Analysis Of Melt Wettability In Microgravity Based On Deep Learning

The calculation of the wetting angle between the solid-liquid interface is of great significance to characterize the interfacial interaction and analyze the material properties.With the continuous development of computer science and image processing technology,the measurement of wetting angle has gradually evolved from the traditional manual observation measurement to the measurement based on computer vision,and the degree of automation has been greatly improved compared with the past,making the operator’s labor greatly reduced.In recent years,the use of image methods to measure the wetting angle has made considerable progress,and many mature technologies and equipment have emerged at home and abroad,which can be used to measure the wetting angle of the droplet and the interface at room temperature,and analyze its wetting effect on the interface.In the experiment of measuring the wetting angle of high-temperature melt,since the high-temperature furnace used in the experiment is closed,scientists can only observe and record the experimental process by placing a camera outside the hightemperature furnace window,and then calculate the wetting angle.However,most of the existing measurement methods are mainly used for the calculation of the wetting angle of droplets at room temperature,and have not achieved very good results in the calculation of the wetting angle of the melt at high temperature.Most of these methods require manual operation to locate the sample.It is not only time-consuming and laborintensive,but also introduces errors introduced by human subjective operations in the subsequent contact angle calculation.In addition,the existing fitting methods mainly use an equation or function to fit the complete contour of the sample,these methods have large errors in calculating the wetting angle,and the robustness is not strong,which makes the measurement results has large deviation,stable and accurate results cannot be obtained in the experiment of the same material,and the reproducibility of the calculation is poor.Therefore,it is necessary to design a wetting angle calculation method that can automatically locate the sample and have high reproducibility.Aiming at the problem of automatic localization and extraction of samples,this paper applies the classical deep learning network U-Net to the localization of samples and the extraction of binary images,and introduces atrous convolution and multi-scale feature fusion structure at the connection between downsampling and upsampling,combined with the residual structure to improve the performance of the U-Net network and complete the automatic positioning of the sample,so that the sample extraction process no longer requires manual cropping and positioning of the sample image,which reduces the consumption of human resources and eliminates the systematic error by human’s subjective operation,especially not affected by the shape,size and position of the sample,can realize the separation of the sample and the base,complete the extraction of the binary image of the sample,and facilitate the subsequent contour extraction and contact angle calculation.For the calculation of the contact angle,it is no longer trying to use an equation to fit the complete sample contour like the traditional method,but after splitting the contour into left and right halves,the iterative weighted least squares ellipse fitting method is used to approximate sample profile,calculate the contact angle of the sample.The method proposed in this paper is applied to the sessile drop measurement system,which can automatically calculate the wetting angle of various materials after melting in a large temperature range.At the same time,the method in this paper is compared with the current mainstream wetting angle calculation methods,the obtained results are not only completely free from errors introduced by manual operations,but also have small standard deviations in repeated experiments and have high reproducibility.