A monitoring and processing system and method for the alarm of a rotating device

Abstract

The present invention relates to a monitoring and processing system and method for the alarm of a rotating device, belonging to the field of data processing. According to the monitored trait type, combined with the rotation mode of the rotating device, historical trait record values are retrieved for partitioning to obtain trait measurement points; graph neural network simulation is performed based on the component distribution topology of the rotating device to construct a reference value configuration unit; according to the variance values of the service life of multiple components, the output means of multiple reference value configuration units are integrated to construct a reference value configuration model, process the service life of multiple components, and obtain the reference value of the trait measurement point; when the monitored trait of the first measurement point is inconsistent with the reference value of the trait measurement point, a delay prompt rule data verification is executed, solving the technical problem that it is difficult to flexibly and accurately perform real-time monitoring and early warning of rotating devices, resulting in insufficient accuracy and reliability of device monitoring.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a monitoring and processing system and method for the alarm of rotating equipment. Background Art

[0002] In industrial production, rotating equipment, such as motors, plays a crucial role, and its operating state directly affects production efficiency and product quality. However, due to the long-term high-load operation of rotating equipment, its various components are easily affected by factors such as wear and fatigue, thus causing various faults. If these faults are not discovered and processed in time, they may lead to equipment shutdown, production line interruption, and even safety accidents, bringing huge economic losses and safety hazards to enterprises. Traditional monitoring methods for rotating equipment often rely on manual inspections and regular maintenance. This method is not only inefficient but also difficult to achieve real-time monitoring and early warning. With the development of industrial Internet and big data technologies, online monitoring and intelligent alarm of rotating equipment have become possible. Usually, early warning settings are made by comparing with a reference value preset by the user, and an error is reported when there is a deviation. However, in actual applications, due to the complexity and diversity of rotating equipment, the differences in different modes of rotating equipment result in dynamic changes in the actual reference value and distribution state. At the same time, the operating environment is also uncertain, making the online monitoring data often contain a large amount of noise and interference, and it is difficult to obtain accurate monitoring error reports, resulting in frequent false alarms and missed alarms. Summary of the Invention

[0003] Aiming at the technical problem in the prior art that it is difficult to flexibly and accurately perform real-time monitoring and early warning of rotating equipment, resulting in insufficient accuracy and reliability of equipment monitoring, the present invention provides a monitoring and processing system and method for the alarm of rotating equipment to solve this problem.

[0004] The technical solution of the present invention to solve the above technical problems is as follows:

[0005] In a first aspect, the present invention provides a monitoring and processing system for rotation equipment alarms, which is applied to a host computer. The system includes: a historical retrieval module for retrieving historical trait record values for partitioning according to the monitoring trait type and in combination with the rotation mode of the rotation equipment to obtain trait measurement points; a network simulation module for performing graph neural network simulation based on the component distribution topology of the rotation equipment to construct a reference value configuration unit, wherein the input nodes of the reference value configuration unit are component distribution nodes, the input is the service life of the components, the output nodes are trait measurement points, and the output is the reference value of the trait index; a model configuration module for integrating the output means of multiple reference value configuration units according to the variance values of the service lives of multiple components to construct a reference value configuration model, processing the service lives of the multiple components to obtain the reference value of the trait measurement point; and a data verification module for performing data verification of the delayed prompt rule when the monitored trait of the first measurement point is inconsistent with the reference value of the trait measurement point.

[0006] In a second aspect, the present invention provides a method for monitoring and processing rotation equipment alarms, which is applied to a host computer. The method includes: retrieving historical trait record values for partitioning according to the monitoring trait type and in combination with the rotation mode of the rotation equipment to obtain trait measurement points; performing graph neural network simulation based on the component distribution topology of the rotation equipment to construct a reference value configuration unit, wherein the input nodes of the reference value configuration unit are component distribution nodes, the input is the service life of the components, the output nodes are trait measurement points, and the output is the reference value of the trait index; integrating the output means of multiple reference value configuration units according to the variance values of the service lives of multiple components to construct a reference value configuration model, processing the service lives of the multiple components to obtain the reference value of the trait measurement point; and performing data verification of the delayed prompt rule when the monitored trait of the first measurement point is inconsistent with the reference value of the trait measurement point.

[0007] The beneficial effects of the present invention are as follows: By combining the rotation mode of the rotation equipment and historical trait records, and using graph neural network simulation to construct a reference value configuration model, accurate prediction and reference value setting of the trait index of the rotation equipment are realized flexibly. When a trait anomaly is detected, data verification of the delayed prompt rule is performed, improving the accuracy and reliability of the alarm and effectively preventing the occurrence of equipment failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a schematic structural diagram of a monitoring and processing system for rotation equipment alarms provided by the present invention.

[0009] Figure 2 It is a schematic flow diagram of a method for monitoring and processing rotation equipment alarms provided by the present invention.

[0010] Description of the reference numerals: Historical retrieval module 11, network simulation module 12, model configuration module 13, data verification module 14. Detailed implementation manners

[0011] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0012] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined.

[0013] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily construed as being more preferred or advantageous than other embodiments. In order for any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without the use of these specific details. In other instances, well-known structures and processes are not described in detail to avoid unnecessary details from obscuring the description of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope that conforms to the principles and features disclosed in the present invention.

