Intelligent servo drive fault warning method and system based on deep learning

Abstract

The present invention provides an intelligent servo drive fault warning method and system based on deep learning, which relates to the field of deep learning technology. By deeply analyzing the changes in the state logic network of the servo drive during operation, efficient warning of the servo drive fault is achieved. First, a priori operating state logic network set before the servo drive triggers the target operating state logic network is obtained. Then, by calculating the operating state logic offset between the prior operating state logic networks and the operating state logic offset between the target operating state logic network and its associated operating state logic network, first fault basis data and second fault basis data are respectively determined. Finally, based on these two types of fault basis data, the present invention can accurately analyze whether the servo drive has a target fault warning activity when the target operating state logic network is triggered, thereby effectively improving the operating reliability and safety of the servo drive.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a deep learning-based intelligent servo drive fault warning method and system. Background Art

[0002] As a key component in modern automation control systems, the stable operation of servo drives is crucial to ensuring the reliability and performance of the entire system. However, in actual applications, servo drives are often subject to various internal and external factors, such as load changes, ambient temperature fluctuations, unstable power supply, etc., and various faults often occur. If these faults are not discovered and handled in time, they often lead to equipment downtime, reduced production efficiency, and even safety accidents.

[0003] Traditional servo drive fault detection usually relies on manual inspections or simple threshold judgments. These methods have disadvantages such as low detection efficiency and poor accuracy, and are difficult to meet the needs of modern industry for efficient and intelligent fault warnings. 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 intelligent servo drive fault warning method based on deep learning, the method comprising:

[0005] Acquire a priori operating state logic network set corresponding to the target operating state logic network triggered by the servo drive; the priori operating state logic network set includes x priori operating state logic networks triggered by the servo drive before triggering the target operating state logic network, wherein x is a positive integer, and the node interval between the triggering node of the priori operating state logic network and the triggering node of the target operating state logic network meets the set interval requirement;

[0006] Determining first fault basis data based on an operation state logic offset between every two logically associated a priori operation state logic networks in the a priori operation state logic network set;

[0007] Determine the second fault basis data based on the operation state logic offset between the target operation state logic network and the associated operation state logic network corresponding to the target operation state logic network; the associated operation state logic network is an operation state logic network associated with the target operation state logic network and triggered by the servo drive before triggering the target operation state logic network;

[0008] Based on the first fault basis data and the second fault basis data, it is analyzed whether a target fault warning activity occurs within a triggering interval of the servo driver when the servo driver triggers the target operating state logic network.

[0009] On the other hand, an embodiment of the present invention also provides an intelligent servo drive fault warning system based on deep learning, including 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.

[0010] Based on the above aspects, the embodiment of the present application realizes efficient early warning of servo drive faults by deeply analyzing the changes in the state logic network of the servo drive during operation. Specifically, the method first obtains the set of prior operation state logic networks before the servo drive triggers the target operation state logic network, and then, by calculating the operation state logic offset between the prior operation state logic networks, and the operation state logic offset between the target operation state logic network and its associated operation state logic network, the first fault basis data and the second fault basis data are determined respectively. Finally, based on these two types of fault basis data, the present invention can accurately analyze whether the servo drive has a target fault warning activity when triggering the target operation state logic network, thereby effectively improving the operation reliability and safety of the servo drive, and reducing the downtime and maintenance costs caused by faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of the execution flow of the intelligent servo drive fault warning method based on deep learning provided in an embodiment of the present invention.

[0012] Figure 2 It is a schematic diagram of the hardware architecture of an intelligent servo drive fault warning system based on deep learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 It is a flow chart of an intelligent servo drive fault warning method based on deep learning provided by an embodiment of the present invention. The intelligent servo drive fault warning method based on deep learning is introduced in detail below.

[0014] Step S110, obtaining a priori operating state logic network set corresponding to the target operating state logic network triggered by the servo drive. The priori operating state logic network set includes x priori operating state logic networks triggered by the servo drive before triggering the target operating state logic network, where x is a positive integer, and the node interval between the triggering node of the priori operating state logic network and the triggering node of the target operating state logic network meets the set interval requirement.

[0015] In this embodiment, the servo driver triggers different operation state logic networks in different working stages. For example, in an automobile manufacturing workshop, the servo driver controls the movement of a robot arm, and the robot arm is performing an automobile parts assembly task.

[0016] Assume that the target operation state logic network is the logic network corresponding to the robot arm when performing a precise installation action of a specific component, which requires extremely high precision and a specific motion trajectory. Before this target operation state logic network is triggered, the servo drive has gone through multiple different operation state logic networks.

[0017] Assume x = 3. The server needs to obtain three prior operation state logic networks that are triggered before this target operation state logic network. The prior operation state logic network may include the logic network in which the robot arm moves from the initial position to the vicinity of the component to be installed (prior operation state logic network 1), the logic network in which the arm posture is adjusted to align with the installation position (prior operation state logic network 2), and the logic network in which the installation tool is close to the component but has not yet started precise installation (prior operation state logic network 3).

[0018] The node interval here meets the set interval requirements, for example, the set interval requirements are in time sequence and the operations between adjacent logic networks have a certain consistency and order. In this scenario, moving from the initial position to the vicinity of the component to be installed, then adjusting the posture, and then approaching the component, these operations are performed in sequence and are coherent in time, so the intervals between their trigger nodes meet the set requirements. The server obtains the relevant information of these prior operation state logic networks through the communication interface with the servo drive or from the database storing the servo drive operation data, thereby constructing a set of prior operation state logic networks.

[0019] Step S120 : determining first fault basis data based on the operation state logic offset between every two logically associated a priori operation state logic networks in the a priori operation state logic network set.

[0020] In this embodiment, for the three a priori operating state logic networks mentioned above, the server needs to construct two a priori operating state logic network combinations. A priori operating state logic network combination 1 is composed of a priori operating state logic network 1 and a priori operating state logic network 2, and a priori operating state logic network combination 2 is composed of a priori operating state logic network 2 and a priori operating state logic network 3.

[0021] For the prior operation state logic network combination 1, the server first performs graph encoding representation on the prior operation state logic network 1 and the prior operation state logic network 2. Taking the prior operation state logic network 1 as an example, the server identifies the network node information in this logic network based on the deep learning model. For example, in the process of the robot arm moving from the initial position to the vicinity of the parts to be installed, the node information may include the starting position, speed, acceleration and other nodes of each joint of the arm, and the connection relationship information is the synergistic relationship between the movements of each joint, such as how the movement of a joint affects the movement trajectory of other joints.

[0022] Based on the network node information, the server abstracts the features of each node. For example, for the starting position node of the arm joint, according to its role and status in the entire movement process, a set of abstract feature labels are defined, such as "initial position reference node", which is used to describe the role of this node in the entire prior operation state logic network 1. Then, based on these abstract node feature sets, a coding framework is constructed. The top level of this coding framework is the overall classification framework for the entire node feature set, such as position class, speed class, acceleration class, etc., and each category is further subdivided into subcategories, such as the position class is further subdivided according to different joints. The bottom layer is the coding rules and positions corresponding to each specific node feature label.

[0023] According to the constructed coding framework, each node in the network node information is coded and mapped according to the corresponding abstract node features. For example, the starting position node of the arm joint is converted into a coding value according to its corresponding coding rule. This coding value reflects the characteristics of the node and its position in the entire network and its relationship with other nodes. Based on the node set after preliminary coding, for each connection relationship in the connection relationship information, according to the coding values ​​of the two connected nodes and the properties of the connection itself, the corresponding coding rules are constructed to generate the corresponding connection relationship coding. Finally, the node set after preliminary coding and the coded connection relationship set are integrated to generate the network graph coding feature data corresponding to the prior operation state logic network 1.

