The Research Of Fast Prediction Method Of Casting Shrinkage Defects Based On Fully Convolutional Networks

The traditional numerical simulation method can effectively calculate the distribution of shrinkage defects that may occur in the casting process plan.However,due to the huge amount of calculation of the numerical simulation method,it takes a long time to calculate the distribution of shrinkage defects in the casting.In response to the above problems,this paper introduces machine learning algorithms into the prediction of shrinkage defects in casting,and uses targeted data preprocessing to process the casting process data.At the same time,multi-modal data input is used to increase the data dimension,and the fully connected convolutional neural network is used to quickly predict the shrinkage defects of the casting.A fast prediction method of the cast shrinkage defects based on the fully connected convolutional neural network algorithm is proposed.The main contents include:(1)Introduce the concept of 3D image and fully connected convolutional neural network algorithm to transform the computational task of casting shrinkage into a 3D image semantic segmentation task.This paper treat the casting,riser,cold iron and other components in the casting process plan as different pixel values in the 3D image,and use the category relationship and relative position relationship between the pixels in the 3D image to describe the various parts in the casting process plan.The 3D image and shrinkage defect distribution results were used as training samples of the algorithm,and a rapid prediction model for casting shrinkage defects was obtained by training.(2)Aiming at the application scene of the casting process,a specialized data preprocessing was introduced in the task.Through normalization and data enhancement of casting shrinkage defect data,the data usability had been improved,and the scale of the data had been expanded;meanwhile,lightweight techniques such as 3D data matrix and 2D matrix slicing had been proposed in order to solve the engineering problem of large-capacity data;(3)Aiming at the cause of the shrinkage defects in the casting process,multi-modal data input method was adopted.The comprehensive influence of factors such as the shape,size,and relative position of the casting process,as well as the pouring temperature and pouring height were considered.Factors such as casting temperature and pouring height of the shrinkage defect distribution in casting were equivalent to 3D image data of different modalities,and the multi-modal data input was introduced in the task.The various features in the casting process were extracted comprehensively,which lays the data foundation for the rapid calculation method of casting shrinkage cavity shrinkage defects;(4)By analyzing the casting shrinkage defects of bearing housings and oil cylinders,the effectiveness and efficiency of the rapid prediction method for casting shrinkage defects based on the fully connected convolution neural network algorithm proposed in this paper was verified.The rapid prediction method of casting shrinkage defects proposed in this paper can significantly reduce the calculation time of casting shrinkage defects compared with traditional numerical simulation algorithms.The completed research results provide a new idea for the calculation of casting shrinkage defects,and provide an important reference for the relevant researchers of enterprises to quickly analyze the advantages and disadvantages of the casting process plan.