[0014] Embodiment 1:

[0015] As Figure 1 shown, the embodiment of the present invention provides a monitoring and processing system for rotational equipment alarm, which is applied to a host computer. The system includes:

[0016] A historical retrieval module 11, configured to retrieve historical trait record values for partitioning according to the monitored trait type in combination with the rotational mode of the rotational equipment, so as to obtain trait measurement points.

[0017] Exemplarily, the host computer refers to the control center or data processing center of an industrial monitoring system, which is responsible for receiving data from rotational equipment monitoring sensors and executing corresponding monitoring and processing methods. By processing and analyzing these data, the host computer can judge the operating state of the rotational equipment, timely discover abnormal situations and take corresponding processing measures, thereby ensuring the normal operation of the equipment and production safety. A monitoring and processing system for rotational equipment alarm in this application is applied to the host computer to achieve intelligent monitoring.

[0018] Optionally, in the monitoring system of rotating equipment, according to the types of monitored traits of concern, including but not limited to vibration, temperature, rotational speed, etc., the system will combine the specific rotational modes of the rotating equipment, that is, the rotation methods and characteristics of the equipment during operation, and then conduct data retrieval and analysis. The rotational modes here involve the startup, stable operation, shutdown, etc. of the equipment, and they jointly determine the different trait characteristics that the equipment may exhibit during operation. In order to accurately evaluate the operating state of the equipment, the system then retrieves historical trait record values, which are obtained through long-term monitoring and recording and reflect the trait performance of the equipment under different operating conditions. By retrieving and analyzing these historical data, the current monitoring data can be compared with past records, thereby more accurately judging the current state of the equipment. After retrieving the historical trait record values, the system will perform zoning processing, that is, classify the monitoring data according to certain rules or standards to more clearly display the distribution and trend of the data. Zoning processing can help the system more quickly locate abnormal data or potential problem areas, thereby improving the efficiency and accuracy of monitoring. Finally, through this process, the system can obtain trait measurement points. Trait measurement points refer to those data points that can reflect the key trait characteristics of the equipment and are crucial for evaluating the operating state of the equipment. By continuously monitoring and analyzing these measurement points, abnormal conditions of the equipment can be detected in a timely manner and corresponding treatment measures can be taken to ensure the normal operation of the equipment and production safety. In summary, through a comprehensive analysis method that combines monitored trait types, rotational modes, historical data retrieval, and zoning processing, the accuracy and efficiency of rotating equipment monitoring are improved.

[0019] The network simulation module 12 is used to perform graph neural network simulation based on the component distribution topology of the rotating equipment and construct a reference value configuration unit, where the input nodes of the reference value configuration unit are component distribution nodes, the input is the component service life, the output nodes are trait measurement points, and the output is the reference value of the trait index.

[0020] Furthermore, the graph neural network (GNN) simulation technology is adopted for component layout analysis. This technology utilizes the distribution topology of the internal components of the rotating equipment, that is, the connection relationship and position distribution among the components, to construct a neural network model. Through this method, the interactions and influences among the internal components of the equipment can be captured and represented more accurately. A reference value configuration unit is constructed through the graph neural network model. The function of this unit is to provide a reference value for each characteristic measurement point of the equipment. This reference value can be used as a reference standard for evaluating whether the equipment is operating normally. The input nodes of the reference value configuration unit correspond to the component distribution nodes of the rotating equipment. These input nodes represent the positions and connection relationships of the various internal components of the equipment, and they are interconnected through the edges in the graph neural network. The data input to these nodes is the service life of the components, that is, the time elapsed from the installation of the components to the current moment. In the reference value configuration unit, after the input data is processed and analyzed by the graph neural network, it will be output to the output nodes. These output nodes correspond to the characteristic measurement points of the equipment, and they are the key positions for monitoring the operating state of the equipment. The data output by the output nodes is the reference value of the characteristic index. These reference values represent the characteristic index values that each characteristic measurement point of the equipment should exhibit under normal operating conditions. By constructing such a reference value configuration unit, an accurate reference value can be provided for each characteristic measurement point of the rotating equipment. When the deviation occurs between the actually monitored characteristic index value and the reference value, the system can detect it in time and issue an alarm, thus ensuring the normal operation and production safety of the equipment. To sum up, by utilizing the distribution topology of the rotating equipment components and the graph neural network simulation technology, an accurate reference value is provided for the characteristic measurement points of the equipment by constructing a reference value configuration unit, which provides strong support for the monitoring and evaluation of the equipment.

[0021] The model configuration module 13 is used to integrate the output means of multiple reference value configuration units according to the variance values of the service lives of multiple components, construct a reference value configuration model, process the service lives of the multiple components, and obtain the reference values of the characteristic measurement points.

[0022] Specifically, in the monitoring and evaluation of rotating equipment, the service life of different components will affect their performance, and further affect the overall operating state of the equipment. To more accurately evaluate the state of the equipment, it is necessary to consider the influence of the service life of components on the reference values of characteristic measurement points. To achieve this goal, first, the variance value is calculated based on the service lives of multiple components. The variance value reflects the degree of dispersion among the service lives of components, that is, the magnitude of the differences in the service lives of different components. This variance value is crucial for systematically understanding the impact of the service life of components on the performance of the equipment. Next, these variance values are used to integrate the output means of multiple reference value configuration units. These reference value configuration units are constructed based on the component distribution topology of the rotating equipment and graph neural network simulation, and they can provide preliminary reference values for each characteristic measurement point. However, due to the differences in the service lives of different components, these preliminary reference values may not be completely accurate. Therefore, a more accurate reference value configuration model is constructed by integrating the output means of multiple reference value configuration units and combining the variance values of the service lives of components. This model can consider the influence of the service life of components on the reference values, thereby providing more accurate reference values. When processing the service lives of multiple components, the model calculates the reference values of each characteristic measurement point based on the service life and variance value of each component, and these reference values can be used as a reference standard for evaluating whether the operating state of the equipment is normal. In summary, by using the variance values of the service lives of components to integrate the output means of multiple reference value configuration units, a more accurate reference value configuration model is constructed, which can provide more accurate reference values for characteristic measurement points, thereby better evaluating the operating state of the equipment.

