CN-122018897-A - Visual AI workflow arrangement method and system

CN122018897ACN 122018897 ACN122018897 ACN 122018897ACN-122018897-A

Abstract

The invention discloses a visual AI workflow arrangement method and a visual AI workflow arrangement system, which are characterized in that a task dependent flow chart is established by extracting application scene requirements of manufacturing industry and combining a prompt word template library, and a low code arrangement standard is determined; the method comprises the steps of identifying AI node types, detecting coupling strength among nodes, generating a dynamic workflow topology map, obtaining multimode interface configuration parameters, carrying out capability matching, forming a scheduling opportunity map through a difference compensation factor, realizing multimode collaborative scheduling, determining a dynamic analysis triggering condition based on an execution bottleneck position, identifying failed nodes, carrying out path reconstruction, generating a self-adaptive execution sequence, carrying out node execution efficiency analysis through manufacturing application verification, constructing a graphical node library, and generating dragging type arrangement interface data. The invention reduces the arrangement threshold of the AI workflow, improves the multi-model cooperation efficiency and enhances the self-adaptive optimization capability of the workflow.

Inventors

Peng Qianxing
CHEN LONGMING
HU SUMING
YANG XINYUAN

Assignees

无锡芯软智控科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251208

Claims (10)

1. A method of visual AI workflow orchestration, comprising: acquiring application scene requirements and prompt word template library resources of manufacturing industry, extracting task logic dependency relationships through the application scene requirements, and establishing low code arrangement standards by adopting the task logic dependency relationships and the prompt word template library resources; identifying an AI node type by adopting the low code arrangement standard, detecting the coupling strength of a data flow direction and a control flow direction between nodes according to the AI node type, generating a dynamic workflow topology map by using the coupling strength, and determining a node arrangement priority table according to the dynamic workflow topology map; Acquiring a multi-model interface configuration parameter and a unified agent layer rule, carrying out model capacity matching detection according to the unified agent layer rule to generate a model distribution scheme, and carrying out collaborative analysis on the multi-model interface configuration parameter and the model distribution scheme to form a multi-model agent scheduling attribute; identifying an execution bottleneck position based on the node arrangement priority table and the multi-model agent scheduling attribute, determining a dynamic analysis triggering condition by adopting the execution bottleneck position, and adjusting and generating a self-adaptive execution sequence when the workflow structure is operated by using the dynamic analysis triggering condition; And performing manufacturing application verification by adopting the self-adaptive execution sequence to generate quality inspection response data and maintenance response data, performing node execution efficiency analysis on the quality inspection response data and the maintenance response data to generate an arrangement optimization scheme, constructing a graphical node library by adopting the arrangement optimization scheme, and generating drag type arrangement interface data based on the graphical node library to complete the visual AI workflow arrangement service.
2. The method of claim 1, wherein said employing the task logic dependency relationship to establish a low code orchestration criteria with the hint word template library resource comprises: generating a task dependency flow map according to the task logic dependency relationship; Combining the prompting word template library resource and the task dependent flow chart to form a template-task mapping track; extracting an orchestratable constraint range in the template-task mapping track; the rules of the choreography constraint range are set as low code choreography criteria.
3. The method of claim 1, wherein generating a dynamic workflow topology map using the coupling strengths comprises: decomposing the coupling strength into a data dependent strength and a control dependent strength; constructing a dependency analysis path by combining the data dependency intensity and the control dependency intensity; performing redundant execution branch identification in the dependency analysis path to determine a fault-tolerant standby node; And selecting the node combination with the highest coupling degree in the fault-tolerant standby nodes to generate a dynamic workflow topology map.
4. The method of claim 1, wherein said co-analyzing the multi-model interface configuration parameters with the model allocation scheme to form multi-model proxy scheduling attributes comprises: performing capability difference analysis on the model allocation scheme to obtain a difference compensation factor; Load matching is carried out on the multimode interface configuration parameters according to the difference compensation factors to form load balancing measurement; Generating a scheduling opportunity pattern by time window interpolation aiming at the load balancing measurement; and forming multi-model agent scheduling attributes by using the scheduling opportunity patterns.
5. The method of claim 1, wherein said generating an adaptive execution sequence using said dynamic parsing trigger condition for workflow structure runtime adjustment comprises: detecting the workflow structural change aiming at the dynamic analysis triggering condition to obtain node adjustment requirements; identifying an execution failure node in the node adjustment requirement to generate a failure node mark; performing path reconstruction analysis by using the failure node mark to generate optimized path configuration; And generating an adaptive execution sequence according to the optimized path configuration.
6. The method of claim 1, wherein said employing said adaptive execution sequence for manufacturing application validation generates quality control response data and maintenance response data, comprising: Performing task assignment on the adaptive execution sequence to generate a quality inspection execution instruction; Triggering multi-model parallel detection to generate multi-path quality inspection results by using the quality inspection execution instruction; extracting detection output with highest confidence from the multi-path quality inspection results to form quality inspection response data; and performing equipment state association analysis based on the quality inspection response data to generate maintenance response data.
7. The method of claim 1, wherein said employing the orchestration optimization scheme to build a graphical node library comprises: carrying out node classification and identification by adopting the arrangement optimization scheme to obtain node type labels; adjusting node display attributes by using the node type labels to form a visual configuration rule; Generating interactive response mapping by drag interaction adaptation through the visual configuration rule; and combining the node type label with the interactive response mapping to construct a graphical node library.
8. A method according to claim 3, wherein said performing redundant execution branch identification in said dependency analysis path to determine a fault tolerant standby node comprises: acquiring node execution branches in the dependency analysis path; Identifying functionally equivalent redundant branches among said node execution branches; Performing fault tolerance capability evaluation on the redundant branches to generate fault tolerance priority identifiers; and marking the redundant branches as fault-tolerant standby nodes according to the fault-tolerant priority identifiers.
9. The method of claim 7, wherein the employing the orchestration optimization scheme for node classification identification to obtain node type labels comprises: performing cross-level node tracking on the arrangement optimization scheme to acquire node execution statistical data; analyzing calling frequency characteristics of each node attribute according to the node execution statistical data, and constructing a node priority association table, wherein each node attribute comprises a prompt word engineering node, a model calling node and a logic judging node; and carrying out classified projection on the node priority association table to generate a node type label.
10. A visual AI workflow orchestration system, comprising: The requirement analysis module is used for acquiring application scene requirements and prompt word template library resources of the manufacturing industry, extracting task logic dependency relationships through the application scene requirements, and establishing low code arrangement standards by adopting the task logic dependency relationships and the prompt word template library resources; The topology construction module is used for identifying the type of the AI node by adopting the low code arrangement standard, detecting the coupling strength of the data flow direction and the control flow direction among the nodes according to the type of the AI node, generating a dynamic workflow topology map by using the coupling strength, and determining a node arrangement priority table according to the dynamic workflow topology map; The agent scheduling module is used for acquiring the multi-model interface configuration parameters and the unified agent layer rules, carrying out model capacity matching detection according to the unified agent layer rules to generate a model distribution scheme, and carrying out collaborative analysis on the multi-model interface configuration parameters and the model distribution scheme to form multi-model agent scheduling attributes; The dynamic analysis module is used for identifying the execution bottleneck position based on the node arrangement priority table and the multi-model agent scheduling attribute, determining a dynamic analysis triggering condition by adopting the execution bottleneck position, and adjusting and generating a self-adaptive execution sequence when the workflow structure is operated by using the dynamic analysis triggering condition; And the interface generation module is used for carrying out manufacturing application verification by adopting the self-adaptive execution sequence to generate quality inspection response data and maintenance response data, carrying out node execution efficiency analysis on the quality inspection response data and the maintenance response data to generate an arrangement optimization scheme, constructing a graphical node library by adopting the arrangement optimization scheme, and generating drag type arrangement interface data based on the graphical node library to complete the visual AI workflow arrangement service.

