Multi-Stage Quality Prediction for Manufacturing Parameter Optimization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing manufacturing processes face challenges in optimizing product quality across multiple stages while minimizing resource consumption, as adjustments in one stage can adversely affect quality in subsequent stages.

Innovation Solution

A neural network-based system predicts quality parameters at subsequent stages using process parameters from previous stages, applying constraints to optimize quality across stages and minimize resource use, utilizing a U-shaped barrier function and graph neural networks to adjust parameters.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If multiple quality parameters are optimized simultaneously across stages, then overall product quality improves, but computational complexity increases

Engineering Contradiction:
Improveoverall product qualityVSAvoidcomputational complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The system applies segmentation by dividing the manufacturing process into distinct stages, with each stage represented by a separate module in the neural network. This modular architecture allows the model to process and optimize quality parameters for each stage independently while maintaining the relationships between stages, reducing the computational burden compared to treating the entire process as a single complex system.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system transitions to another dimension by using a neural network that processes data across multiple dimensions simultaneously - handling multiple process parameters, multiple quality parameters, and multiple stages in a unified computational framework. This dimensional approach enables the system to optimize overall product quality by considering all factors together rather than sequentially, managing complexity through parallel processing in the neural network architecture.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4600772A1System for product quality prediction and optimization for multi-stage production processes
Publication Date: 2025.08.13 SIEMENS AG
  • EP4600772A1 patent drawingFigure 1
  • EP4600772A1 patent drawingFigure 2~4
  • EP4600772A1 patent drawingFigure 5

AI summary

A manufacturing system performs a method of manufacturing an item. The system includes a first stage for manufacturing an item, a second stage for manufacturing the item, wherein the first stage precedes the second stage, and a processor. The processor obtains a first process parameter for a first stage of a manufacturing process, creates a model that predicts a quality parameter for the second stage based on the first process parameter, determines, using the model, a value of the first process parameter at which a different between the predicted quality parameter and a target quality parameter is less than a selected criterion, and adjusts the first process parameter to the value. The item is manufactured with the value of the first process parameter.