Device damage intelligent monitoring method and system based on mechanism and working condition big data

Abstract

The present invention relates to the technical field of industrial device detection, and discloses a device damage intelligent monitoring method and system based on mechanism and working condition big data. The monitoring method comprises: first, setting various damage modes of a device, constructing a mechanism model sample of each damage mode on the basis of a technical standard and an engineering case, and constructing a working condition big data sample of each damage mode on the basis of field data; then, using the mechanism model sample and the working condition big data sample as data sets for model training so as to respectively construct a mechanism model and a working condition model; and finally, inputting real-time text data to be predicted and real-time working condition data to be predicted into the mechanism model and the working condition model, respectively, calculating the possibility that the real-time text data belongs to various damage modes and the possibility that the real-time working condition data belongs to various damage modes, and generating a device damage monitoring result on the basis of the two possibilities. The present invention effectively improves the utilization rate of real-time information during damage monitoring of a pressure-bearing device, so that the monitoring result is more accurate.

Description

Intelligent monitoring method and system for equipment damage based on mechanism and working condition big data

[0001] This application claims priority to the “Intelligent monitoring method and system for equipment damage based on mechanism and working condition big data” with application number 202410310116.9 filed on March 19, 2024, and the original accepting agency is China. Technical Field

[0002] The present invention relates to the field of industrial equipment detection technology, and in particular to an intelligent equipment damage monitoring method based on mechanism and working condition big data, and an intelligent damage pattern prediction system applying this method. Background Art

[0003] Pressure-bearing equipment in process industries such as petrochemicals operates under complex processes and fluctuating conditions, and the risk of equipment damage and failure increases accordingly. Monitoring and early warning of possible damage to pressure-bearing equipment based on the application scenarios of the pressure equipment (such as the type of device / equipment, equipment materials, involved process media, etc.) and operating parameters (such as temperature, pressure, pH value, corrosive medium content, etc.) play an important role in ensuring the safe and reliable operation of the equipment.

[0004] Currently, damage prevention and control for pressure-bearing equipment typically relies on passive methods, such as regular or irregular measurements of locations where thinning has occurred and crack monitoring of areas where cracks have occurred. These methods detect damage after it has occurred and are unable to automatically prevent and control damage to operating equipment once it occurs. Furthermore, analyzing damage patterns in pressure-bearing equipment during service requires specialized databases and extensive expert experience, making large-scale automated damage analysis impossible. This severely impacts the safe, long-term operation of pressure-bearing equipment and is therefore an urgent issue that needs to be addressed. Summary of the Invention

[0005] In order to avoid and overcome the technical problems of low real-time information utilization, inaccurate monitoring results and inability to automatically identify damage to pressure equipment in the existing technology, the present invention provides an intelligent equipment damage monitoring method and system based on mechanism and working condition big data.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention discloses an intelligent equipment damage monitoring method based on mechanism and working condition big data, which is used to monitor the damage mode of pressure-bearing equipment; the monitoring method includes steps S1 to S3.

[0008] S1. Set multiple damage modes of the equipment, build a mechanism model sample for each damage mode based on technical standards and engineering cases, and build a working condition big data sample for each damage mode based on field data.

[0009] S2. Use the mechanism model samples and working condition big data samples as data sets for model training to construct the mechanism model and working condition model respectively.

[0010] S3. Input the real-time text data and real-time operating condition data to be predicted into the mechanism model and the operating condition model respectively, calculate the possibility that the real-time text data belongs to various damage modes, and the possibility that the real-time operating condition data belongs to various damage modes, and combine the two possibilities to generate equipment damage monitoring results.

[0011] As a further improvement of the above solution, before step S1, a custom word library and a stop word library related to equipment damage patterns are also constructed.

[0012] Step S1 includes the following specific steps, namely S11 to S14.

[0013] S11. Obtain mechanism model samples for describing each damage mode from technical standards and engineering cases, and obtain operating condition data samples for describing each damage mode from field data.

[0014] S12. Clean, segment, filter stop words, and remove duplicates from the text samples obtained from the technical standards based on the custom vocabulary and stop word vocabulary to obtain a keyword library for the mechanism model samples.

