An orchestration method and system for automating execution of industrial tasks

Abstract

The application discloses an arrangement method and system for automatically executing industrial tasks, and relates to the technical field of automatic execution, which comprises the following steps: an industrial process is disassembled into ability units which can be independently called and encapsulated and registered according to man-machine-material; based on historical records, step fragments are constructed into a sensor and actuator time sequence matrix, the contribution strength of calculation parameters to control actions is calculated to form an intensity indicator, a gate is generated through a suppression network, and the strength influence is suppressed in multi-head attention with a penalty term, so that a time sequence representation representing work content is obtained; a dimensional vector is output by a time sequence encoder and clustering is performed to obtain an ability type, and template input parameters, static parameters and dynamic transmission parameters can be divided. A scheduler selects a device node according to a process script and an agent configuration, and issues an action, an output is parsed according to a template and written into a state storage to realize automatic parameter transmission, and a high-availability daemon guarantees consistency and fault recovery.

Description

Technical Field

[0001] This invention relates to the field of automated execution technology, and in particular to a method and system for automating industrial tasks. Background Technology

[0002] With the accelerated transformation of industry towards intelligent manufacturing, the demand for automation and standardization of task scheduling in fixed-process industrial production is becoming increasingly urgent.

[0003] However, the current industrial task execution mode generally relies on manual intervention to connect multiple processes and configure equipment parameters, which leads to core technology challenges in system operation.

[0004] These issues are mainly reflected in the lack of a unified abstract description and standardized interface for the capabilities of heterogeneous equipment. The communication protocols and interfaces between different types of equipment are incompatible, making it difficult to integrate them into a unified scheduling and control system. At the same time, the efficiency of parameter transmission between multiple processes is low, requiring manual recording and input into the next process. This not only introduces human error and causes high data latency, but also affects the consistency of the final product quality and production efficiency.

[0005] Therefore, automating industrial processes, especially for different equipment and processes, is crucial for efficient production in enterprises.

[0006] While many automated industrial process orchestration methods exist, they are mostly based on hard-coded logic, lacking dynamic adaptability to changes in production processes and unable to adjust equipment allocation according to real-time resource demands. Furthermore, the systems lack pre-detection mechanisms for abnormal parameters, making it difficult to achieve adaptive handling and automatic fault recovery during task execution, significantly reducing system robustness.

[0007] While preliminary practices in industrial task orchestration have been implemented in recent decades, several shortcomings remain: First, capability abstraction is disconnected from the core scheduling framework, lacking standardized interfaces, leading to repetitive adaptation development for capability calls. Second, process definitions lack structured specifications, often relying on linear text descriptions, which fail to accurately express complex process dependencies and parameter passing rules, hindering automated parsing. Third, the reliability of scheduling components is insufficient, unable to effectively handle single points of failure in distributed scenarios, impacting task execution stability. Finally, the mapping relationship for device access is unclear; the correspondence between abstract capabilities in agent program configurations and physical device API endpoints is ambiguous, resulting in scheduling commands not being accurately and efficiently sent to the device's execution end. Summary of the Invention

[0008] In view of the aforementioned existing problems, the present invention is proposed.

[0009] Therefore, this invention provides an automated industrial task orchestration method to solve problems such as difficulty in capability integration, non-standard process definition, insufficient scheduling reliability, and ambiguous equipment docking in existing industrial task orchestration technologies.

[0010] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0011] In a first aspect, the present invention provides a method for orchestrating automated execution of industrial tasks, which includes decomposing the industrial process into multiple capability units according to human-machine-material, abstracting the capabilities, and generating parameter configurations;

[0012] The capability unit is encapsulated into an independently executable program package and its capabilities are registered.

[0013] After completing the environment deployment for scheduling, write the script files for automated execution;

[0014] By using a scheduler to select the corresponding device nodes based on the process script and agent configuration, actions can be issued to achieve automated execution of industrial tasks.

[0015] As a preferred embodiment of the orchestration method for automating industrial tasks according to the present invention, the capability unit includes: the minimum functional implementation steps that can be independently invoked in the industrial process; during operation, parameter configuration is achieved through template input parameters, static parameters, and dynamically transmitted parameters driven by the output of the previous capability unit;

[0016] The template input parameters are editable parameter items used to configure the execution target values ​​of the functional unit steps without modifying the corresponding package code of the capability unit.

