Papers by Keyword: Machine Vision

Abstract: This study proposes an automated framework for online cutting tool wear classification in CNC turning using low-cost optical equipment and Convolutional Neural Networks (CNNs). Longitudinal turning experiments were performed on CK45 medium carbon steel using a HAAS TL1 lathe under dry machining conditions. Tool wear evolution was monitored via a lathe-mounted digital microscope, with images classified into three distinct stages: Low (Vb<160 μm), Medium (160≤Vb≤200 μm), and Critical (Vb>200 μm). A shallow CNN architecture, consisting of three convolutional blocks and a Softmax output layer, was developed to balance model complexity with computational efficiency for potential edge deployment. To enhance robustness against positional changes, data augmentation techniques including random translations and rotations were applied. The results demonstrate good performance, with the model achieving 94.7% accuracy and a weighted F1-score of 95.4% on the testing subset. While the model showed exceptional performance in identifying Low and Medium wear, data scarcity in the Critical wear class remained a limiting factor for recall. Overall, the study confirms that shallow CNNs can accurately capture spatial hierarchies for image-based wear assessment.

Abstract: This paper presents an overview of advanced deep learning techniques and machine vision technologies aimed at automating defect recognition tasks with unparalleled accuracy and efficiency. Various methodologies, including deep random chains combined with adaptive Faster R-CNN, Gradient-weighted Flaw Detecting using Convolutional Neural Networks (CNNs), and established architectures like Faster R-CNN and YOLOv5, are discussed. These methods leverage CNNs’ robustness in image classification tasks and feature extraction capabilities to improve defect detection accuracy on machined components. Furthermore, the integration of machine vision with optical inspection platforms enables rapid defect recognition, classification, and localization, significantly enhancing the overall quality control process in manufacturing environments. Visualizations of defect recognition scores and improvements in accuracy demonstrate the effectiveness of these methodologies, highlighting their potential to drive efficiency and competitiveness in the manufacturing industry. Overall, the continuous evolution and integration of these technologies offer immense potential for transforming quality control practices and driving excellence in defect detection in machined parts.

Abstract: To quickly adapt to the fast-changing conditions in the modern markets and the global economy, manufacturers are adopting digital manufacturing methods and tools, instead of traditional paper-based processes, to release higher quality products more quickly and at a lower cost. The pharmaceutical industry has a high production standard in the world. Delivering a defective product (or package) can lead to customer complaints and may even result in the entire product series being returned in severe cases. To reach out to the tiny space of products and achieve a high pharmaceutical product dimensional standard, manufacturers must introduce commercial vision inspection systems for the quality inspection process. However, conventional commercial inspection systems are often of a high cost, thus making them unaffordable for micro, small, and medium-sized enterprises (MSMEs), particularly in developing countries. This paper proposes a cost-effective vision inspection system that intelligently measures critical plastic bottle dimensions. The system comprises three 4K industrial cameras, two LED lights, a customized measurement platform, and a laptop, making it more affordable for MSMEs. Under the appropriate illumination setting, a plastic bottle is positioned on the stage and viewed by the laptop screen in real-time. The middle camera captures the bottle image, followed by a series of image processing operations to obtain the region of interest (ROI), such as the snap cap radius and height. Then, extract the target bottle edges with the Canny edge detector. Lastly, the system calculates the pixel-based distance and converts it to the measurement results for records or decision-making. The proposed method demonstrates reliable dimensional detection abilities, offering a potential solution to reduce human workload and improve inspection productivity in measuring pharmaceutical bottles.

Abstract: Industrial image processing technology is widely introduced and applied in various fields such as industrial plants, bio-medical industry, agro technology, and also environmental fields. In addition, image acquisition devices are getting more compact and installed onto mobile phones and handy terminals. Which means that we always carry the devices, and can easily take images (still image and moving pictures) of good quality with high resolution, anytime and anywhere. In the research field, 2D code, known as QR code or Data Matrix, has a great potential for industrial applications. As we already know, QR code has been spread and used with various usage likewise ID recognition, URL display on the sticker, and/or the financial transaction's confirmation processing. Not only QR code, but Data Matrix is also frequently used as the printed tag on various material such as metal, wood, or plastic parts. These are used as a product ID or serial number to be recognized by using a mobile terminal. The 2D code technology can be a powerful tool to check and trace the marketing channel of each part of the product. On the other hand, Sustainable Development Goals (SDGs) have been proposed since 2015, researched and introduced to improve our environment and life. This paper describes the proposals of new application technology related to the concept of SDGs. The main contents are based on science and technology, but these have been developed and implemented by the young students of the college. The process and the details are described in the paper.

