A device intelligent health management method and system based on AI

Abstract

This invention relates to the field of data storage, and more particularly to an AI-based intelligent health management method and system for equipment. The method includes: determining the operating status of the target equipment based on its operational stability and operational diversity, and determining the required data segmentation method based on the operating status; determining the target equipment category based on the data segmentation stability within the operational analysis cycle; determining the data processing method based on load characterization values; when optimizing the screening based on equipment coordination reference values, determining the influence depth of the target equipment based on the hierarchical retrieval frequency, and determining whether to add auxiliary screening parameters based on the equipment coordination balance; determining whether there are storage conflicts based on the change frequency, and adjusting the maximum adaptation length for storage media with storage conflicts. This invention improves the storage effectiveness of target equipment operating data.

Description

Technical Field

[0001] This invention relates to the field of data storage, and in particular to an AI-based intelligent health management method and system for devices. Background Technology

[0002] In modern industrial production systems, the stable operation of equipment is the core foundation for ensuring production safety and efficiency. However, with the improvement of equipment monitoring accuracy, the massive amount of time-series monitoring data generated during equipment operation poses a severe data challenge to equipment health management. Traditional storage methods for monitoring data often adopt a single storage strategy and medium, failing to fully consider the differences in the storage value of monitoring data, resulting in low storage efficiency. Therefore, how to achieve efficient storage of equipment operation data, provide a high-quality data foundation for subsequent equipment fault analysis, and thereby improve the overall effectiveness of the equipment health management system has become a core technical problem that urgently needs to be solved.

[0003] Chinese Patent Publication No. CN119939277A discloses a method, system, and storage medium for equipment fault identification. The method includes: dimensionality reduction of the original fault dataset obtained from equipment operation data using principal component analysis; extraction of key features through feature weight initialization and regularization coefficients to obtain a dimensionality-reduced fault feature set, thus reducing feature redundancy; generating synthetic samples using SMOTE oversampling technology for identified rare fault modes; balancing the dataset distribution through feature correlation analysis to obtain a balanced fault dataset, improving feature sparsity; obtaining fault discrimination results from an optimized simulated annealing algorithm; and triggering an early warning mechanism if rare fault modes are found in the discrimination results through feature interpretability analysis, generating targeted maintenance suggestions based on node fault recovery strategies. However, while the above technical solution considers improving computational efficiency through dimensionality reduction and correlation analysis, thereby improving fault identification efficiency, it does not consider the impact of the actual operating state of industrial equipment on the storage method and storage medium of the original collected data, reducing data storage effectiveness and thus reducing the efficiency of equipment health management. Summary of the Invention

[0004] To address this issue, the present invention provides an AI-based intelligent health management method and system for equipment, which overcomes the problem in the prior art that the actual operating status of industrial equipment does not take into account the impact on the storage method and storage medium of the original collected data, thereby reducing the data storage effect and thus reducing the efficiency of equipment health management.

[0005] To achieve the above objectives, the present invention provides an AI-based intelligent health management method for devices, comprising: The AI ​​health assessment model determines the operating status based on the stability and diversity of the target equipment's operating conditions, and determines the required data segmentation method as either baseline segmentation or adaptive segmentation based on the operating status. Based on the data segmentation stability within the operating condition analysis cycle, the target equipment category is determined to be either Class I or Class II target equipment. The load characterization value is determined based on the number of a target device, and the comparison result between the load characterization value and the preset load characterization value determines whether to change the data processing method from the first processing method of optimizing and screening based on device cooperative dispersion to the second processing method of fusion analysis on storage. When optimizing the screening based on the equipment coordination reference value as the screening benchmark in the first processing method, the influence depth of the target equipment is determined based on the hierarchical retrieval frequency of the associated equipment, and whether to add auxiliary screening parameters is determined based on the equipment coordination balance. When executing the second processing method, the fusion analysis pair is determined based on the analysis urgency coefficient of the data slice. The frequency of changes is used to determine whether there are storage conflicts, and the maximum adapt length of the data slice is adjusted for storage media with storage conflicts.

[0006] Furthermore, for operating states where the stability of the operating condition is less than or equal to the preset stability of the operating condition or the diversity of the operating condition is greater than the preset diversity of the operating condition, the data segmentation method is determined to be segmentation based on the adaptive length. The adaptation length is positively correlated with the working condition reference characterization value; The operating condition reference values ​​are determined based on the operating condition stability and operating condition diversity.

