Industrial equipment predictive operation and maintenance method and system based on multi-modal model

Abstract

The application provides an industrial equipment predictive operation and maintenance method and system based on a multi-modal model, and belongs to the technical field of large models, and specifically comprises: wherein a defect identification module is responsible for determining defect identification data of the multi-modal model in different industrial equipment in a production line; an optimization strategy determination module is responsible for determining an update strategy of an operation and maintenance optimization target in the production line, according to monitoring data combination for fault identification processing in different industrial equipment and a preventive operation and maintenance strategy of the industrial equipment; and an operation and maintenance processing module is responsible for performing preventive operation and maintenance processing on monitoring data of a monitoring device of the industrial equipment by using the preventive operation and maintenance strategy, and determining an operation and maintenance processing method of the industrial equipment according to whether the industrial equipment belongs to the operation and maintenance optimization target, thereby improving the operation safety of the industrial equipment.

Description

Technical Field

[0001] This invention belongs to the field of large model technology, and in particular relates to a predictive operation and maintenance method and system for industrial equipment based on multimodal models. Background Technology

[0002] Industrial equipment (such as fans, pump sets, compressors, steam turbines, generators, and mills) are critical assets in industrial production, and unplanned downtime can lead to significant production losses and safety risks. Predictive maintenance (M&A), by monitoring equipment operating status, predicting failure times and remaining lifespan, and scheduling maintenance before failures occur, is a key means of achieving full lifecycle management of equipment.

[0003] To address the aforementioned technical problems, existing solutions involve collecting equipment vibration signals and performing FFT transformation. Typical fault modes such as bearing faults, gear faults, imbalances, and misalignments are identified through spectral characteristics (sidebands, harmonic components, and characteristic frequency amplitudes). This approach offers good fault diagnosis for rotating machinery, effectively identifying and monitoring fault conditions. However, it suffers from the following technical shortcomings: When performing defect identification and processing, the reliability of the identification and processing is relatively high when using multi-source data. However, at the same time, if there are abnormal monitoring data, the reliability of the identification and processing may be affected. Therefore, how to determine the operation and maintenance processing strategies for different devices, and thus determine the reliability of defect identification and processing of multimodal models in different data combinations, and then carry out targeted preventive operation and maintenance processing, has become an urgent technical problem to be solved.

[0004] In summary, to address the aforementioned technical problems, this application provides a predictive operation and maintenance method and system for industrial equipment based on a multimodal model. Summary of the Invention

[0005] To achieve the objectives of this invention, the following technical solution is adopted: Specifically, in the first aspect, this application provides a predictive operation and maintenance system for industrial equipment based on a multimodal model, specifically including: Defect identification module, operation and maintenance strategy determination module, operation and maintenance processing module; The defect identification module is responsible for using the multimodal model identification data to determine the defect identification data of the multimodal model in different industrial equipment in the production line; The optimization strategy determination module is responsible for determining the update strategy of the operation and maintenance optimization target in the production line based on the defect identification data of each industrial equipment in the production line, determining the monitoring data combination for fault identification processing in different industrial equipment based on the fault identification method, determining the matching identification combination group of the industrial equipment based on the monitoring data combination for fault identification processing in different industrial equipment, and determining the preventive operation and maintenance strategy of the industrial equipment based on the matching identification combination in the matching identification combination group in each industrial equipment and the similarity of the monitoring data of the matching identification combination. The operation and maintenance processing module is responsible for using the preventive operation and maintenance strategy to perform preventive operation and maintenance processing on the monitoring data of the industrial equipment monitoring equipment, and determining the operation and maintenance processing method of the industrial equipment based on whether the industrial equipment belongs to the operation and maintenance optimization target.

[0006] Furthermore, the identification data of the multimodal model is determined based on the identification results of the multimodal model when different industrial equipment in the production line has defects.

[0007] Furthermore, the defect identification data under the industrial equipment includes the number of defect identification matching times of the multimodal model under the industrial equipment, wherein the number of defect identification matching times is the number of times the defect identification result is accurate.

[0008] Furthermore, the operation and maintenance methods for the industrial equipment are determined based on whether the industrial equipment falls under the operation and maintenance optimization objectives, specifically including: If the industrial equipment is an operation and maintenance optimization target, then operation and maintenance processing will be carried out as long as the multimodal model identification result of the operation and maintenance optimization target is abnormal at any monitoring time, or when the preset period is reached. If the industrial equipment is not a target of operation and maintenance optimization, then operation and maintenance processing will be carried out as long as the target of operation and maintenance optimization has an abnormal identification result of the multimodal model in more than a certain number of monitoring times within a continuous target period, or when the preset period is reached.

[0009] Furthermore, the preventative maintenance strategy is used to perform preventative maintenance processing on the monitoring data of the industrial equipment monitoring devices, specifically including: Based on the aforementioned preventive maintenance strategy, preventive maintenance processing is performed on the monitoring data in the matching data combination of the monitoring equipment of the industrial equipment according to the second preset cycle, or on all the monitoring data of the monitoring equipment of the industrial equipment according to the second preset cycle.

[0010] Secondly, this application provides a predictive operation and maintenance method for industrial equipment based on a multimodal model, applied to the aforementioned predictive operation and maintenance system for industrial equipment based on a multimodal model, specifically including: S1 uses the identification data of the multimodal model to determine the defect identification data of the multimodal model in different industrial equipment in the production line, and determines the update strategy of the operation and maintenance optimization target in the production line based on the defect identification data of each industrial equipment in the production line. S2 updates the operation and maintenance optimization targets in the industrial equipment based on the update strategy, and determines the fault identification method of the multimodal model based on the update results of the operation and maintenance optimization targets and the update results of the operation and maintenance data of different industrial equipment. S3 determines the monitoring data combination for fault identification processing in different industrial equipment based on the fault identification method, determines the matching identification combination group of the industrial equipment based on the monitoring data combination for fault identification processing in different industrial equipment, and determines the preventive operation and maintenance strategy of the industrial equipment based on the matching identification combination in the matching identification combination group in each industrial equipment and the similarity of the monitoring data of the matching identification combination.

[0011] Furthermore, the method for determining the preventive operation and maintenance strategy for the industrial equipment is as follows: Based on the matching and identification combination groups in each industrial device, the monitoring data combination in the matching and identification combination group in the industrial device is determined and used as the matching data combination. The similarity coefficient between the monitoring data of the matching data combinations is determined based on the degree of similarity between the monitoring data of the matching data combinations. The preventive operation and maintenance strategy for the industrial equipment is determined by the matching data combinations in each industrial equipment and the similarity coefficient of the monitoring data between the matching data combinations.

[0012] The beneficial effects of this invention are as follows: By using defect identification data from various industrial equipment in the production line, the reliability of defect identification for different industrial equipment in the production line is determined. Based on the reliability of defect identification for different industrial equipment, the update strategy for operation and maintenance optimization goals in the production line is determined. This lays the foundation for determining the reliability of multimodal model identification by combining operation and maintenance data with operation and maintenance optimization goals.

[0013] The reliability of defect identification and processing of industrial equipment is determined by the matching data combinations in the matching identification groups of various industrial equipment and the similarity of monitoring data between matching data combinations. Based on the reliability of defect identification and processing of each industrial equipment in the production line, preventive operation and maintenance strategies for industrial equipment are determined. This reduces the difficulty of operation and maintenance processing of monitoring data while ensuring the reliability and efficiency of updating matching data combinations of industrial equipment.

