CN-121980217-A - Industrial model cross-scene application system and method for industrial Internet

CN121980217ACN 121980217 ACN121980217 ACN 121980217ACN-121980217-A

Abstract

The invention relates to the technical field of industrial Internet platforms, and particularly discloses an industrial model cross-scene application system and method for industrial Internet. The application system is based on an industrial Internet platform and comprises an asset center module, a development center module and a system management module, wherein the asset center module is used for carrying out unified nano-tube on accessed data, algorithms, features and models and providing asset multiplexing service, the development center module is used for constructing an inventory model quick optimizing mechanism and an increment model quick developing mechanism and providing asset precipitation service, and the system management module is used for creating, deleting, modifying and inquiring tasks of a user, providing asset authority management service for the asset center module and providing computing resource management service for the development center module. The invention is based on stock industrial models and/or data resources, and can quickly and economically construct industrial models applicable to different target scenes.

Inventors

WANG JIACHEN
ZHOU YANG
CHEN MINGREN
MA YUE
Hu Ruofan
ZHANG HENGWEI

Assignees

东方电气集团数字科技有限公司
东方电气股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260116

Claims (10)

1. An industrial model cross-scene application system for industrial internet, which is characterized in that: The application system is based on an industrial Internet platform and comprises an asset center module, a development center module and a system management module; The asset center module is used for carrying out unified nano-tube on accessed data, algorithm, characteristics and model and providing asset multiplexing service for the development center module; the development center module is used for constructing a stock model rapid optimizing mechanism and an incremental model rapid developing mechanism and providing asset precipitation service for the asset center module; The stock model rapid optimizing mechanism is used for taking the technical requirement of a target scene as a benchmark, adjusting the managed stock industrial model and the data asset with the highest adaptation degree from an asset center module, and performing iterative optimization on the adjusted stock industrial model by a model increment training mechanism until the adjusted stock industrial model is matched with the target scene; the incremental model rapid development mechanism is to develop an incremental model by using data assets, algorithm assets and characteristic assets which are called in an asset center module and accord with the technical characteristics of a target scene; the system management module is used for creating, deleting, modifying and inquiring tasks of a user, providing asset authority management service for the asset center module and providing computing resource management service for the development center module.
2. The industrial model cross-scenario application system for the industrial internet of claim 1, wherein: the asset center module is provided with a data asset module for carrying out unified nano-tube on accessed marked data and unmarked data; The asset center module is provided with an algorithm asset module for carrying out unified nano-tube on the accessed visual development operator and the online algorithm template; the asset center module is provided with a characteristic asset module for carrying out unified nano-tube on the accessed industrial characteristics including the industrial-level characteristics, the product-level characteristics and the atomic-level characteristics; The asset center module has a model asset module that unifies nanotubes to an accessed industrial model including a classified industrial model, a detected industrial model, a predicted industrial model.
3. The industrial model cross-scenario application system for the industrial internet of claim 1, wherein: The application system also comprises an operation center module, and the system management module provides operation service management service for the operation center module; The operations center module serves as a capability verification for a selected industrial model of the asset center module and as a service release for open model capability for a selected industrial model meeting capability requirements.
4. An industrial model cross-scenario application system for the industrial internet according to claim 3, wherein: the operation center module is provided with a model evaluation module and a model reasoning module which are used for carrying out capability verification on the selected industrial model of the asset center module; The model evaluation module evaluates the capacity of the selected industrial model in a data evaluation mode or a fixed model evaluation mode; The model reasoning module performs online reasoning on the selected industrial model in an image reasoning mode or a numerical value prediction mode.
5. The industrial model cross-scene application system for the industrial internet according to claim 3 or 4, wherein: The operation center module is provided with a model deployment module and a model service module for performing service release on the selected industrial model of the asset center module; the model deployment module performs service release on the selected industrial model meeting the capability requirement in a cloud model service release mode or an edge side cooperative deployment mode; the model service module provides overview and detailed multi-view display service for the industrial model of service release.
6. An application method of the industrial model cross-scene application system according to any one of claims 1 to 5, wherein the application method comprises an asset entry creation process after logging on a platform, a new scene industrial model construction requirement access process, a model evaluation task creation process and a model open task creation process; Wherein the asset entry creation process generates an open asset reuse service; The new scene industrial model building demand access process selects available data assets and industrial model resources or available data assets, characteristic assets and algorithm asset resources from the opened asset multiplexing service of the asset item creation process to perform model development; the model evaluation task creation process is used for performing evaluation and reasoning capability verification on the selected industrial model; And the model opening task creation process performs capability opening service release on the selected industrial model.
7. The method of claim 6, wherein the asset entry creation process comprises the steps of: Step 1, creating asset items; Step 2, under the created asset item, selecting the type of the data asset, the algorithm asset, the characteristic asset or the model asset; Step 3, uploading corresponding asset files; Step 4, configuring description information of the asset file; step 5, completing asset creation of the industrial model; And 6, opening the asset multiplexing service.
8. The method for applying the industrial model across-scene application system according to claim 6, wherein the new scene industrial model construction requirement access process comprises the following steps: Step A, starting a new scene industrial model construction by taking technical requirements including functions and performances of a target scene as a benchmark; Step B, screening whether available stock industrial model resources exist from the opened asset multiplexing service in the asset item creation process based on the similarity of the technical requirements; If available stock industrial model resources exist, executing the following steps C-1 to E-1; if no available stock industrial model resources exist, executing the following steps C-2 to E-2; step C-1, creating a stock model rapid optimizing task; Step D-1, introducing available stock industrial models and data asset resources into the tasks created in the step C-1 from the opened asset multiplexing service of the asset entry creation process; E-1, training and configuring the stock industrial model to match a target scene based on the stock industrial model and the data asset resources introduced in the step D-1 by using strategies of feature enhancement, incremental training and parameter training; step C-2, creating an incremental model rapid development task; Step D-2, introducing available data assets, characteristic assets and algorithm asset resources into the task created in the step C-2 from the opened asset multiplexing service of the asset entry creation process; E-2, performing model development and training configuration by using a migration operator library, a general/field feature library and a data/style migration library based on the data asset, the feature asset and the algorithm asset resources introduced in the step D-2; f, configuring based on the results of the steps E-1 and E-2, and executing an industrial model training task; And H, outputting a training result of the industrial model to finish cross-scene development of the industrial model.
9. The method of claim 6, wherein the model evaluation task creation process comprises the steps of: Step a, creating an industrial model assessment task; selecting an industrial model evaluation mode of data evaluation or model evaluation; Step c, performing industrial model evaluation in a selected evaluation mode; step d, obtaining an evaluation result of the industrial model; step e, creating an industrial model reasoning task; step f, selecting an industrial model reasoning mode of image reasoning or numerical value prediction; Step h, performing industrial model reasoning in a selected reasoning mode; Step i, obtaining an inference result of the industrial model; and j, finishing the capability verification of the industrial model.
10. The method for applying an industrial model across-scene application system according to claim 6, wherein the model open task creation process comprises the steps of: step I, creating an open task of the industrial model; step II, selecting an industrial model to be opened; step III, selecting an industrial model capacity opening mode which is published by cloud model service or cooperatively deployed by edge sides; Step IV, executing the industrial model with the service overview and the service details; and V, finishing the opening of the model capacity.