[0017] As a preferred embodiment of the orchestration method for automating industrial tasks according to the present invention, the capability abstraction includes: representing step segments as temporal feature matrices containing sensors and actuators in historical records; using the temporal feature matrix to calculate an intensity indicator representing the intensity level of the step segment, and inputting the intensity indicator into an inhibition network to generate an inhibition gating; after extracting intermediate temporal features from the step segment using a temporal feature extractor, introducing the inhibition gating as an attention penalty term in the multi-head self-attention calculation process to obtain a temporal feature representation with suppressed parameter influence;

[0018] Based on the temporal feature representation with suppressed parameter influence, a temporal encoder is used to output a fixed-dimensional work content vector as an abstract result;

[0019] The parameter configuration generation process is as follows: Clustering is performed based on the implementation process of each step and the abstract results of the parameter configuration in the historical records; and based on the analysis of parameter characteristics during the clustering process, the static parameters and the dynamically transmitted parameters are divided.

[0020] For each step segment of the sample, parameters other than the template input parameters are captured; samples in the same cluster are taken as the same capability unit, and the same parameters of each individual in the cluster are obtained as the static parameters of the capability unit, while other parameters are taken as the dynamic transmission parameters.

[0021] As a preferred embodiment of the orchestration method for automating industrial tasks according to the present invention, each capability unit is treated as an independently executable program package, and each capability unit is configured with an instance configuration file and a capability metadata file.

[0022] The instance configuration file is used to define the input and output parameter specifications of the capability instance;

[0023] The capability metadata file is used to define the type, unique identifier, and logical association of the capability with other capabilities;

[0024] During deployment and startup, the binary program package and related configuration files are placed in a preset directory, and the startup command is executed to enable the capability unit to complete service registration with a registry center and expose a set of standardized management interfaces to the outside world.

[0025] As a preferred embodiment of the orchestration method for automating industrial tasks according to the present invention, the environment deployment includes: a control plane architecture and a data and network plane architecture;

[0026] Control plane architecture: Deploy a distributed key-value storage cluster as a central data repository to persistently store task status and node information; deploy an API server as a system gateway to handle authentication and routing of all internal and external requests, and as the only component that can directly access the storage cluster;

[0027] The deployment scheduler acts as the decision center, continuously monitoring the tasks to be scheduled in the API server and allocating appropriate execution nodes to the tasks based on resource requirements and policies.

[0028] Deploy high-availability guardian components to ensure data consistency and fault recovery capabilities of the storage cluster in a multi-node deployment.

[0029] Data and network plane architecture:

[0030] Deploy agent programs on multiple worker nodes to register node information with the control plane, report node status, listen for tasks assigned to the nodes, and drive local devices to complete actual operations.

[0031] If feasible, deploy network proxy components on each node to build a cluster network plane, which is responsible for maintaining and updating network rules, and realizing service discovery, load balancing and cross-node communication traffic forwarding.

[0032] As a preferred embodiment of the automated industrial task scheduling method of the present invention, the script file includes a three-level tree structure from task to process group to action to define the production process;

[0033] The action layer is used to bind specific device capabilities and input / output parameters;

[0034] The input parameters of the action layer are bound through string addressing expressions, which enables the automatic passing of parameters between actions;

[0035] Output parameters are bound to the data type template of the preset value through key names;

[0036] During task execution, the system retrieves results from the device's capabilities using a preset template and persists the actual runtime data to the state storage, thus enabling the transfer of parameters.

[0037] As a preferred embodiment of the orchestration method for automating industrial tasks according to the present invention, in the script flow, using a convolutional neural network pre-trained with historical records, each capability unit generates the dynamic transmission parameters for this capability unit based on the template input parameters, static parameters and dynamic transmission parameters of the previous capability unit, and the initial values ​​of the dynamic transmission parameters of each capability unit are generated sequentially according to the flow order.

[0038] During the operation phase of the action layer, after the operation layer of each capability unit completes its operation, the data is used for supervised learning of the convolutional neural network, thereby updating the dynamic transfer parameters used for the current operation based on the initial value of the dynamic transfer parameters.

[0039] In this process, supervised learning only applies during the runtime phase. After the task is completed, the convolutional neural network reverts to the state before supervision.

