Visual Detection Of Twist Drill Parameters Based On Deep Learning

Twist drill is a kind of commonly used complex cutting tool,which has many parameters.It is necessary to sample the important parameters in the production process and the completion of production.At present,the factory mostly uses the method of manual testing parameters.The manual testing method has many disadvantages: Low detection efficiency,artificial fatigue leads to unstable detection standards,and the use of measuring tools requires contact with the bit,resulting in scratches on the bit surface.With the rapid evolution of machine vision technology,vision measurement technology has become the main development direction of industrial product detection.Vision measurement technology obtains measurement results by taking vision sensor and vision software as the core,processing image information by developing algorithms,and extracting image edges.We can overcome the shortcoming of manual measurement effectively by using the visual detection technology to the parameter detection of the full-grinding straight shank twist drill.In this paper,the visual detection of twist drill bit based on deep learning is researched.The specific research contents are as follows:Firstly,the visual inspection system of twist drill is built.Set up the platform according to the working scene and select the equipment model.The industrial camera,lens,light source and lighting mode are selected according to the characteristics of the bit surface and the actual needs of the parameters to be measured,and the hardware is installed and positioned through calculation.Subsequently,a variety of image processing schemes are designed to obtain more accurate image edge information and make the measurement more accurate.The function transformation,image smoothing processing and convolution sharpening processing based on grayscale image can optimize the image edge information,which is beneficial to image edge extraction.Then,it construct an edge extraction network according to convolutional neural network.By analyzing the traditional edge detection operator and comparing the edge images of twist bits,the edge extraction method based on convolutional neural network is raised.This method takes RCF network as the main body and adds residual integration structure to carry out fusion of multi-scale feature images,which is the innovation point.Finally,the visual measurement of twist bit parameters is completed.The high-efficiency and high-precision real-time detection of twist bit is realized,and the measurement results can meet the production requirements.It has very important application value for improving the detection efficiency and the intelligentization degree of production management.