[0014] Other features and advantages will be set forth in the following description, and the objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] The above and other features and advantages of the present invention will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0017] Figure 1 This is a framework diagram of a predictive operation and maintenance system for industrial equipment based on a multimodal model; Figure 2 This is a flowchart of a predictive operation and maintenance method for industrial equipment based on a multimodal model; Figure 3 This is a flowchart illustrating the method for determining the update strategy of operation and maintenance optimization goals in the production line; Figure 4 This is a flowchart illustrating the method for determining the fault identification method for multimodal models. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0019] Example 1 like Figure 1 As shown, this application provides a predictive operation and maintenance system for industrial equipment based on a multimodal model, specifically including: Defect identification module, operation and maintenance strategy determination module, operation and maintenance processing module; The defect identification module is responsible for using the multimodal model identification data to determine the defect identification data of the multimodal model in different industrial equipment in the production line; The optimization strategy determination module is responsible for determining the update strategy of the operation and maintenance optimization target in the production line based on the defect identification data of each industrial equipment in the production line, determining the monitoring data combination for fault identification processing in different industrial equipment based on the fault identification method, determining the matching identification combination group of the industrial equipment based on the monitoring data combination for fault identification processing in different industrial equipment, and determining the preventive operation and maintenance strategy of the industrial equipment based on the matching identification combination in the matching identification combination group in each industrial equipment and the similarity of the monitoring data of the matching identification combination. The operation and maintenance processing module is responsible for using the preventive operation and maintenance strategy to perform preventive operation and maintenance processing on the monitoring data of the industrial equipment monitoring equipment, and determining the operation and maintenance processing method of the industrial equipment based on whether the industrial equipment belongs to the operation and maintenance optimization target.

[0020] Furthermore, the identification data of the multimodal model is determined based on the identification results of the multimodal model when different industrial equipment in the production line has defects.

[0021] The construction of the fault identification and processing model first requires building a multimodal training dataset. This dataset consists of multi-source data collected by various monitoring sensors during the operation of industrial equipment, including vibration monitoring data, temperature monitoring data, current monitoring data, pressure monitoring data, noise monitoring data, and speed monitoring data. Simultaneously, corresponding historical maintenance records are obtained as annotation data, representing the abnormal state labels of the monitoring data combinations corresponding to each actual equipment failure. In the data preprocessing stage, various monitoring data are time-aligned, denoised, filtered, and standardized. Different monitoring data items belonging to the same monitoring time are freely combined according to a preset data combination method to form multiple training samples. Each sample consists of a combination of several monitoring data items as input features, and whether a failure actually occurred at that monitoring time is used as the binary classification label. This process constructs supervised learning training and validation datasets.

[0022] At the model architecture level, a multimodal fusion deep learning model is adopted, consisting of three parts: a feature extraction layer, a cross-modal fusion layer, and a classification output layer. The feature extraction layer designs dedicated feature extraction sub-networks for different types of monitoring data. For example, for vibration signals, a one-dimensional convolutional neural network or a time-frequency attention network is used to extract vibration features such as vibration amplitude, dominant frequency, and energy spectrum; for time-series measurement data such as temperature and current, a long short-term memory network is used to extract trend change features. The cross-modal fusion layer fuses the multi-source features extracted by each sub-network through feature concatenation, attention weighting, or gating fusion mechanisms to generate a fused feature vector that expresses the comprehensive operating status of the equipment. The classification output layer receives the fused feature vector and outputs the probability distribution of fault categories through a fully connected layer and a softmax activation function, while also outputting the confidence value corresponding to each category. During model training, a cross-entropy loss function is used, and the fault identification accuracy and false alarm rate on the validation set are used as evaluation metrics for hyperparameter tuning. A regularization strategy is introduced to prevent overfitting.

[0023] After the model is built, during the fault identification and processing phase, various monitoring data items of the industrial equipment are collected in real time. Input samples are constructed according to the determined combination of monitoring data and fed into the trained model for inference. The model outputs classification results for each fault type and their confidence values. When the confidence value of a certain fault type is greater than the preset judgment threshold, the system determines the identification result as a reliable fault judgment and triggers an alarm response. At the same time, the identification results at the monitoring moment and the actual operation and maintenance verification results are continuously fed back to the training dataset, realizing online incremental learning and continuous optimization of the model, thereby continuously improving the accuracy and reliability of fault identification and processing.

[0024] Specifically, the defect identification data under the industrial equipment includes the number of defect identification matching times of the multimodal model under the industrial equipment, wherein the number of defect identification matching times is the number of times the defect identification result is accurate.

[0025] Specifically, the operation and maintenance methods for the industrial equipment are determined based on whether the industrial equipment falls under the operation and maintenance optimization objectives, including: If the industrial equipment is an operation and maintenance optimization target, then operation and maintenance processing will be carried out as long as the multimodal model identification result of the operation and maintenance optimization target is abnormal at any monitoring time, or when the preset period is reached. If the industrial equipment is not a target of operation and maintenance optimization, then operation and maintenance processing will be carried out as long as the target of operation and maintenance optimization has an abnormal identification result of the multimodal model in more than a certain number of monitoring times within a continuous target period, or when the preset period is reached.

[0026] Specifically, the preventative maintenance strategy involves performing preventative maintenance processing on the monitoring data of industrial equipment monitoring devices, including: Based on the aforementioned preventive maintenance strategy, preventive maintenance processing is performed on the monitoring data in the matching data combination of the monitoring equipment of the industrial equipment according to the second preset cycle, or on all the monitoring data of the monitoring equipment of the industrial equipment according to the second preset cycle.

[0027] Example 2 like Figure 2 As shown, this application provides a predictive operation and maintenance method for industrial equipment based on a multimodal model, applied to the aforementioned predictive operation and maintenance system for industrial equipment based on a multimodal model, specifically including: S1 uses the identification data of the multimodal model to determine the defect identification data of the multimodal model in different industrial equipment in the production line, and determines the update strategy of the operation and maintenance optimization target in the production line based on the defect identification data of each industrial equipment in the production line. Specifically, such as Figure 3 As shown, the method for determining the update strategy of the operation and maintenance optimization target in the production line is as follows: In this embodiment, defect identification data from various industrial equipment in the production line is used to determine the reliability of defect identification for different industrial equipment in the production line. Based on the reliability of defect identification for different industrial equipment, the update strategy for operation and maintenance optimization goals in the production line is determined. This lays the foundation for determining the reliability of multimodal model identification by combining operation and maintenance data with operation and maintenance optimization goals.

[0028] The core objective of this embodiment is to determine the defect identification and matching status of industrial equipment in the production line based on multimodal model identification data, thereby determining the update strategy for operation and maintenance optimization goals. Its core logic is to determine the defect identification matching type of each industrial device by the number of defect identification matchings, and to determine the scope of operation and maintenance optimization goals hierarchically based on the distribution of industrial equipment with different matching types in the production line, so as to achieve precise allocation of operation and maintenance resources. The overall logic follows the process of "defect identification data collection → defect matching type determination → matching type distribution evaluation → update strategy determination".

[0029] S11, based on the defect identification data of each industrial equipment in the production line, determine the number of defect identification matching times for each multimodal model under each industrial equipment in the production line.

[0030] The number of defect identification matching refers to the number of times that the multimodal model identifies defects on industrial equipment and the identification result is confirmed to be accurate by manual review or actual maintenance verification, i.e., the number of true positive identifications.

[0031] Assuming a production line contains multiple industrial devices, and a multimodal model generates a batch of defect identification results for each industrial device within a certain operating cycle, the number of times the identification results for each industrial device are verified as accurate is counted as the defect identification matching count for that industrial device.

[0032] Quantifying the recognition performance of multimodal models into comparable numerical indicators provides an objective data basis for subsequent defect identification and matching type determination. This enables the recognition reliability of multimodal models among different industrial equipment to be compared and graded in a unified dimension, thereby supporting the scope division of subsequent operation and maintenance optimization goals.

[0033] S12, determine the defect identification matching type of the industrial equipment by using the number of defect identification matching under the industrial equipment.

[0034] The defect identification matching type refers to classifying industrial equipment into different reliability levels based on a preset numerical range of the number of defect identification matching times. This includes three matching types: Type 1, Type 2, and Type 3.

[0035] If an industrial device has a high number of defect identification matches during the operation of a multimodal model, it falls into the high match count range and is classified as a match type, indicating that the defect identification reliability of the multimodal model of the device is high.

[0036] By classifying and categorizing the number of defect identification matches, the single numerical index is transformed into a category label with operational and maintenance implications. This facilitates subsequent cross-equipment reliability distribution assessment at the production line level and provides a tiered basis for determining the scope of operational and maintenance optimization objectives.

[0037] It should be noted that the defect identification matching types include three types: Type 1, Type 2, and Type 3. The defect identification omission risk of Type 1 matching is lower than that of Type 2 matching, and the defect identification omission risk of Type 2 matching is lower than that of Type 3 matching.

