Industrial process operation monitoring system based on data analysis

Abstract

The present invention relates to the technical field of industrial process equipment operation monitoring, and proposes an industrial process operation monitoring system based on data analysis, including: a data acquisition module, used to acquire image data, voiceprint data, and vibration data during the operation of mechanical equipment, a data optimization module, used to optimize the acquired image data, voiceprint data, and vibration data, a model prediction module, used to predict the optimized image data, voiceprint data, and vibration data, and an algorithm verification module, which uses a cross-validation method to verify the prediction results. The present invention realizes that under the same computing power, by acquiring a small amount of accurate data to reflect the operating status of mechanical equipment in the industrial process, it ensures that data can be acquired in real time, and the data is processed, the operating status of mechanical equipment is monitored and predicted, the accuracy of the prediction of the operating status of mechanical equipment is improved, and the fault can be warned in time before the mechanical equipment fails.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process equipment operation monitoring, and in particular to an industrial process operation monitoring system based on data analysis. Background Art

[0002] With the development and progress of science and technology, the equipment running in the industrial process of enterprises is becoming more and more complex, and at the same time, the degree of automation is getting higher and higher. However, these mechanical equipment, especially large and high-precision equipment, will cause significant economic losses once a failure occurs. These equipment have scheduled maintenance dates, but some failures often occur a few days before the scheduled maintenance date. Therefore, it is particularly important to monitor and warn the equipment running in the industrial process of enterprises, so as to evaluate the reliability of the equipment and optimize the maintenance strategy.

[0003] The operation of mechanical equipment in industrial processes is a dynamic process, but traditional reliability mainly relies on obtaining the operating data of the current acquisition equipment and comparing it with the preset data in the database. It can only solve the reliability problems in the manufacturing stage, and the reliability results obtained are not very accurate. However, if the equipment is monitored in real time, the data to be processed is relatively large, the computing power required is relatively large, and the comparison results may be delayed, resulting in the warning result being issued only after the equipment fails, causing damage to the equipment.

[0004] Therefore, we made improvements to this and proposed an industrial process operation monitoring system based on data analysis. Summary of the invention

[0005] (I) The technical purpose of the present invention is: under the same computing power, by obtaining a small amount of accurate data to reflect the operating status of mechanical equipment in the industrial process, ensuring that the data can be obtained in real time, and processing these data, monitoring and predicting the operating status of mechanical equipment, improving the accuracy of the prediction of the operating status of mechanical equipment, and being able to timely warn of faults before the mechanical equipment fails.

[0006] (II) Technical solution

[0007] In order to achieve the above-mentioned object of the invention, the present invention provides an industrial process operation monitoring system based on data analysis, comprising:

[0008] The data acquisition module is used to obtain image data, voiceprint data, and vibration data during the operation of mechanical equipment. The image data is collected by a high-resolution industrial camera, which can capture clear images of the equipment at different operation stages, providing detailed visual information for subsequent image recognition and fault diagnosis. The voiceprint data is obtained with the help of advanced audio sensors, providing clear voiceprints for the subsequent timely discovery of potential faults corresponding to abnormal sounds. The vibration data is obtained through a precise vibration sensor, which accurately measures the vibration amplitude, frequency, and direction parameters of the equipment, providing an important basis for evaluating the operating stability and health status of the equipment.

[0009] The data optimization module is used to optimize the acquired image data, voiceprint data and vibration data. The infrared image stitching method is used for image data processing, and the grayscale distribution histogram is used to represent the statistical tool of image pixel distribution, so as to intuitively understand the proportion of pixels of different grayscale levels in the image. The noise reduction overlapping method is used for voiceprint data processing, and then the discrete Fourier transform is used to apply the Hamming window to each frame signal before the transform to increase the continuity of the signal at both ends, thereby reducing the distortion caused by the Fourier transform. The Internet of Things technology of the SVM algorithm is used to build a model for vibration data processing to extract the feature space in the vibration data.

