Solid rocket engine lining defect image data set self-making method

Abstract

The invention relates to the manufacturing field of a label data set required for training a neural network, and discloses a solid rocket engine lining defect image data set self-making method. The method comprises the following steps: gradually extracting subset images from an original lining image from a left upper corner to a right lower corner at a specific step length; randomly combining a subset image and a defect graph generated by a defect generator at equal probability, saving a combination result and a corresponding label after completing the combination, taking out another original lining image for combining and saving after completing one original lining image then, and repeating the steps until all the lining images are combined and saved; and finally, performing data enhancement on all the combined results once, wherein the data enhancement can double the number of the images. Through the method, a large number of labeled lining defect images can be obtained and can be used for training a neural network.

Description

technical field [0001] The invention relates to the field of making label data sets required for training neural networks, in particular to a self-made method for solid rocket motor liner defect image data sets. Background technique [0002] As the power plant of missiles or aerospace vehicles, solid rocket motors have specific requirements for the integrity of the lining coating on the inner wall of the engine casing. The main function of the liner is to firmly bond the propellant with the heat insulation layer or the casing, and it is required to be firmly bonded and able to withstand the stress that occurs when the engine is working. If the lining layer is not bonded firmly and debonding occurs, it will cause overheating and loss of strength inside the shell, or cause the combustion surface of the propellant to expand, resulting in engine failure. [0003] Under the existing liner forming technology, there may be defects such as missing coating, sagging, air bubbles, and...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to overcome the above-mentioned defects, and provides a self-made method for a solid rocket motor lining defect image data set. The original lining image is taken out from the upper left corner to the lower right corner with a specific step size, and the subset images are successively taken out. The set image and the defect graph generated by the defect generator are randomly combined with equal probability, which solves the problem of lack of lining defect image data set and the inability to apply deep learning methods to train neural network models

Images