[0038] The defect identification omission risk refers to the probability risk that the multimodal model fails to correctly identify the actual defect, resulting in the defect being missed. The first type of matching has the lowest defect identification omission risk because it has the highest number of defect identification matchings and the most reliable model performance. The third type of matching has the highest defect identification omission risk because it has the lowest number of defect identification matchings and the least reliable model performance.

[0039] If an industrial piece of equipment in a production line is classified into three matching types, its defect identification matching times are relatively few. This means that there may be many missed detections in the model's identification results for this equipment. In actual production scenarios, defects that are not identified may lead to the failure to detect potential equipment failures in a timely manner.

[0040] By defining three matching types and the relationship between their omission risks, the significance lies in constructing a hierarchical mapping from the number of matching identifications to operational risks. This allows the determination of subsequent operational optimization goals to be based on risk levels rather than solely on numerical judgments, thereby improving the rationality and pertinence of operational decisions.

[0041] S13 determines the update strategy for the operation and maintenance optimization target in the production line based on industrial equipment data under different defect identification and matching types in the production line.

[0042] Specifically, based on industrial equipment data under different defect identification and matching types in the production line, an update strategy for the operation and maintenance optimization target in the production line is determined, which includes the following:

[0043] Scenario 1: If the proportion of industrial equipment of the three matching types in the production line does not meet the requirements, then the number of industrial equipment of the three matching types is large, and the risk of omission in identification is high. Therefore, the update strategy for the operation and maintenance optimization target in the production line is to take industrial equipment of the second matching type, industrial equipment of the third matching type, and industrial equipment with identification accuracy less than the preset accuracy threshold as operation and maintenance optimization targets.

[0044] The proportion of industrial equipment belonging to the three matching types refers to the ratio of the number of industrial equipment belonging to the three matching types in the production line to the total number of industrial equipment in the production line. This proportion reflects the proportion of the part of the production line with the worst overall identification reliability.

[0045] If there are multiple industrial devices in the production line, and a high proportion of them are classified as three matching types, it indicates that the model recognition reliability of a considerable number of devices in the production line is at a low level. In this case, attention needs to be paid to them. When the proportion of devices classified as three matching types is high, the scope of the operation and maintenance optimization target is expanded to include industrial devices classified as two matching types. When the number of devices with poor recognition reliability is large, a relatively conservative operation and maintenance strategy is adopted, including devices with moderate recognition reliability in the key operation and maintenance scope to determine the risk of model recognition omission.

[0046] Scenario 2: If the proportion of industrial equipment of the three matching types in the production line meets the requirements, the omission risk weight coefficient corresponding to different defect identification matching types is used, and combined with the proportion of industrial equipment corresponding to the defect identification matching types, the omission risk value is determined. It is then determined whether the omission risk value is greater than the preset omission risk threshold. If so, the update strategy for the operation and maintenance optimization target in the production line is determined to be to take the industrial equipment of the three matching types and the industrial equipment with identification accuracy less than the preset accuracy threshold as the operation and maintenance optimization target. If not, the update strategy for the operation and maintenance optimization target in the production line is determined to be to take the industrial equipment with identification accuracy less than the preset accuracy threshold as the operation and maintenance optimization target.

[0047] The omission risk weight coefficient refers to the coefficient set separately for different defect identification matching types, reflecting the degree of contribution of the type to the overall omission risk of the production line. The omission risk weight coefficient of the first matching type is the smallest, and the omission risk weight coefficient of the three matching types is the largest.

[0048] If the proportion of industrial equipment of the three matching types in the production line is low, it indicates that the overall identification reliability is acceptable. In this case, it is necessary to comprehensively evaluate the distribution and corresponding weight of each matching type of equipment. Specifically, the weight is determined by the sum of the products of the omission risk weight coefficient corresponding to each defect identification matching type and the proportion of industrial equipment corresponding to that defect identification matching type. When the overall omission risk is still high, only the three matching types of equipment and low-accuracy equipment are targeted; when the overall omission risk is controllable, only low-accuracy equipment is targeted.

[0049] When the number of devices of the three matching types is controllable, the operation and maintenance optimization target is further determined based on the comprehensive omission risk classification. When the number of devices with poor reliability is small, the operation and maintenance target is accurately reduced based on the weighted risk assessment results, avoiding unnecessary frequent operation and maintenance for devices of the second matching type, thereby saving operation and maintenance resources while ensuring security.

[0050] The recognition accuracy is determined based on the ratio of the number of defect recognition matches to the number of defect occurrences under the industrial equipment.

[0051] Specifically, if the industrial equipment is an operation and maintenance optimization target, then operation and maintenance processing will be carried out as long as the multimodal model identification result of the operation and maintenance optimization target is abnormal at any monitoring time, or when the preset period is reached.

[0052] Additionally, it is understood that if the industrial equipment is not a target of operation and maintenance optimization, then operation and maintenance processing will be carried out as long as the multimodal model identification results of the target are abnormal in more than a certain number of monitoring times within a continuous target period, or when a preset cycle is reached.

[0053] The number of defects refers to the total number of defects that have been confirmed to occur during the actual operation of industrial equipment. This includes defects correctly identified by the multimodal model and defects that were missed but were discovered through other means. If an industrial equipment actually experiences multiple defect events within a preset period, and some of them are accurately identified by the multimodal model, then its identification accuracy rate is the ratio of the number of accurately identified defects to the total number of defects. This ratio reflects the model's comprehensive identification capability on the equipment.

[0054] In various situations, it serves as a supplementary screening condition for operation and maintenance optimization targets, always including devices with an identification accuracy rate lower than the preset threshold in the operation and maintenance targets, ensuring that the model can accurately determine identification deviations and omissions through operation and maintenance.

[0055] This embodiment starts with the statistical analysis of defect identification matching times for individual devices, proceeds through matching type classification and judgment, and then evaluates the multi-type distribution at the production line level to finally determine the update strategy for operation and maintenance optimization goals. Its core value lies in three aspects: First, it establishes a quantitative mapping mechanism from defect identification matching times to defect identification omission risk, transforming the identification performance of the multimodal model into operable classification labels for operation and maintenance, thus solving the problem of lack of connection between model identification results and actual operation and maintenance decisions. Second, by combining matching type classification and proportional judgment, the update strategy can dynamically adjust the scope of operation and maintenance goals based on the overall identification reliability level of the production line, realizing the transformation from extensive to intensive operation and maintenance. It introduces omission risk weight coefficients and identification accuracy as supplementary judgment conditions, ensuring a safety baseline while avoiding over-maintenance of devices with good identification reliability, thus balancing operation and maintenance effectiveness and resource efficiency.

[0056] S2 updates the operation and maintenance optimization targets in the industrial equipment based on the update strategy, and determines the fault identification method of the multimodal model based on the update results of the operation and maintenance optimization targets and the update results of the operation and maintenance data of different industrial equipment. Specifically, such as Figure 4 As shown, the method for determining the fault identification method of the multimodal model is as follows: In this application, the matching degree of multimodal model verification processing using operation and maintenance data is determined based on the update results of the operation and maintenance optimization target and the update results of operation and maintenance data of different industrial equipment. Based on the matching degree of multimodal model verification processing using operation and maintenance data, the fault identification method of multimodal model is determined, thereby laying the foundation for improving operation and maintenance processing efficiency and ensuring the efficiency and reliability of multimodal model verification processing.

[0057] S21 determines the proportion of the operation and maintenance optimization target in the industrial equipment of the production line based on the updated results of the operation and maintenance optimization target, and uses it as the operation and maintenance target proportion.

[0058] The maintenance target ratio refers to the proportion of industrial equipment identified as maintenance optimization targets among all industrial equipment on the current production line after the update strategy of S1 is determined. This ratio reflects the size of the equipment range covered by the current maintenance task.

[0059] Assuming the production line contains multiple industrial devices, after the matching type assessment and omission risk analysis of S1, a batch of operation and maintenance optimization targets are determined. Then, the ratio of the number of operation and maintenance optimization targets to the total number of industrial devices in the production line is calculated as the operation and maintenance target ratio.

