Machine-Learning Quality Evaluation for Confidential Manufacturing Processes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In complex supply chains, manufacturers lack effective methods to improve production processes without disclosing sensitive production parameters, leading to inefficient improvements based on guesswork and increased workload due to information leakage among competing entities.
Innovation Solution
An evaluation system comprising learning and evaluation apparatuses that generate estimation models to assess the quality impact of production parameters, allowing targeted process improvements without disclosing specific parameters, using machine learning algorithms to identify key factors affecting downstream product quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manufacturers share production parameters to improve process quality, then quality improvement effectiveness is enhanced, but information security and confidentiality are compromised
Solution Approach 1:
The patent introduces a third-party intermediary system that receives production parameters from multiple manufacturers, performs automated quality impact analysis using machine learning models, and returns analysis results without disclosing other manufacturers' parameters. This intermediary acts as a mediator that enables quality improvement collaboration while maintaining information security, as manufacturers can improve their processes based on aggregated insights without exposing sensitive proprietary data.
Solution Approach 2:
The system creates analytical copies of production parameters through machine learning models that simulate quality impacts without requiring actual parameter sharing. Instead of exchanging real production data, the system uses copied representations and statistical models to analyze quality relationships, enabling manufacturers to gain insights from industry-wide patterns while maintaining the confidentiality of their actual production parameters.
2Loss of information
If manufacturers conduct independent quality analysis without information sharing, then information security is maintained, but quality improvement efficiency decreases due to guesswork
Solution Approach 1:
The system enables manufacturers to self-serve quality analysis by submitting their own production parameters to the intermediary system, which automatically performs the quality impact analysis using pre-trained machine learning models. This eliminates the need for manufacturers to manually conduct guesswork-based analysis or share parameters directly with competitors, providing automated, data-driven insights while maintaining information security and improving quality improvement efficiency.
3Measurement precision
If comprehensive production parameters are shared across the supply chain, then overall quality understanding is improved, but workload increases due to data collection and management
Solution Approach 1:
The system segments the quality analysis function into two parts: manufacturers locally collect and prepare their own production parameters (keeping data management simple), while the intermediary system performs the complex quality impact analysis using machine learning models. This segmentation allows comprehensive quality evaluation without requiring any single manufacturer to manage complex supply chain-wide data, as the analysis complexity is offloaded to the centralized system.
Solution Approach 2:
The intermediary system performs preliminary quality impact analysis on submitted parameters before returning results to manufacturers. By pre-processing and analyzing data centrally, the system eliminates the need for manufacturers to conduct separate, complex data collection and analysis operations, reducing their workload while maintaining high measurement precision through sophisticated machine learning models.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided is a learning apparatus comprising: a correspondence receiving unit for receiving a correspondence between each produced object of a target process which is targeted and a quality evaluation of each downstream produced object produced in a downstream process by using each produced object of the target process; a learning processing unit for generating, through learning, an estimation model for estimating the quality evaluation of the downstream produced object from at least one production parameter, by using the at least one production parameter about the production of each produced object of the target process and the quality evaluation of each downstream produced object produced by using each produced object of the target process; a calculating unit for calculating a model evaluation based on at least one of certainty or complexity of the estimation model; and a model evaluation sending unit for sending the model evaluation calculated by the calculating unit, to an evaluation apparatus for evaluating at least one upstream process by using a model evaluation about each of the at least one upstream process which is upstream of the downstream process.