[0024] In the same way, the server generates the network graph encoding feature data corresponding to the prior operation state logic network 2. Then, based on these two network graph encoding feature data, the operation state logic offset data corresponding to the prior operation state logic network combination 1 is determined. The server determines the global state encoding vector corresponding to each prior operation state logic network based on the state encoding vectors corresponding to each state element in the network graph encoding feature data corresponding to each prior operation state logic network. For example, in the prior operation state logic network 1, the state encoding vectors corresponding to each state element such as the position and speed of different joints are subjected to some calculation (such as weighted summation, etc.) to obtain the global state encoding vector. The characteristic distance between the global state encoding vectors corresponding to the prior operation state logic network 1 and the prior operation state logic network 2 is calculated. This characteristic distance is the operation state logic offset data corresponding to the prior operation state logic network combination 1.

[0025] According to the same process, the server calculates the operation state logic offset data corresponding to the prior operation state logic network combination 2. Finally, the average offset result of the operation state logic offset data corresponding to the two prior operation state logic network combinations is calculated, and this average offset result is the first fault basis data.

[0026] Step S130, determining second fault basis data based on an operation state logic offset between the target operation state logic network and an associated operation state logic network corresponding to the target operation state logic network, wherein the associated operation state logic network is an operation state logic network associated with the target operation state logic network and triggered by the servo drive before triggering the target operation state logic network.

[0027] Assume that the associated operating state logic network corresponding to the target operating state logic network is the prior operating state logic network 3 (in the robot arm installation scenario mentioned above, the prior operating state logic network 3 is a logic network in which the installation tool is close to the component but the precise installation has not yet started, and is closely related to the target operating state logic network of precise installation).

[0028] The server performs graph coding representation on the target operating state logic network and the prior operating state logic network 3 respectively. For the target operating state logic network, the network node information is identified, such as the precise position of each joint of the arm during precise installation, the precise torque applied, and other nodes, and the connection relationship information is the coordinated relationship between the precise movements of each joint. Then, similar operations as the previous prior operating state logic network are performed, such as feature abstraction, coding framework construction, coding mapping, etc., to generate network graph coding feature data corresponding to the target operating state logic network. Similarly, corresponding network graph coding feature data is also generated for the prior operating state logic network 3.

[0029] According to the network diagram encoding feature data corresponding to the target operating state logic network and the prior operating state logic network 3 respectively, the second fault basis data is determined. Based on the state encoding vectors corresponding to each state element in the network diagram encoding feature data corresponding to the target operating state logic network, the global state encoding vector corresponding to the target operating state logic network is determined. For example, the state encoding vectors corresponding to the state elements such as the precise position and torque of each joint during precise installation are calculated to obtain the global state encoding vector. Similarly, the global state encoding vector corresponding to the prior operating state logic network 3 is determined. The characteristic distance between the global state encoding vector corresponding to the target operating state logic network and the global state encoding vector corresponding to the prior operating state logic network 3 is calculated. This characteristic distance is the second fault basis data.

[0030] Step S140, analyzing whether a target fault warning activity occurs within a triggering interval of the servo drive when the servo drive triggers the target operating state logic network based on the first fault basis data and the second fault basis data.

[0031] Assume that the first threshold value is set to 0.3, the second threshold value is set to 0.5, and the third threshold value is set to 0.2.

[0032] If the first fault basis data is less than 0.3, for example, the first fault basis data is 0.2. At this time, if the second fault basis data is not less than 0.5, for example, the second fault basis data is 0.6, then the server determines that the target fault warning activity occurs within the triggering interval of the servo drive when the servo drive triggers the target operating state logic network. This is because the first fault basis data is small, indicating that the offset between the previous prior operating state logic network is relatively normal, but the second fault basis data is large, indicating that there is a large logical offset between the target operating state logic network and the associated prior operating state logic network, and there may be a hidden fault.

[0033] If the second fault basis data is less than 0.5, for example, the second fault basis data is 0.4, then the server determines that when the servo drive triggers the target operating state logic network, no target fault warning activity occurs within the triggering interval of the servo drive. Because although the first fault basis data is small, indicating that it was normal before, the second fault basis data is also small, indicating that the logic offset between the target operating state logic network and the associated network is within the normal range.

[0034] If the first fault basis data is not less than 0.3, for example, the first fault basis data is 0.4. At this time, the server calculates the deviation between the second fault basis data and the first fault basis data. Assuming that the second fault basis data is 0.6, the deviation is 0.2. Since the deviation 0.2 is not less than the third threshold value 0.2, the server determines that the target fault warning activity occurs within the triggering interval of the servo drive when the servo drive triggers the target operating state logic network. This means that although the offset between the prior operating state logic networks has deviated somewhat from normal, the logical offset deviation between the target operating state logic network and the associated network is larger, which may indicate that a fault is about to occur.

[0035] If the deviation is less than 0.2, for example, the deviation is 0.1, then the server determines that when the servo drive triggers the target operating state logic network, no target fault warning activity occurs within the trigger interval of the servo drive. This indicates that although there is a certain logic offset change, it is still within an acceptable range and is unlikely to fail.

[0036] Based on the above steps, the embodiment of the present application realizes efficient early warning of servo drive faults by deeply analyzing the changes in the state logic network of the servo drive during operation. Specifically, the method first obtains the set of prior operation state logic networks before the servo drive triggers the target operation state logic network, and then, by calculating the operation state logic offset between the prior operation state logic networks, and the operation state logic offset between the target operation state logic network and its associated operation state logic network, the first fault basis data and the second fault basis data are determined respectively. Finally, based on these two types of fault basis data, the present invention can accurately analyze whether the servo drive has a target fault warning activity when the target operation state logic network is triggered, thereby effectively improving the operation reliability and safety of the servo drive, and reducing the downtime and maintenance costs caused by faults.

[0037] In a possible implementation, step S120 includes:

[0038] Step S121: construct x-1 priori operating state logical network combinations based on the x priori operating state logical networks in the priori operating state logical network set, wherein each priori operating state logical network combination includes two priori operating state logical networks having associated trigger nodes.

[0039] Step S122: for each of the a priori operating state logic network combinations, based on the operating state logic offset between two of the a priori operating state logic networks included in the a priori operating state logic network combination, determine the operating state logic offset data corresponding to the a priori operating state logic network combination.

[0040] Step S123, calculating the average offset result of the operating state logic offset data corresponding to the x-1 prior operating state logic network combinations respectively as the first fault basis data.

[0041] In this embodiment, in the aforementioned automobile manufacturing workshop, the server is responsible for monitoring the operation status logic network related to the servo drive. In this scenario, the servo drive controls the robot arm to perform the assembly operation of automobile parts.

[0042] Assume that x in the set of prior operating state logic networks is 3. These three prior operating state logic networks correspond to the robot arm moving from the initial position to the vicinity of the component to be installed (prior operating state logic network 1), adjusting the arm posture to align with the installation position (prior operating state logic network 2), and bringing the installation tool close to the component but not starting precise installation (prior operating state logic network 3).

[0043] The server constructs two prior operation state logic network combinations based on the three prior operation state logic networks. Prior operation state logic network combination 1 includes prior operation state logic network 1 and prior operation state logic network 2, and prior operation state logic network combination 2 includes prior operation state logic network 2 and prior operation state logic network 3. The trigger nodes of the two prior operation state logic networks in each combination are associated. For example, the completion of the action of prior operation state logic network 1 triggers the start of prior operation state logic network 2. This sequence reflects the association of the trigger nodes.

[0044] For the a priori operating state logic network combination 1, the server needs to determine the operating state logic offset between the a priori operating state logic network 1 and the a priori operating state logic network 2. The server first performs graph encoding on the a priori operating state logic network 1, identifies its network node information, such as the starting position, speed, acceleration, etc. of each joint of the robot arm, and the connection relationship information is the collaborative relationship between the movements of each joint. Based on the network node information, the features of each node are abstracted, feature labels are defined for different types of nodes, and then a coding framework is constructed. The nodes are encoded and mapped according to the abstract features, and then the connection relationship encoding is constructed according to the connection relationship. Finally, the network graph encoding feature data of the a priori operating state logic network 1 is integrated. Similarly, the above operation is performed on the a priori operating state logic network 2 to obtain its network graph encoding feature data.

