Implementation and potentials of a machine vision system in a series production using deep learning and low-cost hardware

Würschinger, Hubert; Mühlbauer, M.; Winter, Michael; Engelbrecht, Michael; Hanenkamp, Nico

doi:10.1016/j.procir.2020.01.121

Cited by 28 publications

(8 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…It is clear that the proposed approach may be applied in coal processing based on the total ash content error and the results obtained. Würschinger et al [31] introduced that maintaining production efficiency while achieving ever-increasing quality requirements is essential in industrial operations. Dealing with these issues, machine vision systems can be employed.…”

Section: Related Workmentioning

confidence: 99%

Research on Machine Vision System Design Based on Deep Learning Neural Network

Tao

Ju-jiang

2022

Wireless Communications and Mobile Computing

View full text Add to dashboard Cite

A machine vision system (MVS) is a technology that can analyze and recognize still or moving pictures using a computer. It is a branch of computer vision that looks like a security camera but can automatically capture, evaluate, and analyze images. The drawbacks are obvious. In the event of a computer vision system failure, firms must have a team of highly trained people with a thorough understanding of the distinctions. Artificial neural networks with numerous layers between the input and output layers are deep neural networks (DNN). Neurons, synapses, weights, biases, and functions are all part of any neural network, regardless of the kind. Many of the challenges in computer vision revolve around using convolutional neural networks (CNN) to categorize images into predefined categories. Convolutional and pooling layers were utilized to decrease the image’s size before feeding the reduced data to fully connected layers. According to the paper, the MVS-CNN algorithm can analyze a picture and determine the value of various characteristics and objects inside it. It is called convolution when combining two functions to create a third function. It is a fusion of two different datasets. A CNN performs convolution on the input data to build a feature map using a filter or kernel. Using a convolutional neural network, an inverted residual block is introduced as the basic block to balance identification accuracy and processing efficiency. The suggested method’s higher inspection performance is achieved with a huge dataset of photos of faulty and defect-free bottles. The result is obtained from the proposed method, the standard deviation ratio is 83.56%, absolute error ratio is 77.26%, trajectory length difference ratio is 82.35%, source pattern radiation amplitude ratio is 86.25%, classification of accuracy ratio is 83.25%, and finally, overall percentage performance ratio is 90.26%.

show abstract

Section: Related Workmentioning

confidence: 99%

Research on Machine Vision System Design Based on Deep Learning Neural Network

Tao

Ju-jiang

2022

Wireless Communications and Mobile Computing

View full text Add to dashboard Cite

show abstract

“…Würschinger et al [17] use the deep learning model and transfer learning method in a computer vision system and implement it to a manufacturing system to increase the quality.…”

Section: B Deep-learning Techniquesmentioning

confidence: 99%

Deep Learning Based Classification of Military Cartridge Cases and Defect Segmentation

et al. 2022

View full text Add to dashboard Cite

The final stage of the production process in the industry is quality control. Quality control answers the question of is there a defect on the surface of the products. Frequently the quality control is performed manually. The disadvantages of manual quality control are high error rate (low accuracy), low product rate (low performance) and high expense rate (high cost). The solution is automatic quality control using machine vision systems. These systems classify the products and segment the defects on their surfaces by processing the images taken by cameras during the production process in real-time. Some products like military cartridge cases have metallic, cylindrical, non-uniform texture and highly reflective surface. So, the quality of images is very important. Another factor that affects the accuracy is the nonuniform texture of the product surface. Distinguishing the product non-uniform texture from defect texture is a challenging problem. In previous works, this problem have been tried to be solved with image processing and deep learning techniques and the accuracy of 97% and 96% have been obtained, appropriately. According to NATO standards, the accuracy of the classification of the military cartridge cases should be above 99%. In this work, the methodology for classification of the military cartridge cases and segmentation of the defects on their surfaces with non-uniform texture is proposed to increase the accuracy. In scope of the proposed methodology the datasets with non-defective, defective and labeled/masked image classes of the cartridge cases were created, the deep learning models to classify the military cartridge cases and segment the defects on their surfaces were proposed, implemented and obtained results were evaluated using the metrics such as Accuracy, Precision, Recall, F1-Score, Jaccard Index (JI) and Mean Intersection over Union (mIoU). Obtained results showed that the proposed methodology increased the accuracy of classification to 100% with the DenseNet169 model and the F1-Score of segmentation to 92.1% with Improved U-Net and ResUnet models.

show abstract

“…Louw and Droomer [3] proposed a Raspberry Pi-based hardware system to defects on toy trains in which OpenCV has been implemented for machine vision operations and Python has been selected for coding statistical classifiers. Würschinger et al [4] applied Convolutional Neural Networks (CNN) algorithm on Raspberry Pi to suggest an enhanced and low-cost deep learning solution for manufacturing lines where they try to detect chips on piston rods via a deep learning algorithm with two classes. In another application by Ardhy and Hariadi [5], Raspberry Pi, Python, and OpenCV integration has been utilized to inspect Printed Circuit Board (PCB) defects and authors suggested Adaptive Gaussian Threshold as the best defect identifier.…”

Section: Machine Vision Supported Quality Controlmentioning

confidence: 99%

Machine Vision Supported Quality Control Applications in Rotary Switch Production by Using Both Process FMEA and Design FMEA

KARACAN¹,

ERDOĞAN²,

İĞDİL³

et al. 2021

Natural and Applied Sciences Journal

View full text Add to dashboard Cite

Emerging in the past few decades, Industry 4.0 has wide effects over production lines with an increasing number of novel applications. These applications implement more than one of the tools of Industry 4.0. These tools include but are not limited to internet of things (IoT), big data, cloud computing, artificial intelligence, augmented reality, virtual reality, machine to machine communication (M2M), smart robot applications, etc. The aim of these efforts is mainly to acquire smarter manufacturing systems. With spreading Industry 4.0 methodologies, the role of sensors became more important to respond to new demands. One of the most important sensor types in this sense is the camera which now has wide variants in different forms. Applying machine vision algorithms via cameras grants optimization of many critical processes. In this perspective, the quality of both product and process should be handled as a key performance indicator that may be continuously enhanced for excellence. Machine vision algorithms may be adapted to check and manage quality in designated control points on the production lines. This study focuses on the control of the quality of rotary switches that are widely used in household appliances like ovens and washing machines. Rotary switches are critical components of an appliance since they direct the flow of electricity within the product. A failure in the functionality of this component directly causes the failure of the main product. Hence, the quality rate of rotary switches should be calculated in defective parts per million (dppm) units. An intense quality control procedure is required to achieve low dppm rates during production. As a real-life application, a camera system is integrated into the rotary switch production line on a selected point. Classification algorithms are developed on a cost-effective platform to perform visual quality checks of the rotary switches and qualify as "Ok" or "Defective". The selected point ensures a high percent check of quality criteria while enabling repair of the defective parts with minor interventions. This control aims to identify a defective rotary switch as soon as possible since most of the defects are irreversible IDUNAS

show abstract

Implementation and potentials of a machine vision system in a series production using deep learning and low-cost hardware

Cited by 28 publications

References 8 publications

Research on Machine Vision System Design Based on Deep Learning Neural Network

Research on Machine Vision System Design Based on Deep Learning Neural Network

Deep Learning Based Classification of Military Cartridge Cases and Defect Segmentation

Machine Vision Supported Quality Control Applications in Rotary Switch Production by Using Both Process FMEA and Design FMEA

Contact Info

Product

Resources

About