[0010] The model prediction module is used to predict the optimized image data, voiceprint data and vibration data. For image data, high-quality infrared images can be obtained by calculating the pixel distribution probability, improving the image contrast and making the image grayscale level smoother and more continuous. For voiceprint data, feature vectors are obtained by dividing the frame signal to form a feature vector group. For vibration data, the optimal hyperplane is used to maximize the boundary of the vibration category, and the Lagrange multiplier method is used to obtain the optimal solution.

[0011] The algorithm verification module uses cross-validation to verify the prediction results;

[0012] The risk assessment module uses a time failure model to assess the risk level of the operating status of mechanical equipment;

[0013] The training optimization module uses a binary tree algorithm model to train and learn data samples to obtain the optimal function parameters.

[0014] Preferably, the industrial process operation monitoring system monitors mechanical equipment including the following steps:

[0015] a. After acquiring the operation image data, voiceprint data and vibration data of the mechanical equipment, the data acquisition module uploads the data to the data optimization module through data transmission technology to optimize the data;

[0016] b. The data optimization module uploads the optimized data to the model prediction module through data transmission technology for data prediction;

[0017] c. The model prediction module uploads the predicted data to the algorithm verification module and the risk assessment module simultaneously through data transmission technology, and judges the future operation of the equipment according to the verification results and the assessment results;

[0018] d. The algorithm verification module and the risk assessment module simultaneously send the verification results and the assessment results to the training optimization module through data transmission technology to perform optimization training and reduce the overall calculation amount.

[0019] Preferably, in step a, when optimizing the image data, the image quality index is selected as image entropy, and the image information amount is calculated. , the formula is , where The grayscale value ratio of the processed image. The grayscale distribution histogram represents the pixel distribution of the image. The value range is 0-255. The image is spliced ​​and the contrast is set. The index represents the color richness of the image, and the formula is , where is the grayscale difference between adjacent pixels, and are adjacent pixels;

[0020] When optimizing the voiceprint data, the noise signal is framed, the adjacent two frames are continuous, and the number of frames for overlapping signals is , the formula is , where is the noise signal length, is the frame length, is the overlap ratio;

[0021] When optimizing vibration data, principal component analysis is used to transform the original data into a new feature space and obtain the data covariance matrix. The eigenvalues ​​and eigenvectors of are given by , , where is the sample point, is the sample mean, is the number of non-missing values, is the feature vector, is the corresponding eigenvalue, is the standard deviation.

[0022] Preferably, in step b, when predicting the optimized image data, the K nearest neighbor algorithm is used to classify the information fused mechanical equipment operation fault location target tracking strategy and calculate the sample spacing value , the formula is , where and Respectively represent the feature coordinate values ​​in the test and training samples, and at the same time, set the information entropy value , using the decision tree algorithm to calculate the information entropy value of different nodes The calculation formula is , where The probability of generating attribute judgment for the node, is the total amount of node attribute scores, when When , the decision tree of the centralized detection target tracking strategy has a higher convergence speed, which is ready for target tracking monitoring.

[0023] Preferably, in step b, when predicting the optimized voiceprint data, characteristic vectors are obtained according to the framed signals to form a characteristic vector group, and based on the cepstrum coefficients in the Mel frequency domain, the characteristic vectors corresponding to the framed signals are used to obtain the corresponding parameters obtained after signal filtering. , the formula is , , where is the number of filters, is the number of FFT points, is the FFT of the preprocessed framed signal, are the filter parameters, , where is the center frequency of the triangular filter, and the After the value is calculated, the logarithm is removed and discrete cosine transformation is performed. Then the PCA algorithm is used to reduce the dimension of the calculated high-dimensional MFCC feature vector, establish the feature vector matrix, and calculate the variance contribution rate. and cumulative variance contribution , the formula is , , monitor and obtain the fault type.

[0024] Preferably, in step b, when predicting the optimized vibration data, the objective function of SVM is defined to maximize the boundary interval, which is divided into a linearly separable case and a nonlinearly separable case. In the linearly separable case, the formulas of the objective function and the constraint conditions are: , , , where is the normal vector of the hyperplane, is the bias term, is the sample point, is the corresponding category label, the objective function The minimization of is equivalent to the interval Maximization of, in the case of nonlinear separability, , , , , where is the penalty parameter, For classification error, the Lagrange multiplier is introduced and , construct the Lagrangian function, the formula is , solve , and The condition that the partial derivative of is 0 is , , , which is transformed The optimization is , , .

