AI Module Fleet Training Using Similar Operating Situations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence (AI) modules require significant training effort and often lack sufficient training data, especially when applied across multiple industrial plants or domains with varying conditions, leading to inefficiencies and performance inconsistencies.
Innovation Solution
A method for managing an installed base of AI modules by aggregating input-output data and configuration parameters from similar operating situations, using a quantitative similarity criterion to form augmented training and configuration parameters, and determining performance indicators to improve training efficiency and consistency across diverse applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If AI modules are trained individually for each plant or domain, then the training can be tailored to specific requirements, but the training effort and data requirements increase significantly
Solution Approach 1:
The patent combines training data and models across multiple plants and domains into a centralized knowledge base. Instead of training separate AI modules for each plant, the system aggregates similar operating situations and training data from all plants to create a unified training resource that can be reused across different applications, significantly reducing redundant training effort.
Solution Approach 2:
The invention creates a universal training platform that serves multiple functions: it stores training data from diverse plants, identifies similar operating situations across different domains, and provides reusable training resources. This universal system enables a single training effort to benefit multiple plants and applications simultaneously.
2Quantity of substance
If training data is collected from multiple plants and domains, then data scarcity is reduced, but the complexity of managing and processing diverse data increases
Solution Approach 1:
The patent segments the complex task of managing diverse training data by first identifying and categorizing similar operating situations across different plants and domains. This segmentation allows the system to organize heterogeneous data into manageable groups based on similarity criteria, making the data more accessible and easier to process while maintaining the benefits of having data from multiple sources.
3Adaptability or versatility
If AI modules are applied across different industrial plants with varying conditions, then the system achieves broader applicability, but performance consistency deteriorates
Solution Approach 1:
The invention applies local quality by identifying specific characteristics of operating situations that are most important for each plant or domain. The system uses similarity criteria to determine which features are locally relevant and weights them appropriately when selecting training data, ensuring that each AI module receives training focused on the specific conditions of its target application while still benefiting from the broader knowledge base.
Data Source
AI summary
A computer-implemented method for managing an installed base of multiple artificial intelligence (AI) modules, each AI module transforming input data into output data by an internal processing chain, the input data including data that characterize a state and/or the behavior, of at least one physical plant, apparatus, or system, the behavior of the internal processing chain being determined by a set of configuration parameters trainable based on training input data and corresponding reference data to which the internal processing chain is to map the training input data, the method including: obtaining, from multiple AI modules and/or domains in which the AI modules are applied, pairs of input data and corresponding output data, and/or pairs of training input data and corresponding reference data, and/or configuration parameters, the data and/or parameters relating to same or sufficiently similar operating situations of the AI modules according to a predetermined quantitative similarity criterion.