[0015] S13. Simulate the mechanism model identification method and expert experience, obtain self-generated technical standard keyword expert samples based on the technical standard text sample keyword library, and combine them with the text samples obtained from the engineering case to jointly construct the mechanism model sample.

[0016] S14. Retain preset types of operating condition parameters in the operating condition data sample and delete the remaining types of operating condition parameters, thereby obtaining the operating condition big data sample.

[0017] As a further improvement of the above solution, in step S11, the technical standard adopts the GB / T 30579 technical standard; wherein the GB / T 30579 technical standard includes the damage mechanism, damage morphology, main influencing factors, prone devices or equipment, and main preventive measures of the damage mode; the engineering case includes the material name, medium and the name of the damaged component.

[0018] As a further improvement of the above scheme, in step S14, the types of operating parameters retained include: pressure, temperature, pH value, H2S content, H2O content, CO2 content, sulfide content, NH3 content, and chloride ion content; among them, when one or more of these media are not contained or provided, the corresponding operating parameters are set to 0.

[0019] As a further improvement to the above solution, in step S13, the process of obtaining the technical standard keyword expert sample is as follows:

[0020] The recognition method of the simulation mechanism model and expert experience are used to analyze the keyword library of the technical standard text samples. The number of generated damage pattern samples for each damage pattern is determined according to the probability of actual occurrence of each damage pattern. Under the premise of not exceeding the total number of keyword libraries of the technical standard text samples for each damage pattern, the conditional random sampling method is used to sample the vocabulary within the preset number of words, and the corresponding damage pattern classification labels are added to the sampled texts for sample enhancement, thereby obtaining self-generated technical standard keyword expert samples.

[0021] As a further improvement of the above solution, in step S2, the method for constructing the mechanism model includes the following steps, namely S2A1 to S2A3.

[0022] S2A1. The vector composed of all damage mode classification labels in the mechanism model sample is The expression formula is:

[0023] Where n is the number of all damage modes, n>1; y n is the classification label of the nth damage pattern; the superscript T is the transpose of the matrix.

[0024] S2A2. Use the TF-IDF algorithm to vectorize the mechanism model samples and convert the text into a sparse matrix form.

[0025] S2A3. The vectorized text data is divided into training and test sets according to a set ratio. The model is trained with the corresponding classification label vectors. Multiple multi-classification machine learning algorithms are used for training. The accuracy, recall rate, and F1 score of the damage pattern prediction results of each algorithm are compared. At the same time, the overall accuracy, average value, and weighted average value of the final prediction results of each damage pattern of different algorithms are compared. Based on the comparison results, the optimal multi-classification machine learning algorithm is selected to construct the mechanism model.

[0026] As a further improvement of the above solution, in step S2, the method for constructing the operating condition model includes the following steps:

[0027] Using working condition big data samples as training sets, combined with corresponding classification label vectors for training, and adopting a variety of multi-classification learning algorithms for training, the accuracy, recall rate, and F1 score of multiple damage pattern prediction results under each algorithm are compared. At the same time, the overall accuracy, average value, and weighted average value of the final results of each damage pattern prediction of different algorithms are compared. Based on the comparison results, the optimal multi-classification machine learning algorithm is selected to construct the working condition model.

[0028] As a further improvement of the above scheme, when constructing the mechanism model, the logistic regression algorithm, SVM, random forest and other algorithms (including but not limited to these algorithms) are used for model training; when constructing the working condition model, the KNN algorithm, random forest algorithm, Bernoulli and other algorithms (including but not limited to these algorithms) are used for model training.

[0029] As a further improvement of the above solution, in step S3, the real-time text data belongs to the first possibility set P of various damage modes. (1) Expressed as: P (1) ={P 11 ,P 12 ,P 13 ,...,P 1n}

[0030] The real-time working condition data belongs to the second possible set P of various damage modes (2) Expressed as: P (2) ={P 21 ,P 22 ,P 23 ,...,P 2n}

[0031] Where, P 1n Indicates the possibility that the real-time text data belongs to the nth damage mode; P 2n Indicates the possibility that the operating condition data belongs to the nth damage mode.