[0007] Furthermore, for operating states where the stability of the operating condition is greater than the preset stability of the operating condition and the diversity of the operating condition is less than or equal to the preset diversity of the operating condition, the data segmentation method is determined to be segmentation based on the baseline length. The raw data collected by the target device is evenly divided into several data slices, and the duration of each data slice is the baseline length.

[0008] Furthermore, when the data segmentation stability is greater than the preset data segmentation stability, the target device is determined to be a Class I target device; When the data segmentation stability is less than or equal to the preset data segmentation stability, the target device is classified as a Class II target device. The data segmentation stability is determined based on the length distribution coefficient and the periodic fluctuation coefficient. The data segmentation stability is positively correlated with the length distribution coefficient, and the data segmentation stability is negatively correlated with the periodic fluctuation coefficient.

[0009] Furthermore, when the load characterization value is greater than the preset load characterization value, the data processing method is determined to be optimized and filtered based on the device coordination dispersion. When the equipment coordination dispersion is greater than the preset equipment coordination dispersion, the selection criterion is the historical failure frequency; When the equipment coordination dispersion is less than or equal to the preset equipment coordination dispersion, the selection criterion is the equipment coordination reference value.

[0010] Furthermore, in response to preset filtering conditions, for target devices whose hierarchical retrieval frequency of associated devices is greater than the preset hierarchical retrieval frequency, the influence depth of the target device is determined as the association depth; For target devices whose hierarchical retrieval frequency is less than or equal to the preset hierarchical retrieval frequency, the influence depth of the target device is determined as the baseline depth. The preset screening criteria are determined by the equipment coordination reference value.

[0011] Furthermore, in response to the auxiliary screening conditions, when the equipment coordination balance degree is greater than the equipment coordination balance degree threshold, it is determined to add auxiliary screening parameters; Among them, node proximity is used as an auxiliary screening parameter for target device screening; The auxiliary screening criteria are to determine the screening benchmark as the equipment coordination reference value and to determine the influence depth of the equipment coordination reference value.

[0012] Furthermore, when the load characterization value is less than or equal to the preset load characterization value, the data processing method is determined to be fusion analysis for storage; The fusion analysis process includes: determining the analysis urgency coefficient of data slices based on slice length and slice anomaly, and then merging the data slices with compensation slices in descending order of analysis urgency coefficient to form fusion analysis pairs; The compensation slice is the data slice that was not included in the slice fusion analysis and has the smallest analysis urgency coefficient.

[0013] Furthermore, when the change frequency exceeds the change frequency threshold, a storage conflict is identified, and the maximum adaptation length is adjusted based on the change difference for the storage medium with the storage conflict. The maximum adaptation length is negatively correlated with the change difference; The change difference is the absolute value of the difference between the change frequency and the change frequency threshold.

[0014] The present invention also provides a system for applying the AI-based intelligent health management method for devices, comprising: The data segmentation unit is used to determine the operating status of the target equipment based on the stability and diversity of the operating conditions according to the AI ​​health assessment model, and to determine the required data segmentation method as the baseline segmentation or adaptive segmentation based on the operating status. The equipment analysis unit, which is connected to the data segmentation unit, is used to determine the target equipment category as either a Class I or Class II target equipment based on the data segmentation stability within the working condition analysis cycle. The slice storage unit, which is connected to the data segmentation unit and the device analysis unit respectively, is used to determine whether to change the data processing method from optimization and screening based on device coordination dispersion to fusion analysis pair storage based on the load characterization value; during the optimization and screening process based on device coordination dispersion, the screening criterion is the historical failure frequency or the device coordination reference value; when storing fusion analysis pairs, the fusion analysis pairs are determined based on the analysis urgency coefficient; A storage compensation unit, connected to the slice storage unit, is used to determine whether there is a storage conflict based on the change frequency, and to adjust the maximum adaptation length for storage media with storage conflicts.

[0015] Compared with the prior art, the beneficial effects of the present invention are that the operating status of the target equipment is determined according to the stability and diversity of the operating conditions, and the data segmentation method is determined according to the operating status as either baseline segmentation or adaptive segmentation. The stability of the operating conditions characterizes the fluctuation of the production equipment during operation, and the diversity of the operating conditions characterizes the proportion of target equipment with large fluctuations. This avoids the problem that the single data segmentation method in the prior art makes it difficult for data storage to effectively adapt to the actual operating status of the equipment, thus affecting the data storage effect and improving the efficiency of subsequent data processing.

