: Where C is a p×p covariance matrix (p is the number of features), X is an n×p standardized feature data matrix, and n is the number of samples. For the i-th eigenvalue, For the corresponding eigenvectors.

[0121] Based on the above steps, the correlation between features and the importance of features are quantified, providing core data support for dimensionality reduction.

[0122] S603. Based on the set variance contribution rate threshold, select a set number of feature vectors to form a transformation matrix.

[0123] Among them, the variance contribution rate threshold is a critical value that measures the proportion of core information of the data that is retained, and the transformation matrix is ​​an orthogonal matrix used for data dimension transformation.

[0124] In this embodiment, the proportion of each feature value to the sum of all feature values ​​(variance contribution rate) is first calculated, then the variance contribution rate is accumulated, and the top k feature vectors whose accumulated values ​​reach the threshold are selected and arranged in order to form a transformation matrix.

[0125] It should be noted that the threshold needs to balance the degree of dimensional simplification with the information retention rate, and is usually set at 85%-95%.

[0126] As an example, the variance contribution rate threshold is set to 90%, and the cumulative contribution rate of the top 3 eigenvalues ​​reaches 92%. The eigenvectors corresponding to these 3 eigenvalues ​​are selected to form a 3×p-dimensional transformation matrix.

[0127] Based on the above steps, core feature vectors are selected to retain key information to the maximum extent while compressing dimensionality.

[0128] S604. Based on the transformation matrix, project the device feature data after removing irrelevant features into a low-dimensional space to obtain low-dimensional device feature data.

[0129] Projection is an operation that maps high-dimensional feature data to a low-dimensional space through linear transformation. The low-dimensional device feature data is a simplified feature set that retains the core information.

[0130] In this embodiment of the application, the device feature data after removing irrelevant features is multiplied with the transformation matrix to complete the mapping of high-dimensional data to low-dimensional space, thereby obtaining low-dimensional device feature data.

[0131] Based on the above steps, the feature dimensions are greatly simplified, reducing the computational complexity of subsequent models.

[0132] Based on the above technical solution, through the entire process of standardizing and unifying data scale, quantifying feature correlation by covariance matrix and eigenvalue decomposition, selecting core feature vectors by variance contribution rate, and dimensionality reduction by projection mapping, the feature dimensions are greatly simplified and the computational cost is reduced, while retaining the core information of equipment fault correlation to the maximum extent. This provides high-quality low-dimensional features for the dual-channel model, improving the efficiency and accuracy of fault prediction.

[0133] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 7 As shown, the fault classification based on low-dimensional equipment feature data in S107 can be implemented through the following steps S701, S702, and S703, which are explained in detail below: S701. Normalize the low-dimensional device feature data and determine whether the low-dimensional device feature data is linearly separable or non-linearly separable.

[0134] Normalization is a standardization operation that maps data to a fixed interval. Linearly separable data can be classified by a hyperplane, while non-linearly separable data cannot be classified by a single hyperplane.

[0135] In this embodiment of the application, a normalization operation is performed on the low-dimensional device feature data, and then a linear test method is used to determine its classifiable type.

[0136] As an example, low-dimensional feature data is normalized to the [0,1] interval, and separability is determined by combining the data scatter plot with the verification of a linear classifier.

[0137] Based on the above steps, we can unify the data scale and clarify the data classification types.

[0138] S702. For linearly separable low-dimensional equipment feature data, construct an objective function and find the optimal hyperplane through Lagrange duality to perform fault classification.

[0139] Among them, the optimal hyperplane is the dividing boundary that maximizes the interval between different categories of data, and Lagrange duality is used to transform the original optimization problem into an easily solvable form.

[0140] In this embodiment, a function is constructed with the goal of maximizing the classification margin. The optimization problem is transformed through Lagrange dual transformation, and the optimal hyperplane is obtained by solving it through quadratic programming. The fault category to which the data belongs is determined based on the hyperplane.

