In Situ Capture of High-Temperature Precipitate Phases in Ti-48Al-2Cr-2Nb Alloy Using Convolutional Neural Networks

TiAl intermetallic alloy is a crucial high-performance material, and its microstructure evolution at high temperatures is closely related to the process parameters. Observing the lamellar structure is key to exploring growth kinetics, and the feature extraction of precipitate phases can provide an effective basis for subsequent evolution studies and process parameter settings. Traditional observation methods struggle to promptly grasp the growth state of lamellar structures, and conventional object detection has certain limitations for clustered lamellar structures. This paper introduces a novel method for high-temperature precipitate phase feature extraction based on the YOLOv5-obb rotational object detection network, and a corresponding precipitate phase dataset was created. The improved YOLOv5-obb network was compared with other detection networks. The results show that the proposed YOLOv5-obb network model achieved a precision rate of 93.6% on the validation set for detecting and identifying lamellar structures, with a detection time of 0.02 s per image. It can effectively and accurately identify γ lamellar structures, providing a reference for intelligent morphology detection of alloy precipitate phases under high-temperature conditions. This method achieved good detection performance and high robustness. Additionally, the network can obtain precise positional information for target structures, thus determining the true length of the lamellar structure, which provides strong support for subsequent growth rate calculations.