[0016] Furthermore, in this invention, the data processing method is determined by comparing the load characterization value with the preset load characterization value. The data processing method is either optimized screening based on device coordination dispersion or fusion analysis for storage. The load characterization value represents the load of the storage medium, and different data processing methods are set accordingly. For cases with large and unbalanced loads, it is determined to optimize screening based on device coordination dispersion. For cases with relatively balanced loads, it is determined to use fusion analysis for storage. This effectively realizes the rationality of data processing and improves the effectiveness of data storage.

[0017] Furthermore, in this invention, the selection criterion for determining the selection based on the comparison result between the device coordination dispersion degree and the preset device coordination dispersion degree is the historical failure frequency or the device coordination reference value. The device coordination dispersion degree characterizes whether the influence range of the target device is balanced. When the influence range is relatively balanced, data storage is prioritized for target devices with frequent failures based on the historical failure frequency. When the balance of the influence range is poor, target devices are selected based on the device coordination reference value. The coordination reference value characterizes the influence range of a single target device, and data storage is prioritized for target devices with a larger influence range. This avoids the indiscriminate storage of device data in the prior art, which leads to data storage redundancy, thereby reducing the availability of data storage and the efficiency of subsequent data analysis, and thus improving the data storage effect. Attached Figure Description

[0018] Figure 1 This is a flowchart of the AI-based intelligent health management method for devices according to the present invention; Figure 2 This is a flowchart illustrating how the present invention determines the data segmentation method based on operating condition stability and operating condition diversity. Figure 3 This is a flowchart illustrating the process of determining the target device category based on data segmentation stability according to the present invention. Figure 4 This is a unit connection diagram of the AI-based intelligent health management system for devices according to the present invention. Detailed Implementation

[0019] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0020] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0021] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0022] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0023] Please see Figures 1 to 3 As shown, the present invention provides an AI-based intelligent health management method for devices, the method comprising: The AI ​​health assessment model determines the operating status based on the stability and diversity of the target equipment's operating conditions, and determines the required data segmentation method as either baseline segmentation or adaptive segmentation based on the operating status. Based on the data segmentation stability within the operating condition analysis cycle, the target equipment category is determined to be either Class I or Class II target equipment. The load characterization value is determined based on the number of a target device, and the comparison result between the load characterization value and the preset load characterization value determines whether to change the data processing method from the first processing method of optimizing and screening based on device cooperative dispersion to the second processing method of fusion analysis on storage. When optimizing the screening based on the equipment coordination reference value as the screening benchmark in the first processing method, the influence depth of the target equipment is determined based on the hierarchical retrieval frequency of the associated equipment, and whether to add auxiliary screening parameters is determined based on the equipment coordination balance. When executing the second processing method, the fusion analysis pair is determined based on the analysis urgency coefficient of the data slice. The frequency of changes is used to determine whether there are storage conflicts, and the maximum adapt length of the data slice is adjusted for storage media with storage conflicts.

[0024] This invention is applied to the real-time monitoring of industrial equipment to obtain raw data. A pre-trained AI health assessment model is used to determine the operating status, and a segmentation method is determined based on the operating status. The raw data is then sliced ​​and stored for analysis. The raw data is the operating data of the target equipment. In this invention, the data slices of the target equipment are stored on an edge storage medium until the preset edge storage duration is reached. The data slices are then migrated to cloud storage, thereby achieving health management of the industrial equipment. This invention sets a working condition analysis cycle for obtaining the operating status of the target equipment. It is easy to understand that the higher the user's requirements for data storage effectiveness, the longer the working condition analysis cycle should be. This invention provides a working condition analysis cycle value of 2, in hours.

[0025] The training process for the AI ​​health assessment model includes: First, the data preparation stage: the training data is randomly divided into two parts in a 7:3 ratio: 70% is used for parameter updates and 30% is used as a validation set.

[0026] Model building phase: The operating status of the target device is acquired through sensors. At the same time, a data acquisition system is established to continuously acquire time series data at a set frequency. Impulse noise is removed by using Hampel filtering, outliers are processed by RobustScaler, and missing values ​​are filled or deleted to preprocess the raw data. The operating data (such as vibration, temperature, and current) of the most recent 128 time points are input into a one-dimensional convolutional neural network (1D-CNN). Abnormal patterns in the signal are extracted through convolutional layers and pooling layers, and the operating data of the device is output to generate stability.

