Process Optimization of GH4169 Diaphragm Micro-Beam TIG Welding Using Orthogonal Test and BP Neural Network

Literature Overview

This 2018 paper by Yu Guo, Yin Yuhuan, Gao Jiashuang, and Guo Lijie from Shanghai Aerospace Equipment Manufacturing Plant, published in the Welding Journal, presents a methodology for optimizing micro-beam TIG welding parameters for GH4169 nickel-based superalloy diaphragms. GH4169 (equivalent to Inconel 718) is a precipitation-strengthened superalloy widely used in turbine blades, discs, and thin-walled diaphragms in aerospace engines. The diaphragm component, typically 0.3-0.8 mm thick, demands exceptional weld quality with minimal distortion and full penetration.

Core Technical Points

Material Challenges of GH4169 Thin-Walled Components

GH4169 derives its strength from γ''-Ni₃Nb and γ'-Ni₃(Al,Ti) precipitates. During welding, the thermal cycle dissolves these precipitates in the HAZ, creating a sensitized zone susceptible to:

• Solidification cracking – due to the wide freezing range and low solid solubility of sulfur and phosphorus
• Hot cracking – interdendritic cracking in the weld metal
• Sensitization – carbide precipitation at grain boundaries in the HAZ (600-800 °C range)
• Low-cycle fatigue – loss of creep resistance due to precipitate coarsening

The thin diaphragm geometry (typically 0.3-0.8 mm) amplifies these challenges because the entire cross-section may be affected by the thermal cycle, leaving no unaffected base metal.

Micro-Beam TIG Process Configuration

The micro-beam TIG process uses a small-diameter tungsten electrode (φ0.6-1.0 mm) with reduced current (30-60 A) to achieve a concentrated, narrow weld bead. The process configuration includes:

Orthogonal Experimental Design

The authors employed a Taguchi L9 orthogonal array to systematically vary four factors (current, voltage, travel speed, and electrode stick-out) at three levels each. The performance metrics included:

1. Weld penetration depth
2. Bead width
3. Surface defect rating (visual)
4. Microhardness profile

The signal-to-noise ratio analysis identified travel speed and current as the most influential factors on weld quality, with an interaction effect between current and travel speed on penetration depth.

BP Neural Network Optimization

A Back Propagation (BP) neural network was trained with the orthogonal test data to predict weld geometry as a function of process parameters. The network architecture consisted of a 4-8-3 structure (4 inputs, 8 hidden neurons, 3 outputs). The training achieved prediction errors below 5% for penetration depth and bead width. The optimization objective was to maximize penetration while minimizing bead width and surface defects, subject to constraints on distortion and hardness.

Process Windows and Optimal Parameters

Defect Analysis

Integration with Engineering Practice

In aerospace engine manufacturing, diaphragms are critical flow-control components that must maintain dimensional accuracy within ±0.05 mm after welding. The micro-beam TIG process offers the narrowest weld bead among arc welding methods, minimizing distortion and preserving the remaining base metal properties. However, the process is highly sensitive to joint fit-up – even a 0.1 mm variation in gap can cause significant penetration variation.

The combination of orthogonal testing with neural network optimization represents a powerful approach for process development. The orthogonal test efficiently identifies the most influential parameters with minimal experimental runs, while the neural network provides continuous prediction capability across the entire parameter space. This hybrid approach reduces the number of physical trials by approximately 60-70% compared to a full factorial design.

Key Reflections

The most valuable insight from this study is the recognition that micro-beam TIG welding of GH4169 diaphragms requires not only optimal electrical parameters but also meticulous joint preparation and gas shielding. The backing gas requirement (typically argon at 4-6 L/min) is non-negotiable for full-penetration welds on thin sections, as even brief exposure of the root to atmosphere causes oxidation that initiates cracking.

From a quality assurance perspective, the BP neural network model can be embedded in the production system as a real-time monitoring tool. By measuring actual arc voltage and current during welding and comparing them with the predicted optimal values, operators can receive immediate feedback on process deviation. This represents a practical application of statistical process control in welding.

The study also underscores a limitation: the neural network model is only as good as its training data. For production use, the model must be validated with additional confirmation welds across the parameter space, particularly near the boundaries where the training data is sparse.

Reference Value and Outlook

This work demonstrates a mature methodology for welding process optimization that combines classical experimental design with modern computational tools. The approach is directly transferable to other thin-walled superalloy components such as turbine blades, fuel nozzles, and combustor liners. Future developments should incorporate in-situ monitoring (arc voltage waveform analysis, acoustic emission) to enable closed-loop control of the welding process, further improving consistency and reducing scrap rates.