Existing technologies make it difficult to detect multi-scale faults on the surface of resin lenses with great accuracy. An intelligent inspection method for resin lenses is presented using four-step phase-shift grating light. The defect feature extraction approach is optimised using the lightweight feature selection network to increase the model's defect information learning capability. Finally, an experimental test is performed on the resin lens surface using the self-constructed multiscale defect dataset. The results demonstrate that the detection system achieves a worldwide accuracy of 96.3% and a microscopic defect detection accuracy of 63.8%, representing a considerable increase in accuracy and efficiency over previous approaches.
Edge extraction of weld pool image is the key to realize on-line detection and control of laser cladding during the process of industrial laser manufacturing,and the accuracy of on-line control of cladding layer quality is largely determined by the accuracy of extracted weld pool edge.In this paper, a laser cladding pool image detection system based on machine vision is built to collect the molten pool image in the process of laser cladding. According to the characteristics of laser cladding pool image, the image is preprocessed. In this paper, an improved multi-structure, multi-scale and multi-directional morphological operator is proposed to process the molten pool image, and the accuracy and anti-noise ability of different edge detection methods are compared and analyzed. the improved mathematical morphology edge detection method can not only suppress noise but also extract the edge of laser cladding pool image more accurately.
A time-series prediction method for the geometric characteristics (length, length, area) of the molten pool in the laser cladding process is studied. This method uses the historical data of the change of the geometric characteristics of the molten pool to predict the geometric characteristics of the molten pool at the next moment, so as to promote the application in the control of the shape of the molten pool and the estimation of the state of the molten pool. In view of the time correlation shown by the time series of the geometric characteristics of the molten pool, a time series prediction method for the geometric characteristics of the molten pool was developed based on the Autoregressive Integrated Moving Average (ARIMA) model, and the dynamic stability index of the molten pool was proposed. The results show that the area of the molten pool has the best prediction accuracy, and the Mean Absolute Percentage Error (MAPE) is only 3.105, while the predicted MAPE of the width of the molten pool is 3.464 and the length of the predicted MAPE is 4.048. The dynamic stability index proposed can reflect the fluctuation of the molten pool.
In the laser cladding process, there are a large number of interference factors such as arc light, metal droplets, powder splash, etc., which makes it difficult to extract the edge of the melt pool. Aiming at the problem that it is difficult to extract the edge of melt pool accurately, a method of edge detection of melt pool based on mathematical morphology and Chan-Vese active contour model is proposed in this paper. After median filtering, the binary image of the melt pool is obtained by threshold segmentation. The initial contour of the melt pool is obtained by mathematical morphology connected domain labeling algorithm. On this basis, the connected region contour is taken as the initial contour, and the Chan Vese active contour model is used to extract the edge of the melt pool. The experimental results show that this method can achieve accurate extraction of melt pool edge, and has good anti-interference performance, which provides a good foundation for subsequent melt pool feature extraction and target recognition.
Aiming at the images of relevant monitored objects in the process of laser cladding, a super resolution algorithm technology was proposed to optimize and enhance the key details of the images, and the enhanced image content was segmented, extracted and counted. First, construct a training of a sub-resolution convolutional neural network (SRCNN) model; the original low resolution is predicted by the weight after training, the image quality evaluation results: peak signal-to-noise ratio (PSNR) is 30.198212, structural similarity (SSIM) is 0.969966; the most based on the maximum entropy dual threshold split algorithm combined with image processing, extracting and statistics on the powder object in the segmentation result image, the number of effective powders and proportion of the original wandering map and the predicted output delay image is [112, 33.6%] and [240, 40.6%]. The research results show that the cladding image output from the original image after the super resolution model has been significantly improved in terms of clarity and quality as well as the optimization and enhancement of the details of the monitored object.