[0025] Preferably, in step c, -fold cross validation, the data set Randomly split into Mutually exclusive subsets , , , , each As a validation set, the rest As training sets, each subset is used as a validation set, and the estimated error of cross-derivative is The average value of the iterative error is given by , where For use As a subset during training The test error on and the parameters of the PBF kernel , define each set of parameters The cross-validation error is given by , where For use under certain parameters When used as a training set, in the subset The test error on the subsequent optimization determines the parameters and The optimal solution when the cross-validation error is the smallest is: .

[0026] Preferably, in step c, the Weibull distribution failure model in the time failure model is used to calculate the probability density function of the failure operation of the mechanical equipment. , the formula is , where For time, is the shape parameter, is the scale parameter, is the reliability function, is the failure rate function, when The mean failure time ,in, is the gamma function, when When , it degenerates into an exponential distribution, .

[0027] Preferably, in step d, the binary tree algorithm model used for the data sample includes a classification task algorithm and a regression task algorithm, and the classification task algorithm calculates the cross entropy function , the formula is , where For the real situation of mechanical equipment failure, When , it means that the mechanical equipment will not fail when it is judged as a failure. When the mechanical equipment is judged as a fault, it will fail. The probability that the model predicts the positive class.

[0028] Preferably, the regression task algorithm calculates the average of the squared differences between the model predictions and the true values , collect a sample set The mean square error formula is , where is the average true value of mechanical equipment failure, The average predicted value for the model predicting failures.

[0029] (III) Beneficial effects

[0030] Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art:

[0031] 1. Real-time acquisition of image data, voiceprint data, and vibration data of the operating status of mechanical equipment in industrial processes, and real-time pre-processing of the acquired data, splicing of images using grayscale distribution histograms, and setting of contrast by grayscale difference The indicator represents the color richness of the image, frames the voiceprint data, merges the overlapping signals, and uses the covariance matrix to extract the eigenvalues ​​and eigenvectors in the vibration data. The acquired image data, voiceprint data, and vibration data can already reflect the operating status of the mechanical equipment, and these data are preprocessed. Compared with the traditional acquisition of various data such as temperature, speed, current, voltage transmission speed, etc. in the mechanical equipment, these data are then processed to judge the current operating status of the mechanical equipment and predict the future operating status. When less data is acquired, the operating status of the mechanical equipment can be fully reflected, and the future operating status can be accurately predicted. Under the same computing power, the processing speed is faster, and preprocessing of these data can further improve the speed of subsequent processing, thereby making the real-time acquisition of data higher and the accuracy of mechanical equipment failure prediction higher.

[0032] 2. By using corresponding models to predict the pre-processed image data, voiceprint data and vibration data, the model can improve the detection accuracy of the operating status of mechanical equipment, shorten the analysis time, and make the prediction results more accurate, providing an effective monitoring and early warning method for the intelligent management of the operating status of mechanical equipment and the efficiency of fault detection.

[0033] 3. The accuracy of the prediction results can be verified by cross-validation. The prediction results of image data, voiceprint data and vibration data can be cross-validated to further determine the judgment results of the impending failure of the operating status of the mechanical equipment. By establishing a time failure model, the failure time of the equipment under different usage conditions can be predicted, providing a decision-making basis for equipment maintenance, repair and replacement.

[0034] 4. The use of the binary tree algorithm model continuously improves the performance and accuracy of the model in predicting the image data, voiceprint data and vibration data in the operating status of mechanical equipment, and makes it easy to compare the effects of different model parameter settings or different algorithms during the training process, so as to select the optimal training strategy and provide the optimal parameters for the optimization model, so that the real-time monitoring of the operating status of mechanical equipment is higher, the accuracy of fault prediction is improved, and the accuracy and timeliness of fault warnings that may occur in the future during the operating status of mechanical equipment are improved, so that the equipment will not be damaged due to faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0036] Figure 1A schematic diagram of the process flow of the industrial process operation monitoring system based on data analysis provided in this application;

[0037] Figure 2 A schematic diagram of the calculated reliability of precision equipment A and B in an enterprise of an industrial process operation monitoring system based on data analysis provided by this application;

[0038] Figure 3 A schematic diagram of the risk level of precision equipment A in an enterprise of an industrial process operation monitoring system based on data analysis provided by this application;

[0039] Figure 4 Schematic diagram of the risk level of precision equipment B in an enterprise of an industrial process operation monitoring system based on data analysis provided in this application. DETAILED DESCRIPTION

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings and other implementation methods can be obtained based on these drawings without creative work.