[0032] The probability that the real-time text data and the real-time working condition data belong to the same damage mode is multiplied to obtain the comprehensive probability set {P1, P2, P3, ..., P n}; where P1 = P 11 *P 21 ,P2=P 12 *P 22 ,P3=P 13 *P 23 ,...,P n =P 1n *P 2n .

[0033] Among them, the first possibility set P (1) , the second possibility set P (2) The respective possibilities of the comprehensive possibility set are sorted from large to small, and the top several damage modes and their possibilities are used as the equipment damage monitoring results.

[0034] The present invention also discloses an intelligent damage pattern prediction system based on mechanism and working condition big data, which applies the above-mentioned equipment damage intelligent monitoring method based on mechanism and working condition big data. The system includes: a sample construction module, a model construction module and a calculation module.

[0035] The sample construction module is used to set various damage modes of equipment, build mechanism model samples for each damage mode based on technical standards and engineering cases, and build working condition big data samples for each damage mode based on field data.

[0036] The model building module is used to use the mechanism model samples and the working condition big data samples as data sets for model training to construct the mechanism model and the working condition model respectively.

[0037] The calculation module is used to input the mechanism model sample data and operating condition data to be predicted into the mechanism model and operating condition model respectively, calculate the possibility that the text data belongs to various damage modes, and the possibility that the operating condition data belongs to various damage modes, and thus combine the two possibilities to generate equipment damage monitoring results.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The intelligent equipment damage monitoring method disclosed in the present invention combines the analysis results of the mechanism model and the operating condition model. It does not only focus on a single evaluation criterion. By deeply understanding the advantages of the mechanism model and the operating condition model, it demonstrates the potential correlation between the mechanism analysis and the operating condition analysis factors, and ultimately provides a more comprehensive and accurate equipment damage pattern assessment model. It uses the sample data of the mechanism model to be predicted and the real-time operating condition data to generate equipment damage monitoring results, effectively improving the real-time information utilization rate and making the monitoring results more accurate.

[0040] 2. When constructing the mechanism model, the present invention includes technical standards and engineering case contents. The technical standards and engineering cases describe the damage mechanism, influencing factors, design specifications and other information of the equipment. The information in the engineering case includes the data collected in real time during the operation of the equipment. After comprehensively considering the above two aspects, when constructing the mechanism model, multiple influencing factors of standards and site can be considered, a more comprehensive mechanism model can be established, and the applicability of the model is expanded.

[0041] 3. When training the mechanism model, the present invention independently generates mechanism model samples by simulating the mechanism model identification method and expert experience, analyzing the mechanism knowledge, and adopting the conditional random sampling method to sample, thereby solving the problem of insufficient mechanism model samples caused by lack of experience and insufficient number of personnel, and improving the accuracy of the mechanism model prediction results.

[0042] 4. The present invention can solve the problem that the mechanism model cannot process non-text when processing working condition big data. By collecting working condition parameters closely related to equipment damage (such as temperature, pressure, pH value and other data), these digital data are converted into a representation form that can be analyzed, and a working condition model is constructed. This makes up for the limitations of damage pattern recognition based solely on standard definitions, and at the same time reflects the important role of working condition parameters in equipment damage monitoring.

[0043] 5. The present invention has the advantages of intelligence, timeliness, and early warning. It does not require a large amount of manual operation. By conducting real-time monitoring and analysis of the equipment status, any abnormal situation or potential equipment damage can be promptly warned to the operator or related management system. Before the equipment fails or is damaged, appropriate measures can be taken to avoid the occurrence of risks, thereby reducing maintenance costs and improving equipment reliability.

[0044] 6. The damage pattern intelligent prediction system based on mechanism and working condition big data disclosed in the present invention can produce the same beneficial effects as the above-mentioned equipment damage intelligent monitoring method by applying the method, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] FIG1 is a flow chart of an intelligent equipment damage monitoring method based on mechanism and working condition big data in Example 1 of the present invention.

[0046] FIG2 is a framework diagram of the damage pattern intelligent prediction system based on mechanism and working condition big data in Example 2 of the present invention.

[0047] FIG3 is a diagram showing a visualization interface to be inputted into the damage pattern intelligent prediction system in Example 2 of the present invention.

[0048] FIG4 is a diagram showing the operation and result interface of the damage pattern intelligent prediction system in Example 2 of the present invention.

