Building Material Board Quality Prediction with Adaptive Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing production systems for building material plates face challenges in predicting quality parameters for low-production-share building material types and maintaining reliable predictions over time, especially due to seasonal changes.
Innovation Solution
A procedure that involves maintaining a mathematical model based on a minimum number of training data sets, which can include data from similar building material types, to predict quality parameters for building material plates. This model is updated by discarding outdated training data sets and incorporating new data to ensure accuracy and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a mathematical model is created specifically for a particular board type, then prediction accuracy for that board type is improved, but the time required to generate the model increases due to the need for a large number of board samples
Solution Approach 1:
The patent combines training data sets from multiple board types to create a unified mathematical model. When a specific board type has insufficient training data, the system merges data from other board types to reach the minimum required number of training samples, enabling model creation without excessive delays while maintaining adequate prediction accuracy.
Solution Approach 2:
The patent creates a universal mathematical model that can be applied across multiple board types. Instead of creating separate specialized models for each board type, the system develops a multi-functional model that serves various board types, reducing the overall time required for model generation while maintaining prediction accuracy.
2Reliability
If a mathematical model is created for a building material panel type with small production share, then prediction capability for that specific type is improved, but the availability of sufficient training data deteriorates
Solution Approach 1:
The patent merges training data sets from multiple board types to compensate for the limited production volume of specific board types. By combining data from different board types, the system accumulates sufficient training samples even for low-production board types, enabling reliable prediction models to be created.
Solution Approach 2:
The patent uses data from high-production board types as an intermediary resource to support model creation for low-production board types. The training data from commonly produced boards serves as a supplemental source that bridges the data gap for rare board types.
3Productivity
If a mathematical model is used for predictions over time, then production efficiency is improved, but prediction reliability deteriorates due to seasonal changes and data aging
Solution Approach 1:
The patent implements periodic updates of the mathematical model by systematically replacing outdated training data sets with new data. The system periodically refreshes the training data to account for seasonal changes and process variations, maintaining prediction reliability while preserving the productivity benefits of using mathematical models.
Solution Approach 2:
The patent discards outdated training data sets that have become obsolete due to seasonal changes or process variations. By removing aged data and incorporating fresh data, the system recovers and maintains prediction accuracy, ensuring the mathematical model remains reliable for ongoing production efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed is, inter alia, a method comprising providing a mathematical model for use in the manufacture of at least one building material panel, wherein the mathematical model is based on a number of training data sets equal to or greater than a minimum number; wherein a training data set for a building material panel type comprises quality parameter data representing at least one value for at least one corresponding quality characteristic of a building material panel sample of the building material panel type, and process parameter data representing values for a plurality of process parameters of a production process for manufacturing the building material panel sample;wherein the number of training data sets includes at least one training data set for a building material panel type of the at least one building material panel, wherein the method further comprises obtaining prediction data based on the mathematical model, wherein the prediction data represent a prediction of at least one quality parameter for the at least one building material panel; and outputting the prediction data.