[0041] like Figure 1-Figure 4 As shown, this embodiment proposes an industrial process operation monitoring system based on data analysis, including:

[0042] The data acquisition module is used to obtain image data, voiceprint data, and vibration data during the operation of mechanical equipment. The image data is collected by a high-resolution industrial camera, which can capture clear images of the equipment at different operation stages, providing detailed visual information for subsequent image recognition and fault diagnosis. The voiceprint data is obtained with the help of advanced audio sensors, providing clear voiceprints for the subsequent timely discovery of potential faults corresponding to abnormal sounds. The vibration data is obtained through a precise vibration sensor, which accurately measures the vibration amplitude, frequency, and direction parameters of the equipment, providing an important basis for evaluating the operating stability and health status of the equipment.

[0043] The data optimization module is used to optimize the acquired image data, voiceprint data and vibration data. The infrared image stitching method is used for image data processing, and the grayscale distribution histogram is used to represent the statistical tool of image pixel distribution, so as to intuitively understand the proportion of pixels of different grayscale levels in the image. The noise reduction overlapping method is used for voiceprint data processing, and then the discrete Fourier transform is used to apply the Hamming window to each frame signal before the transform to increase the continuity of the signal at both ends, thereby reducing the distortion caused by the Fourier transform. The Internet of Things technology of the SVM algorithm is used to build a model for vibration data processing to extract the feature space in the vibration data.

[0044] The model prediction module is used to predict the optimized image data, voiceprint data and vibration data. For image data, high-quality infrared images can be obtained by calculating the pixel distribution probability, improving the image contrast and making the image grayscale level smoother and more continuous. For voiceprint data, feature vectors are obtained by dividing the frame signal to form a feature vector group. For vibration data, the optimal hyperplane is used to maximize the boundary of the vibration category, and the Lagrange multiplier method is used to obtain the optimal solution.

[0045] The algorithm verification module uses cross-validation to verify the prediction results;

[0046] The risk assessment module uses a time failure model to assess the risk level of the operating status of mechanical equipment;

[0047] The training optimization module uses a binary tree algorithm model to train and learn data samples to obtain the optimal function parameters.

[0048] In this embodiment, the industrial process operation monitoring system monitors mechanical equipment including the following steps:

[0049] a. After acquiring the operation image data, voiceprint data and vibration data of the mechanical equipment, the data acquisition module uploads the data to the data optimization module through data transmission technology to optimize the data;

[0050] b. The data optimization module uploads the optimized data to the model prediction module through data transmission technology for data prediction;

[0051] c. The model prediction module uploads the predicted data to the algorithm verification module and the risk assessment module simultaneously through data transmission technology, and judges the future operation of the equipment according to the verification results and the assessment results;

[0052] d. The algorithm verification module and the risk assessment module simultaneously send the verification results and the assessment results to the training optimization module through data transmission technology to perform optimization training and reduce the overall calculation amount.

[0053] In this embodiment, in step a, when optimizing the image data, the image quality index is selected as the image entropy, and the image information amount is calculated. , the formula is , where The grayscale value ratio of the processed image. The grayscale distribution histogram represents the pixel distribution of the image. The value range is 0-255. The image is spliced ​​and the contrast is set. The index represents the color richness of the image, and the formula is , where is the grayscale difference between adjacent pixels, and are adjacent pixels;

[0054] When optimizing the voiceprint data, the noise signal is framed, the adjacent two frames are continuous, and the number of frames for overlapping signals is , the formula is , where is the noise signal length, is the frame length, is the overlap ratio;

[0055] When optimizing vibration data, principal component analysis is used to transform the original data into a new feature space and obtain the data covariance matrix. The eigenvalues ​​and eigenvectors of are given by , , where is the sample point, is the sample mean, is the number of non-missing values, is the feature vector, is the corresponding eigenvalue, is the standard deviation.