[0040] Secondly, the present invention provides an orchestration system for automating industrial tasks, including an abstraction module that decomposes the industrial process into multiple capability units according to human-machine-material, performs capability abstraction, and generates parameter configuration;

[0041] The encapsulation module encapsulates the capability unit into an independently executable program package and performs capability registration;

[0042] The deployment module, after completing the deployment of the scheduled environment, writes script files for automated execution;

[0043] The execution module uses the scheduler to select the corresponding device nodes based on the process script and agent configuration to issue actions, thereby realizing the automated execution of industrial tasks.

[0044] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the orchestration method for automating the execution of industrial tasks as described in the first aspect of the present invention.

[0045] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the orchestration method for automating the execution of industrial tasks as described in the first aspect of the present invention.

[0046] The beneficial effects of this invention are as follows: By breaking down the production process into "human-machine-material" components and abstracting each step into an independently executable capability unit, and in conjunction with a general capability registration mechanism, service registration and standardized interface exposure can be completed simply by placing the program package and configuration file in a preset directory and starting the system, thus reducing the threshold for heterogeneous device access and reuse; by deploying a control plane of distributed key-value storage, API gateway and scheduler, and a data / network plane of multi-node agent and network agent components, service discovery, load balancing and cross-node communication are achieved, while task status, node information and running parameters are persisted and automatic fault recovery is supported, improving the reliability and scalability of the system in a multi-node environment; by binding task-process group-action three-layer scripts with string addressing, automatic reference and transmission of cross-action parameters are achieved, and the output is parsed according to data type templates and written to the status storage, significantly reducing manual wiring errors and debugging costs, thus requiring only a small amount of configuration to quickly complete industrial task orchestration and automated closed-loop execution. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart of a method for orchestrating industrial tasks to be performed automatically.

[0049] Figure 2 A flowchart illustrating the implementation process of a programming method for automating industrial tasks. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0051] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0053] Reference Figure 1 and Figure 2 As one embodiment of the present invention, this embodiment provides a scheduling method for automating the execution of industrial tasks, comprising the following steps:

[0054] S1: The industrial process is broken down into multiple capability units according to the human-machine-material model, and capability abstraction is performed to generate parameter configuration.

[0055] Most of the static parameters are set to fixed values ​​according to the process control plan, while a few dynamic parameters are set as output parameters of the previous capacity unit.

[0056] The capability unit includes the minimum functional implementation steps that can be independently invoked in the industrial process; during operation, parameter configuration is achieved through template input parameters, static parameters, and dynamically transmitted parameters driven by the output of the previous capability unit.

[0057] The template input parameters are editable parameter items used to configure the execution target values ​​of the functional unit steps without modifying the corresponding package code of the capability unit.

[0058] Furthermore, the capability abstraction includes: representing step segments as temporal feature matrices containing sensors and actuators in the historical record; using the temporal feature matrix to calculate an intensity indicator representing the intensity level of the step segment, and inputting the intensity indicator into a suppression network to generate a suppression gating; after extracting intermediate temporal features from the step segment using a temporal feature extractor, introducing the suppression gating as an attention penalty term during the multi-head self-attention calculation process to obtain a temporal feature representation with suppressed parameter influence. Based on the temporal feature representation with suppressed parameter influence, a temporal encoder outputs a fixed-dimensional work content vector as the abstraction result.

[0059] Specifically: First, calculate the "intensity indicator" but do not use it as a feature for clustering; use it only for suppression. Construct and train a parametric action contribution model (through fitting a functional relationship), so that given a functional unit type identifier and parameter sequence, this model outputs the contribution intensity of each parameter to the control action; thus, the intensity sequence is obtained. . This represents the contribution strength of the t-th parameter.

[0060] First, a lightweight temporal feature extractor is used to transform the original signal into intermediate features. It is obtained using 1D-CNN / TCN (one-dimensional convolution or extended convolution). This step is only responsible for suppressing noise and making the shape noticeable, and does not "ignore parameters".

[0061] The intensity-guided attention inhibition module suppresses intensity-related components by directly adding inhibition terms to the logits of attention. First, calculate the multi-head self-attention... : , , Intensity sequence It becomes a suppression gating through a small network. : .