[0060] The scope of the operation and maintenance optimization target is quantified into a proportional indicator, which provides an input basis for judging the relationship between the operation and maintenance target ratio and the preset threshold in step S21. This determines whether a high-efficiency fault identification method can be directly adopted, or whether it is necessary to make a comprehensive judgment based on the update status of operation and maintenance data.

[0061] If the maintenance target ratio is greater than the preset maintenance target ratio threshold, the maintenance efficiency is higher. Therefore, maintenance can quickly achieve an accurate assessment of the model's identification and matching of different types of defects, and perform maintenance quickly. Therefore, the fault identification method of the multimodal model is determined to be based on the existence of at least a preset number of monitoring data in each monitoring data combination. The monitoring data of the industrial equipment monitoring equipment are freely combined to construct monitoring data combinations, and fault identification is performed based on the monitoring data combinations, thereby further improving the efficiency of maintenance.

[0062] The preset operation and maintenance target ratio threshold is a threshold value used to determine whether the coverage of operation and maintenance optimization targets is large enough. When the operation and maintenance target ratio is higher than this threshold, it indicates that the operation and maintenance target coverage is wide and the operation and maintenance efficiency is high.

[0063] If the proportion of maintenance targets exceeds the preset threshold, then most of the equipment in the production line will be key maintenance targets. In this scenario, a more flexible monitoring data combination construction method is adopted, that is, fault identification can be initiated with a small amount of monitoring data, thereby accelerating the reliability of fault identification and improving the efficiency of maintenance.

[0064] When the proportion of operation and maintenance goals is high, choose an efficiency-first fault identification method. When the coverage of operation and maintenance optimization goals is wide, operation and maintenance resources are already relatively concentrated. At this time, fault identification can be triggered with relatively less monitoring data, which can maximize the operation and maintenance response speed under the premise of sufficient operation and maintenance resources.

[0065] It should also be noted that if the operation and maintenance target ratio is not greater than the preset operation and maintenance target ratio threshold, proceed to step S22.

[0066] S22, based on the update results of the operation and maintenance data of different industrial equipment, determine the number of daily operation and maintenance processes of the industrial equipment, and determine the operation and maintenance matching weight value of the industrial equipment based on the number of daily operation and maintenance processes of the industrial equipment.

[0067] The update result of the operation and maintenance data refers to the operation and maintenance record data generated after each operation and maintenance process of industrial equipment in a historical period, including the date and time of the operation and maintenance process, the operation and maintenance type, the processing result, and other information.

[0068] The daily average number of maintenance processes refers to the daily average value obtained by dividing the cumulative number of maintenance processes of a certain industrial equipment within a preset statistical period by the number of natural days in the statistical period, reflecting the frequency of maintenance activities of the equipment.

[0069] The operation and maintenance matching weight value refers to a weight coefficient assigned to the equipment based on the number of daily operation and maintenance processes. The more daily operation and maintenance processes, the larger the operation and maintenance matching weight value, indicating that the equipment has richer operation and maintenance data, which is more conducive to verifying the recognition reliability of multimodal models through operation and maintenance data.

[0070] If an industrial piece of equipment undergoes multiple maintenance processes during the statistical period, and the number of maintenance processes per day is large, then a high maintenance matching weight value is assigned, indicating that the equipment has sufficient maintenance data accumulation and strong support for model verification and fault identification.

[0071] The update results of operation and maintenance data are quantified into operation and maintenance matching weight values. A correlation is established between the operation and maintenance activity of industrial equipment and the model fault identification method. This allows the selection of subsequent fault identification methods to take into account the sufficiency of operation and maintenance data for each device, and avoids the use of overly aggressive simplified identification methods on devices with insufficient operation and maintenance data, which could lead to a decrease in identification reliability.

[0072] Based on the operation and maintenance matching weight value of each industrial equipment, determine whether the average value of the operation and maintenance matching weight value of each industrial equipment is less than the preset weight threshold. If so, in order to speed up operation and maintenance processing, determine that the fault identification method of the multimodal model is based on the fact that there is at least a second preset number of monitoring data in each monitoring data combination, freely combine the monitoring data of the monitoring equipment of the industrial equipment to construct the monitoring data combination, and perform fault identification processing based on the monitoring data combination, thereby further improving the efficiency of operation and maintenance processing. If not, proceed to step S23.

[0073] The mean of the operation and maintenance matching weight values ​​of each industrial equipment refers to the value obtained by calculating the arithmetic mean of the operation and maintenance matching weight values ​​of all industrial equipment in the production line, which reflects the uniformity of the overall operation and maintenance data of the production line.

[0074] The second preset quantity refers to a smaller threshold value for the number of monitoring data than the preset quantity. When the second preset quantity is used, less monitoring data is required for each combination of monitoring data, the combination method is more flexible, and the response speed of fault identification is faster.

[0075] If the maintenance matching weight value of most equipment in the production line is low, and its average value is lower than the preset weight threshold, it indicates that the overall maintenance data accumulation is insufficient. At this time, faster maintenance processing is needed to accumulate data. Therefore, a construction method that requires less monitoring data for each combination is adopted to speed up the update speed of maintenance data.

[0076] When the average weight of operation and maintenance matching is low, a combination of the second preset quantity is adopted. When the overall operation and maintenance data is insufficient, the accumulation of operation and maintenance data is accelerated by lowering the fault identification start threshold, forming a positive cycle of "identification - operation and maintenance - data update - more accurate identification". Otherwise, proceed to step S23 for more refined judgment.

[0077] S23, using the operation and maintenance target ratio and the operation and maintenance matching weight value of each industrial equipment, determine the fault identification method of the multimodal model.

[0078] The fault identification method for the multimodal model, which is determined by the operation and maintenance target ratio and the operation and maintenance matching weight value of each industrial equipment, specifically includes the following:

[0079] S231, using the maintenance matching weight value of each industrial equipment, the industrial equipment with an maintenance matching weight value not less than a preset weight threshold is regarded as reliable maintenance industrial equipment. It is determined whether the proportion of reliable maintenance industrial equipment in the industrial equipment is greater than the reliable equipment proportion threshold. If so, the fault identification method of the multimodal model is determined to be based on the fact that there are at least a preset number of monitoring data in each monitoring data combination. The monitoring data of the monitoring equipment of the industrial equipment are freely combined to construct a monitoring data combination, and the fault identification processing is based on the monitoring data combination. If not, proceed to step S232.

[0080] The reliable industrial equipment refers to industrial equipment whose operation and maintenance matching weight value reaches or exceeds the preset weight threshold, indicating that the equipment has accumulated sufficient operation and maintenance data and has a strong ability to verify and evaluate the identification results of multimodal models through operation and maintenance data.

[0081] The percentage of reliable maintenance industrial equipment refers to the ratio of the number of reliable maintenance industrial equipment to the total number of industrial equipment in the production line.

[0082] If the maintenance matching weight value of some equipment in the production line is high, and they are identified as reliable maintenance industrial equipment, and their proportion exceeds the reliable equipment proportion threshold, it indicates that the overall maintenance data of the production line is sufficient. At this time, a method of constructing a combination of a preset number of monitoring data can be used to identify faults, taking into account both identification efficiency and data sufficiency.

[0083] When the proportion of reliable maintenance equipment is sufficient, a preset number of combinations can be used. When the proportion of equipment with sufficient maintenance data reaches a certain level, the overall support of maintenance data is sufficient. There is no need to reduce the requirements of the monitoring data combination to achieve efficient fault identification, thereby improving maintenance efficiency while ensuring the reliability of identification.

[0084] S232, based on the proportion of reliable maintenance industrial equipment and the maintenance target proportion, determine the maintenance reliability value, and determine whether the maintenance reliability value is greater than a preset reliability threshold. If not, then for faster maintenance processing, determine that the fault identification method of the multimodal model is based on the fact that each monitoring data combination contains at least a second preset number of monitoring data, and freely combine the monitoring data of the industrial equipment to construct a monitoring data combination. Fault identification processing is performed based on the monitoring data combination, thereby further improving the efficiency of maintenance processing. If yes, then if the industrial equipment belongs to the maintenance optimization target, then determine the multimodal model. The fault identification method of the multimodal model is based on the existence of at least a preset number of monitoring data in each monitoring data combination. The monitoring data of the industrial equipment are freely combined to construct a monitoring data combination. Fault identification processing is performed based on the monitoring data combination. If the industrial equipment does not belong to the operation and maintenance optimization target, then the fault identification method of the multimodal model is determined to be based on the existence of at least a second preset number of monitoring data in each monitoring data combination. The monitoring data of the industrial equipment are freely combined to construct a monitoring data combination. Fault identification processing is performed based on the monitoring data combination, thereby further improving the efficiency of operation and maintenance processing.