[0049] Figure 5 is a logic diagram of the traditional passive risk prevention and control method for pressure-bearing equipment.

[0050] FIG6 is a logic diagram of the active risk prevention and control method for pressure-bearing equipment in Example 3 of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the implementation cases described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] Example 1

[0053] Referring to FIG1 , an embodiment of the present invention provides an intelligent equipment damage monitoring method based on mechanism and working condition big data, which is used to monitor damage patterns of pressure-bearing equipment. The monitoring method includes the following steps:

[0054] First, we build a custom vocabulary and stop word vocabulary related to equipment damage patterns. Some of the custom words are as follows:

[0055] [Catalytic cracking, atmospheric distillation, separatory tank, heavy diesel, austenitic stainless steel...]

[0056] Some stop words are as follows:

[0057] [generally, together, also, whether or not, depends on, happens...]

[0058] S1. Set multiple damage modes of the equipment, build a mechanism model sample for each damage mode based on technical standards and engineering cases, and build a working condition big data sample for each damage mode based on field data.

[0059] In this embodiment, a preset number of damage modes (eg, 20) can be set, covering damage modes such as corrosion thinning, environmental cracking, mechanical damage, and material degradation.

[0060] Step S1 may include the following specific steps, namely S11 to S14.

[0061] S11. Obtain text samples used to describe each damage mode from technical standards and engineering cases, and obtain working condition data samples used to describe each damage mode from field data.

[0062] In step S11, the technical standard adopts GB / T 30579, which includes information on damage mechanisms, damage morphology, primary influencing factors, prone devices or equipment, and primary preventive measures. Of course, in some embodiments, the latest standard for identifying damage patterns in pressure-bearing equipment may also be adopted. The engineering case includes information such as the material name, medium, and the name of the damaged component.

[0063] S12. Based on the custom vocabulary and stop word vocabulary, the text samples obtained from the GB / T 30579 technical standard are cleaned, segmented, stop word filtered, and duplicate removed to obtain a keyword library of the technical standard text samples.

[0064] S13. Simulate the mechanism model identification method and expert experience, and obtain self-generated technical standard keyword expert samples based on the technical standard text sample keyword library.

[0065] The process of obtaining the technical standard keyword expert sample is as follows:

[0066] The recognition method of the simulation mechanism model and expert experience are used to analyze the keyword library of the technical standard text samples. The number of self-generated samples of each damage pattern is determined according to the probability of actual occurrence of each damage pattern. Under the premise of not exceeding the total number of keyword libraries of technical standard text samples of each damage pattern, the conditional random sampling method is used to sample words within a certain number (such as 20-35) of the vocabulary, and the corresponding damage pattern classification labels are added to the sampled texts for sample enhancement, thereby obtaining self-generated technical standard keyword expert samples.

[0067] It should be noted that the mechanism model samples belong to texts, while the mechanism model belongs to semantic models.

[0068] In this embodiment, part of the self-generated technical standard keyword expert sample is shown in Table 1 below.

[0069] Table 1. Partial content of self-generated technical standard keyword expert sample

[0070] Step S13 then combines the text samples obtained from the engineering case to construct the mechanism model sample. Part of the mechanism model sample is shown in Table 2 below:

[0071] Table 2. Partial content of the mechanism model sample

[0072] S14. Retain preset types of operating condition parameters in the operating condition data sample and delete the remaining types of operating condition parameters, thereby obtaining the operating condition big data sample.

[0073] In this embodiment, the types of operating parameters retained include: pressure, temperature, pH value, H2S content, H2O content, CO2 content, sulfide content, NH3 content, and chloride ion content; of course, in some embodiments, the types of operating data retained can also be set according to needs.

[0074] When one or more of the media are not present or not provided, the corresponding operating parameters are set to 0.

[0075] In this embodiment, part of the working condition big data sample is shown in Table 3.

[0076] Table 3. Partial content of working condition big data sample

[0077] S2. Use the mechanism model samples and working condition big data samples as data sets for model training to construct the mechanism model and working condition model respectively.

[0078] The method for constructing the mechanism model includes the following steps, namely S2A1 to S2A3.