[0027] Iterative training phase: The model is trained iteratively using the training set. After each round, the model performance is evaluated using the validation set. If the loss function value on the validation set does not decrease for 10 consecutive rounds, training is stopped early to prevent overfitting.

[0028] Finally, the trained model is used as an AI health assessment model to receive the operating data of the target device and determine the required data segmentation method.

[0029] Specifically, for operating states where the stability of the operating condition is less than or equal to the preset stability of the operating condition or the diversity of the operating condition is greater than the preset diversity of the operating condition, the data segmentation method is determined to be segmentation based on the adaptive length. The adaptation length is determined based on the reference characterization value of the working condition within the working condition analysis cycle, and the adaptation length is positively correlated with the reference characterization value of the working condition. The operating condition reference values ​​are determined based on the operating condition stability and operating condition diversity.

[0030] Operating condition stability = peak value of operating data in the most recent operating condition analysis period - minimum value of operating data; the operating data includes, but is not limited to, the temperature of the target equipment. The operating condition stability of the target equipment is obtained within the most recent several operating condition analysis periods. The number of target equipment with an operating condition stability greater than the preset operating condition stability is recorded as the number of valid target equipment. Operating condition diversity = number of valid target equipment / total number of target equipment; Operating condition reference characterization value = α1 × operating condition stability / preset operating condition stability + α2 × operating condition diversity / preset operating condition diversity; where α1 and α2 are weighting coefficients, α1 + α2 = 1. It is easy to understand that the greater the influence of operating condition stability on the operating condition reference characterization value, the larger the value of α1; the greater the influence of operating condition diversity on the operating condition reference characterization value, the larger the value of α2. This invention provides a set of values ​​for α1 and α2, where α1 is 0.5 and α2 is 0.5.

[0031] The number of operating condition analysis cycles used to obtain operating condition stability can be understood as follows: the higher the user's requirements for data segmentation effect, the larger the value of the number of operating condition analysis cycles used to obtain operating condition stability. In this invention, the value of the number of operating condition analysis cycles used to obtain operating condition stability is 8, with the unit being individual cycles.

[0032] Adaptation length = Baseline length × (Working condition reference characterization value / Working condition reference characterization threshold); Regarding the value of the working condition reference characterization threshold, it is easy to understand that if the working condition reference characterization value has a greater influence on the adaptation length, then the value of the working condition reference characterization threshold will be larger. This invention provides a value of the working condition reference characterization threshold, in which the value of the working condition reference characterization threshold is 0.85.

[0033] Specifically, for operating conditions where the stability of the operating condition is greater than the preset stability of the operating condition and the diversity of the operating condition is less than or equal to the preset diversity of the operating condition, the data segmentation method is determined to be segmentation based on the baseline length. The raw data collected from the target equipment during the operating condition analysis period is divided into several data slices, and the duration of each data slice is the baseline length.

[0034] In this embodiment of the invention, the baseline length is 5 minutes. It is easy to understand that the higher the user's requirements for data storage effect, the smaller the baseline length will be. For the operating state where the operating stability is greater than the preset operating stability and the operating diversity is less than or equal to the preset operating diversity, the device operating state is stable and simple. At this time, the baseline length is used to divide the data to complete the efficient storage of the collected data and avoid the problem of high data redundancy caused by indiscriminate storage.

[0035] If the duration of each data slice is the baseline length, then it is determined that no data processing or analysis is required, and the data slices are directly stored in the edge storage medium.

[0036] Specifically, when the data segmentation stability is greater than the preset data segmentation stability, the target device is determined to be a Class I target device; When the data segmentation stability is less than or equal to the preset data segmentation stability, the target device is classified as a Class II target device. The data segmentation stability is determined based on the length distribution coefficient and the periodic fluctuation coefficient. The data segmentation stability is positively correlated with the length distribution coefficient, and the data segmentation stability is negatively correlated with the periodic fluctuation coefficient.

[0037] Data segmentation stability = length distribution coefficient / preset length distribution coefficient - periodic fluctuation coefficient / preset periodic fluctuation coefficient; the number of data slices of the target device with the first suitable length extracted within the most recent several operating condition analysis periods is denoted as the number of slices of type I, length distribution coefficient = number of slices of type I / total number of data slices; periodic fluctuation coefficient = ;in, For the first The length distribution coefficient of each working condition analysis cycle for The average value of the length distribution coefficient of each working condition analysis cycle. This represents the number of operating condition analysis cycles.

