Intelligent manufacturing dynamic evolution method based on digital twin architecture

Abstract

The invention relates to an intelligent manufacturing dynamic evolution method based on a digital twinning architecture, belongs to computer aided design, and more particularly relates to the field of digital twinning, and the method comprises the steps: constructing a digital twinning model of a multi-joint robot of a target model; and an AI dynamic evolution model is adopted to intelligently predict the operation fault type of the digital twinborn model in the current time segment according to the sensor data and the visual content of each previous moment before the current moment acquired from the digital twinborn model. Through the method, the digital twin model of the multi-joint robot of the target model can be constructed aiming at the technical problem that the fault type occurring in the use process of the specific intelligent manufacturing product cannot be predicted and handled; an artificial intelligence model designed by a customized structure is adopted to complete intelligent prediction of the operation fault type of a future digital twin model according to the multi-modal data selected in a targeted manner, and then an optimal maintenance strategy is designed, so that the technical problem is solved.

Description

Technical Field

[0001] The computer-aided design of this invention relates more specifically to the field of digital twins, and particularly to a dynamic evolution method for intelligent manufacturing based on a digital twin architecture. Background Technology

[0002] Digital twins are an important category in computer-aided design. Generally, a digital twin fully utilizes data from physical models, sensors, and operational history to integrate multi-disciplinary, multi-physical-quantity, multi-scale, and multi-probabilistic simulation processes, completing a mapping in virtual space to reflect the entire lifecycle of the corresponding physical equipment. A digital twin is a concept that transcends reality; it can be viewed as a digital mapping system of one or more important, interdependent equipment systems.

[0003] Meanwhile, digital twins are a universally applicable theoretical and technological system that can be applied in numerous fields, including product design, product manufacturing, medical analysis, and engineering construction. In China, its most extensive application is in engineering construction, while the field of intelligent manufacturing receives the most attention and research, such as the intelligent manufacturing of multi-joint robots. The aim is to simulate the dynamic evolution of intelligent manufacturing products during their production or use through digital twins, providing valuable reference information for the actual production or use of intelligent manufacturing products.

[0004] For example, Chinese invention patent publication CN116579164A proposes a digital twin intelligent manufacturing system, relating to the field of intelligent manufacturing technology. It solves the technical problem that existing technologies cannot fully guarantee the consistency between digital twin models and actual scenarios, affecting enterprise production processes. The digital twin intelligent manufacturing system combines historical production information to determine whether several simulated objects have suffered damage. If so, it updates several simulated entities in the digital twin model according to a damage mapping model stored in the database. The system can detect the consistency between virtual and reality in real time and update the digital twin model promptly based on the detection results, thereby improving enterprise production processes. When simulated objects suffer damage, the system re-collects the basic parameters of the simulated objects and establishes a correspondence between damage and basic parameters. A damage mapping model is constructed based on this correspondence. The system can obtain corresponding adjustment information based on the damage to several simulated objects, enabling rapid updates to the digital twin model.

[0005] For example, Chinese invention patent publication CN118363354A proposes a digital twin-based intelligent manufacturing optimization system. The system includes: a physical object data acquisition device for collecting physical production line data, including twin construction data, historical production line data, and production plans; a digital twin production line construction module for constructing a digital twin production line based on the twin construction data; a simulation analysis module for performing simulation analysis and optimization of the production plan using the digital twin production line and historical production line data to obtain the optimal manufacturing plan; and a feedback adjustment module for outputting the optimal manufacturing plan to adjust the production process of the physical production line. This digital twin-based intelligent manufacturing optimization system can evaluate the optimal production plan without interfering with the operation of the physical production line, reducing the probability of malfunctions during physical production line operation, improving production efficiency, and reducing trial-and-error costs and risks.

[0006] It is evident that existing digital twin-based intelligent manufacturing solutions cannot reliably predict the types of operational failures that will occur during the future use of specific multi-joint robots. Consequently, they cannot pre-select the optimal maintenance strategy for the types of operational failures that will occur during the future use of multi-joint robots, thus delaying the occurrence of these failures for a longer period. Consequently, the ability of specific intelligent manufacturing products to cope with future failures cannot be effectively improved in actual use. Summary of the Invention

[0007] To address technical issues in related fields, the present invention aims to provide a dynamic evolution method for intelligent manufacturing based on a digital twin architecture. This method constructs a digital twin model of a multi-joint robot of a target model using a co-simulation platform. A customized artificial intelligence model, based on selectively chosen multimodal data, intelligently predicts the types of operational faults occurring within the digital twin model in the current time segment. Furthermore, when the type number corresponding to the operational fault type in the current time segment output by the AI ​​dynamic evolution model is a non-empty character, the digital twin model automatically triggers a maintenance simulation at the current moment to implement the maintenance strategy. The method evaluates the delay time of different maintenance strategies adopted at the current moment for the intelligently predicted operational fault type, obtaining the delay time corresponding to each different maintenance strategy. The maintenance strategy corresponding to the largest delay time is taken as the optimal maintenance strategy for the current moment. This achieves the dynamic evolution of the usage process of a specific intelligent manufacturing product based on a digital twin architecture while delaying the occurrence time of future faults during the use of the specific intelligent manufacturing product as much as possible.