[0079] S2A1. The vector composed of all damage mode classification labels in the mechanism model sample is The expression formula is:

[0080] Where n is the number of all damage modes, n=20; y n is the classification label of the nth damage pattern; the superscript T is the transpose of the matrix.

[0081] S2A2. Use the TF-IDF algorithm to vectorize the mechanism model samples and convert the text into a sparse matrix form.

[0082] S2A3. Divide the vectorized text data into training set and test set according to a certain ratio. In this embodiment, 80% is used as training set and 20% is used as test set. Combined with the corresponding classification label vector Model training was performed using algorithms such as logistic regression, support vector machines (SVMs), and random forests. The accuracy, recall, and F1 scores of the damage pattern prediction results for each algorithm were compared. The overall accuracy, average, and weighted average of the final prediction results for each damage pattern were also compared. Based on these comparisons, the algorithm that best suited the dataset (with the highest overall accuracy) was selected to construct the mechanism model. Table 4 shows the comparison of the overall accuracy, average, and weighted average of the final prediction results for each damage pattern using algorithms such as logistic regression, SVM, and random forests.

[0083] Table 2. Comparison of the overall accuracy, average value, and weighted average value of the three algorithms

[0084] It can be seen that the overall accuracy of the logistic regression algorithm is the highest. Under the logistic regression algorithm, the accuracy, recall rate and F1 score (F1-Score) of the prediction results of 20 damage patterns are shown in Table 5.

[0085] Table 3. Precision, recall, and F1 score of logistic regression prediction results for 20 damage patterns

[0086] In this embodiment, the method for constructing the operating condition model includes the following steps:

[0087] Use the working condition big data sample as the training set, combined with the corresponding classification label vector Training was performed using the KNN algorithm, random forest algorithm, and Bernoulli algorithm. The accuracy, recall, and F1 score of the damage pattern prediction results for each algorithm were compared. The overall accuracy, average, and weighted average of the final damage pattern prediction results for each algorithm were also compared. Based on the comparison results, the optimal multi-classification machine learning algorithm was selected to construct the working condition model, specifically the algorithm with the highest overall accuracy. Table 6 shows a comparison of the overall accuracy, average, and weighted average of the final damage pattern prediction results for the KNN algorithm (K=6), random forest algorithm, and Bernoulli algorithm.

[0088] Table 6. Comparison of overall accuracy, average, and weighted average of various algorithms

[0089] Under the KNN algorithm (K=6), the accuracy, recall rate and F1-Score values ​​of the prediction results of 20 damage patterns are shown in Table 7.

[0090] Table 7. Accuracy, recall and F1-Score of the KNN algorithm for the prediction results of 20 damage patterns

[0091] After the mechanism model and working condition model are constructed, 400 engineering case texts that did not participate in the training (20 for each damage mode) are used as the test set data for the mechanism model.

[0092] 400 pieces of field working condition data were used as the working condition model test set data (20 pieces for each damage mode, and there are 9 working conditions that make up each damage mode, including pressure, temperature, pH value, H2S content, H2O content, CO2 content, sulfide content, NH3 content, and chloride ion content).

[0093] Then, combining the results of the above-mentioned mechanism model and working condition model calculations, the mechanism model test set data and the working condition model test set belonging to the same damage mode are randomly combined, the mechanism model and working condition model are called, and the probability P of the mechanism model test set and the working condition model test set results belonging to the 20 damage modes is calculated. 11 ,P 12 ,P 13 ,...,P 1n 、P 21 ,P 22 ,P 23 ,...,P 2n .

[0094] Multiply the text prediction probability of the same label in the 20 damage modes by the working condition prediction probability, and we get P1=P 11 *P 21 ,P2=P 12 *P22 ,P3=P 13 *P 23 ,...,P n =P 1n *P 2n This embodiment takes a section of mechanism model sample data as an example, the data is as follows:

[0095] [H2O 16MnR carbon steel water light gasoline heat exchanger stabilizer tower bottom oil cooler tube circulating water]

[0096] The probability that this text data belongs to the 20 damage modes is shown in Table 8:

[0097] Table 8. Probability of text data belonging to 20 damage patterns

[0098] In Table 8, the damage mode with the highest predicted probability for this section of text data is cooling water corrosion, with a predicted probability of 0.579825.