[0062] Apply an "intensity penalty" to attention logits:

[0063]

[0064] It is the suppression coefficient. The greater the strength, the more attentional connections to other locations are suppressed, thus squeezing out the "strength / intensity (parameter influence)" from the representation.

[0065] Characteristics after output suppression: (The residuals and layer normalization are then used to maintain stability).

[0066] Use an encoder Compile into vectors that "only reflect the work content". :

[0067] The attention layer stack above After the layer, global pooling is performed on the time dimension to obtain... The core of the encoder is that TCN is responsible for local morphological features, while Transformer attention is responsible for suppressing intensity and aggregating it into global semantics.

[0068] Training / Constraint (ensuring attention is indeed focused on "inhibition strength" and not just random learning):

[0069] Self-supervised application of "intensity gain enhancement": Amplitude scaling of the same segment. (Only change the intensity, not the task content), requiring the encoding vector to remain unchanged (comparative learning of InfoNCE or simpler methods). ).

[0070] Regularities that drive intensity away from attention:

[0071]

[0072] Minimizing it forces the model to give smaller attention weights to locations with high intensity (i.e., "suppress intensity features").

[0073] Final loss available: (Training is possible without tags).

[0074] X represents the original input data of the same step segment (a multi-channel timing matrix of a segment), usually in the shape of "number of channels × time length". a represents the amplitude scaling factor (a real number), such as 0.8, 1.2, etc., used to strengthen or weaken the entire signal. z represents the vector encoded by the encoder from the original segment X, with fixed dimensions (e.g., 128-dimensional / 256-dimensional). This indicates the vector obtained by the encoder after encoding the enhanced segment aX; "+" indicates "positive sample / enhanced version of the same segment". This indicates "intensity suppression regularization / supervision term".

[0075] Finally, the vectors of all segments When performing clustering (either HDBSCAN or KMeans), each cluster represents a "macro-step / capability type." Because the intensity (parameter influence) has been suppressed by attention, the clusters represent "what was done," not "how aggressively it was done / how big the goal was."

[0076] The parameter configuration generation process is as follows: Clustering is performed based on the implementation process of each step and the abstract results of the parameter configuration in the historical records; and based on the analysis of parameter characteristics during the clustering process, the static parameters and the dynamically transmitted parameters are divided.

[0077] For each step segment of the sample, parameters other than the template input parameters are captured; samples in the same cluster are taken as the same capability unit, and the same parameters of each individual in the cluster are obtained as the static parameters of the capability unit, while other parameters are taken as the dynamic transmission parameters.

[0078] S2: Encapsulate the capability unit into an independently executable program package and register the capability.

[0079] Each capability unit is configured as an independently executable package, and each capability unit is configured with an instance configuration file and a capability metadata file.

[0080] The instance configuration file is used to define the input and output parameter specifications for capability instances.

[0081] The capability metadata file is used to define the type of the capability, its unique identifier, and its logical association with other capabilities.

[0082] During deployment and startup, the binary program package and related configuration files are placed in a preset directory, and the startup command is executed to enable the capability unit to complete service registration with a registry center and expose a set of standardized management interfaces to the outside world.

[0083] By configuring a general capability registration framework, the written capabilities and configuration files can be placed in the appropriate locations. Then, starting the framework will inject the capabilities and expose the corresponding ports to the scheduling framework.

[0084] S3: After completing the environment deployment for scheduling, write the script files for automated execution.

[0085] The environment deployment includes a control plane architecture and a data and network plane architecture. The control plane architecture deploys a distributed key-value storage cluster as a central data repository for persistently storing task states and node information. An API server is deployed as the system gateway, handling authentication and routing of all internal and external requests, and serving as the only component directly able to access the storage cluster. A scheduler is deployed as the decision center, continuously monitoring tasks awaiting scheduling on the API server and allocating appropriate execution nodes to tasks based on resource requirements and policies. A high-availability daemon is deployed to ensure data consistency and fault recovery capabilities of the storage cluster in a multi-node deployment. Specifically, the high-availability daemon ensures the continuous availability of critical services in the orchestration system's control plane by performing health checks, fault diagnosis, and automatic recovery on the distributed key-value storage cluster and core control plane processes, preventing single-point failures from causing task state loss or scheduling interruptions. Specifically, the high availability guardian component continuously monitors the heartbeat and service metrics of each node. When a node crash, network partition, or process anomaly is detected, it triggers a majority arbitration and master election mechanism to prevent multiple master nodes from writing to the outside world at the same time, which would cause a split-brain situation. It also automatically completes the master-slave switchover when a master node fails. At the same time, it isolates, restarts, and rejoins abnormal replicas to synchronize data, ensuring that the task status and node information remain consistent and recoverable. This improves the stability, continuous operation capability, and fault self-healing capability of the system under multi-node deployment.

