Research And Development Of Grain Crack Detection Device Based On Vision

Solid propellant is the power source used in solid motor.It is an important part of the motor.It has many different shapes and colors.When the workers pour,solidify and demould the solid propellant raw materials,the propellant grain will be formed.In the production process of grain,especially in curing cooling and low temperature test,cracks will be produced due to the tensile stress on the inner surface of grain.If the cracked grain is ignited,it may lead to engine explosion and other hazards.Therefore,it is necessary to detect the grain in time and record the crack data information,so as to take corresponding treatment in the later stage.Due to the strong toxicity of the grain itself,the grain is generally scanned by the imaging system in actual production,and then the imaging results are checked manually.Manual detection mainly includes three problems:large detection workload,inconsistent detection standards and difficult detection information management.Therefore,realizing the automation and standardization of solid propellant grain crack detection process is an urgent problem to be solved in the industry.The solid propellant grain studied in this paper is a kind of cylindrical material,and the crack size to be observed ranges from 10 microns to 1000 microns.Aiming at the problems existing in the crack detection of solid propellant grain in the industry,a grain surface crack detection device based on vision is developed in this paper.Specifically,the work of this paper is as follows:(1)The research and development of a special detection device for grain crack is designed and completed.Device development includes hardware device design and software development.The hardware device needs to be selected according to the detection requirements such as the size information of the grain and the size information of the crack,and cooperate to complete the image acquisition task.In order to accurately calculate the crack area,the detection device adopts a telecentric lens whose magnification will not change with the change of object distance.In addition,an auto focusing algorithm with adaptive search step is designed to ensure the image quality and time requirements of image acquisition task.The software part mainly realizes the control interface of the hardware system,displays the status of image acquisition in real time,embeds the crack segmentation algorithm,and records and displays the crack detection results.(2)A crack segmentation algorithm combining threshold and morphology is proposed.Because the grain crack image obtained at present has a single background texture and not rich colors.Therefore,in the initial crack detection,this paper directly models the grain image,and designs a crack segmentation algorithm combining threshold and morphology.The crack image is segmented more easily by using the closed morphological threshold,and then the crack image is segmented more easily by using the closed morphological threshold.(3)A crack segmentation algorithm model STCSNet,which combines statistical texture features and depth semantic features,is proposed.In the second sampling inspection,because the grain surface image has been collected in the first crack detection,a more complex depth learning model can be appropriately selected to deal with the crack segmentation task.STCSNet model changes the convolution of each layer in u-net structure except input layer and output layer into deep separable convolution,which effectively improves the reasoning speed of the model;The statistical texture feature extraction module is added to adapt to more complex crack scenes and improve the performance of model segmentation.Finally,experiments show that STCSNet model can not only show excellent segmentation performance when the parameters are much less than u-net,but also meet the requirements of grain crack detection task.