[0023] The data verification module 14 is configured to perform delayed prompt rule data verification when the monitored characteristic of the first measurement point is inconsistent with the reference value of the characteristic measurement point.

[0024] Specifically, during actual operation, due to the influence of various external factors (such as environmental changes, load fluctuations) or internal factors (such as component aging, wear), the monitoring characteristics of a certain measurement point may temporarily deviate from its reference value. This deviation does not necessarily mean that the equipment has a serious fault, but it requires attention. Therefore, when the system detects that the monitoring characteristics of the first measurement point (i.e., a certain key measurement point of concern) are inconsistent with its reference value, it does not immediately trigger an alarm, but instead executes a process of delaying the prompt rule data verification. This verification process includes further analysis of the monitoring data to confirm whether this inconsistency is caused by accidental factors or short-term fluctuations. It will consider the time series characteristics of the monitoring data, such as the change trend of the data, the fluctuation range, etc., as well as the correlation with the data of other relevant measurement points. If, after verification, the system believes that this inconsistency is temporary and does not affect the overall operation of the equipment, then it will choose not to trigger an alarm, or only issue a low-level early warning to remind the operation and maintenance personnel to pay attention to the subsequent changes of this measurement point. However, if the system determines that this inconsistency may indicate a potential fault or an impending fault in the equipment, then it will immediately trigger an alarm to remind the operation and maintenance personnel to take necessary maintenance measures to prevent the occurrence or expansion of the fault. In summary, when the monitoring characteristics of the first measurement point are inconsistent with the reference value, the system can reduce false alarms and unnecessary downtime while ensuring the safe operation of the equipment by executing a flexible delayed prompt rule data verification, improving the accuracy and reliability of the monitoring system.

[0025] In a preferred embodiment, when the monitoring characteristics of the first measurement point are inconsistent with the reference value of the measurement point characteristics, a delayed prompt rule data verification is executed, and the execution steps include: The delayed prompt rule includes: Rule A: being in an inconsistent state for at least N consecutive seconds starting from the inconsistent moment; Rule B: the proportion of the data volume being in an inconsistent state for at least M consecutive seconds starting from the inconsistent moment exceeding a preset ratio; Rule C: when any one of Rule A and Rule B is satisfied, it is regarded as an anomaly and a prompt is executed, otherwise, continue to monitor.

[0026] Specifically, to ensure that the system can accurately determine whether such an inconsistency actually represents a potential fault of the device before issuing an exception prompt, the device detection data is processed through three rules configured flexibly. Rule A: The inconsistent state lasts for at least N consecutive seconds starting from the moment of inconsistency. This rule requires that starting from the moment when the trait of the first measurement point is detected to be inconsistent with the reference value, the system needs to continuously monitor for at least N seconds to confirm whether this inconsistent state persists. If within these N seconds, the trait of the measurement point is always inconsistent with the reference value, then it can be considered that this inconsistent state is stable and worthy of further attention and warning. For example, assume that N is set to 10 seconds. This means that when the system first detects that the trait of the first measurement point is inconsistent with the reference value, it will continue to monitor for the next 10 seconds. If within these 10 seconds, the trait of the measurement point always deviates from the reference value, then Rule A is satisfied. Rule B: The proportion of the data volume in the inconsistent state for at least M consecutive seconds starting from the moment of inconsistency exceeds a preset ratio. Rule B is more flexible as it considers the proportion of the inconsistent state over a period of time. Starting from the moment of detecting the inconsistency, the system needs to continuously monitor for at least M seconds and calculate the proportion of the data volume in which the measurement point is in the inconsistent state during this period. If this proportion exceeds the preset ratio (such as 50%), then it can also be considered that this inconsistent state is significant and an exception prompt needs to be issued. For example, assume that M is set to 20 seconds and the preset ratio is 60%. This means that if the time when the measurement point is in the inconsistent state exceeds 12 seconds (i.e., 60% of 20 seconds) within the continuously monitored 20 seconds, then Rule B is satisfied. Rule C: When any one of Rule A and Rule B is satisfied, it is regarded as an abnormal execution prompt; otherwise, continue to monitor. Rule C is a comprehensive judgment of Rule A and Rule B. If the system detects that the trait of the first measurement point is inconsistent with the reference value and satisfies any one of Rule A or Rule B, then it can be considered that the device may have a potential fault and the system should issue an exception prompt. If neither rule is satisfied, then the system will continue to monitor the trait of this measurement point. By implementing flexible delayed prompt rule data verification, it is possible to reduce false alarms and unnecessary downtime while ensuring the safe operation of the device, improving the accuracy and reliability of the monitoring system.

