WAAM Welding Bead Modeling With Dynamic Parameters and Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for modeling welding beads in wire-arc additive manufacturing, such as orthogonal experiments and response surface methods, require numerous experiments and are prone to errors, failing to establish a reliable mathematical relationship between input and output parameters.
Innovation Solution
A deep learning model using a neural network is employed to express the relationship between forming process parameters and welding bead shape, combined with a dynamic parameter method for off-line training, reducing the need for multiple experiments by dynamically changing parameters within a single welding bead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If orthogonal experiment method is used to model welding bead, then the relationship between input and output parameters can be studied, but the number of experiments increases and it is difficult to obtain the mathematical relationship
Solution Approach 1:
The patent applies preliminary action by pre-collecting and organizing welding process data from historical manufacturing records before modeling. This pre-prepared dataset eliminates the need for time-consuming orthogonal experiments, allowing the neural network to be trained directly on existing comprehensive welding data, thus significantly reducing experimental time while maintaining model reliability
Solution Approach 2:
The patent uses neural network to create a virtual copy of the welding process that replicates the complex relationships between input parameters and welding bead outcomes. This digital twin approach replaces physical experiments with simulated predictions, eliminating the need for numerous actual welding experiments while preserving the ability to study parameter relationships
2Manufacturing precision
If response surface method is used to model welding bead, then the mathematical relationship between input and output can be obtained, but the number of experiments increases and parameter errors are amplified
Solution Approach 1:
The patent employs neural network to create a computational model that copies the complex nonlinear relationships in welding processes without requiring response surface experiments. This approach achieves high precision in parameter relationships by learning from data patterns rather than through iterative experimental design, eliminating time consumption while maintaining or improving precision
Solution Approach 2:
The patent transforms the modeling approach from experimental parameter variation (response surface method) to data-driven parameter learning (neural network). By changing from active experimental manipulation to passive data analysis, the method avoids amplifying parameter errors while still obtaining precise mathematical relationships through the network's ability to handle complex parameter interactions
3Ease of manufacture
If conventional modeling methods are used, then the welding bead model can be built, but complex regression equation selection and parameter optimization are required
Solution Approach 1:
The patent replaces the mechanical/mathematical system of regression equation selection and parameter optimization with a neural network system. Instead of manually selecting and tuning regression models, the neural network automatically learns the appropriate mathematical relationships from data, substituting complex analytical methods with a more straightforward data-driven approach that reduces modeling complexity
Solution Approach 2:
The neural network performs self-service by automatically selecting optimal parameter relationships and optimizing itself during training without requiring external intervention for equation selection or parameter tuning. This self-organizing capability eliminates the complex manual processes required by conventional methods, making the modeling process easier while maintaining high accuracy
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 approach provides sufficient training data for neural network modeling while minimizing experimental costs and times, avoiding complex regression equation selection and parameter optimization.
Implementation Method 1
acquiring the temperature of the welding bead in real time by an infrared thermal imager
Implementation Method 2
using a line laser sensor for scanning to obtain the segmented profile of the welding bead
Data Source
AI summary
A welding bead modeling method for wire-arc additive manufacturing, a device therefor and a system therefor including using a dynamic parameter method, and using different welding process parameters in the same welding bead in the a wire-arc additive manufacturing process to obtain a welding bead with synchronous and dynamic changes in profile along with the dynamic changes of the welding process parameters. The method further comprising using a line laser sensor for scanning to obtain the segmented profile of the processed welding bead, and corresponding each welding bead profile to the welding process parameters one by one to train the neural network as training data, so as to obtain a welding bead modeling model capable of obtaining the corresponding welding bead profile according to the input welding process parameters.