[0038] The values ​​of the preset length distribution coefficient and the preset periodic fluctuation coefficient are easily understood. If the length distribution coefficient has a greater impact on the stability of data segmentation, then the value of the preset length distribution coefficient is larger; if the periodic fluctuation coefficient has a greater impact on the stability of data segmentation, then the value of the preset periodic fluctuation coefficient is smaller. This invention provides a set of values ​​for the preset length distribution coefficient and the preset periodic fluctuation coefficient. In this invention, the preset length distribution coefficient is set to 0.6, and the preset periodic fluctuation coefficient is set to 0.12.

[0039] For a single data slice, the adaptive length that is greater than the baseline length is denoted as the first adaptive length, and the adaptive length that is less than or equal to the baseline length is denoted as the second adaptive length.

[0040] The value of the number of operating condition analysis cycles used to obtain a type of slice is understood to be that the higher the user's requirements for the effectiveness of data storage, the larger the value of the number of operating condition analysis cycles, so as to obtain more effective data for subsequent equipment fault analysis, thereby improving the effectiveness of data storage. The present invention provides a value for the number of operating condition analysis cycles, in which the number of operating condition analysis cycles is 6.

[0041] The preset data segmentation stability value can be understood to be larger as the user's requirements for data storage accuracy increase. This invention provides a preset data segmentation stability value of 1.05.

[0042] Specifically, when the load characterization value is greater than the preset load characterization value, the data processing method is determined to be optimized and filtered based on the device coordination dispersion. When the equipment coordination dispersion is less than or equal to the preset equipment coordination dispersion, the screening criterion is the frequency of historical failures. When the equipment coordination dispersion is greater than the preset equipment coordination dispersion, the selection criterion is determined to be the equipment coordination reference value.

[0043] The optimization screening based on device coordination dispersion is denoted as the first processing method. The load characterization value = the number of target devices of a certain type / the total number of target devices. The load characterization value represents the load of the storage medium. The larger the load characterization value, the greater the load of the storage medium. At this time, the optimization screening of data slices based on device coordination dispersion is performed to further screen out data with higher storage value, thereby improving the effectiveness of data storage.

[0044] Equipment coordination dispersion = maximum equipment coordination reference value - minimum equipment coordination reference value; the equipment coordination dispersion characterizes the balance of the influence range of each target device. The larger the equipment coordination dispersion, the more unbalanced the influence range of each target device. In this case, the equipment coordination reference value is used as the screening criterion for data screening. When the influence range of each target device is relatively balanced, the historical failure frequency is used as the screening criterion. The equipment coordination reference value characterizes the influence range of the target device, and a graph neural network is constructed with each target device as a node. In this invention, there are nodes with production relationships in the graph neural network. The connection between any two nodes with production relationships is recorded as an edge. For a single target device, the number of edges whose association depth is less than or equal to the preset influence depth is detected, and the number of edges is recorded as the equipment coordination characterization value of the target device. The association depth is the level of the associated device corresponding to the node of the graph neural network where the target device is located. In this invention, the preset influence depth is 3. In this embodiment of the invention, if all the primary associated devices, secondary associated devices, ..., P-level associated devices of the target device meet the preset retrieval conditions, then the influence depth of the target device is the association depth P. The preset retrieval condition is that the hierarchical retrieval frequency is greater than the preset hierarchical retrieval frequency. For primary associated devices, their hierarchical level is 1, and for secondary associated devices, their hierarchical level is 2.

[0045] For a single target device, the number of failures of the target device within a window interval is obtained. The window interval is [i - fixed duration, i], where i is the current time, and i - fixed duration is the start time of the window with a fixed duration distance from the current time. This number of failures is recorded as the historical failure frequency of the target device. The value of the fixed duration is larger when the user has higher requirements for data filtering accuracy. By obtaining the operating data of the target device over a longer period of time, the accuracy of the historical failure frequency of the target device is improved, and the influence of fluctuations in operating data over a short period of time is avoided. This invention provides a fixed duration value, in which the fixed duration is 8 times the duration of a single operating condition analysis cycle.

[0046] Each target device is filtered in descending order of historical failure frequency, with the number of filtered devices being 10% of the total number of target devices. Any non-integer filtered number is rounded up. Data slices of the filtered target devices are stored in edge storage media until the storage time reaches the preset edge storage time, at which point the data slices are migrated to cloud storage.