[0086] Data and network plane architecture: Agent programs are deployed on multiple worker nodes to register node information with the control plane, report node status, listen for tasks assigned to nodes, and drive local devices to complete actual operations. Where feasible, network agent components are deployed on each node to build a clustered network plane, responsible for maintaining and updating network rules, and enabling service discovery, load balancing, and cross-node communication traffic forwarding.

[0087] S4: The scheduler selects the corresponding device nodes based on the process script and agent configuration to issue actions, thereby realizing the automated execution of industrial tasks.

[0088] The script file defines the production process using a three-level tree structure from task to process group to action. The action layer is used to bind specific equipment capabilities and input / output parameters.

[0089] The input parameters of the action layer are bound through string addressing expressions, which enables the automatic transfer of parameters between actions.

[0090] Output parameters are bound to the data type template of the preset value through key names.

[0091] During task execution, the system retrieves results from the device's capabilities using a preset template and persists the actual runtime data to the state storage, thus enabling the transfer of parameters.

[0092] It is also important to note that the agent program files are deployed on multiple worker nodes to enable them to schedule the abstract capabilities of the process devices. The configuration files are written in a layered structure. The agent layer is configured with name, address, heartbeat monitoring and log management information, which is used to schedule the instance name and type of the device abstract capability configured by the capability framework, as well as the API startup port.

[0093] Furthermore, within the script's flow, a convolutional neural network pre-trained using historical records is used. Each capability unit generates its own dynamic transfer parameters based on the template input parameters, static parameters, and dynamic transfer parameters of the previous capability unit. Initial values ​​for the dynamic transfer parameters of each capability unit are generated sequentially according to the flow. During the action layer's execution phase, after each capability unit's execution layer completes its operation, the data is used for supervised learning of the convolutional neural network, thereby updating the dynamic transfer parameters for the current execution based on their initial values. This supervised learning only applies to the execution phase; after the task is completed, the convolutional neural network reverts to its state before supervision.

[0094] This embodiment also provides an orchestration system for automating industrial tasks, including:

[0095] The abstract module breaks down the industrial process into multiple capability units according to the human-machine-material model, performs capability abstraction, and generates parameter configurations.

[0096] The encapsulation module encapsulates the capability unit into an independently executable program package and performs capability registration.

[0097] The deployment module, after completing the deployment of the scheduled environment, writes script files for automated execution.

[0098] The execution module uses the scheduler to select the corresponding device nodes based on the process script and agent configuration to issue actions, thereby realizing the automated execution of industrial tasks.

[0099] This embodiment also provides a computer device suitable for orchestration methods for automating industrial tasks, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the orchestration method for automating industrial tasks as proposed in the above embodiment.

[0100] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0101] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the scheduling method for automating industrial tasks as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0102] In summary, this invention achieves automated, reusable, scalable, and highly reliable execution of industrial tasks with minimal code by: breaking down industrial processes into independently callable capability units based on the human-machine-material model and performing data-driven capability abstraction; encapsulating capability units into independently executable program packages with accompanying instance configuration files and capability metadata files, and using a registry center to achieve capability discovery, version management, and standardized interface exposure; in a distributed scheduling environment composed of the control plane and the data / network plane, the scheduler selects device nodes with corresponding capabilities based on the process script and agent configuration to issue actions, and persists the running status and output parameters to the state storage to achieve automatic parameter transmission across actions; simultaneously, a high-availability guardian component is introduced for health checks, arbitration for leader election, and fault self-healing to ensure state consistency and continuous scheduling; and at the same time, a high-availability guardian component is introduced for health checks, arbitration for leader election, and fault self-healing to ensure state consistency and continuous scheduling.

