Microstructure Identification Of Ni-base Superalloy Based On Deep Learning

Superalloys refer to metal materials that can be used for a long time in harsh environments of high temperature and high pressure,due to their excellent performance,they are widely used in various industries.In order to meet the growing demand,researchers have developed a high-throughput method to prepare superalloys,and obtain the compositions and microstructure images of superalloys by microbeam X-ray fluorescence spectrometer and scanning electron microscope.Microstructure plays a key role in the relationship between composition,structure and properties,and it can reveal the relationship between compositions and phase to further customize the properties of materials.Therefore,the precise identification ofγ′phases in the microstructure,including the segmentation and the extraction of morphological parameters ofγ′phases,has a significant impact on studying and improving the properties of superalloys.The data studied in this paper include microstructure images of nickel-based superalloys collected at three different temperatures,800°C,900°C,and 1000°C,numbered W800,W900,and W1000.Due to different temperatures and compositions,different microstructures are formed,coupled with uneven corrosion effect or different imaging parameters,the traditional threshold segmentation cannot accurately segment theγ′phases.In view of the superior performance of deep learning in computer vision,this paper adopts the deep learning method to segment theγ′phases.The experimental results show that the deep learning models have better segmentation effect than the traditional threshold segmentation method.For the W800 and W900 datasets,the UNet++and Dual-channel models can be used to achieve good segmentation results;for the W1000 dataset,due to the insignificant image contrast,the previous two models dose not achieve good segmentation results,so the improvement model of UNet++is used.The UNet3+model is used for segmentation,and the effect is improved,the F1 score reaches0.90.But the segmentation of the adjacentγ′phases in the image is not ideal,instance segmentation model Mask R-CNN solves this problem,and the F1 score reaches 0.93.After the γ′ phases are segmented by the deep learning models,the morphological parameters are further counted to establish the corresponding relationship between the compositions and the statistical parameters.Then,Lasso regression was used to analyze the correlation between the compositions and morphological parameters,and it is concluded that the four metal elements Co,Ti,Ta and Nb have a significant impact on the precipitation ofγ′phase,which provides a significant guidance for material scientists to design the composition and improve the properties of ni-based superalloys.