[0099] This embodiment also takes a section of operating condition data as an example, and the operating condition data is as follows:

[0100] [0.5 32 7 0 1000000 0 0 0 25]

[0101] The possibility that the data of this section of working condition belong to the 20 damage modes is shown in Table 9.

[0102] Table 9. Probability of the operating condition data belonging to the 20 damage modes

[0103] In Table 9, the damage mode with the highest predicted probability for this section of operating data is cooling water corrosion, with a prediction value of 0.9916673.

[0104] Multiply the prediction results of the same damage mode in Table 8 and Table 9 to obtain the comprehensive probability that the pressure-bearing equipment belongs to different damage modes, as shown in Table 10.

[0105] Table 10. Comprehensive probability of pressure equipment belonging to different damage modes

[0106] The probability P1, P2, P3, ..., P belonging to each damage mode in Table 10 nSorting from largest to smallest, the most likely damage mode, "cooling water corrosion," is the most likely damage mode. At the same time, for cases where the same description points to multiple damage modes in real cases, the top three most likely damage modes, M1, M2, and M3, are provided as the most likely damage modes for the equipment. Since there is a probability of 0, the top three damage modes with the highest predicted probability are M1 for cooling water corrosion, M2 for atmospheric corrosion (no insulation layer), and M3 for nothing. The actual damage mode is cooling water corrosion, and M1 cooling water corrosion appears in the first three predicted damage modes, achieving accurate prediction. At the same time, the damage modes predicted by the 400 test sets were compared with the actual damage modes, and the final accuracy rate was 85.2130%.

[0107] S3. Input the real-time text data and real-time operating condition data to be predicted into the mechanism model and operating condition model that have been trained, tested and meet the requirements, calculate the possibility that the real-time text data belongs to various damage modes, and the possibility that the real-time operating condition data belongs to various damage modes, and combine the two possibilities to generate equipment damage monitoring results.

[0108] Among them, the possibility that real-time text data belongs to a damage mode, the possibility that real-time working condition data belongs to a damage mode, and the comprehensive possibility that pressure equipment belongs to a damage mode are sorted from large to small, and the top three damage modes and their possibilities are used as the equipment damage monitoring results.

[0109] Example 2

[0110] Please refer to Figure 2. This embodiment provides an intelligent prediction system for damage patterns based on mechanism and working condition big data, and applies the intelligent monitoring method for equipment damage based on mechanism and working condition big data of Example 1. The system includes: a sample construction module 100, a model construction module 200 and a calculation module 300.

[0111] The sample construction module 100 is used to set various damage modes of the equipment, construct a mechanism model sample for each damage mode based on technical standards and engineering cases, and construct a working condition big data sample for each damage mode based on field data.

[0112] The model building module 200 is used to use the mechanism model samples and the working condition big data samples as data sets for model training, and to build a mechanism model and a working condition model respectively.

[0113] The calculation module 300 is used to input the real-time text data and operating condition data to be predicted into the mechanism model and the operating condition model respectively, calculate the possibility that the real-time text data belongs to various damage modes, and the possibility that the operating condition data belongs to various damage modes, thereby combining the two possibilities to generate equipment damage monitoring results.

[0114] In this embodiment, the intelligent damage pattern prediction system can also include an interactive module that can visualize the operation process and prediction results. Please refer to Figures 3 and 4. Figure 3 is a visualization interface display of the damage pattern intelligent prediction system to be input. Within the range of 20 damage patterns, text and working conditions need to be entered in the box respectively. Figure 4 is a visualization interface display of the results of the prediction process of text input and working condition input. At the same time, the top three damage patterns predicted, the top three damage patterns predicted by the mechanism model, and the top three damage patterns predicted by the working condition model are displayed, with the corresponding prediction probabilities.

[0115] Example 3

[0116] Please refer to Figure 5, which shows the logic diagram of a traditional passive risk prevention and control approach. As can be seen, this approach only detects damage after it occurs and is unable to automatically identify and provide early warnings for damage to operating equipment. Furthermore, given the large number of pressure-bearing equipment, manual analysis is time-consuming and labor-intensive, and the low level of intelligence makes automated large-scale damage analysis impossible, severely impacting the safe, long-term operation of pressure-bearing equipment.

