Online rapid diagnosis method and system for servo drive fault based on AI

Abstract

The present invention provides an AI-based servo drive fault online rapid diagnosis method and system, which achieves quantitative evaluation of fault data by acquiring candidate system fault monitoring data of the servo drive control system and converting it into a diagnostic difficulty index and fault description knowledge attributes. When the diagnostic difficulty index exceeds the preset index and the fault data lacks dynamic change information, the present invention can further utilize multiple fault penetration derivation data and their matching associated fault knowledge networks, and effectively identify and integrate similar fault modes by calculating the state feature distance between these associated fault knowledge networks, significantly improving the accuracy and efficiency of fault diagnosis, especially in complex and changeable fault scenarios, and can quickly locate the cause of the fault, providing data support for the immediate maintenance and recovery of the servo drive, thereby reducing system downtime and improving the overall operational reliability and stability of the equipment.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an AI-based servo drive fault online rapid diagnosis method and system. Background Art

[0002] As a key component in modern industrial automation systems, the stability and reliability of servo drives are crucial to ensure the continuous operation of production lines. However, in actual operation, servo drives may fail due to various factors (such as harsh environment, improper operation, equipment aging, etc.), which not only cause production line shutdowns, but may also cause safety accidents and economic losses. Therefore, it is particularly important to quickly and accurately diagnose servo drive failures.

[0003] Traditional servo drive fault diagnosis methods mainly rely on manual experience and simple fault code judgment. This method is often incapable of dealing with complex faults, with low diagnostic efficiency and difficulty in ensuring accuracy. With the advent of the Industrial 4.0 era and the development of intelligent manufacturing technology, data-driven fault diagnosis methods have gradually emerged, but existing data-driven methods mostly focus on the extraction and analysis of single fault features, which makes it difficult to fully reflect the full picture of servo drive faults, especially in the early stages of a fault or when the fault mode is complex and changeable, which can easily lead to misdiagnosis or missed diagnosis. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides an online rapid diagnosis method for servo drive faults based on AI, the method comprising:

[0005] Acquire candidate system fault monitoring data of the servo drive control system, and acquire a fault logic vector of the candidate system fault monitoring data, and convert the fault logic vector into a diagnostic difficulty index and a fault description knowledge attribute of the candidate system fault monitoring data;

[0006] If the diagnostic difficulty index is greater than a preset index, the fault description knowledge attribute reflects that the candidate system fault monitoring data does not contain dynamic change information, and the candidate system fault monitoring data includes multiple fault penetration derivation data, obtaining associated fault knowledge networks corresponding to the multiple fault penetration derivation data, and the feature matching degree between the fault penetration derivation data and the corresponding associated fault knowledge network is greater than a preset matching degree;

[0007] Determine the state characteristic distances between the associated fault knowledge networks respectively corresponding to the plurality of fault penetration derivation data;

[0008] If the state characteristic distance is less than the preset characteristic distance, the associated fault knowledge networks respectively corresponding to the plurality of fault penetration derivation data are used as fault diagnosis reference data of the corresponding fault penetration derivation data.

[0009] In a possible implementation of the first aspect, the method further includes:

[0010] If the state characteristic distance is not less than the preset characteristic distance, the plurality of fault penetration derivation data are decomposed into a plurality of candidate system fault monitoring data.

[0011] In a possible implementation manner of the first aspect, the acquiring a fault logic vector of the candidate system fault monitoring data includes:

[0012] Converting the candidate system fault monitoring data into a structured expression data set;

[0013] Performing fault logic deduction based on the structured expression data set through a fault logic deduction model to generate a fault logic vector of the candidate system fault monitoring data;

[0014] The step of converting the fault logic vector into the diagnostic difficulty index and fault description knowledge attributes of the candidate system fault monitoring data includes:

[0015] Recursively processing the fault logic vector through a recursive sub-model in a first fault diagnosis model to generate first recursive vector data, converting the first recursive vector data into first confidences of a plurality of label diagnosis difficulty indexes through a mapping sub-model in the first fault diagnosis model, and using the label diagnosis difficulty index corresponding to the largest first confidence as the diagnosis difficulty index of the candidate system fault monitoring data, wherein the diagnosis difficulty index is used to reflect the evaluation score of the candidate system fault monitoring data that can be used to generate fault diagnosis reference data;

[0016] The fault logic vector is recursively processed through the recursive sub-model in the second fault diagnosis model to generate second recursive vector data, and the second recursive vector data is converted into a second confidence mapping multiple fault description knowledge attributes through the mapping sub-model in the first fault diagnosis model, and the fault description knowledge attribute corresponding to the largest second confidence is used as the fault description knowledge attribute of the candidate system fault monitoring data, and the fault description knowledge attribute includes: including dynamic change information and not including dynamic change information.

[0017] In a possible implementation of the first aspect, the method further includes:

[0018] If the diagnostic difficulty index is greater than the preset index, and the fault description knowledge attribute reflects that the candidate system fault monitoring data contains dynamic change information, the candidate system fault monitoring data is deleted.

[0019] In a possible implementation of the first aspect, the method further includes:

[0020] If the diagnostic difficulty index is not greater than the preset index, the fault diagnosis reference data of the candidate system fault monitoring data is stored in the fault knowledge network sequence according to the diagnostic chronological order, and the different candidate system fault monitoring data are sequentially collected from the servo drive control system.

[0021] In a possible implementation manner of the first aspect, determining the state characteristic distances between the associated fault knowledge networks corresponding to the plurality of fault penetration derivation data respectively includes:

[0022] For each of the associated fault knowledge networks, determine a weighted average result of a state parameter of each fault state node in the associated fault knowledge network, and combine the weighted average results of the state parameters of each fault state node to generate a state feature of the associated fault knowledge network;

[0023] Determine discrete values ​​of state characteristics of associated fault knowledge networks corresponding to the plurality of fault penetration derivation data, and use the discrete values ​​as state characteristic distances between associated fault knowledge networks corresponding to the plurality of fault penetration derivation data.

[0024] In a possible implementation of the first aspect, the method further includes:

[0025] Acquire a sample fault chain data combination sequence, wherein the sample fault chain data combination sequence includes a plurality of sample fault chain data combinations, wherein the sample fault chain data combination includes sample system fault monitoring data and a sample associated fault knowledge network;

[0026] According to the ascending order of multiple set threshold values, determine the search parameter of the sample fault chain data combination sequence under the current threshold value, the search parameter is the ratio of the statistic of the search sample fault chain data combination to the presidential statistic of the multiple sample fault chain data combinations; the feature matching degree between the sample system fault monitoring data and the sample associated fault knowledge network in the search sample fault chain data combination is not less than the current threshold value;

[0027] If the recall parameter under the current threshold value is not less than the set parameter value, outputting the current threshold value as the preset matching degree;

[0028] Determine a discrete value of a state feature of the sample associated fault knowledge network in the search sample fault chain data combination, and use the discrete value of the state feature of the sample associated fault knowledge network in the search sample fault chain data combination as the preset feature distance.

