Welding Defect Prediction Using Combined ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing welding technologies face challenges in accurately predicting defects such as cavities and melting failures due to the complexity of factors influencing welding quality, particularly in additive manufacturing, making it difficult to detect and predict these defects effectively.
Innovation Solution
A defect prediction system utilizing two trained models, one using a random forest algorithm for discrete values and another using Gaussian process regression for continuous values, to predict welding defects by integrating their outputs for enhanced accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single machine learning model is used to predict welding defects, then the prediction process is simple, but the prediction accuracy is insufficient due to the complexity of factors affecting welding defects
Solution Approach 1:
The patent combines multiple machine learning models (first trained model and second trained model) into a unified prediction system. The first prediction unit and second prediction unit process welding parameters through different models, and the third prediction unit integrates their outputs to predict defect size, achieving higher accuracy than a single model could provide
Solution Approach 2:
The prediction system is divided into three distinct prediction units with specialized functions: the first prediction unit handles discrete value predictions, the second prediction unit handles continuous value predictions, and the third prediction unit synthesizes these results. This segmentation allows each unit to optimize for its specific task while contributing to the overall prediction accuracy
2Measurement precision
If multiple trained models are used to predict welding defects, then the prediction accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The welding parameters are pre-processed and prepared before being input to the prediction units. The first and second trained models have already been trained offline on historical welding data, so during actual prediction, only inference is performed. The third prediction unit integrates results using predetermined algorithms, avoiding the need for real-time complex computations
Data Source
AI summary
A defect prediction system for predicting a defect occurring in welding includes: a first prediction unit configured to predict the defect using a first trained model that receives a welding parameter and outputs a parameter indicating a size of the defect; a second prediction unit configured to predict the defect using a second trained model that receives the welding parameter and outputs a parameter indicating a size of the defect; and a third prediction unit configured to predict the size of the defect based on a parameter predicted by the first prediction unit and a parameter predicted by the second prediction unit.