Industrial Information Model Matching with Confidence Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for mapping human-readable industrial plant documentation into machine-readable formats are time-consuming, expensive, and lack reliability due to the uncertainty in training data used by generative models.
Innovation Solution
A computer-implemented method using a trained encoder to transform inputs or information models into an embedding space, comparing representations in this space to candidate models or inputs, and calculating a confidence measure to ensure accurate semantic suitability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If generative machine learning models (LLM/LVM) are used to map documentation to information models, then automation and processing speed are improved, but reliability and controllability of the mapping are worsened due to unknown training data domains
Solution Approach 1:
The patent introduces an intermediary confidence assessment system that mediates between the generative model's output and the final mapping decision. This intermediary layer evaluates the reliability of each mapping by comparing embeddings and calculating confidence scores, allowing the system to accept or reject mappings based on predefined thresholds, thus resolving the contradiction between automation speed and mapping reliability
Solution Approach 2:
The patent replaces manual verification of mapping reliability with an automated embedding-based confidence assessment mechanism. Instead of relying on unknown training data characteristics, the system uses vector space comparisons and similarity metrics to objectively evaluate mapping quality, substituting mechanical/manual processes with computational ones that provide controllable and measurable reliability
2Reliability
If manual methods are used to attribute information to semantic meanings, then reliability and control are improved, but time consumption and cost are worsened
Solution Approach 1:
The patent segments the mapping process into two distinct phases: an automated embedding generation phase that quickly produces candidate mappings, and a confidence assessment phase that selectively verifies reliability. This segmentation allows the system to benefit from both automated speed and manual-level reliability control without fully committing to either extreme, significantly reducing time consumption while maintaining high reliability
Solution Approach 2:
The patent applies partial action by performing full confidence assessment only on mappings that meet certain criteria or require verification, rather than manually reviewing every single mapping. The system processes the majority of mappings automatically with embedded confidence scores and only intervenes partially when needed, thus reducing overall time consumption while maintaining reliability standards
Data Source
Figure 1
Figure 2
AI summary
A computer-implemented method (100) for determining, for a given input (1) containing payload information (1a) about the layout, configuration and/or operating state of an industrial plant or any part thereof, an information model (5) into which the information contained in the given input fits, comprising the steps of: • transforming (110), by a trained encoder (2), the given input (1) into a representation (3) in an embedding space (4); • comparing (120) this representation (3) to representations (6) of multiple candidate information models (5*) in the same embedding space (4), wherein each information model (5*) identifies a collection of related information items (5a) with semantic meanings that at least partially characterizes the layout, configuration and/or operating state of the industrial plant or part thereof; • pre-selecting (130), based on the result (120a) of this comparison, from the multiple candidate information models (5*), one or more candidate information models (5#) into which the information contained in the given input (1) is likely to fit; • computing (140), for each pre-selected information model (5#), using a given confidence measure (7), a confidence (8) of the semantic suitability of the respective pre-selected information model (5#) for the given input (1); and • selecting (150) an information model (5#) with the best confidence (8) of the semantic suitability as the chosen information model (5) into which the information contained in the given input (1) fits. A corresponding method (200) for determining, for a given information model (5), from a plurality of candidate inputs (1*), a suitable input (1) for extracting payload information (1a).