[0008] According to the present invention, a method for dynamic evolution of intelligent manufacturing based on a digital twin architecture is provided, the method comprising:

[0009] A digital twin model of the target model of the multi-joint robot is constructed. The digital twin model integrates the mapping relationship between the geometric structure, dynamic parameters and control logic of the target model of the multi-joint robot and the sensor output parameters.

[0010] Sensor data from previous moments before the current moment is collected from the digital twin model performing the set task. The sensor data at each moment consists of the output parameters of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor collected from the digital twin model at that moment.

[0011] The visualization content of each past moment before the current moment is collected from the digital twin model performing the set task. The visualization content of each moment is the depth value, brightness value and coordinate value of each pixel of the sub-frame occupied by the multi-joint robot in the presentation image taken by the visual sensor directly above the multi-joint robot at that moment from a bird's-eye view.

[0012] The convolutional neural network is trained in each iteration to obtain the completed convolutional neural network as a dynamic evolution model for AI.

[0013] The AI ​​dynamic evolution model is used to intelligently predict the type of operational failure that will occur in the digital twin model in the current time segment based on the model code value of the target model, the task type number of the set task, the preset time length, the sensor data and visualization content collected from the digital twin model that executes the set task at each previous time before the current time, and the duration of the current time segment.

[0014] Joint simulation platform Attached Figure Description

[0015] The embodiments of the present invention will now be described with reference to the accompanying drawings, wherein:

[0016] Figure 1 This is a schematic diagram of the working scenario of the intelligent manufacturing dynamic evolution method based on digital twin architecture according to the present invention.

[0017] Figure 2 This is a flowchart illustrating the steps of a smart manufacturing dynamic evolution method based on a digital twin architecture according to Embodiment 1 of the present invention.

[0018] Figure 3 This is a flowchart illustrating the steps of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to Embodiment 2 of the present invention.

[0019] Figure 4 This is a flowchart illustrating the steps of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to Embodiment 3 of the present invention.

[0020] Figure 5 This is a flowchart illustrating the steps of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to Embodiment 4 of the present invention.

[0021] Figure 6 This is a flowchart illustrating the steps of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to Embodiment 5 of the present invention. Detailed Implementation

[0022] like Figure 1 The diagram illustrates a working scenario of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to the present invention. The computer-aided design of this invention relates more specifically to the field of digital twins.

[0023] The specific technical process of this invention is as follows:

[0024] Technical Process 1: A digital twin model of the target multi-joint robot is constructed using a co-simulation platform. This digital twin model integrates the mapping relationship between the target robot's geometry, dynamic parameters, and control logic and the sensor output parameters. Figure 1 As shown;

[0025] Obviously, the present invention can be used for various types of multi-joint robots, as well as for fixed types of multi-joint robots. The present invention can be used for fixed types of multi-joint robots that perform various tasks.

[0026] Technical Process Two: To intelligently predict the types of operational failures occurring in the digital twin model within the current time segment, a custom-designed AI dynamic evolution model is introduced as the artificial intelligence model, such as... Figure 1 As shown;

[0027] Specifically, the structural customization of the AI ​​dynamic evolution model designed for the digital twin model of the target multi-joint robot is mainly reflected in the following aspects:

[0028] First: The AI ​​dynamic evolution model is a convolutional neural network after each training iteration. The number of pooling layers used in the convolutional neural network architecture is positively correlated with the number of tasks that the digital twin model can perform.

[0029] For example, if a digital twin model can perform 5 types of tasks, the corresponding convolutional neural network uses 3 pooling layers; if a digital twin model can perform 10 types of tasks, the corresponding convolutional neural network uses 4 pooling layers; if a digital twin model can perform 15 types of tasks, the corresponding convolutional neural network uses 5 pooling layers; if a digital twin model can perform 20 types of tasks, the corresponding convolutional neural network uses 6 pooling layers, and so on.

[0030] Second: The number of training operations performed on the convolutional neural network is directly proportional to the total number of joints in the target model of the multi-joint robot;

[0031] For example, the target model of the multi-joint robot has a total of 3 joints and performs 600 training sessions on the convolutional neural network; the target model of the multi-joint robot has a total of 4 joints and performs 800 training sessions on the convolutional neural network; the target model of the multi-joint robot has a total of 5 joints and performs 1000 training sessions on the convolutional neural network; the target model of the multi-joint robot has a total of 6 joints and performs 1200 training sessions on the convolutional neural network, and so on.

[0032] Third: In each training iteration of the convolutional neural network, the type number corresponding to the operational failure type of the digital twin model within a known past time segment is used as the single output of the AI ​​dynamic evolution model. The model code value of the target model, the task type number of the set task, the preset time length, the sensor data and visualization content collected from the digital twin model executing the set task at each past time segment before the specified past time segment, and the duration of the specified past time segment are used as multiple inputs of the AI ​​dynamic evolution model to complete the training, thereby ensuring the training effect of each training iteration of the convolutional neural network.

[0033] In this way, through the customized structural design of the above aspects, different customized AI dynamic evolution models are designed for different digital twin models, ensuring the stability and effectiveness of intelligent prediction results;

[0034] Technical Process 3: To enable intelligent prediction of the types of operational failures occurring in the digital twin model within the current time segment, several basic data sets were introduced;

[0035] Specifically, the multiple basic data include the model code value of the target model, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model that executes the set task at each past moment before the current moment, and the duration of the current time segment.