[0141] Optionally, the objective function satisfies the following formula:

[0142] in, Let b be the hyperplane normal vector, and b be the bias term. For sample feature vectors, Assign sample labels; transform this into a Lagrange dual problem:

[0143] Among them, the elements in Q satisfy e is a vector of all 1s. It is a Lagrange multiplier.

[0144] It should be noted that the dual problem can be solved using a quadratic programming algorithm to find the optimal solution. , To obtain the optimal hyperplane, this application does not impose specific limitations.

[0145] As an example, the optimal hyperplane is obtained by solving for linearly separable equipment normal and bearing wear characteristic data, thus achieving accurate separation of the two types of data.

[0146] Based on the above steps, the fault classification accuracy and generalization ability of linearly separable data are improved by utilizing the maximum interval characteristic.

[0147] S703. For low-dimensional equipment feature data that can be nonlinearly separable, fault classification is performed by introducing a kernel function and solving the dual problem through quadratic programming.

[0148] Among them, the kernel function can implicitly map nonlinear data to a high-dimensional linearly separable space, and quadratic programming is used to efficiently solve constrained dual optimization problems.

[0149] In this embodiment, a radial basis function is introduced to map nonlinear data to a high-dimensional space and transform it into a linearly separable problem. The dual problem is solved by quadratic programming to obtain the classification boundary and complete the fault classification.

[0150] Optionally, the kernel function satisfies: ,in, For kernel function parameters, , Let be the sample feature vector; the dual problem is: The constraints remain the same as in S702.

[0151] Based on the above steps, and adapting to the characteristics of nonlinear data, accurate fault classification can be achieved while controlling computational costs.

[0152] Based on the above technical solution, by first normalizing and unifying the data scale and clarifying the separable data types, and then adopting targeted classification strategies such as maximizing the margin hyperplane for linearly separable data and solving nonlinearly separable data through kernel functions and quadratic programming, the system can fully adapt to low-dimensional feature data with different characteristics. This not only ensures the generalization ability of linear scenarios but also solves the classification problem of nonlinear scenarios, significantly improving the accuracy and applicability of fault classification.

[0153] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 8 As shown, the fault prediction based on low-dimensional equipment feature data in S107 can be specifically implemented through the following S801 and S802, which are explained in detail below: S801: Convert low-dimensional device feature data into a format suitable for LSTM input.

[0154] The LSTM input format is a three-dimensional time sequence format, i.e., [number of samples, time step, feature dimension], which needs to preserve the temporal correlation characteristics of the data.

[0155] In this embodiment of the application, the dimensional arrangement and sequence length of the low-dimensional device feature data are adjusted according to the structural design of the LSTM model and the time series period of the device operation data to form an input format that the model can recognize.

[0156] As an example, the 2D low-dimensional feature data of 1000 samples are reorganized into a 3D input format of [1000, 10, 2] at a time step of 10.

[0157] Based on the above steps, the data is adapted to the LSTM model structure, and the temporal correlation information is fully preserved.

[0158] S802. Input the converted low-dimensional equipment feature data into the second model and output the equipment fault prediction results.

[0159] The second model is an LSTM model that has been trained and validated in advance using historical fault data with fault labels. The historical fault data includes equipment data and corresponding equipment fault labels. The equipment data also undergoes the above-mentioned process of removing outliers, reducing noise interference through filtering, and comprehensively extracting time-domain and frequency-domain features and filtering core information.

[0160] In this embodiment of the application, the low-dimensional device feature data after format conversion is input into the second model. The model outputs the device fault prediction result by learning the historical time series pattern, including whether there is a fault and the probability of the fault occurring.