[0085] The aforementioned operation and maintenance reliability value refers to a comprehensive indicator that quantifies the overall operation and maintenance reliability of the production line from two dimensions: the proportion of industrial equipment with reliable operation and maintenance and the proportion of operation and maintenance targets. Specifically, it is determined based on the average or product of the two.

[0086] When the operation and maintenance reliability is high, if an industrial device is an operation and maintenance optimization target and the operation and maintenance reliability value is higher than the preset reliability threshold, then the device is both a key operation and maintenance target and in an environment where the operation and maintenance data is relatively reliable. In this case, a combination of preset quantities is used. If the industrial device is not an operation and maintenance optimization target, a second preset quantity is used to facilitate the rapid accumulation of data and the discovery of problems.

[0087] When the reliability of operation and maintenance is low, for all industrial equipment, based on the existence of at least a second preset number of monitoring data in each monitoring data combination, the monitoring data of the monitoring equipment of the industrial equipment are freely combined to construct monitoring data combinations. Based on the monitoring data combinations, fault identification and processing are carried out, thereby further improving the efficiency of operation and maintenance.

[0088] By combining the proportion of reliable maintenance equipment with the proportion of maintenance targets through maintenance reliability values, and by differentiating between equipment that belongs to maintenance optimization targets and equipment that does not, different fault identification methods are used. High identification reliability requirements are maintained for key maintenance equipment, while data accumulation is accelerated for non-key equipment to achieve differentiated allocation of maintenance resources.

[0089] The monitoring data combination refers to a set of monitoring data formed by freely combining several monitoring data items selected from various monitoring parameters of industrial equipment. When a monitoring time contains a monitoring data combination whose multimodal model identification result indicates an anomaly, that monitoring time is determined to be a monitoring time where the multimodal model identification result indicates an anomaly. If an anomaly is determined based on the monitoring data of a certain monitoring data combination, but the actual verification shows that the number of times no anomalies are found exceeds a preset anomaly number threshold, then when determining future monitoring times where the multimodal model identification result indicates an anomaly, it is not necessary to consider the monitoring data combination.

[0090] The anomaly count threshold refers to the maximum number of false alarms used to determine whether a certain combination of monitoring data is reliable. When the number of false alarms for a combination exceeds this threshold, the combination is considered unreliable in its ability to distinguish abnormal events.

[0091] Suppose that at multiple monitoring times, a combination of monitoring data consisting of temperature monitoring data and vibration monitoring data is identified as abnormal by the multimodal model. However, after actual inspection and verification, it is found that all of these abnormality identifications are false alarms, and the number of false alarms reaches the preset abnormality threshold. In this case, the monitoring data combination will be excluded from the subsequent determination of whether there is an abnormality at a monitoring time and will no longer be used as the basis for judgment.

[0092] By recording and evaluating the false alarms of each monitoring data combination, unreliable monitoring data combinations are automatically eliminated. The monitoring data combinations used for fault identification are continuously optimized through operation and maintenance data feedback, thereby gradually improving the accuracy of anomaly determination in the multimodal model and reducing the false alarm rate.

[0093] This embodiment starts with the proportion of operation and maintenance targets, filters by the average weight of operation and maintenance matching, and then jointly judges the proportion of reliable operation and maintenance equipment and the operation and maintenance reliability value. Finally, it determines differentiated fault identification methods for different types of industrial equipment, establishes a direct mapping relationship between the proportion of operation and maintenance targets and fault identification methods, so that scenarios with a wide coverage of operation and maintenance optimization targets can quickly match efficient identification methods and shorten fault response time. The introduction of operation and maintenance matching weight value as a quantitative indicator of the sufficiency of operation and maintenance data ensures that the selection of fault identification methods not only focuses on the scope of operation and maintenance targets, but also takes into account the accumulation of operation and maintenance data for each device, avoiding blindly simplifying the identification process when data is insufficient. Through the dual-dimensional joint judgment of operation and maintenance reliability value and the differentiated processing of operation and maintenance optimization targets, it realizes the device-level differentiated configuration of fault identification methods, ensuring that key equipment maintains a high identification standard, while non-key equipment is appropriately relaxed to accelerate the operation and maintenance cycle, thereby maximizing overall operation and maintenance efficiency while ensuring the bottom line of safety.

[0094] During operation, false alarms for each monitoring data combination are recorded. For example, if a monitoring data combination consisting of vibration, temperature, and current monitoring data is flagged as abnormal at multiple consecutive monitoring times, but no actual abnormality is found during verification, and the number of false alarms reaches a preset threshold of 5, then this monitoring data combination will no longer be considered when determining whether an abnormality exists at subsequent monitoring times. Instead, anomaly detection will be based solely on the remaining monitoring data combinations that have not been eliminated, thereby reducing maintenance frequency and minimizing unnecessary maintenance procedures.

[0095] S3 determines the monitoring data combination for fault identification processing in different industrial equipment based on the fault identification method, determines the matching identification combination group of the industrial equipment based on the monitoring data combination for fault identification processing in different industrial equipment, and determines the preventive operation and maintenance strategy of the industrial equipment based on the matching identification combination in the matching identification combination group in each industrial equipment and the similarity of the monitoring data of the matching identification combination.

[0096] Specifically, the matching and identification combination group in the industrial equipment is a group constructed from monitoring data of the industrial equipment whose identification accuracy is greater than a preset accuracy threshold and whose number of accurate defect identifications is greater than a preset identification number threshold.

[0097] Specifically, the method for determining the preventive operation and maintenance strategy for the industrial equipment is as follows: In this application, the reliability of defect identification processing of industrial equipment is determined by the matching data combinations in the matching identification combination groups in various industrial equipment and the similarity of monitoring data between matching data combinations. Based on the reliability of defect identification processing of each industrial equipment in the production line, preventive operation and maintenance strategies for industrial equipment are determined. This reduces the difficulty of operation and maintenance processing of monitoring data while ensuring the reliability and efficiency of updating matching data combinations of industrial equipment.

[0098] S31 determines the monitoring data combination in the matching identification combination group in each industrial device based on the matching identification combination group in each industrial device, and uses it as the matching data combination.

[0099] The matching and identification combination group refers to the set of monitoring data combinations in industrial equipment whose identification accuracy is greater than a preset accuracy threshold and whose number of accurate defect identifications is greater than a preset identification number threshold. The monitoring data combinations in this group have shown reliable and stable performance in historical identification and can be used as the basis for formulating subsequent preventive operation and maintenance strategies.

[0100] The matching data combination refers to each specific monitoring data combination selected from the matching identification combination group. These combinations meet preset conditions in both the identification accuracy and the number of accurate identifications, and represent the verified monitoring data combinations with excellent identification performance in the industrial equipment.

[0101] Suppose that an industrial device has used various combinations of monitoring data for defect identification in multiple monitoring cycles in the past. Some of these combinations have a high accuracy rate and a large number of accurate identifications. These combinations are then included in the matching identification combination group of the device, and each combination in the group is a matching data combination.

[0102] By selecting high-performance monitoring data combinations from the historical identification records of industrial equipment, a reliable data foundation is provided for subsequent similarity analysis between matching data combinations. This ensures that the formulation of preventive maintenance strategies is based on high-quality monitoring data combinations that have been verified in practice, rather than a general analysis of all monitoring data combinations, thereby improving the credibility of the strategy formulation basis.

[0103] S32, determine the similarity coefficient of the monitoring data between the matching data combinations based on the degree of similarity between the monitoring data between the matching data combinations.

[0104] The similarity of monitoring data between the matching data combinations refers to the degree of closeness of the two different matching data combinations in terms of the composition of monitoring data items, reflecting the consistency of the monitoring information on which different matching data combinations rely when identifying defects.

