Predicting Convective Heat Transfer Coefficient in TIG Welding via Adaptive Neuro-Fuzzy Inference System (ANFIS)

• Osemwegie IKPONMWOSA-EWEKA

  Author

• Alexander I. IDEMUDIA

  Author

This study develops an Adaptive Neuro-Fuzzy Inference System (ANFIS) to predict the convective heat transfer coefficient (h) in Tungsten Inert Gas (TIG) welding of mild steel, overcoming limitations of traditional empirical correlations under variable arc conditions. Overtime, TIG welding has been used to produce high-quality welds for aerospace and precision applications. Mild steel plates were cut into 200 coupons (80 × 40 × 10 mm), machined, cleaned, ground, and butt-welded using TIG equipment with 100% pure argon shielding gas. A central composite design matrix via Design-Expert v7.01 generated 20 experimental runs, with five replicates per run yielding 100 data points. Input parameters included welding current (I), voltage (V), and speed (ws). ANFIS integrates fuzzy logic and neural networks for nonlinear mapping from inputs to h. Linguistic variables, current (c: 170-190 A), welding speed (ws: 2.6-3.0 mm/s), voltage (v: 20-22 V), employed triangular membership functions (trimf: low, moderate, high). Output h ranged 1.9-2.5 W/m².K with constant functions. Grid partition generated the fuzzy inference system (FIS); crisp training data optimized rules via ANFIS toolbox. This research finding established that ANFIS model effectively predicts the convective heat transfer coefficient (h) in TIG welding of mild steel using inputs of welding current (170-190 A), voltage (20-22 V), and speed (2.6-3.0 mm/s). Experimental h values ranged from 2.00 to 2.60 W/m².K across 20 runs, with ANFIS predictions matching closely (e.g., 2.20 vs. 2.20, 2.54 vs. 2.54). Regression analysis yielded the equation: Experimental h = 0.3632 + 0.8497 × ANFIS h, achieving R² = 97.47% and S = 0.0766300, confirming high predictive accuracy. Time series plots showed strong correlation trends with minimal drift, validating ANFIS for real-time process optimization.

• FJET_21_79_79

Copyright (c) 2026 FUDMA Journal of Engineering and Technology

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.