Chinese Journal of Lasers, Volume. 52, Issue 8, 0802101(2025)
Adaptive Laser Welding of Aluminum Alloy Profiles for High‐Speed Rail Based on TVCNN
With the rapid development of high-speed trains in China, achieving lightweight designs while maintaining structural strength has become increasingly critical. Aluminum profiles are widely used in high-speed train body structures due to their low density, high specific strength, excellent corrosion resistance, and ease of welding and processing, making them one of the most commonly utilized body structure materials. Laser welding technology has also been extensively adopted in rail transit manufacturing, offering advantages such as high energy density, fast welding speed, minimal deformation, and stable joint quality. However, unlike conventional aluminum alloy profiles, the aluminum alloy profile components of high-speed trains are larger and often require 3?5 m long straight welds on the body to simplify assembly. At this scale, assembly clearance issues become significant. Horizontal and vertical assembly clearances of 0?1 mm frequently occur at welding joints during production, adversely affecting the stability of the welding process, as well as the formation and mechanical properties of the weld joints. Consequently, real-time monitoring and control of the laser welding process through neural networks and control systems are essential to achieve high-quality welds.
This study examines the impact of laser power on weld formation and performance under varying gap conditions through process experiments. A convolutional neural network (CNN) model based on visual information is proposed to predict the state of welded joints. The model introduces horizontal and vertical gaps as input information sources using vector encoding, enhancing the identification of penetration through three types of information. By interfacing with a programmable logic controller (PLC), real-time control of the laser is achieved to adapt to dynamic gap variations in aluminum alloy profiles for medium- and high-speed rail production. The process test welding platform comprises a robot, a multimode continuous fiber laser, a welding workbench, and related equipment. Constant-gap welding tests are conducted under four typical working conditions, with orthogonal experiments simulating different assembly errors encountered in production. The effects of horizontal gap and vertical misalignment on weld formation and performance are analyzed separately. Weld joints formed under the four working conditions are used to construct and train a transfer vector convolutional neural network (TVCNN) model. Based on this model, a real-time adaptive welding system is developed to improve weld quality and production efficiency.
Regarding weld formation and performance, an increase in the assembly gap results in a corresponding increase in weld penetration depth. This expansion of the effective load-bearing area enhances the mechanical properties of the weld. However, excessive laser powers or larger assembly gaps lead to defects such as undercutting and burn-through. Considering both weld shape and mechanical performance, the optimal process parameters (laser powers) are identified as approximately 5500 W for working condition 1, 4500 W for working conditions 2 and 3, and 3500 W for working condition 4. A dataset comprising three types of joints formed under four distinct conditions is utilized for model development. Compared to previous models, the proposed model demonstrates substantial improvement, achieving an accuracy of 91.24%. In adaptive welding systems, the model exhibits high recognition accuracy for the penetration state of the joint. Moreover, it effectively detects abnormal penetration states in real time and provides feedback to adjust laser power accordingly, thereby facilitating the production of high-quality welds.
A laser welding process experiment is conducted on the locking bottom joint of a 5 mm thick aluminum alloy profile. It is observed that as horizontal and vertical gaps increase, the laser power required to form a normal joint decreases. This phenomenon occurs because moderate horizontal gaps enhance melt pool stability and promote increased melt depth, while excessive horizontal gaps can cause joint collapse and degrade surface quality. Additionally, vertical gaps induce lateral flow of the upper melt, affecting the stability of the melt pool. A TVCNN model is proposed for real-time monitoring of the melt state during the welding process. The model incorporates melt pool images, horizontal gaps, and vertical gaps as input features and is trained using welding test data from various working conditions. The trained model achieves a melt state recognition accuracy of 91.24%, demonstrating strong predictive performance. Based on this TVCNN model, a real-time adaptive welding system is developed. The system identifies defects in real time using the TVCNN model and adjusts the laser power through PLC feedback to accommodate different horizontal and vertical gap conditions, ensuring the production of high-quality weld joints.
Get Citation
Copy Citation Text
Yin Ma, Wei Wang, Guolong Ma, Xiaohui Han, Biao Yang, Fuyun Liu, Caiwang Tan, Xiaoguo Song. Adaptive Laser Welding of Aluminum Alloy Profiles for High‐Speed Rail Based on TVCNN[J]. Chinese Journal of Lasers, 2025, 52(8): 0802101
Category: Laser Forming Manufacturing
Received: Nov. 11, 2024
Accepted: Dec. 17, 2024
Published Online: Apr. 8, 2025
The Author Email: Han Xiaohui (13793237339@139.com)
CSTR:32183.14.CJL241337