[0105] The similarity coefficient between the monitoring data of the matching data combinations refers to the numerical index obtained by quantifying the degree of similarity between two matching data combinations. The larger the similarity coefficient, the closer the monitoring data characteristics of the two matching data combinations are, and the worse their ability to substitute for or complement each other in defect identification.

[0106] Suppose there are two matching data combinations in an industrial device, one consisting of vibration monitoring data and temperature monitoring data, and the other consisting of vibration monitoring data, temperature monitoring data, and noise monitoring data. Since the two have overlap in core monitoring data items and similar data change trends, their monitoring data similarity coefficient is high. The specific similarity coefficient between the matching data combination and other matching data combinations is determined based on the proportion of the number of monitoring data in the matching data combination to the number of monitoring data in the matching data combination.

[0107] The similarity between matching data combinations is quantified into a similarity coefficient, which reveals the degree of information redundancy and complementary relationship between different matching data combinations. This provides a quantitative basis for judging whether the identification coverage of the entire group is sufficient. The lower the similarity coefficient between matching data combinations, the more diverse the monitoring perspective of the group and the more comprehensive the identification coverage.

[0108] S33, using the matching data combinations in each industrial device and the similarity coefficient of the monitoring data between the matching data combinations, to determine the preventive operation and maintenance strategy of the industrial device.

[0109] Understandably, this includes the following:

[0110] Scenario 1: If the number of matching data combinations for each industrial device meets the requirements, then the number of matching data combinations for the industrial devices is relatively large. In this case, the preventive maintenance strategy for the industrial devices is to perform preventive maintenance processing on the monitoring data in the matching data combinations of all monitoring devices according to the second preset cycle, thereby ensuring the reliability of the identification of abnormalities in the industrial devices.

[0111] The requirement that the number of matching data combinations for each industrial device meets the requirement means that the number of matching data combinations for each industrial device in the production line reaches or exceeds the preset threshold for the number of matching data combinations.

[0112] If the number of matching data combinations for all industrial equipment in the production line exceeds the preset threshold, it means that each equipment has enough reliable monitoring data combinations to support identification and can effectively handle anomaly identification. In this case, only the monitoring data involved in the matching data combinations needs to be maintained periodically for preventive maintenance, and there is no need to maintain all monitoring data comprehensively.

[0113] When there is sufficient matching data for all industrial equipment, a simplified preventive operation and maintenance strategy is adopted. When the identification support foundation for each device is relatively solid, operation and maintenance resources are concentrated on the verified monitoring data items with excellent identification performance, so as to ensure the reliability of identification and avoid ineffective operation and maintenance of redundant monitoring data.

[0114] Scenario 2: If the number of matching data combinations for each industrial device does not meet the requirements, the industrial devices with the number of matching data combinations that do not meet the requirements are designated as matching deviation industrial devices. If the proportion of matching deviation industrial devices in the production line is above the preset deviation device proportion threshold, the preventive operation and maintenance strategy for the industrial devices is determined to be to perform preventive operation and maintenance processing on the monitoring data of all monitoring devices according to the second preset cycle, thereby ensuring the reliability of identifying abnormalities in industrial devices and the reliability of identifying matching data combinations.

[0115] The "matching deviation industrial equipment" refers to industrial equipment whose number of matched data combinations does not reach the preset threshold for the number of matched data combinations, indicating that the equipment has insufficient number of verified reliable monitoring data combinations.

[0116] The percentage of industrial equipment with matching deviations refers to the ratio of the number of industrial equipment with matching deviations in the production line to the total number of industrial equipment in the production line.

[0117] If there are insufficient matching data combinations for some industrial equipment in the production line, and these equipment account for a relatively high proportion, it indicates that there is a significant gap in the overall coverage of the matching data combinations of the production line. In this case, it is necessary to expand the scope of operation and maintenance and carry out preventive operation and maintenance on all monitoring data according to the second preset cycle to ensure the reliability of identification.

[0118] When the proportion of devices with matching deviations is high, a comprehensive preventive operation and maintenance strategy should be adopted. When the proven and reliable combination of monitoring data is insufficient on many devices, limited operation and maintenance should not be carried out solely based on the existing combination of matching data. Instead, the scope of operation and maintenance should be expanded to all monitoring data. Comprehensive operation and maintenance should be used to ensure the reliability and accuracy of monitoring data, thereby creating conditions for future expansion of the combination of matching data.

[0119] Scenario 3: If the proportion of mismatched industrial equipment in the production line is not above the preset mismatched equipment proportion threshold, the identification matching weight value is determined based on the matching data combinations in each industrial equipment and the similarity coefficient of the monitoring data between the matching data combinations. It is then determined whether the identification matching weight value is greater than the preset matching weight threshold. If not, the preventive maintenance strategy for the industrial equipment is determined to be to perform preventive maintenance processing on the monitoring data of all monitoring equipment according to the second preset cycle, thereby ensuring the reliability of identifying abnormalities in the industrial equipment and the reliability of identifying matching data combinations. If yes, the preventive maintenance strategy for the industrial equipment is determined based on the matching data combinations in the industrial equipment and the similarity coefficient of the monitoring data between the matching data combinations.

[0120] The identification matching weight value refers to the average of the basic matching weight values ​​of different industrial equipment, which comprehensively reflects the overall level of the number of matching data combinations and the degree of similarity between combinations of all industrial equipment in the production line.

[0121] The basic matching weight value is determined by the number of matching data combinations in the industrial equipment and the similarity coefficient of the monitoring data between the matching data combinations. The more matching data combinations there are and the lower the similarity coefficient of the monitoring data between the matching data combinations, the larger the basic matching weight value, indicating that the industrial equipment has a rich number of matching data combinations and covers a wide range of monitoring dimensions, and has a strong overall identification support capability.

[0122] If the proportion of industrial equipment with matching deviation does not exceed the threshold, but the identification matching weight value is not higher than the preset matching weight threshold, it means that although the number of matching data combinations for most equipment meets the standard, the monitoring dimensions covered are narrow or the differentiation between combinations is insufficient. In this case, comprehensive operation and maintenance of all monitoring data is still required.

[0123] When the proportion of devices with matching deviations is controllable, the quality of the matching data combination is further judged by identifying the matching weight value. The significance is that it not only focuses on the quantity of matching data combinations, but also on the diversity and coverage between combinations, ensuring that further refined management is only carried out when both quantity and diversity meet the requirements, and avoiding premature convergence of the operation and maintenance scope when the quality of the matching data combination is insufficient.

[0124] Furthermore, based on the matching data combinations in the industrial equipment and the similarity coefficients of the monitoring data between the matching data combinations, a preventative maintenance strategy for the industrial equipment is determined, specifically including: Based on the number of matching data combinations in the industrial equipment and the similarity coefficient of the monitoring data between the matching data combinations, the basic matching weight value of the industrial equipment is determined. It is then determined whether the basic matching weight value of the industrial equipment is greater than a preset weight threshold. If so, the preventive operation and maintenance strategy of the industrial equipment is determined to be to perform preventive operation and maintenance processing on the monitoring data in the matching data combinations of all monitoring equipment according to a second preset cycle, thereby ensuring the reliability of the identification of anomalies in the industrial equipment. If not, preventive operation and maintenance processing is performed on all monitoring data of the monitoring equipment according to the second preset cycle, thereby ensuring the reliability of the identification of anomalies in the industrial equipment and the reliability of the identification of matching data combinations.

[0125] The basic matching weight value refers to the weight index obtained by combining the number of matching data combinations and the similarity coefficient of the monitoring data between the matching data combinations for a single industrial device. The larger the value, the more matching data combinations the device has and the lower the similarity coefficient, that is, the stronger the diversity and complementarity of the matching data combinations.

[0126] If the basic matching weight value of a certain industrial equipment is greater than the preset weight threshold, it indicates that the matching data combination of the equipment is sufficient and diverse, and preventive maintenance can be carried out on the monitoring data involved in the matching data combination according to the second preset cycle. If the basic matching weight value is not greater than the preset weight threshold, it indicates that the matching data combination of the equipment is insufficient in terms of quantity or diversity, and all monitoring data of the equipment need to be maintained according to the second preset cycle to promote the formation of new matching data combinations.