[0161] It should be noted that when training the LSTM model, the processed historical fault data is first divided into training, validation, and test sets according to time series. The training set is used for model parameter learning, the validation set prevents overfitting, and the test set evaluates model performance. The LSTM model uses input, forget, and output gates to handle long-term dependencies in time series data. During training, at each time step, the model receives low-dimensional device feature data and combines it with the hidden state and memory cell state from the previous time step. The mean squared error (MSE) loss function is used to measure the difference between the predicted value and the true fault label, and the Adam optimizer is used to minimize the loss function. The Adam optimizer dynamically adjusts the learning rate based on the gradient's first and second moments to ensure rapid and stable convergence of the model. In each training cycle, the model performs forward and backward propagation calculations on the training set samples to update the weight parameters. The validation set is used to evaluate performance during training. If the validation set loss function value no longer decreases or increases, it indicates potential overfitting, and training should be stopped or hyperparameters adjusted. After multiple rounds of training, the model learns the temporal patterns of historical fault data and the correlation between equipment characteristics and faults. Inputting low-dimensional data in a converted format, and through forward propagation, it can output whether the equipment is faulty and the probability of fault.

[0162] Based on the above steps, the fault evolution trend in time series data can be quickly mined to enable early prediction of equipment failures.

[0163] Based on the above technical solution, the compatibility between the data and the LSTM model is ensured through format conversion. The trained model is used to fully explore the fault correlation patterns in the time series features. This not only preserves the time series information of equipment operation, but also achieves accurate early prediction of faults, improving the timeliness and reliability of fault prediction.

[0164] The above primarily describes the solutions of the embodiments of this application from the perspective of device implementation. It is understood that each device, such as an industrial equipment fault prediction device, includes at least one of the hardware structures and software modules corresponding to the execution of each function in order to achieve the above-mentioned functions. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0165] This application embodiment can divide the industrial equipment fault prediction device into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0166] When using integrated units, Figure 9 A possible structural schematic diagram of the industrial equipment fault prediction device (referred to as industrial equipment fault prediction device 90) involved in the above embodiments is shown. The industrial equipment fault prediction device 90 includes a processing unit 901 and a communication unit 902, and may also include a storage unit 903. Figure 9 The structural diagram shown can be used to illustrate the structure of the industrial equipment fault prediction device involved in the above embodiments.

[0167] when Figure 9 The schematic diagram shown illustrates the structure of the industrial equipment fault prediction device involved in the above embodiments. The processing unit 901 is used to control and manage the operation of the industrial equipment fault prediction device, the communication unit 902 is used for the industrial equipment fault prediction device to communicate with other devices, and the storage unit 903 is used to store the program code and data of the industrial equipment fault prediction device.

[0168] For example, communication unit 902 is used to acquire device data; the device data includes first data and second data; the first data follows a normal distribution; the second data does not follow a normal distribution.

[0169] Processing unit 901 is used to remove outliers from equipment data to obtain third data; to perform noise filtering on the third data using a moving average filtering algorithm; to extract features from the noise-filtered third data to obtain equipment feature data; the equipment feature data includes time-domain feature data and frequency-domain feature data; to calculate the correlation coefficient between the equipment feature data and the equipment fault label, and to select and remove irrelevant features based on a threshold; to perform principal component analysis on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data; and to input the low-dimensional equipment feature data into a dual-channel model for fault classification and fault prediction; the dual-channel model includes a first model and a second model; the first model is used for fault classification based on the low-dimensional equipment feature data; and the second model is used for fault prediction based on the low-dimensional equipment feature data.

[0170] In one possible implementation, the processing unit 901 is further configured to remove outliers from the device data to obtain third data, including: based on 3 The criteria are used to remove outliers from the first set of data; outliers in the second set of data are then removed based on the interquartile range to obtain the third set of data.

[0171] In one possible implementation, the processing unit 901 is further configured to perform noise filtering on the third data using a moving average filtering algorithm, including: setting the size of the filtering window; performing traversal filtering on the third data based on the filtering window; the traversal filtering is to calculate the average value of the data within the filtering window range with the data point as the center for each data point, and replace the original data point with the average value.

[0172] In one possible implementation, the processing unit 901 is further configured to perform feature extraction on the noise-filtered third data to obtain device feature data, including: performing time-domain feature extraction on the noise-filtered third data to obtain time-domain feature data; the time-domain feature data includes mean, variance, root mean square value and peak factor; and performing frequency-domain feature extraction on the noise-filtered third data to obtain frequency-domain feature data; the frequency-domain feature data includes power spectral density, dominant frequency and amplitude spectrum.