Defect detection methods for photovoltaic (PV) devices based on electroluminescence (EL) imaging technology are crucial to maintain productivity and prolong components’ life. However, due to the lack of feature extraction capability for morphologically complex defects and some small defects, the traditional object detection algorithms perform not well in EL defect detection. Therefore, an improved algorithm based on the Faster Region-based Convolutional Network algorithm (Faster R-CNN) is proposed to improve the detection performance of multi-scale defects. Specifically, an improved residual module based on deformable convolution and attention module is proposed to improve the detection rate of morphologically complex defects. And a Feature Pyramid Network (FPN) is utilized in the proposed algorithm to improve the detection performance of small defects. In addition, the GIOU loss, instead of the original smooth L1 loss, is utilized to improve the boundary box regression accuracy. Experimental results on the detection of the EL defects show the high efficiency of the proposed algorithm. The method is expected to provide more guiding feedback in both practical design and reliable diagnosis of the PV industry.
Stability detection of melt pool has become a challenging task in laser cladding, due to the high temperature and brightness during laser cladding. In this paper, an enhanced mask R-CNN for object instance segmentation of melt pool is proposed to boost the performance of detection. In order to enrich the dataset and improve the generalization of the neural network, the data enhancement method of elastic deformations is used to simulate the irregular deformation of melt pool topography caused by the interference of the external environment. Meanwhile, the MobilenetV2 structure is introduced into mask R-CNN to solve the problem of a large number of parameters and slow running speed of network model, and transformer model was used to replace the classifier of the original network. Experimental results show that the proposed method can improve testing speed by 14.7% without decreasing the segmentation accuracy. Finally, a dynamic stability detection method of melt pool is proposed in this paper.
The variousness as well as the inaccessibility of defect characteristics make the defect detection of monocrystalline silicon solar cells more challenging. To address these problems, a novel domain adaptive target detection algorithm based on pseudo-label learning, which is an efficient and feasible weakly supervised learning method, is proposed in this paper. Firstly, in the early stage of the model, the loss function is improved to solve the problem that the model is not easy to converge due to a large number of false labels in the pseudo labels. Then, to classify the sample space better, entropy regularization is applied to the sample boundary data, and the unlabels of the target domain images are labeled with pseudo labels. Finally, the source domain data and the target domain data are used for training together, and the generalization performance of the model obtained is greatly improved. The results show that in solar cell image detection, the accuracy of the domain adaptive method based on pseudo-label learning can reach over 90%, which is better than the target detection accuracy of using only the source domain dataset.
With the development of modern society, people have higher requirements for the properties of metal materials. However, according to the traditional performance testing method, the prepared samples will be placed under the high-power metallographic microscope for artificial observation and analysis. It has low efficiency and is greatly affected by subjective human factors. In order to finish the task of material recognition and classification of metallographic images, this paper established database and used the deep learning method to research the process and method of convolution neural network, hierarchical learning, transfer learning and so on. The two classification algorithms based on convolution neural network and hierarchical transfer learning have achieved good results for material recognition and grading of metallographic images, respectively and the highest accuracy rate of classification is 98.89%, which provide a good way of thinking and foundation for subsequent metallographic image analysis and detection.
A crucial problem of applying ultrasonic guided wave flexible array transducer to weld detection is signal focusing. Thus, this paper proposes two focusing methods applied for array sensor in ultrasonic guided wave weld testing. By exciting every row of the array sensor simultaneously or not, we evaluate two methods on signal focusing. By comparing waveform and maximum amplitudes of focused waves generated by different number of rows in the transducer array, we find out that two proposed methods both have their advantages in guided wave focusing and indicate where these two methods should be used in real applications respectively. In results, the numerical simulations show that the guided wave focused by simultaneous excitation depends on the transducer array dimension, while the guided wave is independently focused by asynchronous excitation.
Butt welding is the typical welding mode for the fiber laser welding, and penetration status of the weld is critical point to assess welding quality. For the sake of solving the prediction of the penetration status in the fiber laser welding, a sparse representation prediction model was established to monitor the welding process. The sparse representation classification algorithm of using the K-SVD algorithm constructed the dictionary was used to predict the weld penetration status. However, the dictionary trained by K-SVD algorithm was not discriminative and the prediction accuracy was low. A D-KSVD algorithm with the discriminant dictionary learning mode was proposed, and the initialization method of the initial dictionary was improved to enhance the dictionary discriminant performance. The experiment result indicates the average recognition accuracy of the improved D-KSVD algorithm is 4 percentage points higher than the D-KSVD algorithm, and the accuracy of the weld penetration status prediction can reach 0.943, which shows that the recognition accuracy of the DKSVD algorithm is significantly higher than the K-SVD algorithm, and the dictionary learning with adding the discriminant learning can effectively improve the prediction of weld penetration status.