[0103] The following is a specific example of the copper rod manufacturing process:

[0104] This task involves producing an 8mm copper rod from a copper plate through a series of fixed processes. The specific implementation process of this invention is detailed below, including the following steps:

[0105] S1, abstracting capabilities:

[0106] Through two on-site visits and communications with engineers, the first step revealed the steps involved in the copper rod production process, the parameters required for each step, the processing involved, and the resulting effects. This resulted in eight steps: feeding, heating, heat preservation, pouring, casting, rolling, cleaning, and winding. These were understood as eight processes. The second step focused on identifying potential anomalies at each step and how to handle them. These anomalies were categorized into four types: alarms, machine linkage, shutdown, and oxygen-related anomalies. The operations for each step were then organized along the lines of personnel, machines, and materials to better abstract the process. Based on this analysis, these 12 capabilities were documented in a standardized document. This document includes the required parameters for each capability, the processing steps, the output parameters, and the linkage between parameters in each step. This ensures that each piece of equipment is no longer isolated hardware but a functional unit with a clearly defined data interaction interface.

[0107] In the above operation, the input parameters, taking the feeding as an example, include the weight of the copper plate to be produced. The start and end positions need to be specified by the user. For alarm capability, the input data is the temperature of the vertical furnace, obtained in real-time from the temperature sensor in the furnace. For heat preservation capability, the input parameter is the heat preservation temperature, processed by the heating capability in the previous step. The processing part involves calling a video to view the operation process of each step and displaying the parameters processed in each step. The processing part also includes abnormal handling to restore the sensors to the normal range, which is displayed in the visualization window. The output parameters have two types: the first is the parameters after normal process processing. These parameters may often be used in the next step or indicate the processing status of that step. Taking heating as an example, the temperature output parameter can be used as the input for the next heat preservation step, and the completion status parameter shows the details of the processing of that step. The second type is the output of abnormal capability. This capability is mainly because some parameters need to be within the normal range before the normal process is executed. Abnormal handling accepts these inputs, restores them to normal through abnormal processing, and then executes the normal process.

[0108] S2, following S1, provides abstractions of process capabilities and process exception capabilities. Next, the relevant capabilities need to be written and registered. The process is as follows: Following S1, the written documentation encodes each abstract capability, and then the tool encapsulates it into a binary capability package, thereby achieving a comprehensive digital description of the capability unit's attributes and operating environment. These capability packages can be registered into the capability framework for execution. To execute these capabilities, configuration files must first be written for them.

[0109] The configuration file mainly consists of two files: one for the capability class and one for the instance name. Taking feeding as an example, the capability class file specifies the class name of this capability package, which can perform feeding capabilities, defines its accepted parameters such as the weight of the feed, the start and end positions, and its mass percentage, defines its processing function, and the return parameters after processing. The instance file specifies its instance name, which can establish a connection with the scheduling framework, and some configurations for registering with the capability framework, ensuring that even if two devices have the same capabilities, they can be registered in the capability cache at the same time and scheduled by the scheduling framework.

[0110] Place the encapsulated binary program package and related configuration files in a preset directory, start the cache service management capability, and execute the start command to make the capability unit complete the service registration with the registration center; after registration, each capability unit exposes a set of standardized management interfaces to support external calls.

[0111] Each capability's interface includes GET requests ( / api / cr) to query all capability instances and POST requests ( / api / lifecycle-request) with parameters. The POST request uses the UUID configured in the instance file to initiate and terminate connections with the scheduling framework. A GET request ( / api / task / {id}) can be used to retrieve the execution status of each task. A GET request ( / api / ability-heartbeat) can be used to view the port used for each capability's startup and registration. A POST request ( / api / task / start_task) can be sent.

[0112] It can be used to launch tasks that have been registered in the capability cache.

[0113] You can use the command `Get / api / task / {task_id} / status` to view the completion status of a task and its return parameters.

[0114] After S2, each capability has been registered in the capability cache Redis and the relevant URL of each capability has been exposed to the outside world. In order to achieve high scalability, high availability and high concurrency scheduling capabilities, an overall scheduling framework needs to be deployed.

[0115] First, deploy an ETCD distributed key-value storage cluster as the system's central data warehouse to persistently store the system's metadata and task status, ensuring storage reliability.