[0173] In one possible implementation, the processing unit 901 is further configured to calculate the correlation coefficient between the equipment feature data and the equipment fault label, and select to remove irrelevant features based on a threshold, including: setting a correlation threshold; calculating the correlation coefficient between the equipment feature data and the equipment fault label; and removing the corresponding feature from the equipment feature data if the correlation coefficient is less than the correlation threshold.

[0174] In one possible implementation, the processing unit 901 is further configured to perform principal component analysis on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data, including: standardizing the equipment feature data after removing irrelevant features; calculating the covariance matrix, performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; selecting a set number of eigenvectors to form a transformation matrix based on a set variance contribution rate threshold; and projecting the equipment feature data after removing irrelevant features onto a low-dimensional space based on the transformation matrix to obtain low-dimensional equipment feature data.

[0175] In one possible implementation, the first model is the SVM model; the second model is the LSTM model.

[0176] In one possible implementation, the processing unit 901 is further configured to perform fault classification based on low-dimensional equipment feature data, including: normalizing the low-dimensional equipment feature data to determine whether the low-dimensional equipment feature data is linearly separable or nonlinearly separable; for linearly separable low-dimensional equipment feature data, constructing an objective function to find the optimal hyperplane through Lagrange duality and performing fault classification; for nonlinearly separable low-dimensional equipment feature data, introducing a kernel function to solve the dual problem through quadratic programming and performing fault classification.

[0177] In one possible implementation, the processing unit 901 is further configured to perform fault prediction based on low-dimensional device feature data, including: converting the low-dimensional device feature data into a format suitable for LSTM input; inputting the converted low-dimensional device feature data into a second model and outputting device fault prediction results; the second model is a model trained using historical fault data; the historical fault data includes device data and corresponding device fault labels.

[0178] The processing unit 901 can be a processor or a controller, and the communication unit 902 can be a communication interface, transceiver, transceiver circuit, transceiver device, etc. The term "communication interface" is a general term and may include one or more interfaces. The storage unit 903 can be a memory. When the industrial equipment fault prediction device 90 is a chip, the processing unit 901 can be a processor or a controller, and the communication unit 902 can be an input interface and / or an output interface, pins, or circuits, etc. The storage unit 903 can be a storage unit within the chip (e.g., a register, cache, etc.) or a storage unit located outside the chip (e.g., read-only memory (ROM), random access memory (RAM, etc.).

[0179] The communication unit can also be called a transceiver unit. The antenna and control circuit with transceiver functions in the industrial equipment fault prediction device 90 can be considered as the communication unit 902 of the industrial equipment fault prediction device 90, and the processor with processing functions can be considered as the processing unit 901 of the industrial equipment fault prediction device 90. Optionally, the device in the communication unit 902 that implements the receiving function can be considered as a communication unit. The communication unit is used to execute the receiving steps in the embodiments of this application, and the communication unit can be a receiver, a receiver circuit, etc. The device in the communication unit 902 that implements the transmitting function can be considered as a transmitting unit. The transmitting unit is used to execute the transmitting steps in the embodiments of this application, and the transmitting unit can be a transmitter, a transmitter, a transmitting circuit, etc.

[0180] Figure 9If the integrated units in the process are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0181] Figure 9 The units in the process can also be called modules; for example, a processing unit can be called a processing module.

[0182] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0183] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. A method for predicting industrial equipment failures, characterized in that, include: Acquire device data; the device data includes first data and second data; the first data follows a normal distribution; the second data does not follow a normal distribution. Remove outliers from the device data to obtain third data; The third data is filtered for noise using a moving average filtering algorithm. Feature extraction is performed on the noise-filtered third data to obtain device feature data; the device feature data includes time-domain feature data and frequency-domain feature data. Calculate the correlation coefficient between the equipment feature data and the equipment fault labels, and select irrelevant features to remove based on the threshold. Principal component analysis was performed on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data. The low-dimensional equipment feature data is input into a dual-channel model for fault classification and fault prediction; the dual-channel model includes a first model and a second model; the first model is used for fault classification based on the low-dimensional equipment feature data; the second model is used for fault prediction based on the low-dimensional equipment feature data.