[0029] In a possible implementation of the first aspect, the method further includes:

[0030] If the candidate system fault monitoring data is valid monitoring data, obtain multiple matching fault knowledge networks of the candidate system fault monitoring data, determine the fault diagnosis reference data of the candidate system fault monitoring data from the multiple matching fault knowledge networks, and store the fault diagnosis reference data of the candidate system fault monitoring data. The valid monitoring data is monitoring data that does not meet the invalid monitoring data requirements, and the invalid monitoring data requirements include: the diagnostic difficulty index is greater than the preset index, the fault description knowledge attribute reflects that the candidate system fault monitoring data does not contain dynamic change information, and the candidate system fault monitoring data includes multiple fault penetration derivation data; the state feature distance is not less than the preset feature distance; the diagnostic difficulty index is greater than the preset index, and the fault description knowledge attribute reflects that the candidate system fault monitoring data contains dynamic change information.

[0031] In a possible implementation of the first aspect, the method further includes:

[0032] Extracting key fault feature information from the associated fault knowledge network corresponding to each fault penetration derivation data, wherein the key fault feature information includes possible components where the fault occurs, the fault manifestation form, and the association between the fault and other related factors;

[0033] Integrate the fault feature information in the associated fault knowledge network corresponding to each fault penetration derivation data to construct a fault feature association map, wherein in the process of constructing the fault feature association map, connections are made according to the logical relationship between the fault features;

[0034] Performing a deep traversal analysis on the constructed fault feature association map, starting from the starting node in the fault feature association map, searching along the connection relationship between the fault features, identifying the potential fault propagation direction, and determining the weight relationship between different fault features during the search process. The weight relationship is determined according to the probability of the fault occurrence and the degree of influence on the servo drive control system. Thus, by analyzing the connection relationship and weight relationship in the fault feature association map, the propagation path information from the initial fault point to other fault points caused by the fault is determined. The propagation path information includes the order of fault nodes on each propagation path, the association relationship between the fault nodes, and the probability information of propagation along the propagation path;

[0035] According to the determined propagation path information, each fault propagation path is evaluated to determine the potential fault risk area, wherein, specifically for each fault propagation path, each fault node on the fault propagation path and the connection part between the fault nodes are analyzed, and the target fault node is determined in combination with the importance of the component represented by the fault node in the servo drive control system and the impact range of the fault on the entire servo drive control system when the fault occurs at the fault node, the target fault node and the surrounding associated fault nodes are determined as potential fault risk areas, and each potential fault risk area is described to obtain description information including the components included in each potential fault risk area, the propagation direction of the fault, and the impact information caused by the fault;

[0036] After generating a corresponding preventive maintenance strategy based on the determined potential fault risk areas, the preventive maintenance strategy is simulated and verified using the simulation model of the servo drive control system. Specifically, each maintenance measure and maintenance operation in the preventive maintenance strategy is input into the simulation model according to the set time parameters and sequence parameters. During the simulated preventive maintenance process, the running status change data of the servo drive control system is monitored. Based on the simulation results, it is determined whether the preventive maintenance strategy is verified to be effective. The preventive maintenance strategy verified to be effective in the simulation environment is applied to the actual servo drive control system.

[0037] On the other hand, an embodiment of the present invention also provides an AI-based online rapid fault diagnosis system for servo drives, comprising a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0038] Based on the above aspects, the embodiment of the present application achieves a quantitative evaluation of fault data by acquiring candidate system fault monitoring data of the servo drive control system and converting it into a diagnostic difficulty index and fault description knowledge attributes. When the diagnostic difficulty index exceeds the preset index and the fault data lacks dynamic change information, the present invention can further utilize multiple fault penetration derivation data and their matching associated fault knowledge networks, and effectively identify and integrate similar fault modes by calculating the state feature distances between these associated fault knowledge networks, thereby significantly improving the accuracy and efficiency of fault diagnosis, especially in complex and changeable fault scenarios, and being able to quickly locate the cause of the fault, providing data support for the immediate maintenance and recovery of the servo drive, thereby reducing system downtime and improving the overall operational reliability and stability of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1It is a schematic diagram of the execution flow of the AI ​​servo drive fault online rapid diagnosis method provided in an embodiment of the present invention.

[0040] Figure 2 It is a schematic diagram of the hardware architecture of the AI ​​servo drive fault online rapid diagnosis system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 The present invention is a flowchart of an online rapid diagnosis method for an AI servo drive fault provided by an embodiment of the present invention. The online rapid diagnosis method for an AI servo drive fault is introduced in detail below.

[0042] Step S110, obtaining candidate system fault monitoring data of the servo drive control system, and obtaining a fault logic vector of the candidate system fault monitoring data, and converting the fault logic vector into a diagnosis difficulty index and a fault description knowledge attribute of the candidate system fault monitoring data.

[0043] In this embodiment, when performing fault diagnosis related operations on the servo drive control system, candidate system fault monitoring data is first obtained. For example, in a server room of a large data center, there are many servo drive control systems in operation, providing power support and precise control for various devices. The server will collect relevant data from these servo drive control systems as candidate system fault monitoring data. These data may include the measured values ​​of various parameters such as the current value, voltage value, temperature value, speed, torque, etc. of the servo drive at different time points, as well as information such as the status of the control signal and the status of the feedback signal.

[0044] Then, the server needs to obtain the fault logic vectors of these candidate system fault monitoring data. In this process, the server will convert the candidate system fault monitoring data into a structured expression data set. For example, for the current value of the servo drive, the server will organize the current values ​​at different time points in a certain order and format to form a structured expression data set about the current data. At the same time, similar operations are performed on other parameters such as voltage values ​​and temperature values. After that, the fault logic is derived through the fault logic deduction model based on this structured expression data set to generate the fault logic vector of the candidate system fault monitoring data. Taking current data as an example, if the current value fluctuates greatly in a short period of time, and this fluctuation does not conform to the normal working mode, the fault logic deduction model can derive a logic vector related to the current fault according to predefined rules, such as comparison with the normal current fluctuation range, correlation with other related parameters, etc. The logic vector may contain information such as the amplitude and duration of the current fluctuation, and the logical relationship with other parameters.

[0045] Next, the server converts the fault logic vector into the diagnostic difficulty index and fault description knowledge attribute of the candidate system fault monitoring data. The server recursively processes the fault logic vector through the recursive sub-model in the first fault diagnosis model to generate the first recursive vector data. For example, for the current fault-related logic vector mentioned above, the recursive sub-model may process each element in the fault logic vector multiple times according to a certain algorithm to mine deeper information. Then, the first recursive vector data is converted into the first confidence of multiple label diagnostic difficulty indexes through the mapping sub-model in the first fault diagnosis model. Assuming that for current fault data, the mapping sub-model may give several different diagnostic difficulty index labels, such as "easy to diagnose", "difficult to diagnose", "very difficult to diagnose", etc., and assign a confidence to each label. If the confidence of the label "difficult to diagnose" is the highest, then the server uses the diagnostic difficulty index corresponding to this label as the diagnostic difficulty index of the candidate system fault monitoring data (here is the current-related data), and the diagnostic difficulty index reflects the evaluation score that the candidate system fault monitoring data can be used to generate fault diagnosis reference data.