Abstract: Advances in agricultural automation, coupled with a general decline of available labour hasgenerated interest in automated harvesting of various crops. Paramount to the success of such systemsis the development of accurate, robust detection technologies and localization strategies. This paperpresents an overview of sensor technologies used in the detection and localization of green aspara-gus spears for robotic harvesting. Tactile, photoelectric, machine vision and time-of-flight sensors areinvestigated and their applicability for use in robotic asparagus harvesting is evaluated. Investigationof previous asparagus harvesting devices has revealed that no such device has yet achieved commer-cial viability. It was identified that this is likely due to weaknesses in currently employed detectiontechnologies, namely slow response times, high sensitivity to changes in ambient lighting conditionsand requirement for frequent manual calibration. Of the sensor technologies investigated it was foundthat time-of-flight cameras, such as the Microsoft Kinect V2 are the most feasible for the detectionof asparagus spears for robotic harvesting. It was concluded that further research would be conductedinto the application of such sensors into a commercially viable harvester.

Abstract: Microdroplet detection is the key for precision dispensing technology, which is widely applied in precise assembly of miniature parts, microelectronics packaging and biological tests. In order to measure droplet in micro scale for different working conditions, In this paper, an on-demand detection method--which is composed of statistical weighing detection to measure batch dispensing quality for large-scale production, real-time visual detection to measure droplet area and position accuracy for flexible production as well as pressing-plate method to measure nanoliter volume for single microdroplet detection--is presented. An automatic dispensing system to generate microdroplets is developed and the procedure of online detection based on machine vision of this dispensing system is introduced. Experimental results show that dispensing standard deviations measured by on-demand detection are 2.5 µg, 2.8 nl and 0.013 mm² for weighing method, real-time detection and pressing-plate method, respectively. The results also prove that the pressing-plate method can measure single micro droplet of 10 nl.

Abstract: Inspection plays a major role in any production process. The acceptance or rejection decision of the production lot depends on the inspection results. In recent years, the developments in machine vision techniques have made inspection easier. This paper aims at bottle cap inspection using machine vision techniques. Bottle neck may have defects such as absence of the cap, absence of the tamper ring and improper assembly of cap and tamper ring. This paper deals with checking of the above mentioned defects with a single image of the finished product. This system uses backlight technique. The image of the product obtained using camera is processed using image processing software and then the results obtained are used to accept or reject the particular product. This technique could be implemented in industries for a batch produced series of bottle coming in a conveyor. By using machine vision techniques it is ensured that not even a single defect related to bottle cap is left from observation. Thus the bottle cap inspection becomes easy, accurate and done in less time.

Abstract: The article presents a developed orginal model designed to increase the efficiency of the processes of the design of adaptive image analysis methods used for detection of surface defects in materials. The model was created as part of activities related to the Strategic Program – Specialized knowledge bases and expert systems for the simulation of complex processes. The methods presented in this paper make it possible to accelerate design process, including testing of algorithms developed for a wide range of surface defects, such as: cracks, discoloration, loss of materials, geometric distortion, or even the presence of defect agglomerations caused by corrosion, without the need for the acquisition of the physical image of the actual objects. This in many cases can be a significant problem for engineers who design automatic optical inspection systems, because the acquisition of test objects with specified defects which are characterized by a fixed range of values of selected parameters is not always possible. This paper presents a formal model designed to generate material defects on the surface of three-dimensional virtual objects, which is equivalent to the acquisition of actual data from vision systems. The model takes into account various surface characteristics such as their texture or roughness by using mapping by the Blinn method. The results of the use of the system developed for the classification of products represented in digital images for which image analysis algorithms have been based on so-called artificial intelligence in the form of dedicated neural networks are presented. As described in this paper, artificial neural networks are an example of adaptive models, and provide the ability to solve problems for which there are no deterministic models. The models, however, require the use of learning processes (training) with the use of extensive data sets, which in this case were generated with the use of the developed solution model.

Abstract: The authors present examples of the developed unique optomechatronic systems for the monitoring of fatigue of materials. The systems employ different solutions from the area of optomechatronic technologies like machine vision methods with single-camera and dual-camera vision systems. Additionally, the article discusses the most important functional features and application possibilities of the developed apparatus.

Abstract: This paper use the laser sensor to primary mapping the position of the artifacts. Then manipulator carry through accurately positioning on the artifacts by machine vision. Then the manipulator accurately capture the artifact by adjust error automatically. By using laser sensor and machine vision, the robot on industrial production line can more accurately positioning fetching artifacts and the requirements of industrial production line for precise localization of robot.