Classification And Detection Of Milling Surface Roughness Based On Convolutional Neural Network And Simulation Data

Surface roughness refers to the unevenness of the workpiece surface with small spacing and tiny peaks and valleys,which directly affects the performance and lifetime of the product,and the accurate and efficient measurement of its value is of great significance to modern industry.In recent years,with the continuous development of Convolutional Neural Networks(CNN),the research of CNN-based visual classification and detection method for surface roughness has been progressing.In this method,there is no need to design the roughness evaluation index artificially,but the features associated with roughness are automatically extracted in the training of CNN model,so as to achieve the detection of roughness.In this context,the paper constructs two surface roughness classification and detection models based on lightweight convolutional neural networks and deep convolutional neural networks combined with simulation data,respectively.Therefore,the following research work is done in this paper:(1)Milling Surface roughness classification detection based on lightweight convolutional neural network ShuffleNetV2: In response to the problem that current roughness visual measurement methods rely on artificial indicators,this paper proposes a surface roughness classification detection method based on the lightweight convolutional neural network ShuffleNetV2.The simple and efficient network design of the ShuffleNetV2 model greatly reduces the number of model parameters and computational complexity,which not only facilitates the construction of a fast roughness online detection model,but also automatically extracts features highly correlated with surface roughness.When the roughness image data of the model comes from the same lighting environment,compared with the detection model designed by artificial indicators,the model can still maintain high classification accuracy with shorter training time.(2)Classification and detection of milling surface roughness by deep convolutional neural network combined with simulation data: Although ShuffleNetV2 model effectively solves the shortcomings of artificial index design,it has poor generalization ability for data images in different lighting environments and is difficult to adapt to the complex and changeable lighting conditions in industrial environments.The deep neural network Xception model analyzes and processes a large number of milled surface roughness images to effectively enhance its generalization performance,but requires a large amount of sample data.To solve the problem of large data requirements for deep networks,this article uses simulation technology to obtain a large number of milled roughness surface images and uses the transfer learning model Deep CORAL to achieve cross-domain texture matching between simulated and actual samples,in order to use simulated images to expand actual milled workpiece images.The experimental results show that using simulated images for data augmentation can improve the roughness classification detection accuracy of the Xception model.In addition,this article also explores the impact of different lighting data images on the prediction accuracy of the Xception model and finds that it has good robustness,making roughness visual online detection possible.(3)Starting from the requirements of the online detection visual system for milling surface roughness,the construction of hardware and software modules was emphasized to achieve efficient and accurate surface roughness detection.In terms of hardware system design,a detailed analysis was conducted on the camera,precision optical platform,and light source to ensure that the system could capture the workpiece accurately and stably.In terms of software module design,a thorough analysis was carried out from three aspects: image acquisition,image processing,and image classification,to ensure that the system could effectively process and analyze surface image data.