AI Splatter Detection in Laser Processing for Weld Defect Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing algorithms for splatter detection during laser processing are unreliable when splatters deviate from predefined geometrical or brightness features, leading to inaccurate or missed detections.
Innovation Solution
A method using a convolutional neural network in a U-Net architecture for splatter detection, trained with verified recordings, which identifies splatters and weld seam defects independently of temperature, geometry, and speed, enabling rapid and low-performance processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional image processing algorithms with rigidly predefined geometrical or brightness features are used, then the detection process is simple and fast, but the detection reliability deteriorates when splatters differ from predefined features
Solution Approach 1:
The patent transforms the detection approach by changing from fixed parameter matching (geometrical features, brightness features) to learned parameter extraction through neural networks. The system adapts parameters dynamically based on training data, allowing reliable detection of diverse splatter types without requiring explicit parameter definitions for each case.
Solution Approach 2:
The patent replaces conventional image processing algorithms with a neural network-based system. This substitution transitions from rule-based mechanical processing to intelligent, adaptive processing that can handle variability in splatter characteristics while maintaining automation.
2Measurement precision
If conventional image processing algorithms are used, then computational requirements are low, but detection accuracy deteriorates for varied splatter types
Solution Approach 1:
The patent applies preliminary training of the neural network offline using verified recordings of splatters. This preliminary action transfers learning to the model, enabling accurate detection during actual processing with reduced real-time computational burden. The system performs heavy computational work in advance rather than during production.
Solution Approach 2:
The patent uses verified recordings of splatters as training data to create a digital model that copies the characteristics of various splatter types. This allows the system to learn from examples without requiring exhaustive real-time analysis of every possible splatter variation.
3Adaptability or versatility
If rigidly predefined features are used for detection, then the system is easy to implement, but adaptability to different splatter types deteriorates
Solution Approach 1:
The patent implements a universal detection system using neural networks that can handle multiple splatter types through a single model. The trained network generalizes across different geometrical shapes, brightness levels, and splatter characteristics, replacing multiple specialized algorithms with one adaptable system.
Solution Approach 2:
The system changes from fixed parameter definitions to learned parameters that adapt to different splatter types. The neural network automatically adjusts detection parameters based on the training data, providing versatility without requiring manual configuration for each splatter type.
Data Source
AI summary
A method for identifying splatters or weld seam defects during laser processing of a workpiece includes processing the workpiece using a processing laser beam of a laser processing machine, compiling at least one recording of radiation emerging during the processing of the workpiece using an optical sensor that has a plurality of pixels for recording the radiation, and inputting the at least one recording into an evaluation unit. The evaluation unit has a machine learning algorithm configured as a convolutional neural network in a U-Net architecture. The machine learning algorithm has been trained with verified recordings of splatters or weld seam defects. The method further includes identifying one or more splatters or weld seam defects in the processing of the workpiece by running the machine learning algorithm using the at least one recording as input, and outputting an output indicating the identified one or more splatters or weld seam defects.