[0116] Meanwhile, the APIServer is deployed as the system's sole entry gateway, responsible for handling the authentication and routing of all requests. The APIServer is configured as the only component with access to ETCD, thus ensuring data security and compliance at the architectural level by consolidating data access entry points.

[0117] A Scheduler is deployed as the decision center, continuously monitoring tasks awaiting scheduling and dynamically allocating them to the optimal nodes based on resource constraints and preset strategies. Additionally, ETCDkeep is deployed as a high-availability daemon, responsible for real-time monitoring of cluster health and automatic fault recovery in case of node failures, ensuring continuous and stable operation of the control plane in a multi-node environment.

[0118] Agents are deployed on each worker node to register with the control plane, report real-time status, and listen for and claim tasks assigned to the node, thereby driving local devices to complete the operational loop. Simultaneously, Proxy components are deployed to build the cluster network plane, maintaining network forwarding rules to achieve service discovery, load balancing, and cross-node communication, providing a stable underlying network foundation for the scheduling framework.

[0119] S4. Configure the files. This requires configuring two files: the agent file and the agent file. In the copper rod process, taking the vertical furnace as an example, one agent corresponds to one device. The agent will be configured with the capabilities it has registered in the capability framework, such as heating, oxygen anomaly detection, and alarm. The agent's name will also be specified for unique identification in the process configuration file. The timeout time and activity detection frequency of the agent program will also be specified to indicate whether the task can be scheduled normally.

[0120] Configure the process file. In the process file configuration, it needs to be written according to three aspects: task process group and action.

[0121] In the copper rod production process, the task is to produce copper rods. The process group consists of eight steps: feeding, heating, heat preservation, pouring, casting, continuous rolling, cleaning, and winding.

[0122] Each process group corresponds to one or more devices, and each device corresponds to an agent file; each process group includes one or more actions.

[0123] For example, in the material feeding process group, there is only one action capability and one feeding machine. Although there is only one machine, the actual execution of this action may be completed by the feeding machine through multiple operations. In the heating process, there are multiple machines, such as oxygen abnormality handling equipment and heating equipment (vertical furnace). These machines will perform multiple actions to complete this process. This process is sequential. The abnormality detection equipment first handles the alarm capability and oxygen abnormality capability. If an abnormality is found, the relevant operation is performed; otherwise, no action is taken, and the heating action is performed. The handling of other capabilities is the same as above.

[0124] Configure collaboration between multiple processes: use the handling of copper in each process group as a transfer parameter between processes; define a parameter in the output of each process to identify the start of the next capability; the execution of some process capabilities requires the output parameters of abnormal capabilities.

[0125] The above configuration completes the configuration file for executing an industrial process task with a fixed process template. Using this configuration file as input parameters, a POST request / framework / v1 / task?Name={tg}&Namespace={test} is sent to the deployed scheduling framework to start the scheduling framework and execute the industrial process.

[0126] After testing, the copper rod production task can be scheduled normally through the scheduling framework, and the scheduling framework can connect to the capabilities registered in the capability cache and start the scheduling of each capability normally through its exposed URL.

[0127] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for orchestrating automated execution of industrial tasks, characterized in that: include, The industrial process is broken down into multiple capability units based on the human-machine-material model, and capability abstraction is performed to generate parameter configurations; The capability unit is encapsulated into an independently executable program package and its capabilities are registered. After completing the environment deployment for scheduling, write the script files for automated execution; By using a scheduler to select the corresponding device nodes based on the process script and agent configuration, actions can be issued to achieve automated execution of industrial tasks.

2. The orchestration method for automating industrial tasks as described in claim 1, characterized in that: The capability unit includes the minimum functional implementation steps that can be independently invoked in an industrial process; During operation, parameter configuration is achieved through template input parameters, static parameters, and dynamically transmitted parameters driven by the output of the previous capability unit; The template input parameters are editable parameter items used to configure the execution target values ​​of the functional unit steps without modifying the corresponding package code of the capability unit.