[0036] like Figure 1 As shown, the multiple basic data include multi-sensor parameters, customized visual information, and various auxiliary data. The multi-sensor parameters are sensor data collected from the digital twin model performing the set task at each past moment before the current moment. The customized visual information is the visualization content collected from the digital twin model performing the set task at each past moment before the current moment. The various auxiliary data are the model code value of the target model, the task type number of the set task, the preset time length, and the duration of the current time segment.

[0037] More specifically, sensor data from each past moment before the current moment is collected from the digital twin model performing the set task. The sensor data at each moment consists of the output parameters of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor collected from the digital twin model at that moment. Visualization content from each past moment before the current moment is also collected from the digital twin model performing the set task. The visualization content at each moment consists of the depth, brightness, and coordinate values ​​of each pixel in the sub-frame occupied by the multi-joint robot in the view captured by a visual sensor simulating a viewpoint directly above the multi-joint robot at that moment, taken from a bird's-eye view. This provides a tailored data structure for multiple fundamental data points used for intelligent prediction of future operational failure types.

[0038] More specifically, an articulated robot, also known as an articulated arm robot, articulated mechanical arm, or multi-joint robot, is a common form of industrial robot. It belongs to the articulated robot type and is mainly used for industrial automation operations such as automatic assembly, painting, handling, and welding. It is also applied in industrial scenarios such as intelligent warehousing and quality inspection.

[0039] This robot mimics the structure of a human arm, typically consisting of 1-6 joint axes, each containing rotational and lateral degrees of freedom: 1-2 axes form the lower arm, and 3-6 axes form the upper arm. Based on their construction, they can be categorized into five-axis / six-axis articulated robots, tray-mounted robots, and planar articulated robots (SCARA). Six-axis articulated robots are the mainstream type, achieving flexible movements through joint rotation, possessing the ability to navigate obstacles and perform special motions, and relying on computers to solve motion control analysis and synthesis problems.

[0040] In this way, the targeted selection of the above-mentioned basic data further ensures the stability and effectiveness of the intelligent prediction results;

[0041] Technical Process Four: Using the AI ​​dynamic evolution model customized for the structural design of the digital twin model of the target multi-joint robot based on Technical Process Two, and according to multiple basic data selected specifically in Technical Process Three, intelligent prediction of the types of operational faults that will occur in the digital twin model within the current time segment is completed, such as... Figure 1 As shown;

[0042] Specifically, the result obtained by intelligent prediction is the type number corresponding to the type of operational failure that occurred in the digital twin model within the current time segment, for example, in binary numerical form;

[0043] Technical Process 5: Based on the types of operational faults occurring in the digital twin model within the current time segment obtained from Technical Process 4, analyze the corresponding optimal maintenance strategies, such as... Figure 1 As shown;

[0044] Specifically, when the type number corresponding to the operational fault type obtained by intelligent prediction is a non-empty character, the digital twin model is automatically triggered to simulate maintenance at the current moment to implement the maintenance strategy at the current moment, and evaluate the delay time of different maintenance strategies adopted at the current moment for the intelligently predicted operational fault type. The delay time corresponding to each of the different maintenance strategies adopted at the current moment is obtained, and the maintenance strategy corresponding to the delay time with the largest value among the delay time is taken as the best maintenance strategy at the current moment. Thus, the best maintenance strategy is automatically selected to extend the expected operational fault type as much as possible and improve the operational reliability of the digital twin model and even its corresponding multi-joint robot as much as possible.

[0045] In this way, the best maintenance strategy for dealing with the types of operational failures in future time segments can be configured in advance before the physical multi-joint robot performs its tasks, thereby improving the ability to deal with future failures of smart manufacturing products.

[0046] Therefore, through the coordinated operation of the above five technical processes, the present invention can use an artificial intelligence model to intelligently predict the types of operational faults that occur in the digital twin model within the current time segment based on sensor data, visualization content, and other auxiliary data from previous times before the current time. This enables the dynamic evolution processing of multi-joint robots in intelligent manufacturing based on the digital twin architecture, providing valuable reference information for proactively addressing various types of operational faults in the future.

[0047] The key points of this invention are: dynamic evolution processing of multi-joint robots for intelligent manufacturing based on digital twin architecture, advance configuration of optimal maintenance strategies to cope with operational failure types in future time segments, design of AI dynamic evolution models with different customized structures for different digital twin models, and targeted selection of multiple basic data including multi-sensor parameters and customized visual information.

[0048] The present invention will now be described in detail through examples of the intelligent manufacturing dynamic evolution method based on digital twin architecture.

[0049] Example 1

[0050] Figure 2 This is a flowchart illustrating the steps of a smart manufacturing dynamic evolution method based on a digital twin architecture according to Embodiment 1 of the present invention.

[0051] like Figure 2 As shown, the intelligent manufacturing dynamic evolution method based on digital twin architecture includes the following specific steps:

[0052] Step S21: Construct a digital twin model of the target model of the multi-joint robot. The digital twin model integrates the mapping relationship between the geometric structure, dynamic parameters and control logic of the target model of the multi-joint robot and the sensor output parameters. The dynamic parameters of the multi-joint robot include joint torque, joint displacement, joint velocity, joint acceleration and inertial parameters.

