Molding Machine Component Life Inference Using Adaptive Model Selection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for predicting the life of components in molding machines and estimating the degree of abnormality or quality of molding products face challenges in determining optimal prediction models and collecting sufficient data for training.

Innovation Solution

An estimation method that acquires physical quantity data from molding machine components, prepares multiple estimation models using different algorithms, and selects one or more models to estimate the life, abnormality degree, or quality of the components or products based on the acquired data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a single optimal life prediction model is determined and trained, then prediction accuracy is improved, but data collection requirements and model selection complexity increase

Engineering Contradiction:
Improveprediction accuracyVSAvoidmodel selection complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the prediction task by dividing it into multiple independent estimation models, each specialized for different data conditions. Instead of selecting one optimal model, the system creates several models trained on different data subsets or with different algorithmic approaches, then selects the appropriate model based on the current data availability, thereby reducing the complexity of finding a single universal optimal model while maintaining high prediction accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent changes the parameter of model selection by introducing multiple estimation models with different characteristics (e.g., different algorithms, different training data requirements). The system dynamically selects which model to use based on parameters such as data amount and data quality, allowing the prediction system to adapt to varying conditions without requiring a single complex optimal model.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If multiple estimation models are prepared with different algorithms, then estimation accuracy is improved, but system complexity increases

Engineering Contradiction:
Improveestimation accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces dynamics into the model selection process by making the choice of estimation model adaptive rather than static. The system evaluates current conditions (data amount, data quality, component type) and dynamically selects the most appropriate model from the plurality of available models. This dynamic selection approach allows the system to maintain high estimation accuracy across varying conditions without permanently maintaining complex infrastructure for all possible models simultaneously.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system performs self-service by automatically selecting the appropriate estimation model based on the characteristics of the input data and prediction requirements. Rather than requiring external intervention to choose the best model, the system autonomously evaluates the situation and selects the most suitable model from its plurality of models, reducing the operational complexity of managing multiple models while maintaining high estimation accuracy.

Inventive Principle:
Principle #25Self-service

3Reliability

If data collection is increased for model training, then prediction reliability is improved, but data collection time and resources increase

Engineering Contradiction:
Improveprediction reliabilityVSAvoiddata collection time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent applies partial action by training multiple estimation models on different subsets of available data rather than requiring all models to be trained on complete datasets. Each model is trained on a portion of the data that is most suitable for its specific algorithm and prediction task. This approach achieves reliable predictions using partial data collections, reducing the total time and resources required for data collection while maintaining or improving prediction reliability through the diversity of models.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4556881A1Inference method, inference device, and computer program
Publication Date: 2025.05.21 THE JAPAN STEEL WORKS LTD
  • EP4556881A1 patent drawingFigure 1
  • EP4556881A1 patent drawingFigure 2
  • EP4556881A1 patent drawingFigure 3

AI summary

The present invention involves: acquiring physical quantity data relating to the state of a member that constitutes a molding machine; preparing a plurality of inference models for inferring, by different algorithms, the lifetime or abnormality degree of the member or the quality of a molded article; inferring, by using at least one of the inference models selected from the plurality of inference models, the lifetime or abnormality degree of the member or the quality of a molded article on the basis of the acquired physical quantity data.