Automated grain-size characterization of GTAW Ti-6Al-4 V weldments using OpenCV and comparison with ASTM-based intercept methods

Quantitative measurement of grain size in the fusion zone (FZ) of Ti-6Al-4 V GTAW weldments is difficult because the large, uneven grain morphologies can introduce operator bias in conventional metallography. In this study, a calibrated openCV based image-processing workflow was developed, and its results were compared with those of the Jeffries planimetric, the Heyn linear-intercept, and Abrams three-circle methods of ASTM E112. Four GTAW conditions were considered: conventional DC, pulsed DC, ultrasonic vibration-assisted (UAV) DC, and UAV pulsed DC. The workflow comprises grayscale conversion, filtering, boundary detection, segmentation, morphological operations, grain identification and spatial calibration at approximately 3.03 μm/pixel for the FZ micrographs. The OpenCV mean linear intercepts (± standard deviation) were 421.0 ± 21.0, 300.0 ± 20.6, 261.0 ± 31.9 and 243.9 ± 10.4 μm for DC, pulsed DC, UVA DC and UVA pulsed DC, respectively, showing progressive grain refinement with pulsing and ultrasonic vibration. Relative to the planimetric, linear-intercept and three-circle methods, the OpenCV mean biases were − 20.18, -65.13, and − 128.12 μm, with mean absolute percentage errors (MAPE) of 6.47%, 17.04%, and 29.60%, respectively. Although the OpenCV approach enables consistent grain-size determination, it should not be regarded as an exact replacement for ASTM E112 analysis.