[0053] Specifically, a digital twin model of the target model of the multi-joint robot is constructed using a co-simulation platform. Obviously, this invention can be used for various models of multi-joint robots, as well as for multi-joint robots of a fixed model.

[0054] Step S22: Collect sensor data from the digital twin model performing the set task at each previous moment before the current moment. The sensor data at each moment is the output parameters of the joint torque sensor, motor current sensor, infrared temperature sensor and microelectromechanical system vibration sensor collected from the digital twin model at that moment.

[0055] Step S23: Collect visualization content from the digital twin model performing the set task, including the visualization content of each past moment before the current moment. The visualization content of each moment is the depth value, brightness value and coordinate value of each pixel in the sub-frame of the multi-joint robot in the presentation image taken by the visual sensor directly above the multi-joint robot at the time of the presentation from a bird's-eye view.

[0056] Specifically, the visualization content at each moment is the depth value, brightness value, and coordinate value of each pixel in the sub-frame occupied by the multi-joint robot in the presentation image captured by the visual sensor directly above the multi-joint robot at that moment from a bird's-eye view. It also includes the coordinate value of each pixel as the vertical coordinate value and the horizontal coordinate value of the pixel.

[0057] Step S24: Perform training on the convolutional neural network for each iteration to obtain the convolutional neural network after each training iteration as an AI dynamic evolution model;

[0058] Specifically, the convolutional neural network architecture used contains an input layer, a convolutional layer, a pooling layer, and an output layer;

[0059] Step S25: Using an AI dynamic evolution model, based on the target model's model code value, the task type number of the set task, the preset time length, the sensor data and visualization content collected from the digital twin model executing the set task at each past moment before the current moment, and the duration of the current time segment, the model can intelligently predict the type of operational failure that will occur in the digital twin model in the current time segment.

[0060] Specifically, the operational fault types of the digital twin model that occur within the current time segment obtained by intelligent prediction are numerically represented by the type number corresponding to the operational fault type, and the type number can be a binary value;

[0061] Among them, the type of operational failure that occurs in the digital twin model within the current time segment predicted by the intelligent system is one of the following: joint wear, reducer failure, motor overheating, structural fatigue, lubrication failure, and control deviation accumulation.

[0062] Correspondingly, the fault numbers corresponding to different types of operational faults are represented by different binary values;

[0063] Among them, each past moment before the current moment and the current moment together form a complete time interval on the time axis, and the time length between each two adjacent moments is equal to the preset time length. The current time segment starts from the current moment and the duration of the current time segment is an integer multiple of the preset time length.

[0064] For example, if the current time is 9:00 AM, and the current time segment is from 9:00 AM to 9:20 AM, then the duration of the current time segment is 20 minutes. The previous times before the current time are 8:58 AM, 8:56 AM, 8:54 AM, 8:52 AM, 8:50 AM, 8:48 AM, 8:46 AM, 8:44 AM, 8:42 AM, and 8:40 AM, a total of 10 times. The preset duration of the current time segment is 2 minutes. The duration of the current time segment is 10 times the preset duration.

[0065] In this case, the distance from the imaging lens of the simulated visual sensor directly above the multi-joint robot to the top of the multi-joint robot is equal to a preset distance.

[0066] The visualization content at each moment is the depth value, brightness value, and coordinate value of each pixel in the sub-frame occupied by the multi-joint robot in the presentation image captured by the visual sensor directly above the multi-joint robot at that moment from a bird's-eye view. This includes: identifying the multi-joint robot occupying the sub-frame in the presentation image captured by the bird's-eye view at that moment based on the imaging pattern of the multi-joint robot in its factory state from a bird's-eye view.

[0067] Specifically, the identification of the sub-frame occupied by the multi-joint robot in the presentation image taken from the bird's-eye view at the time based on the imaging pattern of the multi-joint robot in the factory state includes: taking the sub-frame with the most pixels among the multiple sub-frames that match the imaging pattern of the multi-joint robot in the factory state in the presentation image as the sub-frame occupied by the multi-joint robot in the presentation image taken from the bird's-eye view at the time.

[0068] Among them, performing training on the convolutional neural network to obtain the convolutional neural network after each training as an AI dynamic evolution model includes: the number of pooling layers used by the convolutional neural network is positively correlated with the number of tasks that the digital twin model can perform, and the number of training operations performed on the convolutional neural network is proportional to the total number of joints of the target model of the multi-joint robot, thereby completing the update of the model parameters of the AI ​​dynamic evolution model.

[0069] For example, the positive correlation between the number of pooling layers used by a convolutional neural network and the number of tasks a digital twin model can perform includes: if a digital twin model can perform 5 tasks, the corresponding convolutional neural network uses 3 pooling layers; if a digital twin model can perform 10 tasks, the corresponding convolutional neural network uses 4 pooling layers; if a digital twin model can perform 15 tasks, the corresponding convolutional neural network uses 5 pooling layers; if a digital twin model can perform 20 tasks, the corresponding convolutional neural network uses 6 pooling layers, and so on.