[0047] The value of the preset edge storage duration varies depending on the user's requirements for data access efficiency. The higher the user's requirements for data access efficiency, the larger the value of the preset edge storage duration. This invention provides a preset edge storage duration value of 30 days.

[0048] The preset load characterization value can be understood to be larger as the user’s requirements for data segmentation accuracy are higher. This invention provides a preset load characterization value of 0.7.

[0049] Specifically, in response to preset filtering conditions, for target devices whose hierarchical retrieval frequency of associated devices is greater than the preset hierarchical retrieval frequency, the influence depth of the target device is determined as the association depth; For target devices whose hierarchical retrieval frequency is less than or equal to the preset hierarchical retrieval frequency, the influence depth of the target device is determined as the baseline depth. The preset screening criteria are determined by the equipment coordination reference value.

[0050] The maximum number of data retrievals by the first associated device within a fixed time period is denoted as the hierarchical retrieval frequency. The reference depth is set to 1. It is easy to understand that if the hierarchical retrieval frequency of the first-level associated device of the target device is less than or equal to the preset hierarchical retrieval frequency, it is determined that when calculating the device cooperative reference value, only the nodes directly connected to the node where the target device is located are considered.

[0051] In graph neural networks, for a node that is directly connected to the node where the target device is located, the device corresponding to that node is denoted as the first-level associated device of the target device. The device corresponding to a node that is directly connected to the node where the first-level associated device of the target device is located and has a production relationship with the target device is denoted as the second-level associated device of the target device, and so on.

[0052] The preset level retrieval frequency value is determined by the user's requirement for higher data filtering accuracy. The higher the preset level retrieval frequency value, the greater the value. This invention provides a preset level retrieval frequency value of 50, in which the unit is times.

[0053] Specifically, in response to auxiliary screening conditions, when the equipment coordination balance degree is greater than the equipment coordination balance threshold, it is determined to add auxiliary screening parameters; Among them, node proximity is used as an auxiliary screening parameter for target device screening; The auxiliary screening criteria are to determine the screening benchmark as the equipment coordination reference value and to determine the influence depth of the equipment coordination reference value.

[0054] The median of the absolute values ​​of the differences between the equipment coordination reference values ​​and the median values ​​of equipment coordination for each target device is denoted as the equipment coordination balance degree. The median value of equipment coordination is the median of the equipment coordination reference values ​​for the target devices. The equipment coordination balance degree characterizes the consistency of the influence range of the target devices. If the equipment coordination balance degree is larger, it indicates that the influence of each target device on other devices is relatively consistent. In this case, it is impossible to screen target devices using the equipment coordination reference values. Therefore, it is determined to add auxiliary screening parameters to screen target devices in order to improve the screening accuracy and thus improve the equipment health management effect.

[0055] For a single target node, the average distance from the target node to each associated node at the influence depth is recorded as the node proximity of the target node. The node proximity characterizes the diffusion rate of the influence of the target node on its associated nodes. It is easy to understand that the larger the node proximity, the closer the distribution of the target node and associated nodes. At this time, if the target node fails, it will quickly affect the associated nodes. In this case, the target node is more important, and the data of the target node is stored in the edge storage medium.

[0056] It is easy to understand that the higher the user's requirement for data filtering accuracy, the larger the value of the device coordination balance threshold. This invention provides a value of 6.2 for the device coordination balance threshold.

[0057] Specifically, when the load characterization value is less than or equal to the preset load characterization value, the data processing method is determined to be fusion analysis for storage. The data processing method for fusion analysis is referred to as the second processing method. The fusion analysis process includes: determining the analysis urgency coefficient of the data slice based on the slice length and slice anomaly, and then fusion the data slice with the compensation slice in descending order of analysis urgency coefficient to form a fusion analysis pair. The compensation slice is the data slice that was not included in the slice fusion analysis and has the smallest analysis urgency coefficient.

[0058] The urgency coefficient is calculated as follows: Slice length / Slice length threshold + Slice anomaly score / Slice anomaly score threshold. The slice length refers to the duration corresponding to each data slice. The slice anomaly score is determined by uniformly setting several benchmark points for each data slice, recording the time of each benchmark point as the detection time, and extracting the runtime data for each detection time. The slice anomaly score is calculated as follows: / ;in, The maximum value of the running data. The minimum value of the running data, for The average value of the running data at each detection time point. The number of detection times, the operational data includes, but is not limited to, the temperature of the target device.

