AI Weld Quality Prediction from Melt Pool and Keyhole Shapes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting weld quality in battery module manufacturing, such as laser welding, lack accuracy and are not applicable to all products, often resulting in weld failures due to poor quality, especially when subjected to vibration.
Innovation Solution
An apparatus and method utilizing a learned artificial intelligence model, including a shape detection model, depth prediction model, and strength prediction model, which uses an image sensor and distance sensor to acquire and analyze images and penetration depth of the welding area, predicting tensile strength through a deep neural network structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional welding quality checking methods (cutting welded samples) are used, then the manufacturing process is simple, but the method cannot be applied to all products and lacks accuracy for real-time quality assessment
Solution Approach 1:
The patent replaces traditional mechanical cutting and physical inspection methods with an optical detection system using cameras and image processing algorithms. The system captures images of the welding area, extracts features from melt pool and keyhole morphology, and uses machine learning models to predict weld quality parameters, eliminating the need for physical sample cutting and enabling non-contact, real-time quality assessment.
Solution Approach 2:
The patent introduces an intermediary computational model that bridges the gap between optical images and weld quality parameters. The system uses image processing to extract morphological features, then applies trained neural network models to predict penetration depth and tensile strength, serving as an intermediary between visual observation and quality assessment that enables accurate prediction without direct physical measurement.
2Measurement precision
If simple image-based detection is used, then the device is simple, but the prediction accuracy of weld quality parameters is insufficient
Solution Approach 1:
The patent segments the weld quality prediction task into multiple independent stages: image acquisition, feature extraction (melt pool shape, keyhole morphology), penetration depth prediction, and tensile strength prediction. Each stage is handled by a specialized computational module or neural network, allowing the system to achieve high accuracy through sequential processing of different aspects of weld quality rather than attempting to predict all parameters simultaneously.
Solution Approach 2:
The patent performs preliminary action by training machine learning models on datasets of welded samples with known properties before actual production use. The system pre-processes image data to extract morphological features and pre-trains neural networks to recognize patterns correlating with penetration depth and tensile strength, so that during production, the pre-trained models can rapidly predict quality parameters without requiring complex real-time computation.
3Productivity
If real-time weld quality prediction is implemented, then productivity is improved through immediate feedback, but the computational requirements and processing time increase
Solution Approach 1:
The patent applies partial action by focusing computational resources on extracting only the most critical features from welding images that have the strongest correlation with weld quality, such as keyhole morphology and melt pool characteristics. Rather than analyzing all image data in full detail, the system identifies and processes only the essential features needed for accurate prediction, reducing computation time while maintaining prediction accuracy.
Data Source
AI summary
An apparatus for predicting weld quality includes an image sensor acquiring an image of a welding portion of a workpiece, a storage storing a shape detection model, a depth prediction model and a strength prediction model, and a controller configured to acquire shapes of a melt pool and a keyhole by receiving the image and inputting the image to the shape detection model, acquire a penetration depth of the keyhole by inputting the shapes of the melt pool and the keyhole to the depth prediction model, and acquire a tensile strength by inputting one of the shape of the melt pool, the shape of the keyhole and the shapes of the melt pool and the keyhole and the penetration depth of the keyhole to the strength prediction model. The weld quality of all products may be predicted during a welding process.