[0027] In a preferred embodiment, according to the monitored trait type, combined with the rotation mode of the rotating equipment, historical trait record values are retrieved for partitioning to obtain trait measurement points. The steps include: taking the monitored trait type and the rotating equipment model as background constraints, and the rotation mode as the foreground dynamic constraint, collecting the first historical trait detection value and the first detection position distribution information until the Qth historical trait detection value and the Qth detection position distribution information; performing neighborhood clustering analysis on the first detection position distribution information based on the first historical trait detection value to obtain the first detection position partitioning result; until performing neighborhood clustering analysis on the Qth detection position distribution information based on the Qth historical trait detection value to obtain the Qth detection position partitioning result; taking the intersection of the first detection position partitioning result to the Qth detection position partitioning result to obtain the detection position partitioning result, configuring a trait measurement point for each block of the detection position partitioning result, and adding it to the trait measurement points.

[0028] Optionally, during the monitoring process, in order to accurately capture the operating state of the equipment and predict potential faults, it is necessary to set and optimize the trait measurement points according to the monitored trait type (such as vibration, temperature, pressure, etc.) and the rotation mode of the rotating equipment (such as startup, stable operation, shutdown, etc.). This requires considering the background characteristics of the equipment (such as model, specifications, etc.) while also considering its dynamic changes during actual operation. Specifically, taking the monitored trait type and the rotating equipment model as background constraints, and the rotation mode as the foreground dynamic constraint, relevant historical trait record values are retrieved from the database, which includes the first historical trait detection value to the Qth historical trait detection value and their corresponding detection position distribution information. These historical data provide the trait performance and its position distribution of the equipment under different operating states. Next, neighborhood clustering analysis is performed on each historical trait detection value and its detection position distribution information. This is a data mining technique used to group similar data points into the same cluster. Through this method, the detection position partitioning result corresponding to each historical trait detection value can be obtained, from the first detection position partitioning result to the Qth detection position partitioning result. These partitioning results reflect the position characteristics of the equipment under different trait performances. Finally, the intersection of all partitioning results is taken to obtain a stable detection position partitioning result. This intersection area is the position where the equipment shows abnormal or key features under different traits and rotation modes. A trait measurement point is configured for each block so that these key information can be accurately captured during future real-time monitoring. For example, a process description is given by a certain type of centrifugal pump, where the monitored trait type is vibration; the rotating equipment model is a certain type of centrifugal pump; the rotation mode is stable operation. Now, the historical vibration trait records of a certain type of centrifugal pump in the stable operation mode are retrieved, including the first to the tenth (Q = 10) historical trait detection values and their detection position distribution information. Neighborhood clustering analysis is performed on each historical trait detection value, and the following partitioning results are obtained:

[0029] Historical trait detection value serial number Detection location zoning result (represented by different area numbers inside the device) 1 Area A, Area B, Area C 2 Area A, Area B 3 Area B, Area C, Area D ... ... 10 Area A, Area C ;

[0030] Take the intersection of the above partition results to obtain the intersection of the detected position partition results as follows. Intersection area: Area A, Area B, and these two areas appear in all partition results. Furthermore, configure a vibration trait measurement point for each block in the intersection area (i.e., Area A and Area B) so that the vibration information of these two areas can be accurately captured during future real-time monitoring.

[0031] In a preferred embodiment, taking the intersection of the first detected position partition result to the Qth detected position partition result, the execution steps include: selecting the center point and the two farthest points for each partition of the first detected position partition result, constructing a three-element array for each partition, and adding it to the first array set; until the Qth array set is obtained; performing pairwise distance evaluation on the first array set to the Qth array set to obtain a partition result distribution distance set; deleting abnormal partition results according to the partition result distribution distance set, and taking the intersection of the remaining detected position partition results.

[0032] Further, after obtaining the detected position partition results of the rotating equipment under different historical trait detection values, it is necessary to further process these partition results to determine a stable and common detected position area for configuring trait measurement points in this area. For each partition of the first detected position partition result, select its center point (representing the average position or the most typical position of the partition) and the two points farthest from the center point (used to evaluate the range and shape of the partition). These three points form a three-element array, reflecting the key features of the partition. Add the three-element arrays of all partitions to a set to form the first array set. Repeat the above steps until the Qth array set is obtained, and each set corresponds to the detected position partition result of a historical trait detection value. Subsequently, perform pairwise distance evaluation on each three-element array from the first array set to the Qth array set, which includes calculating the distance between the center points of the three-element arrays in different sets and the distance between the farthest points in the three-element arrays within the same set. These distance evaluation results form a partition result distribution distance set, reflecting the relative position and shape differences between different partition results. Then, abnormal partition results that deviate significantly from most partition results can be identified according to the partition result distribution distance set. These abnormal results may be caused by data noise, measurement errors, or accidental changes in the equipment state.

[0033] Delete these abnormal partition results from the set to ensure the accuracy of subsequent analysis. For the remaining detected position partition results, take their intersection. This intersection area is the position where the device exhibits key features under different historical trait detection values, so it is an ideal position for configuring trait measurement points.

[0034] In a preferred embodiment, pairwise distance evaluation is performed on the first array set to the Qth array set to obtain a partition result distribution distance set. The execution steps include: extracting the ith array according to the first array set, where the ith array has a first center coordinate, a first edge coordinate, and a second edge coordinate; extracting the jth array with the second center coordinate closest to the first center coordinate from the second array set, where the jth array has a third edge coordinate and a fourth edge coordinate; calculating the ith distribution distance between the line formed by the first edge coordinate and the second edge coordinate and the line formed by the third edge coordinate and the fourth edge coordinate; until the Yth distribution distance is obtained, calculating the mean of the ith distribution distance to the Yth distribution distance, denoted as the first and second array distribution distance, and adding it to the partition result distribution distance set.