[0070] For example, the number of training iterations performed on the convolutional neural network is proportional to the total number of joints of the target model of the multi-joint robot, including: the target model of the multi-joint robot has a total of 3 joints and the number of training iterations of the convolutional neural network is 600; the target model of the multi-joint robot has a total of 4 joints and the number of training iterations of the convolutional neural network is 800; the target model of the multi-joint robot has a total of 5 joints and the number of training iterations of the convolutional neural network is 1000; the target model of the multi-joint robot has a total of 6 joints and the number of training iterations of the convolutional neural network is 1200, and so on.

[0071] The process of training the convolutional neural network (CNN) multiple times to obtain the CNN after each training iteration as the AI ​​dynamic evolution model also includes: in each training iteration of the CNN, using the type number corresponding to the type of operational failure that occurred in the digital twin model within a known past time segment as the single output of the AI ​​dynamic evolution model, and using the model code value of the target model, the task type number of the set task, the preset time length, the sensor data and visualization content collected from the digital twin model executing the set task at each past time segment before the specified past time segment, and the duration of the specified past time segment as multiple inputs of the AI ​​dynamic evolution model to complete the training.

[0072] Example 2

[0073] Figure 3 This is a flowchart illustrating the steps of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to Embodiment 2 of the present invention.

[0074] like Figure 3 As shown, with Figure 2 The only difference in the embodiments is that, in the intelligent manufacturing dynamic evolution method based on digital twin architecture, sensor data from each past moment before the current moment is collected from the digital twin model performing the set task. The sensor data at each moment is before the output parameters of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor collected from the digital twin model at that moment, i.e., before step S22. The method further includes:

[0075] Step S26: Adjust the sampling frequency of the joint torque sensor, motor current sensor, infrared temperature sensor and microelectromechanical system vibration sensor to achieve synchronous acquisition of the output parameters of each of the joint torque sensor, motor current sensor, infrared temperature sensor and microelectromechanical system vibration sensor;

[0076] For example, adjusting the sampling frequency of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor to achieve synchronous acquisition of their respective output parameters includes: performing frequency reduction or frequency increase processing on the sampling frequency of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor to achieve synchronous acquisition of their respective output parameters;

[0077] The process of adjusting the sampling frequency of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor to achieve synchronous acquisition of their respective output parameters includes: using a rectangular waveform of a preset frequency emitted by a quartz oscillator to provide a reference clock signal for the synchronous acquisition of the output parameters of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor.

[0078] Example 3

[0079] Figure 4 This is a flowchart illustrating the steps of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to Embodiment 3 of the present invention.

[0080] like Figure 4 As shown, with Figure 2 The only difference in the embodiments is that, in the intelligent manufacturing dynamic evolution method based on digital twin architecture, after using an AI dynamic evolution model to intelligently predict the type of operational failure that occurs in the digital twin model in the current time segment based on the model code value of the target model, the task type number of the set task, the preset time length, the sensor data and visualization content collected from the digital twin model executing the set task at each past moment before the current moment, and the duration of the current time segment, that is, after step S25, the method further includes:

[0081] Step S27: When the type number corresponding to the type of operational failure that occurred in the digital twin model within the current time segment output by the AI ​​dynamic evolution model is a non-empty character, automatically trigger the digital twin model to simulate maintenance at the current moment to implement the maintenance strategy at the current moment, and evaluate the delay time of different maintenance strategies adopted at the current moment for the intelligently predicted operational failure type. When the type number corresponding to the type of operational failure that occurred in the digital twin model within the current time segment output by the AI ​​dynamic evolution model is an empty character, temporarily suspend the automatic triggering of the digital twin model to simulate maintenance at the current moment.

[0082] Specifically, the types of operational failures that occur in the digital twin model within the current time segment output by the AI ​​dynamic evolution model may be one or more types.

[0083] Example 4

[0084] Figure 5 This is a flowchart illustrating the steps of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to Embodiment 4 of the present invention.

[0085] like Figure 5 As shown, with Figure 4The only difference in the embodiment is that: when the type number corresponding to the operational fault type of the digital twin model in the current time segment output by the AI ​​dynamic evolution model is a non-empty character, the digital twin model is automatically triggered to simulate maintenance at the current moment to implement the maintenance strategy at the current moment, and the delay time of different maintenance strategies adopted at the current moment on the intelligently predicted operational fault type is evaluated. When the type number corresponding to the operational fault type of the digital twin model in the current time segment output by the AI ​​dynamic evolution model is an empty character, the automatic triggering of the digital twin model to simulate maintenance at the current moment is temporarily suspended, that is, after step S27, the method further includes:

[0086] Step S28: Obtain the delay duration corresponding to the different maintenance strategies adopted at the current moment, and take the maintenance strategy corresponding to the delay duration with the largest value as the best maintenance strategy at the current moment.

[0087] For example, an ASIC chip can be used to obtain the delay durations corresponding to different maintenance strategies at the current moment, and the maintenance strategy corresponding to the delay duration with the largest value among the delay durations can be taken as the best maintenance strategy at the current moment.

[0088] Example 5

[0089] Figure 6 This is a flowchart illustrating the steps of a dynamic evolution method for intelligent manufacturing based on a digital twin architecture, according to Embodiment 5 of the present invention.