In this paper, we concerns on the location of the micro manipulator in a micro manipulating system based on SLM and
two CCDs. After getting the depth image of an object and its background by SLM, we stress the importance on how to
separate them and then obtain the position information of the object. Firstly, we obtain HSI model of the micro-vision
image according to the transform function from RGB to HSI. Then, we draw the curve of H component of the image.
According to this image, we separate the object from the background successfully. After that, applying moment method,
we calculate mass centric coordinate of the object. Finally, applying the projective geometry in the stereo vision, we get
centric coordinate of the object combined with the coordinate system of the CCD. Therefore, we realize the location of
the micro object. This method is applied in practical micro-manipulating and fast and high-accurate location of object is
obtained with a locating accuracy less than 1%.
The die or mould used for extruding aluminum wallboard profile is in serious work conditions, so it is easy to appear drawbacks in the mould such as non-uniform stress and strain distributions, crack initiation and propagation, elastic warp, and even plastic distortion. As we know, the extrusion die or mould is subject to complex loads including the extrusion pressure, friction and thermal load, which make the mould complicated and hard to be designed and analyzed by using conventional analytical method. In this paper, we applied Deform-3D, FEA (Finite Element Analysis) software used frequently in all engineering fields, to simulate three-dimensional extruding process of aluminum profile. The simulation results show that the deformation increases gradually from inside to outside. Exterior deformation contour distribution is relative uniform since the influence of inner holes on deformation is small, and the contour form is regular and similar with the shape of the mould. However, the interior deformation contour is irregular as the influence of holes with basically symmetric equivalent curves. At the middle of the mould, the deformation reaches the largest, it reaches 0.633mm. The deformation of the mould can be reduced by increasing the distance between two holes or increasing thickness of the mould. Experiment result accords with simulation. The simulation process and results ensure the feasibility of finite element method, providing the support for mould design and structural optimization.
This paper presents a new way for online multi-dimensional measurement of complex workpieces, which is based on the stereo computer vision. We also advanced a new algorithm to obtain the disparity image for the image matching. In this algorithm we combine the pyramid image-matching strategy and the means of dynamic programming, and then apply the sub-pixel matching. Thus, the characteristic lines and circles of the measured workpiece are reconstructed in the image space. Using Hough transform, we obtain the information of the circles and the lines in the measurand. Then we get the sizes required. Compared with those obtained by CMM, the results have quite high accuracy.
As the rapid market need of micro-electro-mechanical systems engineering gives it the wide development and application ranging from mobile phones to medical apparatus, the need of metal micro-parts is increasing gradually. Microforming technology challenges the plastic processing technology. The findings have shown that if the grain size of the specimen remains constant, the flow stress changes with the increasing miniaturization, and also the necking elongation and the uniform elongation etc. It is impossible to get the specimen material properties in conventional tensile test machine, especially in the high precision demand. Therefore, one new measurement method for getting the specimen material-mechanical property with high precision is initiated. With this method, coupled with the high speed of Charge Coupled Device (CCD) camera and high precision of Coordinate Measuring Machine (CMM), the elongation and tensile strain in the gauge length are obtained. The elongation, yield stress and other mechanical properties can be calculated from the relationship between the images and CCD camera movement. This measuring method can be extended into other experiments, such as the alignment of the tool and specimen, micro-drawing process.
Micro-machining plays an important role in MEMS. In the micro-machining, the influences of the surface effect and the size effect must be considered. Moreover, the deformation and damage mechanics of the materials of the micro parts are different from those of the macro ones. Thus, traditional measuring methods are unsuitable for the micro-machining. In this paper, a new apparatus based on the computer vision is developed to measure the micro tensile strain of foil material in the micro-machining. A three-CCD computer vision system is constructed. Through image processing and calculation the micro-strain of the specimen is obtained. Therefore, on-line measurement of the micro-strain in the micro-machining can be realized by this system.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.