2. The method according to claim 1, characterized in that, After removing outliers from the device data, third data is obtained, including: Based on 3 The criterion is to remove outliers from the first data; The third data is obtained by removing outliers from the second data based on the interquartile range.

3. The method according to claim 1, characterized in that, The noise filtering of the third data using a moving average filtering algorithm includes: Set the size of the filter window; The third data is subjected to traversal filtering based on the filtering window; the traversal filtering is to calculate the average value of the data within the filtering window range with the data point as the center for each data point, and use the average value to replace the original data point.

4. The method according to claim 1, characterized in that, The step of extracting features from the noise-filtered third data to obtain device feature data includes: The third data after noise filtering is subjected to time-domain feature extraction to obtain time-domain feature data; the time-domain feature data includes mean, variance, root mean square value and peak factor. Frequency domain features are extracted from the noise-filtered third data to obtain frequency domain feature data; the frequency domain feature data includes power spectral density, dominant frequency, and amplitude spectrum.

5. The method according to claim 1, characterized in that, Calculate the correlation coefficient between the equipment feature data and the equipment fault label, and select to remove irrelevant features based on a threshold, including: Set the correlation threshold; Calculate the correlation coefficient between the equipment feature data and the equipment fault tags; If the correlation coefficient is less than the correlation threshold, the corresponding feature will be removed from the device feature data.

6. The method according to claim 1, characterized in that, The process of performing principal component analysis on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data includes: The device feature data after removing irrelevant features is standardized. Calculate the covariance matrix, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; Based on a set variance contribution rate threshold, a set number of feature vectors are selected to form a transformation matrix; Based on the transformation matrix, the device feature data after removing irrelevant features is projected into a low-dimensional space to obtain low-dimensional device feature data.

7. The method according to claim 1, characterized in that, The first model is an SVM model; the second model is an LSTM model.

8. The method according to claim 7, characterized in that, Fault classification based on the low-dimensional device feature data includes: The low-dimensional device feature data is normalized to determine whether it is linearly separable or non-linearly separable. For linearly separable low-dimensional device feature data, an objective function is constructed to find the optimal hyperplane through Lagrange duality for fault classification. For low-dimensional equipment feature data that can be nonlinearly separable, fault classification is performed by introducing a kernel function and solving the dual problem through quadratic programming.

9. The method according to claim 7, characterized in that, Fault prediction based on the low-dimensional device feature data includes: The low-dimensional device feature data is converted into a format suitable for LSTM input; The converted low-dimensional equipment feature data is input into the second model, and the equipment fault prediction result is output. The second model is a model trained with historical fault data. The historical fault data includes equipment data and corresponding equipment fault labels.

10. An industrial equipment fault prediction device, characterized in that, The device includes: a communication unit and a processing unit; The communication unit is used to acquire device data; the device data includes first data and second data; the first data follows a normal distribution; the second data does not follow a normal distribution. The processing unit is used to remove outliers from the equipment data to obtain third data; to perform noise filtering on the third data using a moving average filtering algorithm; to extract features from the noise-filtered third data to obtain equipment feature data; the equipment feature data includes time-domain feature data and frequency-domain feature data; to calculate the correlation coefficient between the equipment feature data and the equipment fault label, and to select and remove irrelevant features based on a threshold; to perform principal component analysis on the equipment feature data after removing irrelevant features to obtain low-dimensional equipment feature data; and to input the low-dimensional equipment feature data into a dual-channel model for fault classification and fault prediction; the dual-channel model includes a first model and a second model; the first model is used for fault classification based on the low-dimensional equipment feature data; and the second model is used for fault prediction based on the low-dimensional equipment feature data.

Images