[0035] Further, first extract the i-th array from the first array set. This array represents a specific detection location partition result, which contains three key coordinate points: the first central coordinate (representing the central position of the partition), the first edge coordinate, and the second edge coordinate (representing two points on the edge of the partition, used to describe the shape and size of the partition). Then, find the j-th array in the second array set (or subsequent array sets, depending on the iteration step being performed) that is closest to the first central coordinate of the i-th array. This j-th array also contains three key coordinate points, but here we mainly focus on its third edge coordinate and fourth edge coordinate (also used to describe the shape and size of the partition). Now, calculate the distribution distance between the line formed by the first edge coordinate and the second edge coordinate of the i-th array and the line formed by the third edge coordinate and the fourth edge coordinate of the j-th array. This distribution distance can be obtained by calculating the shortest distance between the two lines, the distance between parallel lines, or some comprehensive distance between the endpoints of the two lines. The specific calculation method depends on the type of relative position relationship between the partition results that we want to measure. Repeat the above steps to evaluate the distance between each array (i.e., each i) in the first array set and the nearest central coordinate array in the second array set (or subsequent array sets) until all possible distribution distances are calculated and labeled as the i-th distribution distance, the (i + 1)-th distribution distance, up to the Y-th distribution distance. Then calculate the mean of these distribution distances to obtain a comprehensive metric representing the relative position relationship between the first array set and the second array set (or subsequent array sets), namely the first-second array distribution distance. Finally, add the first-second array distribution distance to the partition result distribution distance set, which will contain comprehensive metrics for the relative position relationships between all array sets for subsequent analysis and decision-making. Additionally, different distance calculation methods, more coordinate points or features can be considered and adjusted according to specific application scenarios and requirements.

[0036] In a preferred embodiment, based on the element distribution topology of the rotating device, a graph neural network simulation is performed to construct a reference value configuration unit. Among them, the input node of the reference value configuration unit is the element distribution node, the input is the element service life, the output node is the trait measurement point, and the output is the reference value of the trait index. The execution steps include: taking the element service life record data as a constraint, collecting a set of trait characteristic value record data of preset trait measurement points that meet the rotating device model and the rotating mode; performing a same-measurement point central value evaluation on the set of trait characteristic value record data to obtain the true value data of the trait characteristic value; and supervising and training the reference value configuration unit according to the element service life record data and the true value data of the trait characteristic value.

[0037] Specifically, collect the service life record data of the components of the rotating equipment. These data record the service time of each component in the equipment from installation to the current moment, which is an important indicator reflecting the aging degree and performance change of the components. At the same time, it is also necessary to collect the set of characteristic value record data of the preset characteristic measurement points that meet the specific rotating equipment model and rotating mode. These measurement points are pre-selected to monitor the performance state of the equipment, and the characteristic values reflect the performance performance of these measurement points under specific service life. Perform the same measurement point central value evaluation on the collected set of characteristic value record data. The purpose of this step is to remove noise and outliers in the data, and obtain more reliable true characteristic value data of the characteristics by calculating the central tendency (such as mean, median, etc.) of the characteristic values of the same measurement point at different times or under different conditions. Next, construct a graph neural network model according to the component distribution topology of the rotating equipment. In this model, the component distribution nodes serve as input nodes, and the connection relationship between them reflects the physical layout and interaction of the internal components of the equipment. The input to the model is the service life of the components, and this information is used to simulate the performance change of the components over time. Design a reference value configuration unit based on the graph neural network model. The output nodes of this unit correspond to the preset characteristic measurement points, and the output is the reference value of the characteristic index of these measurement points. These reference values are the expected values of the measurement point performance indicators predicted by the model under the given component service life and component distribution topology conditions. Finally, use the preprocessed component service life record data and true characteristic value data of the characteristics to supervise the training of the reference value configuration unit. During the training process, the model will continuously adjust its internal parameters to minimize the difference between the predicted value and the true value data, thereby improving the accuracy and reliability of the reference value configuration. Through the above steps, a reference value configuration unit based on a graph neural network is constructed. This unit can provide accurate reference values of characteristic indicators for specific characteristic measurement points according to the component distribution topology and component service life of the rotating equipment. This not only helps the performance monitoring and optimization of the equipment, but also provides strong data support for the maintenance decision-making of the equipment.

[0038] In a preferred embodiment, according to the variance value of the service life of multiple components, integrate the output means of multiple reference value configuration units to construct a reference value configuration model. The execution steps include: taking 2 as the base and the variance value as the exponent, calculate the integration number of the reference value configuration units; according to the integration number of the reference value configuration units, integrate the output means of multiple reference value configuration units to construct the reference value configuration model.

