Quaternion Neuron Network for Multispectral Welding Image Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional laser welding quality detection methods based on visible light images are prone to interference and lack comprehensive information, making it difficult to accurately analyze the molten pool and identify penetration states during welding.
Innovation Solution
A quaternion multi-degree-of-freedom neuron-based multispectral welding image recognition method using three cameras with glass filters and protective glasses to capture images across different spectral bands, processing these images to extract low-frequency features and establish a quaternion-based edge model, followed by classification using a multi-degree-of-freedom neuron network for improved recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional visible light image methods are used for welding detection, then the detection system is simple, but the identification reliability is low due to arc light interference and lack of comprehensive molten pool information
Solution Approach 1:
The detection system is segmented into multiple independent spectral detection channels (ultraviolet, visible, infrared) with separate cameras and filters for each band. This allows each channel to capture specific spectral information independently, improving reliability through multi-source information fusion while maintaining modular system architecture that manages complexity
Solution Approach 2:
Multiple spectral bands (ultraviolet, visible, infrared) are merged into a unified detection system that captures comprehensive molten pool information simultaneously. The fusion of these different spectral information sources provides redundant and complementary data, significantly improving identification reliability compared to single-band methods
2Loss of information
If single band visual information is used, then the detection system is simple, but the information completeness about the molten pool is insufficient
Solution Approach 1:
The detection system transitions from single-band (one-dimensional spectral information) to multi-band (multi-dimensional spectral information) by adding ultraviolet and infrared dimensions alongside visible light. This dimensional expansion captures comprehensive molten pool characteristics including temperature distribution, chemical composition, and physical state that are invisible in any single band
Solution Approach 2:
The multi-spectral detection system performs multiple detection functions simultaneously: ultraviolet band detects chemical composition and plasma characteristics, visible band captures molten pool morphology and edge information, and infrared band measures temperature distribution. This multi-functionality ensures complete information capture without requiring separate specialized systems
3Measurement precision
If visible light images are used for molten pool detection, then the system is simple, but the detection precision is low due to arc light interference
Solution Approach 1:
The arc light interference, which is harmful to visible light detection, is converted into a beneficial signal by detecting it in the ultraviolet band where arc emission is strongest. The ultraviolet camera captures arc characteristics that provide valuable information about welding process stability and chemical composition, transforming the interference into useful diagnostic data
Solution Approach 2:
Different spectral bands are assigned to detect different local characteristics of the molten pool: ultraviolet for chemical composition and plasma properties, visible for morphology and edge information, infrared for temperature distribution. This localized detection strategy ensures high precision for each specific measurement by using the optimal spectral band for that particular property
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method provides higher recognition accuracy and reduces interference by integrating multi-spectral information, enhancing the detection of welding penetration states with reduced susceptibility to interference compared to traditional methods.
Implementation Method 1
the three cameras equipped with glass filters and protective glasses are arranged side by side and vertically aligned with a portion to be subject to laser welding
Data Source
AI summary
Disclosed is a quaternion multi-degree-of-freedom neuron-based multispectral welding image recognition method, comprising: using three cameras having different wavebands to obtain multispectral weld pool images, and respectively performing pre-processing and edge extraction on the weld pool images having the different wavebands obtained at a same moment by the three cameras; establishing a quaternion-based multispectral weld pool image edge model; extracting low-frequency features after a quaternion discrete cosine transform; using a quaternion-based multi-degree-of-freedom neuron network to perform classification, training and recognition on edge features of the multispectral weld pool images. Compared to traditional means, the present invention has multiple recognition information sources, strong anti-interference capabilities and high recognition accuracy.