[0127] Based on the device, personalized preventive maintenance strategies are determined through basic matching weight values. After identifying that the matching weight values ​​meet the standards, the system further refines the approach to the individual device level. The scope of maintenance is determined based on the actual situation of the matching data combination for each device, thereby achieving differentiated configuration of device-level preventive maintenance. Under the premise that the overall level meets the standards, more comprehensive maintenance support is provided for individual vulnerable devices.

[0128] In a possible specific embodiment, assuming a smart manufacturing production line has 50 industrial devices, within a preset cycle of the multimodal model operation, the number of defect identification matching times and the number of defect occurrences for each industrial device are statistically analyzed as follows: There are 20 industrial devices with a first-class matching type, with all defect identification matching times exceeding 80 times and identification accuracy rates exceeding 0.92; there are 18 industrial devices with a second-class matching type, with defect identification matching times ranging from 45 to 75 times and identification accuracy rates ranging from 0.78 to 0.91; and there are 12 industrial devices with a third-class matching type, with defect identification matching times all below 40 times and identification accuracy rates ranging from 0.55 to 0.77.

[0129] Calculate the proportion of industrial equipment in the three matching types: Proportion of three matching types = 12 ÷ 50 = 0.24. The preset threshold for the proportion of three matching types is 0.30. Since the proportion of three matching types of 0.24 meets the requirement (does not exceed the threshold of 0.30), proceed to the judgment and processing of case 2.

[0130] The omission risk weighting coefficient for Class I matching is 0.10, for Class II matching it is 0.35, and for Class III matching it is 0.55.

[0131] Calculate the proportion of industrial equipment for each matching type: the proportion of matching type 1 = 20 ÷ 50 = 0.40, the proportion of matching type 2 = 18 ÷ 50 = 0.36, and the proportion of matching type 3 = 12 ÷ 50 = 0.24.

[0132] Calculate the omission risk value: Omission risk value = Omission risk weight coefficient of type I match × Proportion of type I match + Omission risk weight coefficient of type II match × Proportion of type II match + Omission risk weight coefficient of type III match × Proportion of type III match.

[0133] Substituting the specific values ​​into the calculation: Omission risk value = 0.10 × 0.40 + 0.35 × 0.36 + 0.55 × 0.24 = 0.040 + 0.126 + 0.132 = 0.298.

[0134] The preset omission risk threshold is 0.250. Since the omission risk value of 0.298 is greater than the preset omission risk threshold of 0.250, the update strategy for determining the operation and maintenance optimization targets is to use industrial equipment of three matching types (12 units) and industrial equipment with an identification accuracy rate less than the preset accuracy rate threshold of 0.80 as operation and maintenance optimization targets. Among the 18 units of the second matching type, 4 units have an identification accuracy rate lower than 0.80, so a total of 16 industrial units of operation and maintenance optimization targets are finally determined. For the 16 industrial units of operation and maintenance optimization targets, operation and maintenance will be carried out as long as the identification result of the multimodal model is abnormal at any monitoring time, or when the preset period of 30 days is reached.

[0135] For industrial equipment that is not part of the operation and maintenance optimization target, operation and maintenance will be carried out if the multimodal model identification results are abnormal in more than 3 monitoring times within a continuous target period of 10 days, or if the preset period of 30 days is reached.

[0136] Following a specific implementation of S1, the production line has a total of 50 industrial devices, and S1 determines that 16 devices are targeted for operation and maintenance optimization. The operation and maintenance target ratio is calculated as follows: Operation and maintenance target ratio = 16 ÷ 50 = 0.32. The preset threshold for the operation and maintenance target ratio is 0.40. Since the operation and maintenance target ratio of 0.32 is not greater than the preset threshold of 0.40, proceed to step S22.

[0137] The number of maintenance processes for each piece of industrial equipment over the past 60 days was statistically analyzed. Among the 16 equipment targeted for maintenance optimization, 5 had a high average daily maintenance processing frequency (above 0.2 times per day), 6 had a medium average daily processing frequency (0.1 to 0.2 times per day), and 5 had a low average daily processing frequency (below 0.1 times per day). Among the 34 equipment not targeted for maintenance optimization, 8 had a high average daily processing frequency, 12 had a medium frequency, and 14 had a low frequency.

[0138] The maintenance matching weight value is determined based on the average number of maintenance processes per day for each device, using a normalized mapping method: the higher the average number of processes per day, the larger the maintenance matching weight value. The mean of the maintenance matching weight values ​​for all 50 industrial devices is calculated.

[0139] The operation and maintenance matching weight values ​​for each industrial device are set as follows (normalized to the range of 0 to 1): For operation and maintenance optimization targets, the weights are 0.82, 0.78, 0.85, 0.80, and 0.76 for 5 devices, 0.52, 0.48, 0.55, 0.50, 0.45, and 0.53 for 6 devices, and 0.22, 0.18, 0.25, 0.20, and 0.30 for 5 devices; for non-operation and maintenance optimization targets, the weights are between 0.72 and 0.88 for 8 devices, between 0.35 and 0.58 for 12 devices, and between 0.10 and 0.28 for 14 devices.

[0140] The mean of the maintenance matching weight value is calculated as: (Sum of maintenance matching weight values ​​for all 50 industrial devices on the production line) ÷ 50. Assuming the sum of all maintenance matching weight values ​​is 25.65, then the mean of the maintenance matching weight value is 25.65 ÷ 50 = 0.513.

[0141] The preset weight threshold is 0.60. Since the mean weight value for operation and maintenance matching (0.513) is less than the preset weight threshold of 0.60, the fault identification method for the multimodal model is determined by freely combining the monitoring data of the industrial equipment monitoring devices based on the premise that each monitoring data combination contains at least a second preset number of monitoring data. The preset number is 3 monitoring data, and the second preset number is 2 monitoring data.

[0142] Following the specific implementation of S2, the production line comprises 50 industrial devices. Over a nearly 90-day operating period, the accuracy rate of each monitoring data combination and the number of accurate defect identifications for each industrial device are recorded. The preset accuracy threshold is 0.85, and the preset identification count threshold is 30.

[0143] For industrial equipment A, there are a total of 8 monitoring data combinations used. Among them, 5 combinations have an accuracy rate greater than 0.85 and more than 30 accurate identifications. These 5 combinations constitute the matching identification combination group for equipment A, and each combination in the group is a matching data combination. The number of matching data combinations for each device is as follows: the number of matching data combinations for devices A to J (10 devices in total) is between 5 and 8; the number of matching data combinations for devices K to R (8 devices in total) is between 4 and 5; and the number of matching data combinations for devices S to Z and the remaining devices (32 devices in total) is between 2 and 4.

[0144] Inspection Scenario 1: Does the number of matching data combinations for all industrial equipment meet the requirement (i.e., all have more than 4 combinations)? Among equipment S and some remaining equipment, 6 devices have only 2 or 3 matching data combinations, which does not meet the requirement. Therefore, the number of matching data combinations for each industrial device does not meet the requirement, proceeding to Scenario 2.

[0145] Industrial equipment with fewer than 4 matching data combinations is considered mismatched industrial equipment. A total of 6 mismatched industrial equipment units are identified. The percentage of mismatched industrial equipment is calculated as: Mismatched industrial equipment percentage = 6 ÷ 50 = 0.12. The preset threshold for the percentage of mismatched equipment is 0.20.

[0146] Since the proportion of industrial equipment with matching deviation of 0.12 is within the preset deviation equipment proportion threshold of 0.20 (i.e. not above the preset deviation equipment proportion threshold), the process proceeds to case 3.

[0147] In scenario 3, calculate the basic matching weight value for each industrial device, and then use this to calculate the overall identification matching weight value for the production line.

[0148] Taking industrial equipment A as an example, its basic matching weight value is calculated. Equipment A has 5 matching data combinations. The similarity coefficient of the monitoring data between the 5 matching data combinations is calculated as follows: the 5 matching data combinations are paired up to form 10 pairs. The similarity coefficient of the monitoring data between each pair is calculated and the average value is taken. The average similarity coefficient of the monitoring data between the matching data combinations of equipment A is 0.35.