[0090] like Figure 6 As shown, with Figure 2 Unlike the previous embodiment, in the intelligent manufacturing dynamic evolution method based on digital twin architecture, after using an AI dynamic evolution model to intelligently predict the type of operational failure that occurs in the digital twin model in the current time segment based on the model code value of the target model, the task type number of the set task, the preset time length, the sensor data and visualization content collected from the digital twin model executing the set task at each past time before the current time, and the duration of the current time segment, that is, after step S25, the method further includes:

[0091] Step S29: Obtain the types of operational faults that occurred in the digital twin model within the current time segment, and wirelessly transmit the types of operational faults that occurred in the digital twin model within the current time segment to the remote intelligent manufacturing monitoring server via a two-way wireless communication link.

[0092] For example, obtaining the types of operational faults that occur in the digital twin model within the current time segment and wirelessly transmitting these types of operational faults to a remote smart manufacturing monitoring server via a two-way wireless communication link further includes: the smart manufacturing monitoring server being a blockchain server or a cloud computing server.

[0093] Next, the various method embodiments of the present invention will be described in detail.

[0094] In a dynamic evolution method for intelligent manufacturing based on a digital twin architecture according to various embodiments of the present invention:

[0095] Constructing a digital twin model of the target model of the multi-joint robot, the digital twin model integrates the geometric structure, dynamic parameters, and mapping relationship between the control logic and sensor parameters of the target model of the multi-joint robot, including: constructing a digital twin model of the target model of the multi-joint robot using a co-simulation platform;

[0096] Specifically, a co-simulation platform is used to construct digital twin models of different types of multi-joint robots, meaning that this invention is applicable to different types of multi-joint robots.

[0097] In a dynamic evolution method for intelligent manufacturing based on a digital twin architecture according to various embodiments of the present invention:

[0098] The number of pooling layers used by the convolutional neural network is positively correlated with the number of tasks that the digital twin model can perform, and the number of training sessions performed on the convolutional neural network is proportional to the total number of joints of the target model of the multi-joint robot. This includes: using a first numerical mapping function to represent the first numerical mapping relationship that is positively correlated with the number of pooling layers used by the convolutional neural network and the number of tasks that the digital twin model can perform, and using a second numerical mapping function to represent the second numerical mapping relationship that is proportional to the number of training sessions performed on the convolutional neural network and the total number of joints of the target model of the multi-joint robot.

[0099] For example, different types of CPLD chips can be selected to test and simulate the implementation process of the first numerical mapping function and the second numerical mapping function respectively;

[0100] The first numerical mapping relationship, which uses a first numerical mapping function to represent the positive correlation between the number of pooling layers used by the convolutional neural network and the number of tasks that the digital twin model can perform, includes: in the first numerical mapping function, the number of tasks that the digital twin model can perform is the input parameter, and the number of pooling layers used by the convolutional neural network corresponding to the number of tasks that the digital twin model can perform is the output parameter.

[0101] The second numerical mapping function, which represents the relationship between the number of training operations performed on the convolutional neural network and the total number of joints of the target multi-joint robot, includes the following: In the second numerical mapping function, the total number of joints of the target multi-joint robot is the input parameter, and the number of training operations performed on the convolutional neural network, which is proportional to the total number of joints of the target multi-joint robot, is the output parameter.

[0102] And in a method for the dynamic evolution of intelligent manufacturing based on a digital twin architecture according to various embodiments of the present invention:

[0103] The AI ​​dynamic evolution model uses the target model's model code value, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model executing the set task at each past moment before the current moment, and the duration of the current time segment to intelligently predict the types of operational faults that will occur in the digital twin model in the current time segment. This includes: when the type number corresponding to the operational fault type of the digital twin model in the current time segment output by the AI ​​dynamic evolution model is a non-empty character, it indicates that the intelligently predicted operational fault type of the digital twin model in the current time segment exists; otherwise, it indicates that the intelligently predicted operational fault type of the digital twin model in the current time segment does not exist.

[0104] For example, if the type number corresponding to the operational failure type of the digital twin model in the current time segment output by the AI ​​dynamic evolution model is a non-empty character, it indicates that the intelligently predicted operational failure type of the digital twin model in the current time segment exists; otherwise, it indicates that the intelligently predicted operational failure type of the digital twin model in the current time segment does not exist. This includes: both non-empty characters and empty characters are represented by binary values.

[0105] The AI ​​dynamic evolution model intelligently predicts the types of operational faults that occur in the digital twin model within the current time segment based on the target model's model code value, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model executing the set task at each past moment before the current moment, and the duration of the current time segment. This also includes: synchronously inputting the target model's model code value, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model executing the set task at each past moment before the current moment, and the duration of the current time segment into the AI ​​dynamic evolution model, and running the AI ​​dynamic evolution model to obtain the type number corresponding to the types of operational faults that occur in the digital twin model within the current time segment, as output by the AI ​​dynamic evolution model.

[0106] In addition, the present invention may also refer to the following technical contents to highlight the significant technical progress of the present invention:

[0107] The previous moments before the current moment and the current moment together form a complete time interval on the time axis, and the time length between any two adjacent moments is equal to the preset time length. The current time segment starts from the current moment and the duration of the current time segment is an integer multiple of the preset time length. This includes the following: the number of previous moments before the current moment has the same trend as the total number of joints of the target model's multi-joint robot.

[0108] Among them, the trend of the number of moments before the current moment being the same as the total number of joints of the target model multi-joint robot includes: using a first numerical change curve to represent the numerical change trend of the total number of joints of the target model multi-joint robot, and using a second numerical change curve to represent the numerical change trend of the number of moments before the current moment.