[0046] At the same time, the server recursively processes the fault logic vector through the recursive sub-model in the second fault diagnosis model to generate second recursive vector data, and then converts the second recursive vector data into a second confidence that maps multiple fault description knowledge attributes through the mapping sub-model in the first fault diagnosis model. Continuing with the current fault data as an example, the fault description knowledge attributes include "containing dynamic change information" and "not containing dynamic change information." If after processing, the confidence of the attribute "not containing dynamic change information" is the highest, then the server uses this attribute as the fault description knowledge attribute of the candidate system fault monitoring data (current-related data). This means that for this current-related candidate system fault monitoring data, from the server's analysis, it does not have obvious dynamic change information and may be a relatively stable fault state manifestation.

[0047] Step S120: If the diagnostic difficulty index is greater than a preset index, the fault description knowledge attribute reflects that the candidate system fault monitoring data does not contain dynamic change information, and the candidate system fault monitoring data includes multiple fault penetration derivation data, the associated fault knowledge networks corresponding to the multiple fault penetration derivation data are obtained, and the feature matching degree between the fault penetration derivation data and the corresponding associated fault knowledge network is greater than the preset matching degree.

[0048] In this embodiment, when the diagnostic difficulty index is greater than the preset index, the fault description knowledge attribute reflects that the candidate system fault monitoring data does not contain dynamic change information, and the candidate system fault monitoring data includes multiple fault penetration derivation data, the corresponding operation will be performed. For example, assuming that the preset index is a certain value, when the server analyzes the candidate system fault monitoring data of a servo drive control system, it is found that the diagnostic difficulty index of the current-related data is higher than the preset index, and the fault description knowledge attribute indicates that the current data does not contain dynamic change information (for example, the current has always been at a value that deviates from the normal range but is stable), and the current-related data can be further decomposed into multiple fault penetration derivation data, such as the source of the current anomaly (which may be the power supply part), the transmission process of the current in the circuit (there may be abnormal line resistance, etc.), the distribution of the current in the internal circuit of the servo drive (there may be a fault in some circuit modules that causes uneven current distribution), and other aspects.

[0049] At this time, the server will obtain the associated fault knowledge network corresponding to multiple fault penetration derivation data. The associated fault knowledge network here is pre-built by the server or obtained from an existing fault knowledge database. Taking the fault penetration derivation data of the current source (power supply part) as an example, the corresponding associated fault knowledge network may include various fault modes of power supply, such as capacitor aging of the power module, transformer failure, etc., as well as the relationship between these fault modes and other related parameters (such as temperature, voltage fluctuation, etc.), and also includes symptoms that may appear under different fault modes (such as unstable output current, voltage drop, etc.). In addition, the feature matching degree between the fault penetration derivation data and the corresponding associated fault knowledge network is greater than the preset matching degree, which is determined by the server through a certain method. For example, the server will obtain a sample fault chain data combination sequence, which includes multiple sample fault chain data combinations, and the sample fault chain data combination includes sample system fault monitoring data and sample associated fault knowledge network. Then, according to the ascending order of multiple set threshold values, the search parameter of the sample fault chain data combination sequence under the current threshold value is determined. The search parameter is the ratio of the statistic of the search sample fault chain data combination to the presidential statistic of multiple sample fault chain data combinations. The feature matching degree between the sample system fault monitoring data and the sample associated fault knowledge network in the search sample fault chain data combination is not less than the current threshold value. If the search parameter under the current threshold value is not less than the set parameter value, the current threshold value is output as the preset matching degree. In this way, it is ensured that there is sufficient matching degree between the acquired associated fault knowledge network and the fault penetration derivation data, which can provide an effective reference for subsequent fault diagnosis.

[0050] Step S130: determining the state characteristic distances between the associated fault knowledge networks corresponding to the plurality of fault penetration derivation data.

[0051] When determining the state characteristic distance between the associated fault knowledge networks corresponding to multiple fault penetration derivation data, the following operations are performed. For each associated fault knowledge network, the server determines the weighted average result of the state parameters of each fault state node in the associated fault knowledge network. For example, in the associated fault knowledge network of the power supply part, the fault state nodes may include the fault states of components such as capacitors, transformers, and rectifiers. For the capacitor fault state node, its state parameters may include the change in the capacity of the capacitor, leakage conditions, etc.; for the transformer, the state parameters may include changes in the transformation ratio, heat generation conditions, etc. The server will assign a certain weight to each state parameter based on factors such as the importance of these components in the entire power supply system, and then calculate the weighted average result. The weighted average results of the state parameters of each fault state node are combined to generate the state characteristics of the associated fault knowledge network.

[0052] Next, the server will determine the discrete values ​​of the state characteristics of the associated fault knowledge networks corresponding to the multiple fault penetration derivation data, and use this discrete value as the state characteristic distance between the associated fault knowledge networks corresponding to the multiple fault penetration derivation data. For example, if there are two associated fault knowledge networks, one is about the power supply part and the other is about the current transmission line part. After the server calculates their state characteristics respectively, it determines the state characteristic distance between them by calculating the discrete value between the two state characteristics. The discrete value can reflect the degree of difference in the fault state performance of the two associated fault knowledge networks.

[0053] Step S140: If the state characteristic distance is less than a preset characteristic distance, the associated fault knowledge networks corresponding to the plurality of fault penetration derivation data are used as fault diagnosis reference data for the corresponding fault penetration derivation data.

[0054] In this embodiment, when it is found that the state characteristic distance is less than the preset characteristic distance, the associated fault knowledge networks corresponding to the multiple fault penetration derivation data can be used as fault diagnosis reference data for the corresponding fault penetration derivation data. For example, assuming that the preset characteristic distance is a specific value, after the server calculates the state characteristic distance between the two associated fault knowledge networks of the power supply part and the current transmission line part, it is found that the state characteristic distance is less than the preset characteristic distance. This means that the two associated fault knowledge networks are similar in the performance of the fault state, and there may be the same or related root causes of the fault. Therefore, the server will use these two associated fault knowledge networks as fault diagnosis reference data for the corresponding fault penetration derivation data (fault penetration derivation data related to the power supply part and the current transmission line part). These fault diagnosis reference data can be used for further fault diagnosis analysis, such as by comparing the fault mode, symptoms and other information in the two associated fault knowledge networks to more accurately locate the fault point in the servo drive control system, so as to take effective repair measures.

[0055] Based on the above steps, the embodiment of the present application achieves a quantitative evaluation of fault data by acquiring candidate system fault monitoring data of the servo drive control system and converting it into a diagnostic difficulty index and fault description knowledge attributes. When the diagnostic difficulty index exceeds the preset index and the fault data lacks dynamic change information, the present invention can further utilize multiple fault penetration derivation data and their matching associated fault knowledge networks, and effectively identify and integrate similar fault modes by calculating the state feature distances between these associated fault knowledge networks, thereby significantly improving the accuracy and efficiency of fault diagnosis, especially in complex and changeable fault scenarios, and being able to quickly locate the cause of the fault, providing data support for the immediate maintenance and recovery of the servo drive, thereby reducing system downtime and improving the overall operational reliability and stability of the equipment.

[0056] In a possible implementation, if the state characteristic distance is not less than the preset characteristic distance, the plurality of fault penetration derivation data are decomposed into a plurality of candidate system fault monitoring data.

