Laser Machining Analysis Using Neural Networks for Weld Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional laser machining monitoring systems lack the ability to non-destructively determine physical properties of machining results, such as strength and conductivity, requiring post-process material testing that often damages workpieces and cannot be applied to all products.
Innovation Solution
A method using a trained neural network to quantify physical properties of laser machining processes by analyzing sensor data sets, allowing for real-time prediction of values like tensile strength, conductivity, and welding depth without destructive measurements, using a transfer function formed by a deep convolutional neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional monitoring systems use in-process monitoring with sensor data acquisition and reference curve comparison, then process monitoring capability is improved, but the ability to non-destructively determine physical properties like strength and conductivity deteriorates
Solution Approach 1:
A neural network acts as an intermediary between sensor data and physical property determination. The neural network is trained to map sensor data patterns to physical properties such as strength and conductivity, enabling non-destructive measurement without direct contact with the workpiece. This intermediary system bridges the gap between process monitoring and physical property assessment.
Solution Approach 2:
The system transforms sensor data parameters into physical property predictions through neural network processing. By changing the processing approach from simple threshold comparison to complex pattern recognition, the system can infer physical properties that were previously only measurable through destructive testing.
2Measurement precision
If post-process material testing is used to determine physical properties, then measurement accuracy is improved, but workpiece destruction and inability to test all products deteriorates
Solution Approach 1:
The system replaces mechanical destructive testing with optical/electromagnetic sensor-based monitoring. Instead of physically testing the workpiece after machining, the system uses sensor data acquired during the machining process and processes it through a neural network to predict physical properties, eliminating the need for destructive mechanical testing.
Solution Approach 2:
The neural network creates a virtual model or copy of the physical testing process. By training the network on data from both sensor measurements and corresponding destructive test results, the system learns to predict physical properties from sensor data alone, creating a non-destructive copy of the measurement capability.
3Device complexity
If conventional monitoring systems analyze data independently with expert-defined parameters, then system complexity is reduced, but monitoring performance and reliability deteriorates
Solution Approach 1:
The system changes from using simple, expert-defined parameters to using complex, multi-dimensional sensor data patterns processed by a neural network. This parameter transformation enables the system to capture subtle relationships in the data that improve monitoring reliability while the neural network handles the complexity automatically.
Solution Approach 2:
The neural network performs self-learning and automatic parameter optimization without requiring continuous expert intervention. The system automatically adjusts its internal parameters and weightings based on the sensor data patterns it observes, enabling reliable monitoring while reducing the need for complex manual configuration.
Data Source
AI summary
A method for analyzing a laser machining process for machining workpieces includes the steps of acquiring at least one sensor data set for the laser machining process and determining a value of at least one physical property of a machining result of the laser machining process based on the at least one sensor data set using a transfer function. The transfer function is formed by a trained neural network. A system for analyzing a laser machining process and a laser machining system including such a system are also disclosed.