[0045] Based on the two network graph encoding feature data, the server determines the operation state logic offset data corresponding to the prior operation state logic network combination 1. The server calculates its global state encoding vector through the state encoding vector corresponding to each state element in the network graph encoding feature data corresponding to the prior operation state logic network 1, and performs the same operation on the prior operation state logic network 2. Then, the feature distance between the two global state encoding vectors is calculated, and this feature distance is the operation state logic offset data corresponding to the prior operation state logic network combination 1.

[0046] According to the same method, the server determines the operating state logic offset data corresponding to the prior operating state logic network combination 2. Finally, the server calculates the average offset result of the operating state logic offset data corresponding to the two prior operating state logic network combinations. This average offset result comprehensively considers the logic offset between each prior operating state logic network combination and uses it as the first fault basis data. This first fault basis data can reflect the overall logic offset state between the prior operating state logic networks during the operation of the robot arm, and provides an important basis for subsequent judgment of whether there is a fault warning activity. For example, if the average offset result is small, it means that the logical relationship between the prior operating state logic networks is relatively stable; if the average offset result is large, it may imply that there are some potential problems or abnormal conditions before reaching the target operating state logic network.

[0047] In a possible implementation, the step of determining the operating state logic offset between the two associated a priori operating state logic networks includes:

[0048] Step A110: Based on the deep learning model, the two prior operating state logic networks are respectively represented by graph coding, and network graph coding feature data corresponding to the two prior operating state logic networks are generated.

[0049] Step A120, determining the operation state logic offset data between the two a priori operation state logic networks according to the network graph coding feature data respectively corresponding to the two a priori operation state logic networks.

[0050] Step S130 includes:

[0051] Step S131, performing graph coding representation on the target operation state logic network and the associated operation state logic network respectively, and generating network graph coding feature data corresponding to the target operation state logic network and the associated operation state logic network respectively.

[0052] Step S132, determining the second fault basis data according to the operation state logic offset between the network diagram encoding feature data respectively corresponding to the target operation state logic network and the associated operation state logic network.

[0053] In this embodiment, in the scenario of an automobile manufacturing workshop, the operation status logic network related to the servo drive is continuously monitored and analyzed.

[0054] For the determination of the operating state logic offset between two associated prior operating state logic networks, take prior operating state logic network 1 (the robot arm moves from the initial position to the vicinity of the parts to be installed) and prior operating state logic network 2 (adjusting the arm posture to align with the installation position) as examples. The server first performs graph encoding on the prior operating state logic network 1. The server will identify the network node information in the prior operating state logic network 1, which covers various parameters of the joints of the robot arm in the process of moving from the initial position to the vicinity of the parts to be installed, such as the starting position node of each joint, which contains precise coordinate information; the moving speed node of each joint, which reflects the speed change in this process; the acceleration node of each joint, which reflects the speed change, etc. At the same time, the connection relationship information is also crucial, which describes the synergistic relationship between the movements of each joint, such as how the movement of a joint affects the movement trajectory of other joints, which may involve complex mechanical kinematics principles.

[0055] Based on the information of these network nodes, the server abstracts the features of each node. For the starting position node, according to its role and status in the entire prior operating state logic network 1, a set of abstract feature labels are defined, such as "initial position reference node", which is used to accurately describe the specific meaning of this node in the entire logical network. Then, a coding framework is constructed based on these abstract node feature sets. The top level of this coding framework is the overall classification framework for the entire node feature set, such as position, speed, acceleration and other major categories. Subcategories are further subdivided under each major category. Taking the position category as an example, it may be further subdivided according to different joints. The bottom layer is the coding rules and positions corresponding to each specific node feature label.

[0056] According to the constructed coding framework, each node in the network node information is coded and mapped according to the corresponding abstract node features. For example, the starting position node of a joint of the arm is converted into a coding value according to its corresponding coding rule. This coding value not only reflects the characteristics of the node itself, but also reflects its position in the entire prior operation state logic network 1 and its relationship with other nodes. Based on the node set after preliminary coding, for each connection relationship in the connection relationship information, according to the coding values ​​of the two connected nodes and the properties of the connection itself, the corresponding coding rules are constructed to generate the corresponding connection relationship coding. Finally, the node set after preliminary coding and the encoded connection relationship set are integrated to generate the network graph coding feature data corresponding to the prior operation state logic network 1.

[0057] The same process is applied to the prior operating state logic network 2. The server identifies its network node information (such as the target position of each joint when adjusting the posture, the adjustment speed and other nodes, and the connection relationship information such as the collaborative relationship of each joint when adjusting the posture), performs feature abstraction, constructs a coding framework, and performs coding mapping, and finally obtains the network graph coding feature data corresponding to the prior operating state logic network 2.

[0058] According to the network graph encoding feature data corresponding to the two prior operation state logic networks, the server determines the operation state logic offset data between the two prior operation state logic networks. The server determines the global state encoding vector corresponding to the prior operation state logic network 1 based on the state encoding vectors corresponding to each state element in the network graph encoding feature data corresponding to the prior operation state logic network 1. For example, the global state encoding vector is obtained by calculating the state encoding vectors corresponding to the state elements such as the position, velocity, acceleration, etc. of each joint according to a specific calculation method (such as weighted summation, etc.). Similarly, the corresponding global state encoding vector is determined for the prior operation state logic network 2. Then the characteristic distance between the two global state encoding vectors is calculated, and this characteristic distance is the operation state logic offset data between the two prior operation state logic networks.

[0059] For the process of determining the second fault basis data based on the operating state logic offset between the target operating state logic network and the associated operating state logic network corresponding to the target operating state logic network, it is assumed that the target operating state logic network is the logic network for the robot arm to perform precise component installation, and its associated operating state logic network is the previously mentioned prior operating state logic network 3 (the installation tool is brought close to the component but precise installation has not yet begun).

[0060] The server performs graph encoding on the target operation state logic network. The server identifies the network node information in the target operation state logic network, which includes the precise position of each joint of the arm during precise installation, the precise torque applied, and other node information. The connection relationship information is the synergistic relationship between the precise movements of each joint. Then feature abstraction is performed. For example, for the precise position node of a joint during precise installation, an abstract feature label is defined according to its importance and role in the target operation state logic network, such as "critical node of precise installation position". Then, a coding framework is constructed, the top layer of which is the overall classification framework (such as position class, torque class, etc.), and the subclasses are subdivided below and the coding rules and positions corresponding to each node feature label are determined at the bottom layer. After that, coding mapping is performed to convert the nodes into coding values ​​according to the rules, and then the connection relationship coding is constructed according to the connection relationship. Finally, the network graph coding feature data corresponding to the target operation state logic network is integrated.

[0061] Similarly, the above operation is performed on the associated operating state logic network (prior operating state logic network 3) to obtain its corresponding network graph encoding feature data.

[0062] The second fault basis data is determined based on the operating state logic offset between the network diagram encoding feature data corresponding to the target operating state logic network and the associated operating state logic network. The server determines the global state encoding vector corresponding to the target operating state logic network based on the state encoding vectors corresponding to each state element in the network diagram encoding feature data corresponding to the target operating state logic network, for example, by calculating the state encoding vectors corresponding to the state elements such as the precise position and precise torque of each joint. The global state encoding vector corresponding to the associated operating state logic network (prior operating state logic network 3) is also determined. Finally, the characteristic distance between the global state encoding vector corresponding to the target operating state logic network and the global state encoding vector corresponding to the associated operating state logic network is calculated, and this characteristic distance is the second fault basis data. This second fault basis data can reflect the logical offset between the target operating state logic network and the associated operating state logic network, thereby providing a key basis for determining whether the servo drive has a fault warning activity when the target operating state logic network is triggered.

[0063] In a possible implementation, step A120 includes:

[0064] Step A121, based on the state coding vectors corresponding to each state element in the network graph coding feature data corresponding to each of the prior operation state logic networks, determine the global state coding vector corresponding to each of the prior operation state logic networks.

[0065] Step A122, calculating the characteristic distance between the global state encoding vectors corresponding to the two prior operation state logic networks respectively as the operation state logic offset data.