[0056] Specifically, the image data, voiceprint data and vibration data of the operating status of mechanical equipment in the industrial process are obtained in real time, and the obtained data is pre-processed in real time, and the image is spliced ​​using the grayscale distribution histogram, and the contrast is set by the grayscale difference. The indicator represents the color richness of the image, frames the voiceprint data, merges the overlapping signals, and uses the covariance matrix to extract the eigenvalues ​​and eigenvectors in the vibration data. The acquired image data, voiceprint data, and vibration data can already reflect the operating status of the mechanical equipment, and these data are preprocessed. Compared with the traditional acquisition of various data such as temperature, speed, current, voltage transmission speed, etc. in the mechanical equipment, these data are then processed to judge the current operating status of the mechanical equipment and predict the future operating status. When less data is acquired, the operating status of the mechanical equipment can be fully reflected, and the future operating status can be accurately predicted. Under the same computing power, the processing speed is faster, and preprocessing of these data can further improve the speed of subsequent processing, thereby making the real-time acquisition of data higher and the accuracy of mechanical equipment failure prediction higher.

[0057] In this embodiment, in step b, when predicting the optimized image data, the K nearest neighbor algorithm is used to classify the information fused mechanical equipment operation fault location target tracking strategy and calculate the sample spacing value. , the formula is , where and Respectively represent the feature coordinate values ​​in the test and training samples, and at the same time, set the information entropy value , using the decision tree algorithm to calculate the information entropy value of different nodes The calculation formula is , where The probability of generating attribute judgment for the node, is the total amount of node attribute scores, when When the decision tree of the centralized detection target tracking strategy has a higher convergence speed, it is ready for target tracking monitoring, can improve the contrast of the image, make the details of each pixel in the image more obvious, better show the subtle changes of the image, and at the same time, improve the entropy of the image, enhance the processing degree of the total amount of information, make the image content rich, can obtain more feature points, and make the image processing smoother.

[0058] In this embodiment, in step b, when predicting the optimized voiceprint data, characteristic vectors are obtained according to the framed signals to form a characteristic vector group. Based on the cepstrum coefficients in the Mel frequency domain, the characteristic vectors corresponding to the framed signals are used to obtain the corresponding parameters after signal filtering. , the formula is , , where is the number of filters, is the number of FFT points, is the FFT of the preprocessed framed signal, is the filter parameter, , where is the center frequency of the triangular filter, and the After the value is calculated, the logarithm is removed and discrete cosine transformation is performed. Then the PCA algorithm is used to reduce the dimension of the calculated high-dimensional MFCC feature vector, establish the feature vector matrix, and calculate the variance contribution rate. and cumulative variance contribution , the formula is , , monitoring to obtain the fault type, and extracting features from the voiceprint data can achieve continuous and uninterrupted monitoring of the operation of mechanical equipment, which can fully reflect the changes in the operating status of the mechanical equipment during its life cycle, provide effective data for correct early warning, and improve the accuracy of early warning.

[0059] In this embodiment, in step b, when predicting the optimized vibration data, the objective function of SVM is defined to maximize the boundary interval, which is divided into a linearly separable case and a nonlinearly separable case. In the linearly separable case, the formulas of the objective function and the constraint conditions are: , , , where is the normal vector of the hyperplane, is the bias term, is the sample point, is the corresponding category label, the objective function The minimization of is equivalent to the interval Maximization of, in the case of nonlinear separability, , , , , where is the penalty parameter, For classification error, the Lagrange multiplier is introduced and , construct the Lagrangian function, the formula is , solve , and The condition that the partial derivative of is 0 is , , , which is transformed The optimization is , , The added penalty parameters and classification errors can make the prediction of the results more accurate when processing vibration data. The introduction of the Lagrange multiplier method can ensure the optimal solution of the constrained optimization problem under certain conditions, making the prediction results more reliable in both theoretical research and practical applications.