[0057] In this embodiment, taking the previously mentioned servo drive control system as an example, it is assumed that after the server calculates the state characteristic distance of the associated fault knowledge network corresponding to the fault penetration derivation data of the power supply part and the current transmission line part, it is found that this distance is not less than the preset characteristic distance. This means that the two parts have large differences in fault state performance, and it is difficult to use them as a whole for fault diagnosis reference. At this time, the server will disassemble these fault penetration derivation data into multiple candidate system fault monitoring data. For example, the power supply part may be further disassembled into monitoring data of various components (capacitors, transformers, rectifiers, etc.) inside the power module, and the current transmission line part may be disassembled into monitoring data such as current and voltage at different nodes of the line. These re-disassembled candidate system fault monitoring data will be re-analyzed in the subsequent analysis process to more accurately find the root cause of the fault.

[0058] In a possible implementation, step S110 includes:

[0059] Step S111: convert the candidate system fault monitoring data into a structured expression data set.

[0060] Step S112, performing fault logic deduction based on the structured expression data set through a fault logic deduction model to generate a fault logic vector of the candidate system fault monitoring data.

[0061] Step S113, recursively process the fault logic vector through the recursive sub-model in the first fault diagnosis model to generate first recursive vector data, convert the first recursive vector data into first confidence levels of multiple label diagnosis difficulty indexes through the mapping sub-model in the first fault diagnosis model, and use the label diagnosis difficulty index corresponding to the largest first confidence level as the diagnosis difficulty index of the candidate system fault monitoring data, wherein the diagnosis difficulty index is used to reflect the evaluation score of the candidate system fault monitoring data that can be used to generate fault diagnosis reference data.

[0062] Step S114, recursively process the fault logic vector through the recursive sub-model in the second fault diagnosis model to generate second recursive vector data, convert the second recursive vector data into a second confidence mapping multiple fault description knowledge attributes through the mapping sub-model in the first fault diagnosis model, and use the fault description knowledge attribute corresponding to the largest second confidence as the fault description knowledge attribute of the candidate system fault monitoring data, wherein the fault description knowledge attribute includes: including dynamic change information and not including dynamic change information.

[0063] In this embodiment, the temperature monitoring data of the servo drive is used as the candidate system fault monitoring data for illustration, and the server converts these temperature data into a structured expression data set. For example, the server arranges the temperature values ​​collected at different times in chronological order, and records other related information, such as the working mode and load condition of the servo drive when the temperature value is collected, to form a structured temperature data set. Then, based on this structured expression data set, the fault logic deduction model is used to perform fault logic deduction to generate a fault logic vector of the candidate system fault monitoring data (here, the temperature data). If the temperature data shows a trend of gradually increasing and exceeding the normal operating temperature range, the fault logic deduction model will take into account the relationship between the temperature increase and the heat dissipation system, workload, ambient temperature and other factors of the servo drive, and derive a fault logic vector containing the correlation between these factors and the temperature change characteristics.

[0064] Furthermore, the server will recursively process the fault logic vector through the recursive sub-model in the first fault diagnosis model. Continuing with the fault logic vector of temperature data as an example, the recursive sub-model will perform multiple in-depth processes on the elements in this vector according to a specific algorithm to dig out more potential information, thereby generating first recursive vector data. Afterwards, the first recursive vector data is converted into the first confidence of multiple label diagnostic difficulty indexes through the mapping sub-model in the first fault diagnosis model. Assuming that there are several diagnostic difficulty index labels such as "simple", "medium" and "difficult", the mapping sub-model will calculate a confidence for each label based on the first recursive vector data. If the confidence of the "medium" label is the highest, the server will use the diagnostic difficulty index corresponding to this label as the diagnostic difficulty index of the temperature data, the candidate system fault monitoring data. This index reflects the evaluation score of the temperature data for generating fault diagnosis reference data.

[0065] At the same time, the server will also recursively process the fault logic vector through the recursive sub-model in the second fault diagnosis model to generate second recursive vector data. For example, for the fault logic vector of temperature data, the second recursive vector data is obtained after being processed by the recursive sub-model in the second fault diagnosis model. Then, the second recursive vector data is converted into a second confidence that maps multiple fault description knowledge attributes through the mapping sub-model in the first fault diagnosis model. The fault description knowledge attributes include "containing dynamically changing information" and "not containing dynamically changing information". If, after calculation, the confidence of the attribute "containing dynamically changing information" is the highest, then the server will use this attribute as the fault description knowledge attribute of the temperature data, the candidate system fault monitoring data. This shows that the temperature data has a dynamic change trend, and this dynamic change factor needs to be considered during fault diagnosis.

[0066] In a possible implementation, if the diagnostic difficulty index is greater than the preset index, and the fault description knowledge attribute reflects that the candidate system fault monitoring data contains dynamic change information, the candidate system fault monitoring data is deleted.

[0067] Assuming that the preset index is set to a certain value, for the torque monitoring data in a servo drive control system, the server calculates through the previous process that its diagnostic difficulty index is greater than the preset index, and the fault description knowledge attribute indicates that the torque data contains dynamic change information (for example, the torque fluctuates frequently and significantly in a short period of time). In this case, the torque monitoring data is too complex and unstable for fault diagnosis and is difficult to be effectively used to generate fault diagnosis reference data, so the server will delete this candidate system fault monitoring data to avoid unnecessary processing in the subsequent fault diagnosis process, thereby improving the efficiency and accuracy of fault diagnosis.

[0068] In one possible implementation, if the diagnostic difficulty index is not greater than the preset index, the fault diagnosis reference data of the candidate system fault monitoring data is stored in the fault knowledge network sequence according to the timing order of the diagnosis, and different candidate system fault monitoring data are collected sequentially from the servo drive control system.

[0069] Continuing with the speed monitoring data of the servo drive as an example, if the server calculates that its diagnostic difficulty index is not greater than the preset index, this indicates that the speed monitoring data is relatively easy to use to generate fault diagnosis reference data. The server will store the fault diagnosis reference data related to the speed monitoring data in the fault knowledge network sequence according to the diagnostic timing order. For example, the fault diagnosis reference data corresponding to the speed monitoring data collected at different time points will be stored in the fault knowledge network sequence in the order of collection. If the fault diagnosis reference data corresponding to the speed data collected at a certain moment indicates that the speed is slightly lower than the normal range and is related to a possible fault mode, this data will be stored in the fault knowledge network sequence in the order of collection, so that in the subsequent fault diagnosis analysis, the development process of the fault can be better analyzed according to this timing sequence, providing more basis for accurately finding the root cause of the fault. As a result, various candidate system fault monitoring data and related fault diagnosis reference data collected from the servo drive control system can be systematically managed and utilized to improve the efficiency and accuracy of the entire fault diagnosis process.

[0070] In a possible implementation, step S130 includes:

[0071] Step S131, for each of the associated fault knowledge networks, determine the weighted average result of the state parameters of each fault state node in the associated fault knowledge network, and combine the weighted average results of the state parameters of each fault state node to generate the state characteristics of the associated fault knowledge network.

[0072] Step S132, determining discrete values ​​of state characteristics of associated fault knowledge networks corresponding to the plurality of fault penetration derivation data, and using the discrete values ​​as state characteristic distances between associated fault knowledge networks corresponding to the plurality of fault penetration derivation data.