[0066] Step S132 includes:

[0067] Step S1321, based on the state coding vectors corresponding to the state elements in the network graph coding feature data corresponding to the target operation state logic network, determine the global state coding vector corresponding to the target operation state logic network.

[0068] Step S1322: determining a global state coding vector corresponding to the associated operation state logic network based on the state coding vectors corresponding to the state elements in the network graph coding feature data corresponding to the associated operation state logic network.

[0069] Step S1323, calculating a characteristic distance between a global state encoding vector corresponding to the target operating state logic network and a global state encoding vector corresponding to the associated operating state logic network as the second fault basis data.

[0070] For example, step A110 includes:

[0071] Step A111: Based on the deep learning model, identify the network node information and connection relationship information in the prior operating state logic network.

[0072] Step A112, based on the network node information, perform feature abstraction on each node in the prior operating state logic network, define a set of abstract feature labels for each type of node, and the feature labels are used to describe the role and status of the node in the entire prior operating state logic network, thereby generating an abstract node feature set.

[0073] Step A113, construct a coding framework based on the abstracted node feature set, the top layer of the coding framework is the overall classification framework for the entire node feature set, and each category in the overall classification framework is further subdivided into subcategories, the subcategories are divided according to the logical differences between the node features, and the bottom layer of the coding framework is the coding rules and positions corresponding to each specific node feature label.

[0074] Step A114, based on the constructed coding framework, encode and map each node in the network node information according to the corresponding abstract node features, so that the various attributes and functional characteristics of each node can be converted into corresponding coding values ​​according to the coding rules in the coding framework, and a preliminary encoded node set is obtained, which includes the encoded node information, and the coding value is used to reflect the characteristics of the node and the position of the node in the entire prior operating state logic network and its relationship with other nodes.

[0075] Step A115, based on the preliminary encoded node set, for each connection relationship in the node connection relationship information, construct corresponding encoding rules according to the encoding values ​​of the two connected nodes and the attributes of the connection itself to generate the corresponding connection relationship code, and generate the encoded connection relationship set.

[0076] Step A116, integrating the initially encoded node set and the encoded connection relationship set to generate the network graph encoding feature data.

[0077] In this embodiment, the determination of the operation state logic offset data between the prior operation state logic networks is carried out by taking the prior operation state logic network 1 (the robot arm moves from the initial position to the vicinity of the component to be installed) and the prior operation state logic network 2 (adjusting the arm posture to align with the installation position) as examples. First, the server determines the operation state logic offset data between the two prior operation state logic networks based on the network graph encoding feature data generated by the graph encoding representation of the two prior operation state logic networks.

[0078] When the prior operating state logic network is represented by graph encoding, taking the prior operating state logic network 1 as an example, the server needs to identify the network node information and connection relationship information in detail based on the deep learning model. The network node information contains many key elements, such as the specific position coordinates, velocity values, acceleration values, etc. of each joint of the robot arm in the process of moving from the initial position to the vicinity of the parts to be installed. These values ​​accurately describe the state of each joint at this stage. The connection relationship information reflects the synergistic relationship between the movements of each joint, which involves complex mechanical motion principles, such as how the movement of a joint affects the motion trajectory of other joints in terms of angle, speed, acceleration, etc. This is a complex system with multiple interrelated parameters.

[0079] Based on these network node information, the server performs feature abstraction on each node in the prior operation state logic network 1. For the starting position node of each joint, a set of abstract feature labels are defined for it according to its unique role and status in the entire prior operation state logic network 1, such as "initial position reference node-joint 1", which clearly indicates the starting position of a specific joint. This label accurately describes the role of the node in the entire logic network. For the speed node, it may be defined as a similar label such as "moving speed-joint 2", which clearly indicates which joint's speed it is and its significance in the entire network. Then, a coding framework is constructed based on these abstract node feature sets. The top level of this coding framework is the overall classification framework for the entire node feature set, such as "position class", "speed class", "acceleration class" and other main categories. Under the "position class", subcategories are further subdivided according to different joints, such as "joint 1 position", "joint 2 position", etc.; under the "speed class", it is also subdivided according to joints. The lowest level is the coding rules and positions corresponding to each specific node feature label. For example, for the label "initial position reference node - joint 1", its coding value is specified to be obtained according to a specific coordinate transformation formula, and it is clear that the position of this coding value in the entire coding system corresponds to the coding of other joint positions.

[0080] According to the constructed coding framework, each node in the network node information is coded and mapped according to the corresponding abstract node features. For example, for the starting position node of the robot arm joint 1, according to its corresponding coding rules, its actual coordinate value is converted into a coding value through coordinate transformation and specific mathematical operations. This coding value not only reflects the characteristics of the node, that is, the starting position of joint 1, but also reflects its position in the entire prior operation state logic network 1, such as the relative relationship with the starting positions of other joints, and the relationship with other nodes (such as speed nodes, acceleration nodes, etc.). Based on the node set after preliminary coding, for each connection relationship in the node connection relationship information, according to the coding values ​​of the two connected nodes and the properties of the connection itself, a corresponding coding rule is constructed to generate the corresponding connection relationship code. For example, if there is a certain correlation between the starting position node code value of joint 1 and the speed node code value of joint 2 (possibly based on the constraint relationship of mechanical kinematics), a coding rule is constructed according to this correlation to generate a specific connection relationship code. Finally, the node set after preliminary coding and the coded connection relationship set are integrated to generate the network graph coding feature data corresponding to the prior operation state logic network 1.

[0081] The same process is applied to the prior operating state logic network 2 to identify its network node information (such as the target position of each joint when adjusting the posture, the adjustment speed, etc.) and connection relationship information (the mutual influence relationship of each joint when adjusting the posture), perform feature abstraction, build a coding framework, and perform coding mapping operations to obtain the network graph coding feature data corresponding to the prior operating state logic network 2.

[0082] Based on the state coding vectors corresponding to each state element in the network graph coding feature data corresponding to each prior operation state logic network, the global state coding vector corresponding to each prior operation state logic network is determined. For the prior operation state logic network 1, the state coding vectors corresponding to each state element are extracted from its network graph coding feature data, such as the coding vectors corresponding to the state elements such as the position, speed, acceleration, etc. of each joint. Then, according to a specific calculation method, such as a weighted summation method, the global state coding vector corresponding to the prior operation state logic network 1 is determined. The weight of this weighted summation may be determined based on the importance of each state element in the entire prior operation state logic network 1. For example, the position information may be given a higher weight because the accuracy of the position has an important impact on subsequent operations during the movement of the robot arm. Similarly, the global state coding vector corresponding to the prior operation state logic network 2 is determined in the same way.

[0083] The characteristic distance between the global state encoding vectors corresponding to the two prior operation state logic networks is calculated as the operation state logic offset data. This characteristic distance can be calculated using a variety of mathematical methods, such as the Euclidean distance calculation method. By calculating the Euclidean distance between the global state encoding vectors corresponding to the prior operation state logic network 1 and the prior operation state logic network 2, a numerical value is obtained, which represents the operation state logic offset data between the two prior operation state logic networks. This offset data can reflect the degree of change in the logic state between the two processes from the robot arm moving to the vicinity of the component to be installed to adjusting the arm posture to align with the installation position. If this value is large, it means that there is a large difference in the logic state between the two processes, which may indicate that there are some potential problems in this conversion process; if the value is small, it means that the logic conversion between the two processes is relatively smooth and conforms to the normal operation logic.

[0084] For the process of determining the second fault basis data between the target operating state logic network (the robot arm performs precise component installation) and its associated operating state logic network (prior operating state logic network 3 - the installation tool is brought close to the component but precise installation has not yet begun), the server also needs to encode the operating state logic offset between the feature data based on their corresponding network diagrams.

[0085] When the server represents the target operating state logic network through graph encoding, it identifies the network node information, which includes key information such as the precise position of each arm joint during precise installation and the precise torque applied. The precise position information is accurate to millimeters or even smaller units to ensure the accurate installation of parts; the precise torque is set according to the assembly requirements of the parts, and different parts may require different torque values ​​to ensure the firmness of the installation. The connection relationship information is the coordinated relationship between the precise movements of each joint. For example, when applying torque, how the movements of each joint cooperate with each other to achieve precise installation actions.