Description

Visual AI workflow arrangement method and system Technical Field The invention relates to the technical field of artificial intelligence application, in particular to a visual AI workflow arrangement method and system. Background With the deep advancement of intelligent transformation in manufacturing industry, more and more enterprises begin to try to apply AI technology such as large language models to production scenes such as intelligent quality inspection, predictive maintenance and process optimization. However, these AI applications often involve a combined orchestration of multiple task nodes, including multiple links such as data collection, model reasoning, result analysis, flow control, etc., with complex data dependency and control flow relationships between the different nodes. The traditional AI workflow construction mode relies on professional algorithm engineers to write codes manually, so that the development period is long, the cost is high, the flexibility is lacking, and the quick response to the change of the service requirement is difficult. While existing workflow orchestration tools provide some visualization capabilities, there are significant shortcomings in AI task orchestration. These tools are mainly directed to general business processes, and lack of node classification and arrangement specifications for AI task characteristics leads to confusion of workflow structures and difficulty in maintenance. In the aspect of multi-model scheduling, the prior art generally does not support collaborative scheduling of a plurality of large language models, cannot dynamically allocate according to model capacity difference and load conditions, and is easy to cause performance bottleneck. More importantly, once deployed and operated, the workflow is fixed and cannot be adjusted in operation according to the bottleneck position and failure condition in the execution process, and the self-adaptive optimization capability is lacking. There is therefore a need for a method to address at least one of the above problems. Disclosure of Invention The invention discloses a visual AI workflow arrangement method and a visual AI workflow arrangement system, which aim to realize intelligent scheduling and load balancing of multiple models by establishing low code arrangement standards and unified agent layer rules, generate a self-adaptive execution sequence by dynamically analyzing execution bottlenecks and reconstructing paths during operation, reduce the arrangement threshold of AI workflows by constructing a graphical node library and a dragging arrangement interface, enable business personnel to quickly construct, deploy and optimize AI application workflows of manufacturing scenes without coding, and provide efficient, flexible and easy arrangement services for intelligent quality inspection, predictive maintenance and other applications. The first aspect of the present invention provides a visual AI workflow editing method, comprising the steps of: acquiring application scene requirements and prompt word template library resources of manufacturing industry, extracting task logic dependency relationships through the application scene requirements, and establishing low code arrangement standards by adopting the task logic dependency relationships and the prompt word template library resources; identifying an AI node type by adopting the low code arrangement standard, detecting the coupling strength of a data flow direction and a control flow direction between nodes according to the AI node type, generating a dynamic workflow topology map by using the coupling strength, and determining a node arrangement priority table according to the dynamic workflow topology map; Acquiring a multi-model interface configuration parameter and a unified agent layer rule, carrying out model capacity matching detection according to the unified agent layer rule to generate a model distribution scheme, and carrying out collaborative analysis on the multi-model interface configuration parameter and the model distribution scheme to form a multi-model agent scheduling attribute; identifying an execution bottleneck position based on the node arrangement priority table and the multi-model agent scheduling attribute, determining a dynamic analysis triggering condition by adopting the execution bottleneck position, and adjusting and generating a self-adaptive execution sequence when the workflow structure is operated by using the dynamic analysis triggering condition; And performing manufacturing application verification by adopting the self-adaptive execution sequence to generate quality inspection response data and maintenance response data, performing node execution efficiency analysis on the quality inspection response data and the maintenance response data to generate an arrangement optimization scheme, constructing a graphical node library by adopting the arrangement optimization scheme, and generating drag type arrangeme