[0073] In a possible implementation, the method further includes:

[0074] Step A110, obtaining a sample fault chain data combination sequence, wherein the sample fault chain data combination sequence includes a plurality of sample fault chain data combinations, and the sample fault chain data combinations include sample system fault monitoring data and a sample associated fault knowledge network.

[0075] Step A120, according to the ascending order of multiple set threshold values, determine the completeness parameter of the sample fault chain data combination sequence under the current threshold value, the completeness parameter is the ratio of the statistic of the complete sample fault chain data combination to the presidential statistic of the multiple sample fault chain data combinations. The feature matching degree between the sample system fault monitoring data and the sample associated fault knowledge network in the complete sample fault chain data combination is not less than the current threshold value.

[0076] Step A130: If the recall parameter under the current threshold value is not less than the set parameter value, the current threshold value is output as the preset matching degree.

[0077] Step A140, determining a discrete value of a state feature of the sample associated fault knowledge network in the searched sample fault chain data combination, and using the discrete value of the state feature of the sample associated fault knowledge network in the searched sample fault chain data combination as the preset feature distance.

[0078] In this embodiment, the power supply module and the motor drive module in the servo drive control system are taken as examples. The server first processes each associated fault knowledge network. For the associated fault knowledge network of the power supply module, it contains multiple fault state nodes, such as capacitor fault state nodes, transformer fault state nodes, etc. For the capacitor fault state node, its state parameters include the actual capacity of the capacitor, the leakage current size, the equivalent series resistance, etc. The server assigns different weights according to the importance of these parameters in the judgment of capacitor faults. For example, the actual capacity of the capacitor is more critical for judging whether the capacitor is faulty, and a higher weight may be assigned. The state of the capacitor fault state node is comprehensively represented by calculating the weighted average results of these state parameters. For the transformer fault state node, its state parameters include transformation ratio, output voltage stability, heating conditions, etc., and weights are also assigned according to importance and weighted average results are calculated. The weighted average results of each fault state node in the power supply module associated fault knowledge network are combined to generate the state characteristics of the associated fault knowledge network.

[0079] Similarly, for the associated fault knowledge network of the motor drive module, the fault state nodes therein are such as the power transistor fault state node, the drive circuit fault state node, etc. The state parameters of the power transistor fault state node include on-resistance, breakdown voltage, etc., and the state parameters of the drive circuit fault state node include signal transmission delay, circuit gain, etc. These parameters are weighted according to their respective importance and the weighted average result is calculated, and then combined to obtain the state characteristics of the associated fault knowledge network of the motor drive module.

[0080] After obtaining the state characteristics of the associated fault knowledge network of the power supply module and the motor drive module, the server needs to determine the discrete values ​​of the two state characteristics. The calculation of the discrete value involves multiple algorithms, such as calculating the Euclidean distance between the state characteristic vectors or other suitable distance measurement methods. The discrete value is used as the state characteristic distance between the associated fault knowledge networks corresponding to the two fault penetration derivation data of the power supply module and the motor drive module. The distance can reflect the degree of difference in the fault state performance of the two modules, thereby providing a basis for subsequent fault diagnosis decisions.

[0081] In determining the preset matching degree and the preset feature distance, the server must first obtain a sequence of sample fault chain data combinations, which contains multiple sample fault chain data combinations, and each sample fault chain data combination includes sample system fault monitoring data and sample associated fault knowledge network. Taking multiple servo drive control systems in a data center as an example, these sample fault chain data combinations may be collected from different fault cases in the past. For example, a sample system fault monitoring data may be a collection of current, voltage, temperature and other monitoring data of a servo drive under a specific workload, and the corresponding sample associated fault knowledge network contains information such as possible fault modes, faulty components and the relationship between them determined based on the fault monitoring data.

[0082] Then, the server determines the completeness parameter of the sample fault chain data combination sequence under the current threshold value according to the ascending order of multiple set threshold values. The completeness parameter is the ratio of the statistic of the complete sample fault chain data combination to the presidential statistic of multiple sample fault chain data combinations. For each threshold value, the server checks whether the feature matching degree between the sample system fault monitoring data and the sample associated fault knowledge network in the sample fault chain data combination is not less than the current threshold value. For example, for a threshold value of 0.8, the server checks the sample fault chain data combination one by one and calculates the feature matching degree between the sample system fault monitoring data and the sample associated fault knowledge network. If this matching degree is not less than 0.8, this sample fault chain data combination is included in the statistic of the complete sample fault chain data combination. When the server has traversed all the sample fault chain data combinations, the completeness parameter under this threshold value can be calculated.

[0083] If the recall parameter under a certain current threshold value is not less than the set parameter value, the server will output the current threshold value as the preset matching degree. Assuming the set parameter value is 0.9, when the recall parameter calculated by the server under a certain threshold value reaches or exceeds 0.9, this threshold value is determined as the preset matching degree, which will be used to subsequently determine whether the matching degree between the fault penetration derivation data and the associated fault knowledge network is sufficient for fault diagnosis reference.

[0084] At the same time, the server also needs to determine the preset feature distance. The server will determine the discrete value of the state characteristics of the sample associated fault knowledge network in the complete sample fault chain data combination. Continuing with the example of the data center, for the sample associated fault knowledge network in the complete sample fault chain data combination, according to the method of calculating the state characteristics of the associated fault knowledge network described above, the state characteristics of each sample associated fault knowledge network are calculated respectively, and then the discrete value between these state characteristics is calculated. The discrete value is used as the preset feature distance. The preset feature distance will be used to determine whether the distance between the associated fault knowledge networks corresponding to multiple fault penetration derivation data is small enough, so as to decide whether to use these associated fault knowledge networks as fault diagnosis reference data. In this way, the preset matching degree and the preset feature distance can be accurately determined based on the historical fault data, thereby improving the accuracy and reliability of fault diagnosis.

[0085] In a possible implementation, if the candidate system fault monitoring data is valid monitoring data, multiple matching fault knowledge networks of the candidate system fault monitoring data are obtained, fault diagnosis reference data of the candidate system fault monitoring data is determined from the multiple matching fault knowledge networks, and the fault diagnosis reference data of the candidate system fault monitoring data is stored, and the valid monitoring data is monitoring data that does not meet the invalid monitoring data requirements, and the invalid monitoring data requirements include: the diagnostic difficulty index is greater than the preset index, the fault description knowledge attribute reflects that the candidate system fault monitoring data does not contain dynamic change information, and the candidate system fault monitoring data includes multiple fault penetration derivation data. The state feature distance is not less than the preset feature distance. The diagnostic difficulty index is greater than the preset index, and the fault description knowledge attribute reflects that the candidate system fault monitoring data contains dynamic change information.

[0086] In this embodiment, taking the servo drive control system of the data center as an example, the server continuously evaluates a large number of candidate system fault monitoring data. The valid monitoring data is determined after a series of screenings, and it does not meet the invalid monitoring data requirements. The invalid monitoring data requirements include a variety of complex situations.