[0086] Based on these network node information, feature abstraction is performed on each node in the target operation state logic network. For the precise position node of a joint during precise installation, an abstract feature label is defined according to its importance and role in the target operation state logic network, such as "critical node of precise installation position - joint 3", which clearly refers to the position node of a specific joint during precise installation. Then a coding framework is constructed, the top layer of which is the overall classification framework, such as "position class", "torque class", etc. Under the "position class", it is subdivided according to different joints, such as "joint 3 position", etc.; under the "torque class", it is also subdivided according to relevant factors. The bottom layer is the coding rules and positions corresponding to each specific node feature label. For example, for the label "critical node of precise installation position - joint 3", it is stipulated that its coding value is obtained according to specific coordinate accuracy and assembly requirements, and it is clear that the position of this coding value in the entire coding system corresponds to the coding of other related nodes.

[0087] According to the constructed coding framework, each node in the network node information is coded and mapped according to the corresponding abstract node features to obtain a preliminary coded node set. Then, for each connection relationship in the node connection relationship information, according to the coding values ​​of the two connected nodes and the properties of the connection itself, a corresponding coding rule is constructed to generate the corresponding connection relationship coding and generate the coded connection relationship set. Finally, the preliminary coded node set and the coded connection relationship set are integrated to obtain the network diagram coding feature data corresponding to the target operation state logic network.

[0088] Similarly, the above operation is performed on the associated operating state logic network (prior operating state logic network 3) to obtain its corresponding network graph encoding feature data.

[0089] Based on the state coding vectors corresponding to each state element in the network graph coding feature data corresponding to the target operation state logic network, the global state coding vector corresponding to the target operation state logic network is determined. For example, the state coding vectors corresponding to the state elements such as the precise position and precise torque of each joint are extracted from its network graph coding feature data, and then the global state coding vector corresponding to the target operation state logic network is determined according to a specific calculation method (such as weighted summation, and the weight is determined according to the importance of each state element in the precise installation process). Similarly, based on the state coding vectors corresponding to each state element in the network graph coding feature data corresponding to the associated operation state logic network, the global state coding vector corresponding to the associated operation state logic network is determined.

[0090] Calculate the characteristic distance between the global state encoding vector corresponding to the target operating state logic network and the global state encoding vector corresponding to the associated operating state logic network as the second fault basis data. This characteristic distance can also be calculated using the Euclidean distance calculation method. This second fault basis data can reflect the degree of change in the logic state from bringing the installation tool close to the component to the precise installation process. If this characteristic distance is large, it means that there may be a large difference in the logic state during this conversion process, which may indicate that there is a risk of failure in the upcoming precise installation process; if this characteristic distance is small, it means that the conversion process is relatively smooth and conforms to the normal operating logic, and the precise installation process may proceed smoothly.

[0091] By calculating and analyzing the operating state logic offset data between the prior operating state logic networks and the operating state logic offset data between the target operating state logic network and its associated operating state logic network, the server can more accurately monitor the operating status of the servo drive during the entire automotive parts assembly process and promptly discover potential failure risks, thereby ensuring the high production efficiency of the automobile manufacturing workshop and the stability of product quality.

[0092] In a possible implementation, step A120 may further include:

[0093] A first state element combination is constructed based on state elements in the same operating position in the network graph encoding feature data corresponding to the two prior operating state logic networks.

[0094] The operating state logic offset data is determined based on the encoding feature distances corresponding to each combination of the first state elements.

[0095] Step S132 may further include:

[0096] A second state element combination is constructed based on the state elements in the same operation position in the network graph encoding feature data corresponding to the target operation state logical network and the network graph encoding feature data corresponding to the associated operation state logical network.

[0097] The second fault basis data is determined based on the coding feature distances corresponding to each combination of the second state elements.

[0098] The encoding feature distance corresponding to the state element combination is the feature distance between the state encoding vectors corresponding to two state elements.

[0099] In this embodiment, the operating state logic offset data is determined based on the network graph encoding feature data corresponding to two prior operating state logic networks, and prior operating state logic network 1 (the robot arm moves from the initial position to the vicinity of the component to be installed) and prior operating state logic network 2 (adjusting the arm posture to align with the installation position) are used as examples to illustrate.

[0100] After the server performs graph coding on the two prior operation state logic networks and obtains the respective network graph coding feature data, it starts to determine the operation state logic offset data. The server first constructs a first state element combination based on the state elements in the same operation position in the network graph coding feature data corresponding to the two prior operation state logic networks. For example, in these two logic networks, the position information of the joint 1 of the robot arm is in the same logical operation position in their respective networks, so the state element corresponding to the position of joint 1 in the prior operation state logic network 1 and the state element corresponding to the position of joint 1 in the prior operation state logic network 2 are combined together to form a first state element combination; similarly, such combination operations are also performed on the state elements of the joint speed, acceleration, etc. that are in the same operation position in their respective networks.

[0101] Then, the operation state logic offset data is determined based on the coding feature distances corresponding to each first state element combination. For each first state element combination, two state elements correspond to their own state coding vectors, and the feature distance between the two state coding vectors is calculated. This feature distance is the coding feature distance corresponding to the first state element combination. For example, for the first state element combination at the joint 1 position, first obtain the state coding vector corresponding to the state element at the joint 1 position in the prior operation state logic network 1 and the state coding vector corresponding to the state element at the joint 1 position in the prior operation state logic network 2, and then use a suitable mathematical method (such as the Euclidean distance formula) to calculate the feature distance between the two vectors. The same operation is performed for other first state element combinations (such as combinations corresponding to the speed and acceleration of the joint). Finally, the operation state logic offset data is determined by comprehensively considering the coding feature distances corresponding to all first state element combinations. This operating state logic offset data can reflect the degree of state difference between the two prior operating state logic networks at the same operating position. If the coding feature distance corresponding to a certain first state element combination is large, it means that there is a large logic offset between the two logic networks in the aspect represented by this state element (such as the position of joint 1), which may indicate that an abnormality occurred in this aspect during the process from moving the robot arm to the vicinity of the component to be installed to adjusting the arm posture to align with the installation position. If the coding feature distances corresponding to each first state element combination are small, it means that the transition between the two logic networks in these key state elements is relatively smooth, which conforms to the normal operating logic.

[0102] For the process of determining the second fault basis data based on the operating state logic offset between the network graph encoding feature data corresponding to the target operating state logic network (the robot arm performs precise component installation) and the associated operating state logic network (prior operating state logic network 3 - the installation tool is brought close to the component but precise installation has not yet started), the server also follows similar steps.

[0103] The server constructs a second state element combination based on the state elements in the same operating position in the network diagram encoding feature data corresponding to the target operating state logic network and the network diagram encoding feature data corresponding to the associated operating state logic network. For example, the torque information of joint 3 during precise installation in the target operating state logic network and the torque-related state elements of joint 3 when close to the parts in the associated operating state logic network (although the torque requirements at this time are different from those during precise installation, they are in the same logical operating position related to the joint torque), these two state elements are combined together to form a second state element combination; similarly, such a combination operation is also performed on other state elements in the same operating position, such as the position of the joint.

[0104] Next, the second fault basis data is determined based on the coding feature distances corresponding to each second state element combination. For each second state element combination, the characteristic distance between the state coding vectors corresponding to two state elements is calculated, and this characteristic distance is the coding feature distance corresponding to the second state element combination. For example, for the second state element combination of the torque of joint 3, the state coding vector corresponding to the torque state element of joint 3 in the target operation state logic network and the state coding vector corresponding to the torque-related state element when joint 3 is close to the component in the associated operation state logic network are obtained, and then the characteristic distance between the two vectors is obtained by a specific mathematical calculation (such as Euclidean distance calculation). The same operation is performed for other second state element combinations (such as combinations corresponding to joint positions, etc.). Finally, the second fault basis data is determined by combining the coding feature distances corresponding to all second state element combinations. This second fault basis data can reflect the degree of logical change in the state elements at the same operating position from bringing the installation tool close to the component to precise installation. If the coding feature distance corresponding to a certain combination of second-state elements is large, it means that there is a large logical offset in the aspect represented by this state element (such as the torque of joint 3) during this conversion process, which may indicate that there is a risk of failure in the precise installation process; if the coding feature distances corresponding to each combination of second-state elements are small, it means that the conversion process is relatively smooth in terms of these key state elements, conforms to the normal operating logic, and the precise installation process may proceed smoothly.

