Research On Roughness Measurement Of Milled Workpieces Based On Deep Transfer Learning And Object Detection

Surface roughness is often used in the process of assessing the surface quality of industrial products.It affects various performance indicators such as wear resistance,service life,and corrosion resistance of the product.With the continuous development of artificial intelligence,the application of machine vision technology in roughness measurement has received much attention from related researchers.In response to the problems currently existing in practical applications of machine vision-based roughness measurement methods,this paper introduces transfer learning and object detection algorithms based on deep neural networks(DNNs),and conducts research on roughness measurement based on deep learning with milled workpieces as the research object.The following are the main research contents of this paper:In this paper,we propose a deep transfer learning-based algorithm to recognize the surface roughness level of milled workpieces,"Deep Alex CORAL",to address the problems that many machine vision-based roughness measurement methods rely on artificially designed features that cannot accurately characterize the surface roughness of workpieces and DNNs that require a large number of samples and consistent data distribution between training and test samples.This algorithm not only reduces the data requirement of DNNs,but also reduces the difference of data distribution between training set and test set,and automatically extracts more general roughness features.Experimental results show that Deep Alex CORAL exhibits excellent robustness to various lighting environments.Deep Alex CORAL solves some of the problems of current roughness measurement methods,but it does not allow for parallel roughness measurement of multiple parts in an image.In application scenarios where a large number of workpieces need to be measured,the method that can only measure the roughness of a single workpiece at a time greatly limits the detection efficiency.To address this problem,a multi-object roughness level detection method based on Faster R-CNN is proposed in this paper.The method uses a convolutional neural network(CNN)to extract the image features of the milled workpiece,based on which the target region where the workpiece may exist is inferred and the roughness level of the workpiece in the region is calculated.Also,the method reduces the time required for training by a deep transfer learning algorithm.The experimental results show that the proposed method can accurately detect the area where the workpiece is present in the image and the corresponding roughness level of the workpiece for a test set of milled workpieces with different placement angles and shooting conditions.The roughness detection algorithm based on Faster R-CNN fails to meet the positioning and roughness detection requirements of milled workpieces in certain situations.Specifically,the algorithm does not consider cases where some workpieces are not fully within the camera’s field of view,resulting in insufficient model accuracy.The essence of this problem is that the algorithm has not learned how to detect the roughness of these types of workpieces.To solve this problem,this paper proposes adding images of such workpieces to the training set of the model to obtain knowledge required by deep learning algorithms.Experimental results show a significant improvement in both workpiece positioning and roughness detection after adding the images of workpieces that are not captured as a whole in the training set.The research work in this paper solves some problems of current roughness measurement methods,expands the application scenarios of deep learning algorithms for roughness measurement,and promotes the development of roughness measurement technologies.