[0059] Regarding the value of the number of benchmark points, the higher the user's requirements for data processing effect, the larger the value of the number of benchmark points, so as to obtain more running data for slice anomaly analysis, thereby improving the accuracy of slice anomaly determination. This invention provides a value for the number of benchmark points, in which the value of the number of benchmark points is 10, and the unit is points.

[0060] In this embodiment of the invention, the slice length threshold is 10 (in minutes), and the slice anomaly threshold is 62.5 (in degrees Celsius). It is easy to understand that if the slice length has a greater impact on the urgency coefficient of the analysis, then the slice length threshold will be larger; if the slice anomaly has a greater impact on the urgency coefficient of the analysis, then the slice anomaly threshold will be larger.

[0061] It is worth noting that if the number of data slices is 2g+1, then the g+1th data slice is counted as a separate fusion analysis pair.

[0062] Specifically, when the change frequency exceeds the change frequency threshold, a storage conflict is identified, and the maximum adaptation length is adjusted based on the change difference for the storage medium with the storage conflict. The maximum adaptation length is negatively correlated with the change difference; The change difference is the absolute value of the difference between the change frequency and the change frequency threshold.

[0063] Extract a slice from the most recent several operating condition analysis cycles, with a change frequency of = / (End time of the last Class I slice in the time series - Start time of the first Class I slice in the time series); For the first The duration of a slice of a certain type The maximum fit length is the number of slices in a class, and the maximum fit length = fit length × 1 / |change frequency - change frequency threshold|. The value of the frequency change threshold can be understood as follows: the higher the user's requirements for data storage performance, the larger the value of the frequency change threshold. This invention provides a value for the frequency change threshold, in which the value of the frequency change threshold is 0.65.

[0064] Please see Figure 4 As shown, the present invention also provides a system applied to the AI-based intelligent health management method for devices, the system comprising: The data segmentation unit is used to determine the operating status of the target equipment based on the stability and diversity of the operating conditions according to the AI ​​health assessment model, and to determine the required data segmentation method as the baseline segmentation or adaptive segmentation based on the operating status. The equipment analysis unit, which is connected to the data segmentation unit, is used to determine the target equipment category as either a Class I or Class II target equipment based on the data segmentation stability within the working condition analysis cycle. The slice storage unit, which is connected to the data segmentation unit and the device analysis unit respectively, is used to determine whether to change the data processing method from optimization and screening based on device coordination dispersion to fusion analysis pair storage based on the load characterization value; during the optimization and screening process based on device coordination dispersion, the screening criterion is the historical failure frequency or the device coordination reference value; when storing fusion analysis pairs, the fusion analysis pairs are determined based on the analysis urgency coefficient; A storage compensation unit, connected to the slice storage unit, is used to determine whether there is a storage conflict based on the change frequency, and to adjust the maximum adaptation length for storage media with storage conflicts.

[0065] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An AI-based intelligent health management method for devices, characterized in that, include: The AI ​​health assessment model determines the operating status based on the stability and diversity of the target equipment's operating conditions, and determines the required data segmentation method as either baseline segmentation or adaptive segmentation based on the operating status. Based on the data segmentation stability within the operating condition analysis cycle, the target equipment category is determined to be either Class I or Class II target equipment. The load characterization value is determined based on the number of a target device, and the comparison result between the load characterization value and the preset load characterization value determines whether to change the data processing method from the first processing method of optimizing and screening based on device cooperative dispersion to the second processing method of fusion analysis on storage. When optimizing the screening based on the equipment coordination reference value as the screening benchmark in the first processing method, the influence depth of the target equipment is determined based on the hierarchical retrieval frequency of the associated equipment, and whether to add auxiliary screening parameters is determined based on the equipment coordination balance. When executing the second processing method, the fusion analysis pair is determined based on the analysis urgency coefficient of the data slice. The frequency of changes is used to determine whether there are storage conflicts, and the maximum adapt length of the data slice is adjusted for storage media with storage conflicts.

2. The AI-based intelligent health management method for devices according to claim 1, characterized in that, For operating conditions where the stability is less than or equal to the preset stability or the diversity is greater than the preset diversity, the data segmentation method is determined to be segmentation based on the adaptive length. The adaptation length is positively correlated with the working condition reference characterization value; The operating condition reference values ​​are determined based on the operating condition stability and operating condition diversity.