[0060] Specifically, by using corresponding models to predict the preprocessed image data, voiceprint data and vibration data, the model can improve the detection accuracy of the operating status of mechanical equipment, shorten the analysis time, and make the prediction results accurate, providing an effective monitoring and early warning means for the intelligent management of the operating status of mechanical equipment and the efficiency of fault detection.

[0061] In this embodiment, in step c, -fold cross validation, the data set Randomly split into Mutually exclusive subsets , , , , each As a validation set, the rest As training sets, each subset is used as a validation set, and the estimated error of cross-derivative is The average value of the iterative error is given by , where For use As a subset during training The test error on and the parameters of the PBF kernel , define each set of parameters The cross-validation error is given by , where For use under certain parameters When used as a training set, in the subset The test error on the subsequent optimization determines the parameters and The optimal solution when the cross-validation error is the smallest is: , determine the optimal parameters and After that, you can use the parameters in the entire dataset Revalidate and train the SVM model on.

[0062] In this embodiment, in step c, the Weibull distribution failure model in the time failure model is used to calculate the probability density function of the failure operation of the mechanical equipment. , the formula is , where For time, is the shape parameter, is the scale parameter, is the reliability function, is the failure rate function, when The mean failure time ,in, is the gamma function, when When , it degenerates into an exponential distribution, , by analyzing historical data or based on theoretical assumptions, determining the appropriate failure time distribution function, the reliability and life of mechanical equipment can be evaluated.

[0063] Specifically, the accuracy of the prediction results can be verified through cross-validation. The prediction results of image data, voiceprint data and vibration data can be cross-validated to further determine the judgment results of the impending failure of the operating status of the mechanical equipment. By establishing a time failure model, the failure time of the equipment under different usage conditions can be predicted, providing a decision-making basis for equipment maintenance, repair and replacement.

[0064] In this embodiment, in step d, the binary tree algorithm model used for the data sample includes a classification task algorithm and a regression task algorithm. The classification task algorithm calculates the cross entropy function , the formula is , where For the real situation of mechanical equipment failure, When , it means that the mechanical equipment will not fail when it is judged as a failure. When the mechanical equipment is judged as a fault, it will fail. The probability that the model predicts the positive class.

[0065] In this embodiment, the regression task algorithm calculates the average of the squared differences between the model prediction value and the true value , collect a sample set The mean square error formula is , where is the average true value of mechanical equipment failure, The average predicted value for the model predicting failures.

[0066] Specifically, the use of the binary tree algorithm model enables the model's performance and accuracy to continuously improve in the prediction tasks of image data, voiceprint data, and vibration data in the operating status of mechanical equipment, and makes it easy to compare the effects of different model parameter settings or different algorithms during the training process, so as to select the optimal training strategy and provide the optimal parameters for the optimization model, so that the real-time monitoring of the operating status of mechanical equipment is higher, the accuracy of fault prediction is improved, and the accuracy and timeliness of fault warnings that may occur in the future during the operating status of mechanical equipment are improved, so that the equipment will not be damaged due to faults.

[0067] like Figure 2-Figure 4 As shown in the figure, the curve is the operating feudal threshold determined based on the experience of the manufacturing company's experts and engineering technicians, taking into account the different operating time and operating environment of equipment A and B. When the risk value is below the curve, it means that the equipment is in a good operating state. Figure 3 and Figure 4 Points B1 and B2 indicate that the equipment is predicted to have deteriorated or is about to fail, and it is recommended to overhaul or replace the equipment. After the equipment is shut down for overhaul or replacement, point B3 indicates that the equipment is in good operating condition and does not need to be shut down for overhaul. Therefore, the monitoring method proposed in the present invention can effectively predict and warn mechanical equipment, evaluate the reliability of equipment operation, optimize maintenance strategies, prevent unnecessary shutdowns for overhaul and untimely warnings, and reduce maintenance costs.

[0068] in, Figure 2 The horizontal axis in the middle represents the working time of the equipment, and the vertical axis represents the reliability value of the equipment's operating status. The larger the reliability value, the higher the reliability of the equipment's operation. Figure 3 and Figure 4 In the figure, the horizontal axis represents the working time of the equipment, and the vertical axis represents the risk value of the equipment operation status. The lower the risk value, the smaller the possibility of failure during equipment operation.