[0105] During the entire automotive parts assembly process, through this precise logic offset analysis, the server can more carefully monitor the state changes of the servo drive during the logic network conversion process in different operating states, thereby more effectively discovering potential fault hazards and ensuring the efficiency of the automobile manufacturing process and the stability of product quality.

[0106] When performing these operations, the server needs to accurately process a large amount of data information to ensure that the encoding of each state element is accurate, and that the calculation of the state encoding vector and the characteristic distance conform to precise mathematical logic. At the same time, for different parts assembly tasks and different robot arm operations, the server can accurately analyze according to the preset logical network structure and encoding rules, which is crucial for the complex automated production process of automobile manufacturing workshops. Any small logical deviation, if ignored, may cause serious problems in the subsequent production process, such as inaccurate installation of parts and equipment damage. Therefore, the server ensures the smooth progress of the entire production process through this meticulous logical deviation analysis.

[0107] In a possible implementation, the step of determining the operating state logic offset between the two associated a priori operating state logic networks includes:

[0108] The logical knowledge representation model is used to perform logical knowledge representation on the two priori operating state logic networks respectively, and logical knowledge representation data corresponding to the two priori operating state logic networks are generated.

[0109] The operation state logic offset data between the two a priori operation state logic networks is determined according to the logic knowledge representation data respectively corresponding to the two a priori operation state logic networks.

[0110] Step S130 may further include:

[0111] The target operation state logic network and the associated operation state logic network are respectively represented by logic knowledge using the logic knowledge representation model to generate logic knowledge representation data corresponding to the target operation state logic network and the associated operation state logic network. The second fault basis data is determined based on the logic knowledge representation data corresponding to the target operation state logic network and the associated operation state logic network.

[0112] In a possible implementation, the step of determining the operating state logic offset between the two associated a priori operating state logic networks includes:

[0113] The operation state logic offset prediction model is used to determine the operation state logic offset data between the two prior operation state logic networks based on the two prior operation state logic networks. The operation state logic offset prediction model is used to determine the operation state logic offset data reflecting the offset degree between the two operation state logic networks for the two input operation state logic networks.

[0114] Step S130 may further include:

[0115] The second fault basis data is determined by using the operation state logic deviation prediction model based on the target operation state logic network and the associated operation state logic network.

[0116] In this embodiment, we first look at the case of using the logical knowledge representation model to determine the operation state logic offset between two associated prior operation state logic networks. Take prior operation state logic network 1 (the robot arm moves from the initial position to the vicinity of the component to be installed) and prior operation state logic network 2 (adjusting the arm posture to align with the installation position) as examples.

[0117] The server uses the logical knowledge representation model to perform logical knowledge representation on the prior operation state logic network 1 and the prior operation state logic network 2 respectively. For the prior operation state logic network 1, this model will analyze the various logical relationships contained therein. For example, in the process of the robot arm joints moving from the initial position to the vicinity of the parts to be installed, the logical relationship of the movement sequence of different joints, the time logical relationship of how long after a joint starts to move before other joints start to move in coordination, and the logical relationship between the joint movement speed and the overall movement trajectory. These logical relationships are converted into a specific logical knowledge representation data through the logical knowledge representation model. In this process, the model will identify the logical rules contained in the node information and connection relationship information in the network. For example, the change in the movement direction of joint 1 may have a specific logical association with the speed adjustment of joint 2, which is extracted by the model and converted into a part of the logical knowledge representation data. Similarly, for the prior operation state logic network 2, the model will also analyze the various logical relationships in the process of adjusting the arm posture to align with the installation position, such as the logical order of different joints in adjusting postures, the logical relationship between the posture adjustment amplitude and the final alignment accuracy, etc., and then generate the logical knowledge representation data corresponding to the prior operation state logic network 2.

[0118] Based on the logical knowledge representation data corresponding to the two prior operating state logic networks, the server determines the operating state logic offset data between the two prior operating state logic networks. In this process, the server will compare the logical elements in the two logical knowledge representation data. For example, compare the parts about the logical relationship of the joint movement sequence in the prior operating state logic network 1 and the prior operating state logic network 2. If in network 1, joint 1 moves before joint 2, and in network 2, joint 2 moves before joint 1, this indicates that there is a large logical offset in this logical element. For another example, for the logical relationship between the joint movement speed and the overall movement trajectory, there may be differences in the representation in the two networks, and the quantitative evaluation of this difference constitutes part of the operating state logic offset data. By comparing and quantitatively analyzing multiple such logical elements, comprehensive operating state logic offset data between the two prior operating state logic networks is finally obtained.

[0119] For the case where the second fault basis data is determined based on the operating state logic offset between the target operating state logic network (the robot arm performs precise component installation) and the associated operating state logic network corresponding to the target operating state logic network (prior operating state logic network 3 - the installation tool is brought close to the component but precise installation has not yet begun), the logical knowledge characterization model is also used.

[0120] The server uses the logical knowledge representation model to perform logical knowledge representation on the target operation state logic network and the associated operation state logic network respectively. For the target operation state logic network, the model will deeply analyze the various logical relationships in the process of precise component installation. For example, the logical order of torque applied by each joint during precise installation, the logical relationship between different torque values ​​and the degree of firmness of component installation, and the logical relationship between the precise position adjustment of each joint and the final installation accuracy. These logical relationships are converted into logical knowledge representation data corresponding to the target operation state logic network. For the associated operation state logic network, the model will also analyze the logical relationship in the process of bringing the installation tool close to the component, such as the logical adjustment of the joint movement speed during the approach process, the logical relationship between the relative position changes between the joints, etc., thereby generating the logical knowledge representation data corresponding to the associated operation state logic network.

[0121] Based on the logical knowledge representation data corresponding to the target operating state logic network and the associated operating state logic network, the server determines the second fault basis data. Similar to the previous example, the server will compare the logical elements in the two logical knowledge representation data. For example, compare the difference in the logical adjustment of the joint movement speed during precise installation and approaching the parts. If the joint movement speed follows a uniform deceleration logic when approaching the parts, and a complex, torque-related logical adjustment of the joint movement speed is required during precise installation, this logical difference will be reflected in the second fault basis data. By comparing and quantifying multiple logical elements, the accurate second fault basis data is determined to determine whether there is a risk of failure in the process from bringing the installation tool close to the parts to precise installation.

[0122] Now let’s look at the situation of using the operating state logic deviation prediction model to determine the operating state logic deviation.

[0123] For two associated prior operation state logic networks, such as prior operation state logic network 1 and prior operation state logic network 2, the server uses the operation state logic offset prediction model to determine the operation state logic offset data between them. This operation state logic offset prediction model is specifically used to determine the operation state logic offset data reflecting the offset degree between the two input operation state logic networks. When the prior operation state logic network 1 and the prior operation state logic network 2 are input, the model will analyze based on its internal algorithm and pre-trained parameters. For example, the model may consider factors such as the initial state parameters of each joint of the robot arm in network 1, the motion trajectory planning, and the target posture adjustment requirements in network 2. According to the relationship between these factors, a value that can reflect the offset degree between the two networks is calculated. This value is the operation state logic offset data. In this process, the operation state logic offset prediction model may comprehensively consider multiple factors, such as the dynamic factors of joint movement, the coupling relationship between the joints, etc., to accurately obtain the operation state logic offset data.

[0124] For the case where the second fault basis data is determined based on the operating state logic offset between the target operating state logic network and the associated operating state logic network corresponding to the target operating state logic network, the operating state logic offset prediction model is also used.