[0087] First, consider the case where the diagnostic difficulty index is greater than the preset index, the fault description knowledge attribute reflects that the candidate system fault monitoring data does not contain dynamic change information, and the candidate system fault monitoring data includes multiple fault penetration derivation data. For example, for the current monitoring data of a servo drive, if its diagnostic difficulty index is judged to be greater than the preset index, this means that based on the existing fault diagnosis model and data characteristics, it is difficult to derive useful fault diagnosis reference data from the current monitoring data. At the same time, if the fault description knowledge attribute indicates that this current monitoring data does not contain dynamic change information, that is, the current value is in a relatively stable but abnormal state, and this current monitoring data can be further disassembled into multiple fault penetration derivation data, such as fault penetration derivation in multiple aspects such as from power input to current distribution in different circuit modules, then the current monitoring data in this case is invalid monitoring data.

[0088] Let's look at the case where the diagnostic difficulty index is greater than the preset index and the fault description knowledge attribute reflects that the candidate system fault monitoring data contains dynamic change information. Taking the speed monitoring data of the servo drive as an example, if the server determines that its diagnostic difficulty index exceeds the preset index, it means that it is difficult to obtain accurate fault diagnosis reference data from this speed monitoring data, and if the fault description knowledge attribute shows that this speed monitoring data contains dynamic change information, such as the speed fluctuates irregularly and rapidly within a certain range, this speed monitoring data is also invalid monitoring data.

[0089] There is also a situation where the state feature distance is not less than the preset feature distance. Assuming that when analyzing the associated fault knowledge network corresponding to the fault penetration derivation data of the power supply part and the motor drive part, the calculated state feature distance is not less than the preset feature distance, this indicates that the two parts have large differences in fault state performance and it is difficult to unify them as a reference for fault diagnosis. At this time, the relevant monitoring data may also be judged as invalid monitoring data.

[0090] When the candidate system fault monitoring data is valid monitoring data, the server will perform the following operations. The server will obtain multiple matching fault knowledge networks for the candidate system fault monitoring data. For example, for a valid temperature monitoring data, the server will search for multiple matching fault knowledge networks from the existing fault knowledge database. These matching fault knowledge networks may contain knowledge information under different temperature-related fault scenarios. In the servo drive control system of the data center, there may be different heat dissipation fault scenarios, component overheating fault scenarios and other related fault knowledge networks, which may match the temperature monitoring data.

[0091] The server then determines the fault diagnosis reference data for the candidate system fault monitoring data from multiple matching fault knowledge networks. For temperature monitoring data, the server selects the most suitable part as fault diagnosis reference data from many matching fault knowledge networks. If the fault knowledge network of a certain heat dissipation fault scenario clearly points out the association between a specific temperature range and a cooling fan fault, and this association is highly consistent with the current temperature monitoring data, then the knowledge related to this heat dissipation fault scenario may be determined as the fault diagnosis reference data for the temperature monitoring data.

[0092] Finally, the server will store the fault diagnosis reference data of the candidate system fault monitoring data. The server will establish a special storage area or data structure for each valid candidate system fault monitoring data to store these fault diagnosis reference data. For example, the fault diagnosis reference data of temperature monitoring data will be stored in the server's storage system in a certain format. In this way, in the subsequent fault diagnosis process, when encountering similar temperature monitoring data again, the server can quickly call these stored fault diagnosis reference data to improve the efficiency and accuracy of fault diagnosis. Through such a process, the server can effectively process effective monitoring data and provide a reliable basis for fault diagnosis of servo drive control systems.

[0093] In a possible implementation, the method further includes:

[0094] Step S150, extracting key fault feature information from the associated fault knowledge network corresponding to each fault penetration derivation data, wherein the key fault feature information includes possible components where the fault occurs, fault manifestations, and associations between the fault and other related factors.

[0095] Step S160, integrating the fault feature information in the associated fault knowledge network corresponding to each fault penetration derivation data, and constructing a fault feature association map, wherein in the process of constructing the fault feature association map, connections are made according to the logical relationship between the fault features.

[0096] Step S170, perform a deep traversal analysis on the constructed fault feature association map, start from the starting node in the fault feature association map, search along the connection relationship between the fault features, identify the potential fault propagation direction, and determine the weight relationship between different fault features during the search process. The weight relationship is determined according to the probability of the fault occurrence and the degree of influence on the servo drive control system. Thus, by analyzing the connection relationship and weight relationship in the fault feature association map, the propagation path information from the initial fault point to other resulting fault points is determined. The propagation path information includes the order of fault nodes on each propagation path, the association relationship between fault nodes, and the probability information of propagation along the propagation path.

[0097] Step S180, based on the determined propagation path information, each fault propagation path is evaluated to determine a potential fault risk area, wherein, specifically for each fault propagation path, each fault node on the fault propagation path and the connection parts between the fault nodes are analyzed, and the target fault node is determined in combination with the importance of the component represented by the fault node in the servo drive control system and the impact range of the fault on the entire servo drive control system when the fault occurs at the fault node, the target fault node and the surrounding associated fault nodes are determined as potential fault risk areas, and each potential fault risk area is described to obtain descriptive information including the components contained in each potential fault risk area, the propagation direction of the fault, and the impact information caused by the fault.

[0098] Step S190, after generating a corresponding pre-maintenance strategy according to the determined potential fault risk area, the pre-maintenance strategy is simulated and verified using the simulation model of the servo drive control system. Specifically, each maintenance measure and maintenance operation in the pre-maintenance strategy is input into the simulation model according to the set time parameters and sequence parameters, and the operating status change data of the servo drive control system is monitored during the simulated pre-maintenance process. Based on the simulation results, it is determined whether the pre-maintenance strategy is verified to be effective, and the pre-maintenance strategy verified to be effective in the simulation environment is applied to the actual servo drive control system.

[0099] In this embodiment, taking the servo drive control system of the data center as an example, it is assumed that there is fault penetration derivation data related to the power supply, and its corresponding associated fault knowledge network contains rich fault-related information. The server will extract the possible components where the fault occurs. For example, in the power supply module, the possible fault components are capacitors, transformers, rectifiers, etc. For the capacitor component, the fault manifestation may be capacitor leakage, which is specifically manifested as the leakage current at both ends of the capacitor exceeding the normal range. This manifestation can be obtained by analyzing the voltage and current monitoring data at both ends of the capacitor. At the same time, the fault is related to other related factors. For example, capacitor leakage may be related to excessively high ambient temperature. When the ambient temperature is at a high level for a long time, the electrolyte inside the capacitor may change, thereby causing the leakage phenomenon to intensify; it may also be related to the length of time the capacitor is used. Long-term use will gradually reduce the performance of the capacitor and increase the risk of leakage.

[0100] For the associated fault knowledge network corresponding to the fault penetration derivation data of the motor drive part, the possible components where the fault occurs include power transistors, drive circuits, etc. The fault manifestation of the power transistor may be an increase in on-resistance, which will cause abnormal motor drive current and make the motor unable to work normally. This kind of fault is related to the load condition of the motor. When the motor is under high load operation for a long time, the heat dissipation demand of the power transistor increases. If there is a problem with the heat dissipation system, it is easy to cause the power transistor temperature to be too high, which in turn causes the on-resistance to increase.