[0069] Table 1 Comparison table of mechanical equipment operation status monitoring in industrial processes

[0070] group Actual number of failures Determine the number of faults Correct rate % 1 268 268 100 2 268 295 90.85 3 268 249 92.91 4 268 236 88.05 5 268 275 97.45 6 268 231 86.19

[0071] It can be seen from Table 1 that this is a comparison table of all fault predictions and actual maintenance of all equipment within the enterprise within one year from April 1, 2023 to March 31, 2024. Among them, Group 1 is the monitoring system provided in the present invention, and Groups 2-6 are the detection systems used in the enterprise over the years. It is found through the fault judgment accuracy that Group 1 has the highest fault judgment accuracy, reaching 100%, and other groups have misjudgments, which shows that in the present invention, a relatively accurate fault identification, judgment and warning is achieved, and a good monitoring effect is achieved.

[0072] The above embodiments are only used to illustrate the present invention, but not to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that various combinations, modifications or equivalent substitutions of the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and should be included in the scope of the claims of the present invention.

Claims

1. Industrial process operation monitoring system based on data analysis, characterized in that: include: The data acquisition module is used to obtain image data, voiceprint data, and vibration data during the operation of mechanical equipment. The image data is collected by a high-resolution industrial camera, which can capture clear images of the equipment at different operation stages, providing detailed visual information for subsequent image recognition and fault diagnosis. The voiceprint data is obtained with the help of advanced audio sensors, providing clear voiceprints for the subsequent timely discovery of potential faults corresponding to abnormal sounds. The vibration data is obtained through a precise vibration sensor, which accurately measures the vibration amplitude, frequency, and direction parameters of the equipment, providing an important basis for evaluating the operating stability and health status of the equipment. The data optimization module is used to optimize the acquired image data, voiceprint data and vibration data. The infrared image stitching method is used for image data processing, and the grayscale distribution histogram is used to represent the statistical tool of image pixel distribution, so as to intuitively understand the proportion of pixels of different grayscale levels in the image. The noise reduction overlapping method is used for voiceprint data processing, and then the discrete Fourier transform is used to apply the Hamming window to each frame signal before the transform to increase the continuity of the signal at both ends, thereby reducing the distortion caused by the Fourier transform. The Internet of Things technology of the SVM algorithm is used to build a model for vibration data processing to extract the feature space in the vibration data. The model prediction module is used to predict the optimized image data, voiceprint data and vibration data. For image data, high-quality infrared images can be obtained by calculating the pixel distribution probability, improving the image contrast and making the image grayscale level smoother and more continuous. For voiceprint data, feature vectors are obtained by dividing the frame signal to form a feature vector group. For vibration data, the optimal hyperplane is used to maximize the boundary of the vibration category, and the Lagrange multiplier method is used to obtain the optimal solution. The algorithm verification module uses cross-validation to verify the prediction results; The risk assessment module uses a time failure model to assess the risk level of the operating status of mechanical equipment; The training optimization module uses a binary tree algorithm model to train and learn data samples to obtain the optimal function parameters; The industrial process operation monitoring system includes the following steps for monitoring mechanical equipment: a. After acquiring the operation image data, voiceprint data and vibration data of the mechanical equipment, the data acquisition module uploads the data to the data optimization module through data transmission technology to optimize the data; b. The data optimization module uploads the optimized data to the model prediction module through data transmission technology for data prediction; c. The model prediction module uploads the predicted data to the algorithm verification module and the risk assessment module simultaneously through data transmission technology, and judges the future operation of the equipment according to the verification results and the assessment results; d. The algorithm verification module and the risk assessment module simultaneously send the verification results and the assessment results to the training optimization module through data transmission technology to perform optimization training and reduce the overall calculation amount; In step a, when optimizing the image data, the image quality index is selected as the image entropy, and the image information volume is calculated. , the formula is , where The grayscale value ratio of the processed image. The grayscale distribution histogram represents the pixel distribution of the image. The value range is 0-255. The image is spliced ​​and the contrast is set. The index represents the color richness of the image, and the formula is , where is the grayscale difference between adjacent pixels, and are adjacent pixels; When optimizing the voiceprint data, the noise signal is framed, the adjacent two frames are continuous, and the number of frames for overlapping signals is , the formula is , where is the noise signal length, is the frame length, is the overlap ratio; When optimizing vibration data, principal component analysis is used to transform the original data into a new feature space to obtain the data covariance matrix. The eigenvalues ​​and eigenvectors of are given by , , where is the sample point, is the sample mean, is the number of non-missing values, is the feature vector, is the corresponding eigenvalue, is the standard deviation; In the step c, using -fold cross validation, the data set Randomly split into Mutually exclusive subsets , , , , each As a validation set, the rest As training sets, each subset is used as a validation set, and the estimated error of cross-derivative is The average value of the iterative error is given by , where For use As a subset during training The test error on and the parameters of the PBF kernel , define each set of parameters The cross-validation error is given by , where For use under certain parameters When used as a training set, in the subset The test error on the subsequent optimization determines the parameters and The optimal solution when the cross-validation error is the smallest is: , using the Weibull distribution failure model in the time failure model to calculate the probability density function of mechanical equipment failure operation , the formula is , where For time, is the shape parameter, is the scale parameter, is the reliability function, is the failure rate function, when The mean failure time ,in, is the gamma function, when When , it degenerates into an exponential distribution, ; In step d, the binary tree algorithm model used for the data sample includes a classification task algorithm and a regression task algorithm, and the classification task algorithm calculates the cross entropy function , the formula is , where For the real situation of mechanical equipment failure, When , it means that the mechanical equipment will not fail when it is judged as a failure. When the mechanical equipment is judged as a fault, it will fail. The probability that the model predicts the positive class.