[0125] The server inputs the target operating state logic network and the associated operating state logic network into the operating state logic offset prediction model. For the target operating state logic network, the operating state logic offset prediction model considers various key parameters and logical relationships in the process of precise component installation, such as the precise position of each joint, torque value, movement speed, etc. For the associated operating state logic network, the operating state logic offset prediction model considers the relevant parameters and logical relationships in the process of bringing the installation tool close to the component. Then, the operating state logic offset prediction model analyzes the difference between the two networks based on its own algorithm and pre-trained parameters, and determines a value reflecting the degree of offset between them. This value is the second fault basis data. Through this value, the server can determine whether there is a risk of failure in the process from approaching the component to precise installation. This method of using the operating state logic offset prediction model can quickly and accurately derive the degree of offset between the logic networks, providing an effective basis for the server to monitor the operating status of the servo drive, helping to timely discover potential fault hazards and ensure the smooth production process of the automobile manufacturing workshop.

[0126] In a possible implementation, step S140 includes:

[0127] When the first fault basis data is less than the first threshold value, if the second fault basis data is not less than the second threshold value, it is determined that when the servo drive triggers the target operating state logic network, a target fault warning activity occurs within the triggering interval of the servo drive; if the second fault basis data is less than the second threshold value, it is determined that when the servo drive triggers the target operating state logic network, no target fault warning activity occurs within the triggering interval of the servo drive.

[0128] When the first fault basis data is not less than a first threshold value, a deviation between the second fault basis data and the first fault basis data is calculated.

[0129] If the deviation is not less than the third threshold value, it is determined that when the servo drive triggers the target operating status logic network, a target fault warning activity occurs within the triggering interval of the servo drive; if the deviation is less than the third threshold value, it is determined that when the servo drive triggers the target operating status logic network, no target fault warning activity occurs within the triggering interval of the servo drive.

[0130] In this embodiment, firstly, when analyzing whether a target fault warning activity occurs within the triggering interval of the servo drive when the servo drive triggers the target operating state logic network based on the first fault basis data and the second fault basis data, there are two different situations.

[0131] The first case is when the first fault basis data is less than the first threshold value. In the example of an automobile manufacturing workshop, it is assumed that the first threshold value is a value set based on a large amount of normal operating data and engineering experience. This value represents the upper limit of the logic deviation between the prior operating state logic network during relatively stable operation. For example, in the process of the robot arm moving from the initial position to the vicinity of the part to be installed, adjusting the arm posture to align with the installation position, and moving the installation tool close to the part (these processes correspond to the prior operating state logic network), after long-term statistics and analysis of normal operating data, it is found that the logic deviation between these processes is normal within a certain range, and the upper limit of this range is set as the first threshold value.

[0132] When the first fault basis data is less than the first threshold value, it means that in the conversion process between the prior operating state logic networks, the overall logic offset is in a relatively normal range. Now let's look at the relationship between the second fault basis data and the second threshold value. The second threshold value is also set based on a large amount of engineering data and experience. It represents a critical value of the logic offset between the target operating state logic network and the associated operating state logic network. This critical value distinguishes between normal and possible fault warning states. For example, in the process from bringing the installation tool close to the component (associated operating state logic network) to performing precise component installation (target operating state logic network), under normal circumstances, the logic offset between the two should be within a certain range.

[0133] If the second fault basis data is not less than the second threshold value at this time, this means that when converting from the relatively normal prior operating state logic network to the target operating state logic network, a large logical offset occurs between the target operating state logic network and the associated operating state logic network. In the scenario of automobile manufacturing, this may mean that when the robot arm switches from the state of being close to the parts to the state of accurately installing the parts, some key logical relationships have undergone abnormal changes. For example, there may be a large difference between the precise position of each joint during precise installation, the precise torque applied, etc. and the related logical relationships when approaching the parts. This difference exceeds the normal range, so the server determines that when the servo drive triggers the target operating state logic network, a target fault warning activity occurs within the triggering interval of the servo drive.

[0134] On the contrary, if the second fault basis data is less than the second threshold value, this indicates that in the conversion process from the previous prior operation state logic network to the target operation state logic network, although the first fault basis data shows that the prior operation state logic network is normal, the logic offset between the target operation state logic network and the associated operation state logic network is also within the normal range. For example, in the operation process of the robot arm, from approaching the parts to accurately installing the parts, the changes in the relationship between the motion logic, position, torque, etc. of each joint are within normal expectations, so the server determines that when the servo driver triggers the target operation state logic network, no target fault warning activity occurs within the triggering interval of the servo driver.

[0135] The second situation is when the first fault basis data is not less than the first threshold value. This means that a relatively large logic deviation has occurred in the conversion process between the prior operation state logic networks. For example, in the process of the robot arm moving from the initial position to the vicinity of the component to be installed, and then adjusting the arm posture to align with the installation position and bringing the installation tool close to the component, the changes in certain logical relationships have exceeded the normal range, and this situation beyond the normal range is reflected by the first fault basis data.

[0136] At this time, the server will calculate the deviation between the second fault basis data and the first fault basis data. The calculation of this deviation is to further analyze the changing trend of the logic offset during the conversion process from the prior operating state logic network to the target operating state logic network. For example, if the first fault basis data shows that the logic offset between the prior operating state logic network is already large, and the deviation between the second fault basis data and the first fault basis data is also large, this may mean that the problem of logic offset is constantly accumulating or a new and more serious logic offset situation has appeared throughout the process.

[0137] The third threshold value is also set based on engineering experience and a large amount of normal operation data. It is used to determine whether this deviation has reached the level where a fault warning may occur. If this deviation is not less than the third threshold value, it indicates that the change in logic offset has reached a dangerous level during the conversion process from the prior operation state logic network to the target operation state logic network. In the automotive manufacturing scenario, this may mean that during the operation of the robot arm, when the problem in the prior operation state logic network further develops to the target operation state logic network, a more serious disorder of logic relationships occurs. For example, in the process from bringing the installation tool close to the component to the precise installation of the component, since the joint motion logic in the prior operation state logic network has deviated, the precise position, torque, etc. of each joint during the precise installation have a greater deviation from the expected logical relationship, so the server determines that when the servo drive triggers the target operation state logic network, a target fault warning activity occurs within the triggering interval of the servo drive.

[0138] On the contrary, if the deviation is less than the third threshold value, it means that although there is a certain logic offset problem between the prior operation state logic network, the change of logic offset is still within an acceptable range when converting to the target operation state logic network. For example, during the operation of the robot arm, although some logic offsets occurred in the process of moving from the initial position to the vicinity of the component to be installed, adjusting the arm posture to align the installation position, and bringing the installation tool close to the component, these logic offsets did not further deteriorate to the extent that may cause failure in the process from bringing the installation tool close to the component to accurately installing the component, so the server determines that when the servo drive triggers the target operation state logic network, no target fault warning activity occurs within the triggering interval of the servo drive.

[0139] Through this analysis method based on the first fault basis data and the second fault basis data, combined with different threshold values, the server can accurately determine whether the servo drive has a target fault warning activity within the trigger interval when the target operating state logic network is triggered, so as to take corresponding measures in time, such as issuing an alarm to notify maintenance personnel to check the equipment, or adjusting the parameters of the servo drive to avoid possible failures, thereby ensuring the production efficiency and product quality of the automobile manufacturing workshop. This monitoring and analysis method is of vital importance in the automobile manufacturing industry, which has extremely high requirements for precision and reliability.

[0140] Figure 2 The hardware structure of the intelligent servo drive fault warning system 100 based on deep learning for implementing the intelligent servo drive fault warning method based on deep learning provided by an embodiment of the present invention is shown as follows: Figure 2 As shown, the deep learning-based intelligent servo drive fault warning system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0141] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the deep learning-based intelligent servo drive fault warning system 100 uses to execute or use to complete the exemplary method described in the present invention.

[0142] 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 intelligent servo drive fault warning method based on deep learning in 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.

[0143] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned deep learning-based intelligent servo drive fault warning system 100. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0144] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned intelligent servo drive fault warning method based on deep learning is implemented.