[0117] To solve the above problems, this embodiment provides a method for proactive risk control of pressure-bearing equipment. Referring to FIG6 , the method for proactive risk control of pressure-bearing equipment provided in this embodiment includes the following steps:

[0118] Step 1: When the pressure-bearing equipment is in operation and subjected to fluctuating working conditions, perform intelligent damage monitoring on all equipment.

[0119] Among them, the intelligent damage monitoring in step one can be implemented by using the equipment damage intelligent monitoring method based on mechanism and working condition big data in Example 1, or the damage pattern intelligent prediction system in Example 2 can be directly applied to obtain the equipment damage monitoring results.

[0120] Step 2: Based on the equipment damage monitoring results obtained in step 1, determine whether an early warning is needed for the equipment; when an early warning is needed for the equipment, a warning or operation suggestion is made, and all equipment is continuously intelligently monitored for damage.

[0121] It can be seen that the active risk prevention and control method disclosed in this embodiment has the advantages of intelligence, timeliness, and early warning. It does not require a large amount of manual operation. Through real-time monitoring model analysis of equipment status, abnormal situations or potential equipment damage can be promptly warned to operators or related management systems. Before the equipment fails or is damaged, corresponding measures can be taken to avoid the occurrence of risks, thereby reducing maintenance costs and improving equipment reliability.

[0122] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An intelligent equipment damage monitoring method based on mechanism and working condition big data, characterized by: Used to monitor damage modes of pressure-bearing equipment; the monitoring method comprises the steps of: S1. Define multiple damage modes for equipment, construct mechanism model samples for each damage mode based on technical standards and engineering cases, and construct working condition big data samples for each damage mode based on field data; S2. Use the mechanism model samples and the working condition big data samples as data sets for model training to construct the mechanism model and the working condition model respectively; S3. Input the real-time text data and real-time operating condition data to be predicted into the mechanism model and the operating condition model respectively, calculate the possibility that the real-time text data belongs to various damage modes, and the possibility that the real-time operating condition data belongs to various damage modes, and combine the two possibilities to generate equipment damage monitoring results.

2. The intelligent equipment damage monitoring method based on mechanism and working condition big data according to claim 1 is characterized in that: Before step S1, a custom word library and a stop word library related to the equipment damage pattern are also constructed; Step S1 includes the following specific steps: S11. Obtain mechanism model samples for describing each damage mode from technical standards and engineering cases, and obtain operating condition data samples for describing each damage mode from field data; S12. Clean, segment, filter stop words, and remove duplicates from the text samples obtained from the technical standards according to the custom vocabulary and stop word vocabulary to obtain a keyword library of technical standard text samples; S13. Simulation mechanism model identification method and expert experience, according to the technical standard text sample keyword library to obtain self-generated technical standard keyword expert samples, combined with samples obtained from engineering cases to jointly construct the mechanism model sample; S14. Retain preset types of operating condition parameters in the operating condition data sample and delete the remaining types of operating condition parameters, thereby obtaining the operating condition big data sample.

3. The intelligent equipment damage monitoring method based on mechanism and working condition big data according to claim 2 is characterized in that: In step S11, the technical standard adopts the GB / T 30579 technical standard; wherein the GB / T 30579 technical standard includes the damage mechanism, damage morphology, main influencing factors, prone devices or equipment, and main preventive measures of the damage mode; the engineering case includes the material name, medium and the name of the damaged component.

4. The intelligent equipment damage monitoring method based on mechanism and working condition big data according to claim 2 is characterized in that: In step S14, the types of operating parameters retained include: pressure, temperature, pH value, H2S content, H2O content, CO2 content, sulfide content, NH3 content, and chloride ion content; among them, when one or more of these media are not contained or provided, the corresponding operating parameters are set to 0.

5. The intelligent equipment damage monitoring method based on mechanism and working condition big data according to claim 2 is characterized in that: In step S13, the process of obtaining the technical standard keyword expert sample is as follows: The recognition method of the simulation mechanism model and expert experience are used to analyze the keyword library of the technical standard text samples. The number of self-generated samples of each damage pattern is determined according to the probability of actual occurrence of each damage pattern. Under the premise of not exceeding the total number of keyword libraries of technical standard text samples of each damage pattern, the conditional random sampling method is used to sample the vocabulary within the preset number of words, and the corresponding damage pattern classification labels are added to the sampled texts for sample enhancement, thereby obtaining self-generated technical standard keyword expert samples.