2. The industrial process operation monitoring system based on data analysis according to claim 1 is characterized in that: In step b, when predicting the optimized image data, the K nearest neighbor algorithm is used to classify the information fused mechanical equipment operation fault location target tracking strategy and calculate the sample spacing value. , the formula is , where and Respectively represent the feature coordinate values ​​in the test and training samples, and at the same time, set the information entropy value , using the decision tree algorithm to calculate the information entropy value of different nodes The calculation formula is , where The probability of generating attribute judgment for the node, is the total amount of node attribute score, when When , the decision tree of the centralized detection target tracking strategy has a higher convergence speed, which is ready for target tracking monitoring.

3. The industrial process operation monitoring system based on data analysis according to claim 1 is characterized in that: In the step b, when predicting the optimized voiceprint data, characteristic vectors are obtained according to the framed signals to form a characteristic vector group. Based on the cepstrum coefficients in the Mel frequency domain, the characteristic vectors corresponding to the framed signals are used to obtain the corresponding parameters after signal filtering. , the formula is , , where is the number of filters, is the number of FFT points, is the FFT of the preprocessed framed signal, are the filter parameters, , where is the center frequency of the triangular filter, and the After the value is calculated, the logarithm is removed and discrete cosine transformation is performed. Then the PCA algorithm is used to reduce the dimension of the calculated high-dimensional MFCC feature vector, establish the feature vector matrix, and calculate the variance contribution rate. and cumulative variance contribution , the formula is , , monitor and obtain the fault type.

4. The industrial process operation monitoring system based on data analysis according to claim 1 is characterized in that: In step b, when predicting the optimized vibration data, the objective function of SVM is defined to maximize the boundary interval, which is divided into the linear separable case and the nonlinear separable case. In the linear separable case, the formula of the objective function and the constraint condition is: , , , where is the normal vector of the hyperplane, is the bias term, is the sample point, is the corresponding category label, the objective function Minimizing is equivalent to the interval Maximization of, in the case of nonlinear separability, , , , , where is the penalty parameter, For classification error, the Lagrange multiplier is introduced and , construct the Lagrangian function, the formula is , solve , and The condition that the partial derivative of is 0 is , , , which is transformed The optimization is , , .

5. The industrial process operation monitoring system based on data analysis according to claim 1 is characterized in that: The regression task algorithm calculates the average of the squared differences between the model predictions and the true values , collect a sample set The mean square error formula is , where is the average true value of mechanical equipment failure, The average predicted value for the model predicting failures.

Images