[0145] 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, figure or description thereof.

Claims

1. An intelligent servo drive fault warning method based on deep learning, characterized in that: The method comprises: Acquire a priori operating state logic network set corresponding to the target operating state logic network triggered by the servo drive; the priori operating state logic network set includes x priori operating state logic networks triggered by the servo drive before triggering the target operating state logic network, wherein x is a positive integer, and the node interval between the triggering node of the priori operating state logic network and the triggering node of the target operating state logic network meets the set interval requirement; Determining first fault basis data based on an operation state logic offset between every two logically associated a priori operation state logic networks in the a priori operation state logic network set; Determine the second fault basis data based on the operation state logic offset between the target operation state logic network and the associated operation state logic network corresponding to the target operation state logic network; the associated operation state logic network is an operation state logic network associated with the target operation state logic network and triggered by the servo drive before triggering the target operation state logic network; Analyzing, based on the first fault basis data and the second fault basis data, whether a target fault warning activity occurs within a triggering interval of the servo driver when the servo driver triggers the target operating state logic network; The step of determining the operation state logic offset between the two associated a priori operation state logic networks comprises: Based on the deep learning model, the two prior operation state logic networks are respectively represented by graph coding, and network graph coding feature data corresponding to the two prior operation state logic networks are generated; Determine the operation state logic offset data between the two a priori operation state logic networks according to the network graph coding feature data respectively corresponding to the two a priori operation state logic networks; The determining of the second fault basis data based on the operation state logic offset between the target operation state logic network and the associated operation state logic network corresponding to the target operation state logic network includes: Performing graph coding representation on the target operation state logic network and the associated operation state logic network respectively, and generating network graph coding feature data corresponding to the target operation state logic network and the associated operation state logic network respectively; The second fault basis data is determined according to the operating state logic offset between the network diagram encoding feature data respectively corresponding to the target operating state logic network and the associated operating state logic network.

2. The intelligent servo drive fault early warning method based on deep learning according to claim 1 is characterized in that: The determining of the first fault basis data based on the operation state logic offset between each two logically associated a priori operation state logic networks in the a priori operation state logic network set includes: According to the x a priori operating state logic networks in the a priori operating state logic network set, x-1 a priori operating state logic network combinations are constructed; each of the a priori operating state logic network combinations includes two a priori operating state logic networks in which trigger nodes are associated; For each of the a priori operating state logic network combinations, based on the operating state logic offset between two of the a priori operating state logic networks included in the a priori operating state logic network combination, determining the operating state logic offset data corresponding to the a priori operating state logic network combination; The average offset result of the operating state logic offset data respectively corresponding to the x-1 prior operating state logic network combinations is calculated as the first fault basis data.

3. The intelligent servo drive fault early warning method based on deep learning according to claim 1 is characterized in that: The determining of the operation state logic offset data between the two a priori operation state logic networks according to the network graph encoding feature data respectively corresponding to the two a priori operation state logic networks comprises: Determine a global state encoding vector corresponding to each of the a priori operation state logic networks based on the state encoding vectors corresponding to each state element in the network graph encoding feature data corresponding to each of the a priori operation state logic networks; Calculating the characteristic distance between the global state encoding vectors respectively corresponding to the two prior operation state logic networks as the operation state logic offset data; The determining the second fault basis data according to the operation state logic offset between the network diagram encoding feature data respectively corresponding to the target operation state logic network and the associated operation state logic network includes: Determine a global state encoding vector corresponding to the target operation state logic network based on the state encoding vectors corresponding to each state element in the network graph encoding feature data corresponding to the target operation state logic network; Determine a global state encoding vector corresponding to the associated operation state logic network based on the state encoding vectors corresponding to each state element in the network graph encoding feature data corresponding to the associated operation state logic network; Calculating a characteristic distance between a global state encoding vector corresponding to the target operation state logic network and a global state encoding vector corresponding to the associated operation state logic network as the second fault basis data; The step of performing graph coding representation on the two prior operation state logic networks based on the deep learning model and generating network graph coding feature data corresponding to the two prior operation state logic networks respectively comprises: Based on the deep learning model, identifying the network node information and connection relationship information in the prior operation state logic network; Based on the network node information, feature abstraction is performed on each node in the prior operation state logic network, and a set of abstract feature labels is defined for each type of node, wherein the feature labels are used to describe the role and status of the node in the entire prior operation state logic network, and an abstract node feature set is generated; A coding framework is constructed based on the abstracted node feature set. The top layer of the coding framework is the overall classification framework for the entire node feature set, and each category in the overall classification framework is further subdivided into subcategories. The subcategories are divided according to the logical differences between the node features. The bottom layer of the coding framework is the coding rules and positions corresponding to each specific node feature label. According to the constructed coding framework, each node in the network node information is coded and mapped according to the corresponding abstract node features, so that various attributes and functional features of each node can be converted into corresponding coding values ​​according to the coding rules in the coding framework, and a preliminary coded node set is obtained, the node set includes the coded node information, and the coding value is used to reflect the characteristics of the node and the position of the node in the entire prior operation state logic network and the relationship with other nodes; Based on the node set after the preliminary coding, for each connection relationship in the node connection relationship information, according to the coding values ​​of the two connected nodes and the attributes of the connection itself, a corresponding coding rule is constructed to generate a corresponding connection relationship code, thereby generating a coded connection relationship set; The preliminary encoded node set and the encoded connection relationship set are integrated to generate the network graph encoding feature data.

4. The intelligent servo drive fault early warning method based on deep learning according to claim 1 is characterized in that: The determining of the operation state logic offset data between the two a priori operation state logic networks according to the network graph encoding feature data respectively corresponding to the two a priori operation state logic networks comprises: Constructing a first state element combination according to the state elements in the same operation position in the network diagram encoding feature data corresponding to the two prior operation state logic networks respectively; Determine the operation state logic offset data based on the encoding feature distances corresponding to the respective combinations of the first state elements; The determining the second fault basis data according to the operation state logic offset between the network diagram encoding feature data respectively corresponding to the target operation state logic network and the associated operation state logic network includes: Constructing a second state element combination according to the state elements in the same operation position in the network graph encoding feature data corresponding to the target operation state logical network and the network graph encoding feature data corresponding to the associated operation state logical network; Determining the second fault basis data based on the coding feature distances corresponding to each combination of the second state elements; The encoding feature distance corresponding to the state element combination is the feature distance between the state encoding vectors corresponding to two state elements.

5. The intelligent servo drive fault early warning method based on deep learning according to claim 1 or 2, characterized in that: The step of analyzing, based on the first fault basis data and the second fault basis data, whether a target fault warning activity occurs within a triggering interval of the servo driver when the servo driver triggers the target operating state logic network comprises: When the first fault basis data is less than the first threshold value, if the second fault basis data is not less than the second threshold value, it is determined that when the servo drive triggers the target operating state logic network, a target fault warning activity occurs within the triggering interval of the servo drive; if the second fault basis data is less than the second threshold value, it is determined that when the servo drive triggers the target operating state logic network, no target fault warning activity occurs within the triggering interval of the servo drive.

6. The intelligent servo drive fault early warning method based on deep learning according to claim 1 or 2, characterized in that: The step of analyzing, based on the first fault basis data and the second fault basis data, whether a target fault warning activity occurs within a triggering interval of the servo driver when the servo driver triggers the target operating state logic network comprises: When the first fault basis data is not less than a first threshold value, calculating a deviation between the second fault basis data and the first fault basis data; If the deviation is not less than the third threshold value, it is determined that when the servo drive triggers the target operating status logic network, a target fault warning activity occurs within the triggering interval of the servo drive; if the deviation is less than the third threshold value, it is determined that when the servo drive triggers the target operating status logic network, no target fault warning activity occurs within the triggering interval of the servo drive.

7. An intelligent servo drive fault warning system based on deep learning, characterized in that: The deep learning-based intelligent servo drive fault warning 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 deep learning-based intelligent servo drive fault warning method described in any one of claims 1 to 6.

Images