[0039] Optionally, for each component in the rotating equipment, calculate the variance value of its service life. Variance is an important indicator to measure the dispersion of data. Here, it reflects the fluctuation of the component's service life relative to the average value. Next, use a formula with base 2 and the variance value as the exponent to calculate the integrated number of the reference value configuration units. The purpose of this formula is to adjust the number of integrated units according to the dispersion degree of the component's service life. When the variance is large, it means that the difference in the component's service life is significant, and more reference value configuration units may be needed to capture this difference; conversely, when the variance is small, fewer units may be sufficient. It should be noted that this formula is a heuristic method aimed at providing a reasonable estimate of the integrated number. In practical applications, it may still need to be adjusted and optimized according to specific situations. Subsequently, construct the corresponding number of reference value configuration units based on the integrated number calculated in the previous step. These units can be based on the same graph neural network structure, but their initial parameters and training processes are independent to capture different data characteristics and patterns. For each reference value configuration unit, input the same component service life data and obtain the reference values of the trait indicators output by each of them. Then, calculate the mean of these output values as the output of the final reference value configuration model. In this way, the output integration of multiple reference value configuration units is achieved, thereby improving the robustness and accuracy of the model. The integration strategy utilizes the diversity of multiple models and reduces the bias and overfitting risks that may exist in a single model.

[0040] In a preferred embodiment, according to the component service life record data and the true value data of the trait characteristics, supervise and train the reference value configuration unit. The steps include: constructing a loss function:

[0041] ,

[0042] where represents the loss value, represents the predicted lower limit value at the k-th measurement point, represents the true lower limit value at the k-th measurement point, represents the predicted upper limit value at the k-th measurement point, represents the true upper limit value at the k-th measurement point, H represents the number of measurement points, and k represents the measurement point serial number.

[0043] Specifically, in the process of constructing the reference value configuration unit, supervised training is a crucial step, which ensures that the model can accurately predict the reference value and its range of the trait index of a specific measurement point based on the data of the component service duration record and the true value data of the trait characteristics. In order to supervise the training of this reference value configuration unit, a loss function needs to be defined to quantify the difference between the model prediction value and the true value. The loss function is the objective to be optimized during the training process, and the smaller its value, the more accurate the model prediction. Here, the loss function is designed to include two parts: one part corresponds to the difference between the predicted lower limit value and the true lower limit value of the measurement point, and the other part corresponds to the difference between the predicted upper limit value and the true upper limit value of the measurement point. For each measurement point k (k ranges from 1 to H, where H is the total number of measurement points), the difference between its predicted lower limit value and the true lower limit value, and the difference between the predicted upper limit value and the true upper limit value are calculated respectively, and these differences are accumulated in a certain way (such as the sum of squares) to form the final loss value. Specifically, the specific formula of the loss function in this application is:

[0044] . After having the loss function, the supervised training process can begin. During the training process, the model will continuously adjust its internal parameters to minimize the value of the loss function, which is usually achieved through optimization algorithms such as gradient descent. The algorithm will gradually adjust the parameter values according to the gradient information of the loss function with respect to the parameters until the preset training stop condition is reached (such as the loss value no longer decreases significantly, or the preset number of training epochs is reached). After the training is completed, the model needs to be evaluated to verify whether its performance meets the requirements, which usually includes calculating indicators such as the prediction accuracy and stability of the model on an independent test dataset. If the model performance is not good, it may be necessary to return to the training stage and adjust the model structure, loss function, or training parameters, etc., to optimize the model performance. In summary, through this process, a model can be obtained that can accurately predict the reference value and its range of the trait index of the measurement point according to the component service duration.

[0045] The monitoring and processing system for rotational equipment alarm provided by the embodiment of the present invention has at least the following technical effects:

[0046] 1. By dynamically partitioning the historical trait record values by combining the rotational mode of the rotational equipment and the monitored trait type, the key trait measurement points are accurately identified. This method not only considers the differences in equipment models but also incorporates the dynamic characteristics during the rotation process, making the monitoring more detailed and targeted. In addition, when it is detected that the trait of the first measurement point is inconsistent with the reference value, a delayed prompt rule is adopted for data verification, effectively avoiding false alarms caused by instantaneous fluctuations or mismeasurements, and significantly improving the accuracy and reliability of the alarm. This innovation combines dynamic analysis and intelligent decision-making, providing a strong guarantee for the stable operation of rotational equipment.

[0047] 2. Using the graph neural network simulation technology, a reference value configuration unit is constructed according to the component distribution topology of the rotating equipment, realizing accurate modeling of the complex relationship between the service life of components and trait indicators. Further, by integrating the output means of multiple reference value configuration units according to the variance value of the component service life, a more robust reference value configuration model is constructed. This multi-element integration strategy not only improves the adaptability of the model to complex working conditions, but also enhances the accuracy and generalization performance of the prediction.

[0048] 3. When supervising and training the reference value configuration unit, a loss function including the loss of the prediction lower limit value and the upper limit value is designed. This loss function not only considers the error between the predicted value and the true value, but also particularly focuses on the accuracy of the prediction range, so as to be able to more comprehensively evaluate the performance of the model. By optimizing this loss function, the training process can converge more effectively, improving the training efficiency and prediction accuracy of the model.

[0049] Embodiment 2:

[0050] As Figure 2 shown, based on the same inventive concept as the monitoring and processing system for rotating equipment alarm provided in Embodiment 1, the present invention embodiment also provides a monitoring and processing method for rotating equipment alarm, which is applied to the host computer, and the method includes:

[0051] According to the monitored trait type, combined with the rotation mode of the rotating equipment, retrieve the historical trait record values for partitioning to obtain trait measurement points.