3. The orchestration method for automating industrial tasks as described in claim 2, characterized in that: The capability abstraction includes representing step segments in historical records as a temporal feature matrix containing sensors and actuators; Using the temporal feature matrix, an intensity indicator representing the intensity level of a step segment is calculated, and the intensity indicator is input into the suppression network to generate a suppression gate; After extracting intermediate temporal features from the step segment using a temporal feature extractor, the suppression gate is introduced as an attention penalty term during the multi-head self-attention calculation process to obtain a temporal feature representation with suppressed parameter influence. Based on the temporal feature representation with suppressed parameter influence, a temporal encoder is used to output a fixed-dimensional work content vector as an abstract result; The parameter configuration generation process is as follows: Clustering is performed based on the implementation process of each step and the abstract results of the parameter configuration in the historical records; and based on the analysis of parameter characteristics during the clustering process, the static parameters and the dynamically transmitted parameters are divided. For each step segment of the sample, parameters other than the template input parameters are captured; samples in the same cluster are taken as the same capability unit, and the same parameters of each individual in the cluster are obtained as the static parameters of the capability unit, while other parameters are taken as the dynamic transmission parameters.

4. The scheduling method for automating industrial tasks as described in claim 3, characterized in that: Each capability unit is configured as an independently executable package, and each capability unit is configured with an instance configuration file and a capability metadata file; The instance configuration file is used to define the input and output parameter specifications of the capability instance; The capability metadata file is used to define the type, unique identifier, and logical association of the capability with other capabilities; During deployment and startup, the binary program package and related configuration files are placed in a preset directory, and the startup command is executed to enable the capability unit to complete service registration with a registry center and expose a set of standardized management interfaces to the outside world.

5. The scheduling method for automating industrial tasks as described in claim 4, characterized in that: The environment deployment includes: a control plane architecture and a data and network plane architecture; Control plane architecture: Deploy a distributed key-value storage cluster as a central data repository to persistently store task status and node information; deploy an API server as a system gateway to handle authentication and routing of all internal and external requests, and as the only component that can directly access the storage cluster; The deployment scheduler acts as the decision center, continuously monitoring the tasks to be scheduled in the API server and allocating appropriate execution nodes to the tasks based on resource requirements and policies. Deploy high-availability guardian components to ensure data consistency and fault recovery capabilities of the storage cluster under multi-node deployment; Data and network plane architecture: Deploy agent programs on multiple worker nodes to register node information with the control plane, report node status, listen for tasks assigned to the nodes, and drive local devices to complete actual operations. If feasible, deploy network proxy components on each node to build a cluster network plane, which is responsible for maintaining and updating network rules, and realizing service discovery, load balancing and cross-node communication traffic forwarding.

6. The orchestration method for automating industrial tasks as described in claim 5, characterized in that: The script file includes a three-level tree structure that defines the production process from task to process group to action; The action layer is used to bind specific device capabilities and input / output parameters; The input parameters of the action layer are bound through string addressing expressions, which enables the automatic passing of parameters between actions; Output parameters are bound to the data type template of the preset value through key names; During task execution, the system retrieves results from the device's capabilities using a preset template and persists the actual runtime data to the state storage, thus enabling the transfer of parameters.

7. The orchestration method for automating industrial tasks as described in claim 6, characterized in that: In the script's flow, a convolutional neural network pre-trained using historical records is used. Each capability unit generates the dynamic transfer parameters for this capability unit based on the template input parameters, static parameters, and dynamic transfer parameters of the previous capability unit. The initial values ​​of the dynamic transfer parameters for each capability unit are generated sequentially according to the flow order. During the operation phase of the action layer, after the operation layer of each capability unit completes its operation, the data is used for supervised learning of the convolutional neural network, thereby updating the dynamic transfer parameters used for the current operation based on the initial value of the dynamic transfer parameters. In this process, supervised learning only applies during the runtime phase. After the task is completed, the convolutional neural network reverts to the state before supervision.

8. A scheduling system for automating industrial tasks, based on the scheduling method for automating industrial tasks according to any one of claims 1 to 7, characterized in that: include, The abstract module breaks down the industrial process into multiple capability units according to the human-machine-material model, and performs capability abstraction to generate parameter configurations; The encapsulation module encapsulates the capability unit into an independently executable program package and performs capability registration; The deployment module, after completing the deployment of the scheduled environment, writes script files for automated execution; The execution module uses the scheduler to select the corresponding device nodes based on the process script and agent configuration to issue actions, thereby realizing the automated execution of industrial tasks.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the orchestration method for automating industrial tasks as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the orchestration method for automating industrial tasks as described in any one of claims 1 to 7.

Images