[0101] The server integrates the fault feature information in the associated fault knowledge network corresponding to each fault penetration derivation data to construct a fault feature association map. During the construction process, connections are made strictly according to the logical relationship between the fault features. For example, in the above-mentioned example of power supply and motor drive, capacitor leakage may cause unstable power supply output voltage, and unstable power supply output voltage will further affect the input voltage of the motor drive circuit, thereby affecting the working state of the power transistor. In the fault feature association map, the fault feature of capacitor leakage will be connected with the unstable power supply output voltage, and then the unstable power supply output voltage will be connected with the abnormal working state of the power transistor. This connection relationship reflects the conduction path of the fault in the system and is established based on the physical structure and working principle of the system.

[0102] Next, the server conducts a deep traversal analysis of the constructed fault feature association map. Starting from the starting node in the fault feature association map, this starting node may be a fault feature that is most prone to failure or has a greater impact on the system, such as capacitor leakage in the power supply module. The server searches along the connection relationship between the fault features to identify the potential fault propagation direction. During the search process, the weight relationship between different fault features is determined, and the weight relationship is determined based on the probability of the fault occurring and the degree of impact on the servo drive control system. For example, the probability of capacitor leakage may be affected by factors such as ambient temperature and usage time. If the data shows that the probability of capacitor leakage is high in a high temperature environment, and once capacitor leakage occurs, it will have a serious impact on the entire power supply module and even the entire servo drive control system, such as causing a sudden power outage in the system or unstable output, then the fault feature of capacitor leakage will be given a higher weight in the association relationship with other fault features.

[0103] Through such analysis, the server determines the propagation path information from the initial fault point to other fault points caused by it. For the propagation path starting from capacitor leakage, the order of fault nodes may be capacitor leakage-unstable power supply output voltage-abnormal power transistor working state-abnormal motor drive-abnormal motor working. The correlation between fault nodes clearly shows how each node is affected by the previous node, such as unstable power supply output voltage is caused by capacitor leakage, abnormal power transistor working state is because the unstable power supply output voltage leads to abnormal input voltage, etc. The probability information propagated along the propagation path is calculated based on the occurrence probability of each fault feature mentioned above and the correlation between them. For example, the propagation probability from capacitor leakage to unstable power supply output voltage may be based on the statistical data of voltage fluctuations after capacitor leakage.

[0104] Then, the server evaluates each fault propagation path according to the determined propagation path information to determine the potential fault risk area. For the above fault propagation path, the server analyzes each fault node on the path and the connection part between the fault nodes. For the capacitor, the fault node is extremely important in the servo drive control system because it is a key component of the power supply module. Once a capacitor has a leakage fault, the impact on the entire system is very large and may cause the entire system to fail to operate normally, so the capacitor is the target fault node. The capacitor and the surrounding related fault nodes, such as the circuit part connected to the capacitor and other components that may be affected by the capacitor leakage, are identified as potential fault risk areas. When describing this potential fault risk area, it will be clearly pointed out that the components included therein are capacitors, related circuit connection components, etc. The propagation direction of the fault is from the capacitor leakage to the power output, motor drive, etc. The impact information caused by the fault includes possible system power failure and abnormal motor operation.

[0105] For the power transistor, a fault node, it plays a key role in the motor drive part. If the power transistor fails, it will directly affect the normal drive of the motor. When the on-resistance of the power transistor increases, it has a greater impact on the motor drive circuit, which may cause the motor to fail to run at the expected speed and torque. Therefore, the power transistor is also a target fault node, and it and the surrounding associated drive circuit elements are identified as potential fault risk areas. When describing this area, the component information therein will be included, such as power transistors, resistors and capacitors in the drive circuit, etc. The fault propagation direction is from the power transistor to the motor. The impact of the fault is that the motor cannot work normally, which may affect the operation of the equipment that relies on the motor in the entire system.

[0106] Finally, the server generates a corresponding preventive maintenance strategy based on the identified potential failure risk areas. For potential failure risk areas where capacitors are located, the preventive maintenance strategy may include regularly testing the performance parameters of capacitors, such as capacitance, leakage current, etc., increasing the detection frequency in high temperature environments, and replacing capacitors in advance when the usage time reaches a certain threshold. For potential failure risk areas where power transistors are located, the preventive maintenance strategy may be to strengthen the monitoring of the power transistor temperature, optimize its heat dissipation conditions, and regularly check the on-resistance of the power transistors.

[0107] Afterwards, the server uses the simulation model of the servo drive control system to simulate and verify the pre-maintenance strategy. The specific operation is to input each maintenance measure and maintenance operation in the pre-maintenance strategy into the simulation model according to the set time parameters and sequence parameters. For example, for the pre-maintenance strategy of the capacitor, the performance detection of the capacitor is simulated in the simulation model according to the set detection cycle, and the capacitor replacement operation is simulated when the replacement threshold is reached; for the pre-maintenance strategy of the power transistor, the monitoring of its temperature and the adjustment of the heat dissipation conditions are simulated according to the set time interval. During the simulation of pre-maintenance, the server closely monitors the operating status change data of the servo drive control system, such as the stability of the power supply output, the operating parameters of the motor, etc. According to the simulation results, it is judged whether the pre-maintenance strategy is verified to be effective. If in the simulation, after the pre-maintenance strategy is executed, the failure rate of the system is significantly reduced and the system operation is more stable, then it is determined that the pre-maintenance strategy is verified to be effective in the simulation environment and applied to the actual servo drive control system to improve the reliability and stability of the system, reduce the possibility of failure, and reduce maintenance costs.

[0108] Figure 2 The hardware structure of the system 100 for online rapid diagnosis of faults based on AI servo drive provided by an embodiment of the present invention for implementing the above-mentioned online rapid diagnosis method of faults based on AI servo drive is shown as follows: Figure 2 As shown, the AI-based servo drive fault online rapid diagnosis system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0109] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the AI ​​servo drive fault online rapid diagnosis system 100 uses to execute or use to complete the exemplary method described in the present invention.

[0110] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the online rapid diagnosis method of AI servo drive fault based on the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0111] The specific implementation process of the processor 110 can refer to the various method embodiments executed based on the above-mentioned AI servo drive fault online rapid diagnosis system 100. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0112] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When the processor executes the computer executable instructions, the above-mentioned AI-based online rapid diagnosis method for servo drive faults is implemented.