[0052] Perform graph neural network simulation based on the component distribution topology of the rotating equipment to construct a reference value configuration unit, wherein the input node of the reference value configuration unit is the component distribution node, the input is the component service life, the output node is the trait measurement point, and the output is the trait indicator reference value.

[0053] According to the variance values of the service lives of multiple components, integrate the output means of multiple reference value configuration units to construct a reference value configuration model, process the service lives of the multiple components, and obtain the trait measurement point reference value.

[0054] When the monitored trait of the first measurement point is inconsistent with the trait measurement point reference value, perform delayed prompt rule data verification.

[0055] Further, when the monitored trait of the first measurement point is inconsistent with the trait measurement point reference value, perform delayed prompt rule data verification, including: The delayed prompt rule includes: Rule A: being in an inconsistent state for at least N consecutive seconds starting from the inconsistent moment; Rule B: the proportion of the data volume being in an inconsistent state for at least M consecutive seconds starting from the inconsistent moment exceeds a preset proportion; Rule C: when any one of Rule A and Rule B is satisfied, it is regarded as an anomaly and a prompt is executed, otherwise continue to monitor.

[0056] Further, according to the monitored trait type and in combination with the rotational mode of the rotating equipment, retrieve the historical trait recorded values for partitioning to obtain trait measurement points, including: taking the monitored trait type and the rotating equipment model as background constraints and the rotational mode as a foreground dynamic constraint, collecting the first historical trait detection value and the first detected position distribution information until the Qth historical trait detection value and the Qth detected position distribution information; performing neighborhood clustering analysis on the first detected position distribution information based on the first historical trait detection value to obtain the first detected position partitioning result; until performing neighborhood clustering analysis on the Qth detected position distribution information based on the Qth historical trait detection value to obtain the Qth detected position partitioning result; taking the intersection of the first detected position partitioning result to the Qth detected position partitioning result to obtain the detected position partitioning result, configuring a trait measurement point for each block of the detected position partitioning result, and adding it to the trait measurement points.

[0057] Further, taking the intersection of the first detected position partitioning result to the Qth detected position partitioning result includes: selecting the center point and the two farthest points for each partition of the first detected position partitioning result, constructing a ternary array for each partition, and adding it to the first array set; until obtaining the Qth array set; performing pairwise distance evaluation on the first array set to the Qth array set to obtain the partition result distribution distance set; deleting abnormal partition results according to the partition result distribution distance set, and taking the intersection of the remaining detected position partitioning results.

[0058] Further, performing pairwise distance evaluation on the first array set to the Qth array set to obtain the partition result distribution distance set includes: extracting the ith array according to the first array set, where the ith array has a first center coordinate, a first edge coordinate, and a second edge coordinate; extracting the jth array with the second center coordinate closest to the first center coordinate from the second array set, where the jth array has a third edge coordinate and a fourth edge coordinate; calculating the ith distribution distance between the line formed by the first edge coordinate and the second edge coordinate and the line formed by the third edge coordinate and the fourth edge coordinate; until obtaining the Yth distribution distance, calculating the mean of the ith distribution distance to the Yth distribution distance, setting it as the pairwise array distribution distance, and adding it to the partition result distribution distance set.

[0059] Further, based on the component distribution topology of the rotating equipment, a graph neural network simulation is performed to construct a reference value configuration unit. Among them, the input nodes of the reference value configuration unit are component distribution nodes, the input is the service life of the component, the output nodes are trait measurement points, and the output is the reference value of the trait index, including: taking the service life record data of the component as a constraint, collecting a set of record data of the trait characteristic values of the preset trait measurement points that meet the rotating equipment model and the rotating mode; performing a centralized value evaluation of the same measurement points on the set of record data of the trait characteristic values to obtain the true value data of the trait characteristic values; and supervising and training the reference value configuration unit according to the service life record data of the component and the true value data of the trait characteristic values.

[0060] Further, according to the variance values of the service lives of multiple components, the output means of multiple reference value configuration units are integrated to construct a reference value configuration model, including: calculating the integration number of the reference value configuration units with 2 as the base and the variance value as the exponent; and integrating the output means of multiple reference value configuration units according to the integration number of the reference value configuration units to construct the reference value configuration model.

[0061] Further, supervising and training the reference value configuration unit according to the service life record data of the component and the true value data of the trait characteristic values includes: constructing a loss function: where represents the loss value, represents the predicted lower limit value of the k-th measurement point, represents the lower limit true value of the k-th measurement point, represents the predicted upper limit value of the k-th measurement point, represents the upper limit true value of the k-th measurement point, H represents the number of measurement points, and k represents the measurement point serial number.

