Strength Prediction of Aluminum-Stainless Steel Pulsed TIG Welding-Brazing Joints Using RSM and ANN

Literature Overview

This 2014 publication in Acta Metallurgica Sinica (English Letters) by researchers from the State Key Laboratory of Advanced Welding and Joining at Harbin Institute of Technology presents a comprehensive investigation into the mechanical strength prediction of dissimilar metal joints produced by pulsed TIG welding-brazing. The work combines experimental design methodology with advanced modeling techniques, specifically Response Surface Methodology (RSM) and Artificial Neural Networks (ANN), to establish predictive relationships between welding parameters and joint strength.

Core Technical Framework

The aluminum-stainless steel joint represents a classic dissimilar metal joining challenge where direct fusion welding is impractical due to the formation of brittle intermetallic compounds (IMCs) and the large difference in thermal properties. The welding-brazing approach solves this by melting only the aluminum side while the stainless steel side is heated to brazing temperature, using a filler alloy that wets the stainless steel surface.

The pulsed TIG configuration provides superior control over heat input compared to continuous TIG, enabling precise management of the thermal cycle at the dissimilar interface. Key process parameters investigated include:

Pulse current (Ip): 150–250 A
Background current (Ib): 50–100 A
Pulse frequency: 5–20 Hz
Pulse width ratio: 0.2–0.6
Travel speed: 40–80 mm/min
Shielding gas flow rate: 10–20 L/min

RSM and ANN Modeling Approach

The study employs both RSM and ANN to model the relationship between process parameters and joint strength. RSM provides an explicit mathematical model based on polynomial regression, while ANN offers a data-driven approach capable of capturing complex nonlinear relationships.

Modeling Approach	Advantages	Limitations
RSM	Explicit equations, interpretable coefficients, identifies parameter interactions	Limited to polynomial relationships, may miss higher-order effects
ANN	Captures complex nonlinearities, higher prediction accuracy	Black-box model, requires large training datasets, less interpretable

The research demonstrates that ANN generally provides superior prediction accuracy compared to RSM, with mean absolute errors approximately 20–30% lower for joint strength prediction. However, RSM offers valuable insights into parameter sensitivity and interaction effects that guide process optimization.

Interface Microstructure and Strength Mechanisms

The joint strength is primarily governed by the interfacial reaction layer formed between the aluminum alloy and stainless steel substrate. During pulsed TIG welding-brazing, the following microstructural features develop:

Aluminum side fusion zone: Solid solution with filler alloy elements, exhibiting good ductility
Interfacial reaction zone: Thin layer of intermetallic compounds (typically Al-Fe, Al-Cr phases)
Stainless steel heat-affected zone: Minimal microstructural change due to sub-solidus heating
Filler alloy bond zone: Wetting layer providing mechanical interlocking

The pulsed current configuration effectively controls the interfacial reaction layer thickness by modulating peak temperatures. Lower pulse widths and frequencies result in thinner IMC layers and higher joint strength, while higher thermal inputs promote excessive IMC growth and joint embrittlement.

Process Optimization Results

The combined RSM and ANN analysis reveals the following optimization insights:

Pulse current is the most significant factor affecting joint strength, followed by travel speed and pulse frequency
Optimal pulse width ratio falls between 0.3–0.4, providing sufficient melting without excessive thermal input
Travel speed has a threshold effect: too low promotes IMC growth, too high results in incomplete wetting
The interaction between pulse frequency and width ratio significantly influences joint morphology

Typical optimal parameters for a 2 mm aluminum alloy on 2 mm stainless steel lap joint include: pulse current of 200 A, background current of 70 A, pulse frequency of 12 Hz, pulse width ratio of 0.35, and travel speed of 60 mm/min. Under these conditions, shear strength values of 60–75 MPa are achievable, representing a significant improvement over conventional TIG welding-brazing without pulsing.

Engineering Applications and Quality Control

The predictive models developed in this study have direct applications in manufacturing environments where rapid process optimization is required. For production shops, the ANN model can serve as a digital twin component, predicting joint quality based on process parameters before actual welding occurs.

Quality control considerations for aluminum-stainless steel welding-brazing joints include:

Visual inspection of joint width and surface morphology
Shear or tensile testing for joint strength verification
Metallographic examination of interface thickness (target: 10–30 μm IMC layer)
Hardness profiling across the joint cross-section
Thermal cycling tests for long-term reliability assessment

Study Insights and Reflections

This research exemplifies the power of combining experimental design with computational modeling in welding technology development. The dual approach of RSM and ANN provides complementary strengths: RSM offers interpretability for process understanding, while ANN delivers superior prediction accuracy for complex parameter spaces.

For engineering practice, the key insight is that pulsed TIG welding-brazing represents a viable production technology for aluminum-stainless steel joining, particularly in automotive, aerospace, and marine applications where weight reduction and corrosion resistance are critical. The predictive models enable rapid qualification of new welding procedures, reducing the trial-and-error approach that traditionally dominates welding development.

The study also highlights the importance of understanding interfacial metallurgy in dissimilar metal joining. The balance between sufficient bonding and limited IMC growth represents the fundamental challenge that pulsed current technology addresses through precise thermal management.

Reference Value and Outlook

The work provides a validated methodology for process optimization and quality prediction in dissimilar metal welding-brazing. Future research directions include extending the modeling approach to other dissimilar metal combinations, incorporating real-time sensor feedback for adaptive control, and validating long-term joint performance under cyclic loading conditions. The integration of data analysis techniques with physical metallurgy knowledge continues to advance the field of welding process development.