Description

Industrial model cross-scene application system and method for industrial Internet Technical Field The invention relates to the technical field of industrial Internet platforms, in particular to an industrial model cross-scene application system and method for an industrial Internet platform. Background Industrial artificial intelligence is implemented based on an industrial model (industrial AI model), and thus development of the industrial model directly determines implementation of the industrial artificial intelligence. Currently, in the aspect of application scene construction of industrial artificial intelligence, technical pain points of poor application effect and high model migration cost of industrial models across scenes are commonly existed. The following is an analytical description taking the application of the welding defect intelligent detection model as an example. As a key link in industrial production, quality inspection work directly determines whether product quality defects can be found in time, and is a core process for preventing defective products from flowing into markets to cause safety production accidents. The traditional quality inspection relies on the manual work in a large number, and has large workload and individual subjective difference, so that consistency of working quality is difficult to ensure, and phenomena such as missed inspection and false inspection are easy to occur. In the aspect of quality detection of welding operation, by applying an artificial intelligence technology, welding defect samples under a target scene are collected, and an industrial model for identifying welding quality defects is trained by adopting a neural network, so that a better application effect can be obtained under the target scene, the artificial dependence degree of welding quality detection is reduced to a certain extent, and the working efficiency is remarkably improved. However, in the intelligent detection model for welding quality defects, a large number of defect negative film samples are required to be collected for training, and meanwhile, a large number of preprocessing works such as image processing and defect labeling are required to be carried out by professionals so as to construct a high-quality training data set. Therefore, the intelligent welding quality defect detection model with high precision and high efficiency is constructed with long period, low efficiency and high cost. Although the industrial model can obtain a better detection effect under the original target scene, when the industrial model is applied across scenes, the application effect of the industrial model is often degraded to different degrees due to factors such as data distribution/characteristic difference, weak model generalization capability and the like existing in different industrial scenes. Therefore, in order to meet the detection accuracy and efficiency requirements of the application scene, the detection model needs to be reconstructed according to the technical requirements of the new scene, which tends to result in repeated construction of the model capability and significant improvement of modeling cost. Disclosure of Invention Aiming at the development of the industrial model and the specificity in the cross-scene application, the technical aim of the invention is to provide an industrial model cross-scene application system for an industrial Internet platform, which is used for rapidly and economically constructing industrial models applicable to different target scenes by using stock industrial models and/or data resources, and an application method thereof. The technical aim of the invention is achieved by the following technical scheme, namely an industrial model cross-scene application system for industrial Internet, wherein the application system is based on an industrial Internet platform and comprises an asset center module, a development center module and a system management module; The asset center module is used for carrying out unified nano-tube on accessed data, algorithm, characteristics and model and providing asset multiplexing service for the development center module; the development center module is used for constructing a stock model rapid optimizing mechanism and an incremental model rapid developing mechanism and providing asset precipitation service for the asset center module; The stock model rapid optimizing mechanism is used for taking the technical requirement of a target scene as a benchmark, adjusting the managed stock industrial model and the data asset with the highest adaptation degree from an asset center module, and performing iterative optimization on the adjusted stock industrial model by a model increment training mechanism until the adjusted stock industrial model is matched with the target scene; the incremental model rapid development mechanism is to develop an incremental model by using data assets, algorithm assets and characteristic assets wh