[0109] Among them, the trend of the number of past moments before the current moment being the same as the total number of joints of the target model multi-joint robot also includes: performing length-based numerical normalization processing on the first numerical change curve and the second numerical change curve respectively to obtain the first normalized curve and the second normalized curve respectively, and the curvature values ​​at uniform intervals on the first normalized curve are equal to the curvature values ​​at uniform intervals on the second normalized curve respectively.

[0110] For example, after performing length-based numerical normalization on the first numerical change curve and the second numerical change curve respectively, a first normalized curve and a second normalized curve are obtained respectively. The curvature values ​​at uniform intervals on the first normalized curve are equal to the curvature values ​​at uniform intervals on the second normalized curve. This includes: optionally using a programmable logic device to perform length-based numerical normalization on the first numerical change curve and the second numerical change curve respectively.

[0111] The foregoing description of exemplary embodiments of the invention is provided for illustrative and descriptive purposes. It is not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations will obviously be apparent to those skilled in the art. Exemplary embodiments were chosen and described in order to best illustrate the principles of the invention and its practical application, thereby enabling others skilled in the art to understand the various embodiments and variations of the invention suitable for the contemplated particular purpose. The scope of the invention is intended to be defined by the appended claims and their equivalents.

Claims

1. A dynamic evolution method for intelligent manufacturing based on a digital twin architecture, characterized in that, The method includes: A digital twin model of the target model of the multi-joint robot is constructed. The digital twin model integrates the mapping relationship between the geometric structure, dynamic parameters and control logic of the target model of the multi-joint robot and the sensor output parameters. Sensor data from previous moments before the current moment is collected from the digital twin model performing the set task. The sensor data at each moment consists of the output parameters of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor collected from the digital twin model at that moment. The visualization content of each past moment before the current moment is collected from the digital twin model performing the set task. The visualization content of each moment is the depth value, brightness value and coordinate value of each pixel of the sub-frame occupied by the multi-joint robot in the presentation image taken by the visual sensor directly above the multi-joint robot at that moment from a bird's-eye view. The convolutional neural network is trained in each iteration to obtain the completed convolutional neural network as a dynamic evolution model for AI. The AI ​​dynamic evolution model is used to intelligently predict the type of operational failure that will occur in the digital twin model in the current time segment based on the model code value of the target model, the task type number of the set task, the preset time length, the sensor data and visualization content collected from the digital twin model that executes the set task at each previous time before the current time, and the duration of the current time segment.

2. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in claim 1, characterized in that: The type of operational fault that occurs in the digital twin model within the current time segment predicted by the intelligent system is one of the following: joint wear, reducer failure, motor overheating, structural fatigue, lubrication failure, and control deviation accumulation. The operational fault numbers corresponding to different types of operational faults are different. Among them, each past moment before the current moment and the current moment together form a complete time interval on the time axis, and the time length between each two adjacent moments is equal to the preset time length. The current time segment starts from the current moment and the duration of the current time segment is an integer multiple of the preset time length. In this case, the distance from the imaging lens of the simulated visual sensor directly above the multi-joint robot to the top of the multi-joint robot is equal to a preset distance. The visualization content at each moment is the depth value, brightness value, and coordinate value of each pixel in the sub-frame occupied by the multi-joint robot in the presentation image captured by the visual sensor directly above the multi-joint robot at that moment from a bird's-eye view. This includes: identifying the multi-joint robot occupying the sub-frame in the presentation image captured by the bird's-eye view at that moment based on the imaging pattern of the multi-joint robot in its factory state from a bird's-eye view.

3. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in claim 2, characterized in that: The process of training a convolutional neural network (CNN) and obtaining the CNN after each training iteration to serve as an AI dynamic evolution model includes: the number of pooling layers used in the CNN is positively correlated with the number of tasks that the digital twin model can perform, and the number of training iterations performed on the CNN is proportional to the total number of joints of the target model of the multi-joint robot. The process of training the convolutional neural network (CNN) to obtain the trained CNN as the AI ​​dynamic evolution model further includes: in each training iteration of the CNN, using the type number corresponding to the type of operational failure that occurred in the digital twin model within a known past time segment as the single output of the AI ​​dynamic evolution model, and using the model code value of the target model, the task type number of the set task, the preset time length, the sensor data and visualization content collected from the digital twin model executing the set task at each past time segment before the specified past time segment, and the duration of the specified past time segment as multiple inputs to the AI ​​dynamic evolution model to complete the training.

4. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in claim 3, characterized in that, The method further includes acquiring sensor data from a digital twin model performing a set task, up to each previous moment, where the sensor data at each moment is prior to the output parameters of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor acquired from the digital twin model at that moment. The sampling frequencies of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor are adjusted to achieve synchronous acquisition of the output parameters of each sensor. The process of adjusting the sampling frequency of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor to achieve synchronous acquisition of their respective output parameters includes: using a rectangular waveform of a preset frequency emitted by a quartz oscillator to provide a reference clock signal for the synchronous acquisition of the output parameters of the joint torque sensor, motor current sensor, infrared temperature sensor, and microelectromechanical system vibration sensor.

5. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in claim 3, characterized in that, After employing an AI dynamic evolution model to intelligently predict the type of operational failure that may occur in the digital twin model within the current time segment based on the target model's model code value, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model executing the set task at each past moment before the current moment, and the duration of the current time segment, the method further includes: When the type number corresponding to the operational fault type that occurred in the digital twin model within the current time segment output by the AI ​​dynamic evolution model is a non-empty character, the digital twin model is automatically triggered to simulate maintenance at the current moment to implement the maintenance strategy at the current moment, and to evaluate the delay time of different maintenance strategies adopted at the current moment on the intelligently predicted operational fault type. When the type number corresponding to the operational fault type that occurred in the digital twin model within the current time segment output by the AI ​​dynamic evolution model is an empty character, the automatic triggering of the digital twin model to simulate maintenance at the current moment is temporarily suspended.

6. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in claim 5, characterized in that, When the type number corresponding to the operational fault type occurring in the digital twin model within the current time segment output by the AI ​​dynamic evolution model is a non-empty character, the digital twin model is automatically triggered to simulate maintenance at the current moment to implement the maintenance strategy at the current moment, and the delay time of different maintenance strategies adopted at the current moment on the intelligently predicted operational fault type is evaluated. When the type number corresponding to the operational fault type occurring in the digital twin model within the current time segment output by the AI ​​dynamic evolution model is an empty character, the automatic triggering of the digital twin model to simulate maintenance at the current moment is temporarily suspended. The method further includes: Obtain the delay duration corresponding to different maintenance strategies adopted at the current moment, and take the maintenance strategy corresponding to the delay duration with the largest value as the best maintenance strategy at the current moment.

7. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in claim 3, characterized in that, After employing an AI dynamic evolution model to intelligently predict the type of operational failure that may occur in the digital twin model within the current time segment based on the target model's model code value, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model executing the set task at each past moment before the current moment, and the duration of the current time segment, the method further includes: The system obtains the types of operational faults that occur in the digital twin model within the current time segment and wirelessly transmits these types of operational faults to the remote intelligent manufacturing monitoring server via a two-way wireless communication link. The process of obtaining the operational fault types of the digital twin model within the current time segment and wirelessly transmitting these fault types to a remote intelligent manufacturing monitoring server via a two-way wireless communication link includes: the two-way wireless communication link being based on frequency division duplex communication mode.

8. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in any one of claims 3-7, characterized in that: A digital twin model of the target type of multi-joint robot is constructed. The digital twin model integrates the geometric structure, dynamic parameters, and mapping relationship between the control logic and sensor parameters of the target type of multi-joint robot. This includes constructing a digital twin model of the target type of multi-joint robot using a co-simulation platform.

9. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in any one of claims 3-7, characterized in that: The number of pooling layers used by the convolutional neural network is positively correlated with the number of tasks that the digital twin model can perform, and the number of training sessions performed on the convolutional neural network is proportional to the total number of joints of the target model of the multi-joint robot. This includes: using a first numerical mapping function to represent the first numerical mapping relationship that is positively correlated with the number of pooling layers used by the convolutional neural network and the number of tasks that the digital twin model can perform, and using a second numerical mapping function to represent the second numerical mapping relationship that is proportional to the number of training sessions performed on the convolutional neural network and the total number of joints of the target model of the multi-joint robot. The first numerical mapping relationship, which uses a first numerical mapping function to represent the positive correlation between the number of pooling layers used by the convolutional neural network and the number of tasks that the digital twin model can perform, includes: in the first numerical mapping function, the number of tasks that the digital twin model can perform is the input parameter, and the number of pooling layers used by the convolutional neural network corresponding to the number of tasks that the digital twin model can perform is the output parameter. The second numerical mapping function, which represents the relationship between the number of training operations performed on the convolutional neural network and the total number of joints of the target multi-joint robot, includes the following: In the second numerical mapping function, the total number of joints of the target multi-joint robot is the input parameter, and the number of training operations performed on the convolutional neural network, which is proportional to the total number of joints of the target multi-joint robot, is the output parameter.

10. The intelligent manufacturing dynamic evolution method based on digital twin architecture as described in any one of claims 3-7, characterized in that: The AI ​​dynamic evolution model uses the target model's model code value, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model executing the set task at each past moment before the current moment, and the duration of the current time segment to intelligently predict the types of operational faults that will occur in the digital twin model in the current time segment. This includes: when the type number corresponding to the operational fault type of the digital twin model in the current time segment output by the AI ​​dynamic evolution model is a non-empty character, it indicates that the intelligently predicted operational fault type of the digital twin model in the current time segment exists; otherwise, it indicates that the intelligently predicted operational fault type of the digital twin model in the current time segment does not exist. The AI ​​dynamic evolution model intelligently predicts the types of operational faults that occur in the digital twin model within the current time segment based on the target model's model code value, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model executing the set task at each previous moment before the current moment, and the duration of the current time segment. This also includes: synchronously inputting the target model's model code value, the task type number of the set task, the preset time length, sensor data and visualization content collected from the digital twin model executing the set task at each previous moment before the current moment, and the duration of the current time segment into the AI ​​dynamic evolution model, and running the AI ​​dynamic evolution model to obtain the type number corresponding to the types of operational faults that occur in the digital twin model within the current time segment, as output by the AI ​​dynamic evolution model.

Images