6. The intelligent equipment damage monitoring method based on mechanism and working condition big data according to claim 5 is characterized in that: In step S2, the method for constructing the mechanism model includes the following steps: S2A1. The vector composed of all damage mode classification labels in the mechanism model sample is The expression formula is: Where n is the number of all damage modes, n>1; y n is the classification label of the nth damage pattern; the superscript T is the transpose of the matrix; S2A2. Use the TF-IDF algorithm to vectorize the mechanism model samples and convert the text into a sparse matrix form; S2A3. Divide the vectorized text data into training set and test set according to the set ratio, and combine the corresponding classification label vector Model training was carried out using a variety of multi-classification machine learning algorithms. The accuracy, recall rate, and F1 score of multiple damage pattern prediction results under each algorithm were compared. At the same time, the overall accuracy, average value, and weighted average value of the final results of each damage pattern prediction by different algorithms were compared. Based on the comparison results, the optimal multi-classification machine learning algorithm was selected to construct the mechanism model.

7. The intelligent equipment damage monitoring method based on mechanism and working condition big data according to claim 6 is characterized in that: In step S2, the method for constructing the operating condition model includes the following steps: Using working condition big data samples as training sets, combined with corresponding classification label vectors for training, and adopting a variety of multi-classification learning algorithms for training, the accuracy, recall rate, and F1 score of multiple damage pattern prediction results under each algorithm are compared. At the same time, the overall accuracy, average value, and weighted average value of the final results of each damage pattern prediction of different algorithms are compared. Based on the comparison results, the optimal multi-classification machine learning algorithm is selected to construct the working condition model.

8. The intelligent equipment damage monitoring method based on mechanism and working condition big data according to claim 7 is characterized in that: When constructing the mechanism model, multiple multi-classification learning algorithms include logistic regression algorithm, SVM, and random forest; when constructing the working condition model, multiple multi-classification learning algorithms include KNN algorithm, random forest algorithm, and Bernoulli.

9. The intelligent equipment damage monitoring method based on mechanism and working condition big data according to claim 6 is characterized in that: In step S3, the real-time text data belongs to the first possibility set P of various damage modes (1) Expressed as: P (1) ={P 11 ,P 12 ,P 13 ,...,P 1n } The real-time working condition data belongs to the second possible set P of various damage modes (2) Expressed as: P (2) ={P 21 ,P 22 ,P 23 ,...,P 2n } Where, P 1n Indicates the possibility that the real-time text data belongs to the nth damage mode; P 2n Indicates the possibility that the working condition data belongs to the nth damage mode; The probability that the real-time text data and the real-time working condition data belong to the same damage mode is multiplied to obtain the comprehensive probability set {P1, P2, P3, ..., P n }; Where P1 = P 11 *P 21 ,P2=P 12 *P 22 ,P3=P 13 *P 23 ,...,P n =P 1n *P 2n ; Among them, the first possibility set P (1) , the second possibility set P (2) The respective possibilities of the comprehensive possibility set are sorted from large to small, and the top several damage modes and their possibilities are used as the equipment damage monitoring results.

10. The intelligent damage pattern prediction system based on mechanism and working condition big data is characterized by: The intelligent equipment damage monitoring method based on mechanism and working condition big data according to any one of claims 1 to 9 is applied, wherein the system comprises: The sample construction module is used to set various damage modes of equipment, build mechanism model samples for each damage mode based on technical standards and engineering cases, and build working condition big data samples for each damage mode based on field data; A model building module, which is used to use the mechanism model samples and the working condition big data samples as data sets for model training to respectively build a mechanism model and a working condition model; and The calculation module is used to input the real-time text data and operating condition data to be predicted into the mechanism model and the operating condition model respectively, calculate the possibility that the real-time text data belongs to various damage modes, and the possibility that the operating condition data belongs to various damage modes, thereby combining the two possibilities to generate equipment damage monitoring results.

Images