3. The AI-based intelligent health management method for devices according to claim 1, characterized in that, For operating conditions where the stability of the operating condition is greater than the preset stability of the operating condition and the diversity of the operating condition is less than or equal to the preset diversity of the operating condition, the data segmentation method is determined to be segmentation based on the baseline length. The raw data collected by the target device is evenly divided into several data slices, and the duration of each data slice is the baseline length.

4. The AI-based intelligent health management method for devices according to claim 2 or 3, characterized in that, When the data segmentation stability is greater than the preset data segmentation stability, the target device is classified as a Class I target device. When the data segmentation stability is less than or equal to the preset data segmentation stability, the target device is classified as a Class II target device. The data segmentation stability is determined based on the length distribution coefficient and the periodic fluctuation coefficient. The data segmentation stability is positively correlated with the length distribution coefficient, and the data segmentation stability is negatively correlated with the periodic fluctuation coefficient.

5. The AI-based intelligent health management method for devices according to claim 1, characterized in that, When the load characterization value is greater than the preset load characterization value, the data processing method is determined to be optimized and filtered based on the device coordination dispersion. When the equipment coordination dispersion is greater than the preset equipment coordination dispersion, the selection criterion is the historical failure frequency; When the equipment coordination dispersion is less than or equal to the preset equipment coordination dispersion, the selection criterion is the equipment coordination reference value.

6. The AI-based intelligent health management method for devices according to claim 5, characterized in that, In response to preset filtering conditions, for target devices whose hierarchical retrieval frequency of associated devices is greater than the preset hierarchical retrieval frequency, the influence depth of the target device is determined as the association depth; For target devices whose hierarchical retrieval frequency is less than or equal to the preset hierarchical retrieval frequency, the influence depth of the target device is determined as the baseline depth. The preset screening criteria are determined by the equipment coordination reference value.

7. The AI-based intelligent health management method for devices according to claim 6, characterized in that, In response to auxiliary screening conditions, when the equipment coordination balance is greater than the equipment coordination balance threshold, it is determined to add auxiliary screening parameters; Among them, node proximity is used as an auxiliary screening parameter for target device screening; The auxiliary screening criteria are to determine the screening benchmark as the equipment coordination reference value and to determine the influence depth of the equipment coordination reference value.

8. The AI-based intelligent health management method for devices according to claim 1, characterized in that, When the load characterization value is less than or equal to the preset load characterization value, the data processing method is determined to be fusion analysis against storage; The fusion analysis process includes: determining the analysis urgency coefficient of data slices based on slice length and slice anomaly, and then merging the data slices with compensation slices in descending order of analysis urgency coefficient to form fusion analysis pairs; The compensation slice is the data slice that was not included in the slice fusion analysis and has the smallest analysis urgency coefficient.

9. The AI-based intelligent health management method for devices according to claim 1, characterized in that, When the change frequency exceeds the change frequency threshold, a storage conflict is identified, and the maximum adaptation length is adjusted based on the change difference for the storage medium with the storage conflict. The maximum adaptation length is negatively correlated with the change difference; The change difference is the absolute value of the difference between the change frequency and the change frequency threshold.

10. A system applying the AI-based intelligent health management method for devices according to any one of claims 1 to 9, characterized in that, include: The data segmentation unit is used to determine the operating status of the target equipment based on the stability and diversity of the operating conditions according to the AI ​​health assessment model, and to determine the required data segmentation method as the baseline segmentation or adaptive segmentation based on the operating status. The equipment analysis unit, which is connected to the data segmentation unit, is used to determine the target equipment category as either a Class I or Class II target equipment based on the data segmentation stability within the working condition analysis cycle. The slice storage unit is connected to the data segmentation unit and the device analysis unit respectively. It is used to determine the load characterization value based on the number of a type of target device, and to determine whether to change the data processing method from the first screening method based on the device coordination dispersion to the second screening method of fusion analysis on storage based on the load characterization value. When executing the first screening method, the screening criterion based on the equipment coordination dispersion is the historical failure frequency or the equipment coordination reference value; When storing fusion analysis pairs, the fusion analysis pairs are determined based on the analysis urgency coefficient of the data slices; A storage compensation unit, connected to the slice storage unit, is used to determine whether there is a storage conflict based on the change frequency, and to adjust the maximum adaptable length of the data slice for storage media with storage conflicts.

Images