[0113] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A method for rapid online diagnosis of servo drive faults based on AI, characterized in that: The method comprises: Acquire candidate system fault monitoring data of the servo drive control system, and acquire a fault logic vector of the candidate system fault monitoring data, and convert the fault logic vector into a diagnostic difficulty index and a fault description knowledge attribute of the candidate system fault monitoring data; If the diagnostic difficulty index is greater than a preset index, the fault description knowledge attribute reflects that the candidate system fault monitoring data does not contain dynamic change information, and the candidate system fault monitoring data includes multiple fault penetration derivation data, obtaining associated fault knowledge networks corresponding to the multiple fault penetration derivation data, and the feature matching degree between the fault penetration derivation data and the corresponding associated fault knowledge network is greater than a preset matching degree; Determine the state characteristic distances between the associated fault knowledge networks respectively corresponding to the plurality of fault penetration derivation data; If the state characteristic distance is less than the preset characteristic distance, the associated fault knowledge networks corresponding to the plurality of fault penetration derivation data are used as fault diagnosis reference data of the corresponding fault penetration derivation data; The acquiring the fault logic vector of the candidate system fault monitoring data comprises: Converting the candidate system fault monitoring data into a structured expression data set; Performing fault logic deduction based on the structured expression data set through a fault logic deduction model to generate a fault logic vector of the candidate system fault monitoring data; The step of converting the fault logic vector into the diagnostic difficulty index and fault description knowledge attributes of the candidate system fault monitoring data includes: Recursively processing the fault logic vector through a recursive sub-model in a first fault diagnosis model to generate first recursive vector data, converting the first recursive vector data into first confidences of a plurality of label diagnosis difficulty indexes through a mapping sub-model in the first fault diagnosis model, and using the label diagnosis difficulty index corresponding to the largest first confidence as the diagnosis difficulty index of the candidate system fault monitoring data, wherein the diagnosis difficulty index is used to reflect the evaluation score of the candidate system fault monitoring data that can be used to generate fault diagnosis reference data; Recursively process the fault logic vector through the recursive sub-model in the second fault diagnosis model to generate second recursive vector data, convert the second recursive vector data into a second confidence mapping multiple fault description knowledge attributes through the mapping sub-model in the first fault diagnosis model, and use the fault description knowledge attribute corresponding to the largest second confidence as the fault description knowledge attribute of the candidate system fault monitoring data, wherein the fault description knowledge attribute includes: including dynamic change information and not including dynamic change information; The determining of the state characteristic distances between the associated fault knowledge networks respectively corresponding to the plurality of fault penetration derivation data includes: For each of the associated fault knowledge networks, determine a weighted average result of a state parameter of each fault state node in the associated fault knowledge network, and combine the weighted average results of the state parameters of each fault state node to generate a state feature of the associated fault knowledge network; Determine discrete values ​​of state characteristics of associated fault knowledge networks corresponding to the plurality of fault penetration derivation data, and use the discrete values ​​as state characteristic distances between associated fault knowledge networks corresponding to the plurality of fault penetration derivation data; The method further comprises: Extracting key fault feature information from the associated fault knowledge network corresponding to each fault penetration derivation data, wherein the key fault feature information includes possible components where the fault occurs, the fault manifestation form, and the association between the fault and other related factors; Integrate the fault feature information in the associated fault knowledge network corresponding to each fault penetration derivation data to construct a fault feature association map, wherein in the process of constructing the fault feature association map, connections are made according to the logical relationship between the fault features; Performing a deep traversal analysis on the constructed fault feature association map, starting from the starting node in the fault feature association map, searching along the connection relationship between the fault features, identifying the potential fault propagation direction, and determining the weight relationship between different fault features during the search process. The weight relationship is determined according to the probability of the fault occurrence and the degree of influence on the servo drive control system. Thus, by analyzing the connection relationship and weight relationship in the fault feature association map, the propagation path information from the initial fault point to other fault points caused by the fault is determined. The propagation path information includes the order of fault nodes on each propagation path, the association relationship between the fault nodes, and the probability information of propagation along the propagation path; According to the determined propagation path information, each fault propagation path is evaluated to determine the potential fault risk area, wherein, specifically for each fault propagation path, each fault node on the fault propagation path and the connection part between the fault nodes are analyzed, and the target fault node is determined in combination with the importance of the component represented by the fault node in the servo drive control system and the impact range of the fault on the entire servo drive control system when the fault occurs at the fault node, the target fault node and the surrounding associated fault nodes are determined as potential fault risk areas, and each potential fault risk area is described to obtain description information including the components included in each potential fault risk area, the propagation direction of the fault, and the impact information caused by the fault; After generating a corresponding preventive maintenance strategy based on the determined potential fault risk areas, the preventive maintenance strategy is simulated and verified using the simulation model of the servo drive control system. Specifically, each maintenance measure and maintenance operation in the preventive maintenance strategy is input into the simulation model according to the set time parameters and sequence parameters. During the simulated preventive maintenance process, the running status change data of the servo drive control system is monitored. Based on the simulation results, it is determined whether the preventive maintenance strategy is verified to be effective. The preventive maintenance strategy verified to be effective in the simulation environment is applied to the actual servo drive control system.

2. The method for online rapid diagnosis of AI servo drive faults according to claim 1 is characterized in that: The method further comprises: If the state characteristic distance is not less than the preset characteristic distance, the plurality of fault penetration derivation data are decomposed into a plurality of candidate system fault monitoring data.

3. The method for online rapid diagnosis of AI servo drive faults according to claim 1 is characterized in that: The method further comprises: If the diagnostic difficulty index is greater than the preset index, and the fault description knowledge attribute reflects that the candidate system fault monitoring data contains dynamic change information, the candidate system fault monitoring data is deleted.

4. The method for online rapid diagnosis of AI servo drive faults according to claim 2 is characterized in that: The method further comprises: If the diagnostic difficulty index is not greater than the preset index, the fault diagnosis reference data of the candidate system fault monitoring data is stored in the fault knowledge network sequence according to the diagnostic chronological order, and the different candidate system fault monitoring data are sequentially collected from the servo drive control system.

5. The method for online rapid diagnosis of servo drive faults based on AI according to claim 1 is characterized in that: The method further comprises: Acquire a sample fault chain data combination sequence, wherein the sample fault chain data combination sequence includes a plurality of sample fault chain data combinations, wherein the sample fault chain data combination includes sample system fault monitoring data and a sample associated fault knowledge network; According to the ascending order of multiple set threshold values, determine the search parameter of the sample fault chain data combination sequence under the current threshold value, the search parameter is the ratio of the statistic of the search sample fault chain data combination to the presidential statistic of the multiple sample fault chain data combinations; the feature matching degree between the sample system fault monitoring data and the sample associated fault knowledge network in the search sample fault chain data combination is not less than the current threshold value; If the recall parameter under the current threshold value is not less than the set parameter value, outputting the current threshold value as the preset matching degree; Determine a discrete value of a state feature of the sample associated fault knowledge network in the search sample fault chain data combination, and use the discrete value of the state feature of the sample associated fault knowledge network in the search sample fault chain data combination as the preset feature distance.

6. The method for online rapid diagnosis of servo drive faults based on AI according to any one of claims 1 to 5, characterized in that: The method further comprises: If the candidate system fault monitoring data is valid monitoring data, obtain multiple matching fault knowledge networks of the candidate system fault monitoring data, determine the fault diagnosis reference data of the candidate system fault monitoring data from the multiple matching fault knowledge networks, and store the fault diagnosis reference data of the candidate system fault monitoring data. The valid monitoring data is monitoring data that does not meet the invalid monitoring data requirements, and the invalid monitoring data requirements include: the diagnostic difficulty index is greater than the preset index, the fault description knowledge attribute reflects that the candidate system fault monitoring data does not contain dynamic change information, and the candidate system fault monitoring data includes multiple fault penetration derivation data; the state feature distance is not less than the preset feature distance; the diagnostic difficulty index is greater than the preset index, and the fault description knowledge attribute reflects that the candidate system fault monitoring data contains dynamic change information.

7. An AI-based servo drive fault online rapid diagnosis system, characterized in that: The AI-based servo drive fault online rapid diagnosis system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the AI-based servo drive fault online rapid diagnosis method described in any one of claims 1 to 6.

Images