Welding stress-strain rapid prediction control method and system based on LSTM (Long Short Term Memory) network

The welding stress-strain prediction system based on LSTM network enables efficient and real-time monitoring and control of stress and strain during laser welding, solving the problems of low efficiency and high cost in traditional methods and improving welding quality and production efficiency.

CN122172726APending Publication Date: 2026-06-09HUAZHONG UNIV OF SCI & TECH
2 Cites 0 Cited by

Patent Information

Application Number
CN202610176680.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-02-06
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing methods for measuring and calculating welding stress and strain suffer from low efficiency, complex operation, high cost, and inability to effectively control them. In particular, in laser welding, traditional methods rely on experience or complex simulation calculations, making it difficult to achieve real-time monitoring and efficient control.

Method used

A time-series prediction system for welding stress and strain is constructed by using an LSTM network-based approach to organize multi-source information on welding conditions, identify key locations, predict stress and strain using a long short-term memory network, and perform quality assessment and control based on the prediction results.

Benefits of technology

It significantly reduces computation time and cost, improves prediction accuracy, achieves high-fidelity stress calculation and real-time control, reduces welding scrap rate, adapts to various working conditions, and has a cross-working condition recognition accuracy of over 80%, meeting industrial needs.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses the field of welding stress, and discloses a welding stress-strain rapid prediction control method and system based on an LSTM (Long Short Term Memory) network. Welding process parameters such as laser power, welding speed, defocusing amount and light spot diameter and parameters such as a welded workpiece are used as input, and corresponding welding stress and strain are used as predicted output to train an LSTM proxy model. The method can be applied to various welding butt joint states and can adapt to various states, the accuracy rate of cross-working-condition recognition exceeds 80%, and the normal working requirement can be met. According to the method, the stress and strain of welding are rapidly calculated based on the LSTM network training model, the welding process is detected to facilitate backtracking adjustment, and the welding process is dynamically adjusted based on the model advantages to achieve closed-loop optimization of the welding process.
Need to check novelty before this filing date? Find Prior Art