[0062] It should be noted that the above sequence of the embodiments of the present application is only for description and does not represent the advantages or disadvantages of the embodiments. And the above description of specific embodiments of this specification has been made. The processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0063] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0064] This specification and the accompanying drawings are merely exemplary descriptions of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. A monitoring and processing system for rotating equipment alarms, characterized in that: Applied to a host computer, the system includes: A historical retrieval module is used to retrieve historical trait record values ​​for partitioning and obtain trait measurement points according to the monitored trait type and the rotation mode of the rotating equipment; A network simulation module is used to perform graph neural network simulation based on the component distribution topology of the rotating equipment and construct a reference value configuration unit, wherein the input node of the reference value configuration unit is the component distribution node, the input is the component service time, the output node is the property measurement point, and the output is the property indicator reference value; A model configuration module, for integrating the output means of multiple reference value configuration units according to the variance values ​​of the service time of multiple components, constructing a reference value configuration model, processing the service time of the multiple components, and obtaining the reference value of the property measurement point; A data verification module, used for executing delayed prompt rule data verification when the monitored characteristic of the first measuring point is inconsistent with the benchmark value of the characteristic measuring point; According to the monitoring property type and the rotation mode of the rotating equipment, the historical property record values ​​are retrieved for partitioning to obtain the property measurement points. The execution steps include: Taking the monitored trait type and the rotating equipment model as background constraints and the rotating mode as foreground dynamic constraints, collecting the first historical trait detection value and the first detection position distribution information, until the Qth historical trait detection value and the Qth detection position distribution information; Performing neighborhood cluster analysis on the first detection position distribution information based on the first historical trait detection value to obtain a first detection position partition result; Until a neighborhood cluster analysis is performed on the Qth detection position distribution information based on the Qth historical trait detection value to obtain a Qth detection position partition result; Taking the intersection of the first detection position partition result to the Qth detection position partition result, obtaining the detection position partition result, configuring a property measurement point for each block of the detection position partition result, and adding the property measurement point; The step of taking the intersection of the first detection position partition result to the Qth detection position partition result includes: For each partition of the first detected position partition result, a center point and two points with the farthest distance are selected to construct a ternary array for each partition, and the array is added to the first array set; Until the Qth array set is obtained; Perform pairwise distance evaluation on the first array set to the Qth array set to obtain a partition result distribution distance set; Deleting abnormal partition results according to the partition result distribution distance set, and taking the intersection of the retained detection position partition results; The step of performing pairwise distance evaluation on the first array set to the Qth array set to obtain a partition result distribution distance set includes: Extracting an i-th array according to the first array set, wherein the i-th array has a first center coordinate, a first edge coordinate, and a second edge coordinate; Extracting a j-th array whose second center coordinate is closest to the first center coordinate from the second array set, wherein the j-th array has a third edge coordinate and a fourth edge coordinate; Calculate an i-th distribution distance between a straight line formed by the first edge coordinate and the second edge coordinate and a straight line formed by the third edge coordinate and the fourth edge coordinate; Until the Yth distribution distance is obtained, the mean of the i-th distribution distance until the Yth distribution distance is calculated, set as a two-dimensional array distribution distance, and added to the partition result distribution distance set.

2. The system according to claim 1, characterized in that When the monitored characteristic of the first measuring point is inconsistent with the characteristic measuring point reference value, the delayed prompt rule data verification is performed, and the execution steps include: The delay prompt rules include: Rule A: The inconsistent state lasts for at least N seconds starting from the inconsistent moment. Rule B: The proportion of data that has been in an inconsistent state for at least M consecutive seconds since the inconsistent moment exceeds the preset ratio; Rule C: When any one of Rule A and Rule B is met, it is regarded as an abnormal execution prompt, otherwise monitoring continues.

3. The system according to claim 1, characterized in that Based on the component distribution topology of the rotating equipment, a graph neural network simulation is performed to construct a benchmark value configuration unit, wherein the input node of the benchmark value configuration unit is the component distribution node, the input is the component service time, the output node is the property measurement point, and the output is the property index benchmark value. The execution steps include: Taking the service time record data of the component as a constraint, collecting a property characteristic value record data set of preset property measurement points that meet the rotating equipment model and the rotating mode; Performing a same-measurement-point centralized value evaluation on the trait characteristic value record data set to obtain trait characteristic value true value data; The benchmark value configuration unit is supervised and trained according to the service time record data of the component and the true value data of the property characteristic value.

4. The system according to claim 1, characterized in that According to the variance values ​​of the service time of multiple components, the output means of multiple reference value configuration units are integrated to construct a reference value configuration model. The execution steps include: With 2 as base and the variance value as exponent, calculate the number of base value configuration unit integration; According to the integration quantity of the reference value configuration units, the output mean values ​​of a plurality of reference value configuration units are integrated to construct the reference value configuration model.

5. The system according to claim 3, characterized in that According to the service time record data of the component and the true value data of the property characteristic value, the benchmark value configuration unit is supervised and trained, and the execution steps include: Construct the loss function: ， in, Characterize the loss value, Characterizes the predicted lower limit value of the kth measurement point, Characterizes the lower limit true value of the kth measurement point, Characterizes the upper limit of the prediction of the kth measurement point, represents the upper limit true value of the kth measuring point, H represents the number of measuring points, and k represents the sequence number of the measuring point.

6. A method for monitoring and processing rotating equipment alarms, characterized in that: Applied to a host computer, the method is applied to a monitoring and processing system for rotating equipment alarms according to any one of claims 1 to 5, and the method comprises: According to the monitored trait type and the rotation mode of the rotating equipment, the historical trait record values ​​are retrieved for partitioning to obtain the trait measurement points; Based on the component distribution topology of the rotating equipment, a graph neural network simulation is performed to construct a benchmark value configuration unit, wherein the input node of the benchmark value configuration unit is the component distribution node, the input is the component service time, the output node is the property measurement point, and the output is the property index benchmark value; According to the variance values ​​of the service time of multiple components, the output means of multiple reference value configuration units are integrated to construct a reference value configuration model, process the service time of the multiple components, and obtain the reference value of the property measurement point; When the monitored characteristic of the first measuring point is inconsistent with the characteristic measuring point reference value, a delayed prompt rule data check is performed.

Images