Part machining surface prediction method based on space-time full convolutional recurrent neural network

Abstract

The invention relates to the technical field of image processing, and discloses a part machining surface prediction method based on a space-time full convolutional recurrent neural network, which aims at solving the problem that information cannot be shared between layers in a convolutional neural network in the prior art, and comprises the following steps of: (1) generating a grayscale image; (2) obtaining a sub-image set; (3) making spatial autocorrelation analysis; (4) segmenting a sliding window of the grayscale image; (5) making time autocorrelation analysis; and (6) constructing the space-time full convolutional recurrent neural network. Global features and local detail features are extracted by using a convolutional layer of a full convolutional network, the extracted global features and local detail features are fused through a skip layer structure, and a ConvLSTM unit is introduced into the full convolutional network (FCN). The network has certain memorability on the basis that global feature and local feature extraction and fusion can be completed, and then the part machining surface prediction precision is improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for predicting the processing surface of a part based on a spatio-temporal fully convolutional cyclic neural network. Background technique [0002] Part surface topography is a comprehensive reflection of process parameters and process information, and is also the key to ensuring product quality and realizing design functions. For the same processing condition, the continuous measurement of the surface topography data of multiple parts contains the evolution information of the processing process, which can realize the topography prediction. By mining the process information of the processing system contained in the surface topography data of historical parts, the change rule of part topography can be grasped, and the surface topography of future processed parts can be predicted, and then the surface tolerance and tool wear status can be analyzed according to the...

AI Technical Summary

Problems solved by technology

[0006] The present invention aims to overcome the problem that information cannot be shared between layers in the convolutional neural network in the prior art, and provides a part processing surface prediction method based on the spatio-temporal full convolutional neural network, using the convolution of the full convolutional network The layer extracts the global features and local detail features, and fuses the extracted global features and local detail features through the skip layer structure, and introduces the convolutional long short-term memory network (ConvLSTM) unit into the full convolutional network (FCN), so that the network can complete Based on the extraction and fusion of global features and local features, it has a certain memory, thereby improving the prediction accuracy of the machined surface of the part

Images