[0149] The basic matching weight value is calculated as follows: Basic matching weight value = Number of matching data combinations after removing units ÷ (Average similarity coefficient of monitoring data between matching data combinations + Adjustment constant). The adjustment constant is set to 0.10 to avoid division by zero. The basic matching weight value for device A = 5 ÷ (0.35 + 0.10) = 5 ÷ 0.45 ≈ 11.11.

[0150] The basic matching weight values ​​for all 50 industrial devices were calculated using this method. The basic matching weight values ​​for the 10 devices with the most matching data combinations ranged from 8.50 to 12.30, the basic matching weight values ​​for the 8 devices with a medium number of matching data combinations ranged from 5.20 to 8.40, and the basic matching weight values ​​for the 32 devices with the fewest matching data combinations ranged from 2.10 to 5.10.

[0151] Calculate the identification matching weight value (the average of the basic matching weight values ​​of all industrial equipment): Identification matching weight value = Sum of basic matching weight values ​​of all equipment ÷ 50. Assuming the sum of all basic matching weight values ​​is 316.50, then the identification matching weight value = 316.50 ÷ 50 = 6.33.

[0152] The preset matching weight threshold is 5.80. Since the identification matching weight value of 6.33 is greater than the preset matching weight threshold of 5.80 (i.e., the "if" branch in S33 case 3), a preventive operation and maintenance strategy is determined for each device based on the matching data combinations in the industrial equipment and the similarity coefficient of the monitoring data between the matching data combinations.

[0153] Taking device A as an example, its basic matching weight value is 11.11, and the preset weight threshold is 6.00. Since the basic matching weight value of 11.11 is greater than the preset weight threshold of 6.00, the preventive maintenance strategy for device A is determined to perform preventive maintenance processing on the monitoring data in its matching data combination according to the second preset cycle. The matching data combination of device A involves five types of monitoring data: vibration monitoring data, temperature monitoring data, current monitoring data, pressure monitoring data, and noise monitoring data. The second preset cycle is 60 days, that is, preventive maintenance processing (sensor calibration, data channel detection, signal quality assessment, etc.) is performed on these five types of monitoring data every 60 days.

[0154] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0155] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0156] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A predictive operation and maintenance system for industrial equipment based on a multimodal model, characterized in that, Specifically, it includes: Defect identification module, operation and maintenance strategy determination module, operation and maintenance processing module; The defect identification module is responsible for using the multimodal model identification data to determine the defect identification data of the multimodal model in different industrial equipment in the production line; The optimization strategy determination module is responsible for determining the update strategy of the operation and maintenance optimization target in the production line based on the defect identification data of each industrial equipment in the production line, determining the monitoring data combination for fault identification processing in different industrial equipment based on the fault identification method, determining the matching identification combination group of the industrial equipment based on the monitoring data combination for fault identification processing in different industrial equipment, and determining the preventive operation and maintenance strategy of the industrial equipment based on the matching identification combination in the matching identification combination group in each industrial equipment and the similarity of the monitoring data of the matching identification combination. The operation and maintenance processing module is responsible for using the preventive operation and maintenance strategy to perform preventive operation and maintenance processing on the monitoring data of the industrial equipment monitoring equipment, and determining the operation and maintenance processing method of the industrial equipment based on whether the industrial equipment belongs to the operation and maintenance optimization target.

2. The predictive operation and maintenance system for industrial equipment based on a multimodal model as described in claim 1, characterized in that, The identification data of the multimodal model is determined based on the identification results of the multimodal model in different industrial equipment in the production line when defects exist.

3. The predictive operation and maintenance system for industrial equipment based on a multimodal model as described in claim 1, characterized in that, The defect identification data under the industrial equipment includes the number of defect identification matching times of the multimodal model under the industrial equipment, wherein the number of defect identification matching times is the number of times the defect identification result is accurate.

4. The predictive operation and maintenance system for industrial equipment based on a multimodal model as described in claim 1, characterized in that, The operation and maintenance methods for the industrial equipment are determined based on whether the industrial equipment falls under the operation and maintenance optimization targets, specifically including: If the industrial equipment is an operation and maintenance optimization target, then operation and maintenance processing will be carried out as long as the multimodal model identification result of the operation and maintenance optimization target is abnormal at any monitoring time, or when the preset period is reached. If the industrial equipment is not a target of operation and maintenance optimization, then operation and maintenance processing will be carried out as long as the target of operation and maintenance optimization has an abnormal identification result of the multimodal model in more than a certain number of monitoring times within a continuous target period, or when the preset period is reached.

5. The predictive operation and maintenance system for industrial equipment based on a multimodal model as described in claim 1, characterized in that, The preventative maintenance strategy involves performing preventative maintenance processing on the monitoring data of industrial equipment, specifically including: Based on the aforementioned preventive maintenance strategy, preventive maintenance processing is performed on the monitoring data in the matching data combination of the monitoring equipment of the industrial equipment according to the second preset cycle, or on all the monitoring data of the monitoring equipment of the industrial equipment according to the second preset cycle.

6. A predictive operation and maintenance method for industrial equipment based on a multimodal model, applied to the predictive operation and maintenance system for industrial equipment based on a multimodal model as described in any one of claims 1-5, characterized in that, Specifically, it includes: Using the identification data of the multimodal model, the defect identification data of the multimodal model in different industrial equipment in the production line is determined. Based on the defect identification data of each industrial equipment in the production line, the update strategy of the operation and maintenance optimization target in the production line is determined. The operation and maintenance optimization objectives in the industrial equipment are updated based on the update strategy. Based on the update results of the operation and maintenance optimization objectives and the update results of the operation and maintenance data of different industrial equipment, a fault identification method for the multimodal model is determined. Based on the fault identification method, a combination of monitoring data for fault identification processing in different industrial equipment is determined. Based on the combination of monitoring data for fault identification processing in different industrial equipment, a matching identification combination group for the industrial equipment is determined. Based on the matching identification combination in the matching identification combination group in each industrial equipment and the similarity of the monitoring data of the matching identification combination, a preventive operation and maintenance strategy for the industrial equipment is determined.

7. The predictive operation and maintenance method for industrial equipment based on a multimodal model as described in claim 6, characterized in that, The method for determining the update strategy of the operation and maintenance optimization goals in the production line is as follows: Based on the defect identification data of each industrial equipment in the production line, determine the number of defect identification matching times for each industrial equipment in the production line using a multimodal model. The defect identification matching type of the industrial equipment is determined by the number of defect identification matching times under the industrial equipment. Based on industrial equipment data under different defect identification and matching types in the production line, an update strategy for the operation and maintenance optimization target in the production line is determined.

8. The predictive operation and maintenance method for industrial equipment based on a multimodal model as described in claim 7, characterized in that, The defect identification matching type of the industrial equipment is determined based on the number of defect identification matching times of the industrial equipment, specifically based on the preset matching type corresponding to the number of defect identification matching times.

9. The predictive operation and maintenance method for industrial equipment based on a multimodal model as described in claim 7, characterized in that, Based on industrial equipment data under different defect identification and matching types in the production line, an update strategy for the operation and maintenance optimization goals in the production line is determined, specifically including: If the proportion of industrial equipment of the three matching types in the production line does not meet the requirements, the update strategy for determining the operation and maintenance optimization target in the production line is to take industrial equipment of the second matching type, industrial equipment of the third matching type, and industrial equipment with an identification accuracy rate less than a preset accuracy threshold as operation and maintenance optimization targets.

10. The predictive operation and maintenance method for industrial equipment based on a multimodal model as described in claim 6, characterized in that, The method for determining the preventive operation and maintenance strategy for the industrial equipment is as follows: Based on the matching and identification combination groups in each industrial device, the monitoring data combination in the matching and identification combination group in the industrial device is determined and used as the matching data combination. The similarity coefficient between the monitoring data of the matching data combinations is determined based on the degree of similarity between the monitoring data of the matching data combinations. The preventive operation and maintenance strategy for the industrial equipment is determined by the matching data combinations in each industrial equipment and the similarity coefficient of the monitoring data between the matching data combinations.

Images