Detection and Segmentation of Manufacturing Defects with Convolutional Neural Networks and Transfer Learning

Ferguson, Max; Ak, Ronay; Lee, Yung-Tsun Tina; Law, Kincho H.

doi:10.1520/ssms20180033

Cited by 188 publications

(92 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Transfer learning can also reduce training data requirements by transferring pre-trained model weights [8], [9]. Ferguson et al [8] applied transfer learning to the CNNbased inspection of casting defects in X-ray images. This model was first trained using two large public image datasets and then optimized with a relatively small casting dataset.…”

Section: Related Workmentioning

confidence: 99%

“…Various approaches to representative feature collection have been proposed in the literature, to compensate for a lack of defect samples. This has included partitioning image patches to augment samples [7], re-using pre-trained models with transfer learning [8], [9], and using pixel-level labels to increase sample quantities [10] in a supervised model. For example, He et al [11] developed a defect classification model for steel surfaces using a semi-supervised network.…”

Section: Introductionmentioning

confidence: 99%

“…These methods have produced viable results but they also have specific limitations [7]. For example, identifying defect regions by partition image patches can be time-consuming in the forward process [8], [9]. Transfer learning is effective but imposes restrictions on network structure design [10].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Weakly Supervised Surface Defect Detection Based on Convolutional Neural Network

Liang

Deng

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Surface defect detection is a critical task in product quality assurance for manufacturing lines. The deep learning-based methods recently developed for defect detection are typically trained using a supervised learning strategy and large defect sample sets. Conventional methods often require additional pixel-level labeling or bounding boxes to predict the location of defects. However, the number of required samples and the time-intensive annotation process limits the practical use of these algorithms. As such, this study proposes a weakly supervised detection framework in which a CNN model is trained to identify surface cracks in motor commutators. The model was trained using small subsets of defect samples (∼5-30) and does not require a pre-trained network. This approach consists of localization and decision networks that simultaneously predict both the location and probability of defects. A new loss function was also developed to identify abnormal regions in a sample with accessible image-level labels. A collaboration learning strategy was then applied to utilize the loss function and compensate for imbalances at the pixel level. Experimental results using a small number of image-level training labels from a real industrial dataset exhibited a 99.5% recognition accuracy, which is comparable to relevant methods using pixel-level labels.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Weakly Supervised Surface Defect Detection Based on Convolutional Neural Network

Liang

Deng

et al. 2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…There are several studies that use sound data with a CNN to measure defects in gears to determine if they are defective. As another example, vibration data from bearings is used with a CNN [29][30][31]. CNNs have been applied to various manufacturing fields in addition to image processing.…”

Section: Image-based Failure Detection By Machine Vision For the Me Pmentioning

confidence: 99%

Image-based failure detection for material extrusion process using a convolutional neural network

Kim

Lee

Kim

et al. 2020

Int J Adv Manuf Technol

View full text Add to dashboard Cite

The material extrusion (ME) process is one of the most widely used 3D printing processes, especially considering its use of inexpensive materials. However, the error known as the “spaghetti-shape error,” related to filament tangling, is a common problem associated with the ME process. Once occurring, this issue, which consumes both time and materials, requires a restart of the entire process. In order to prevent this, the user must constantly monitor the process. In this research, a failure detection method which uses a webcam and deep learning is developed for the ME process. The webcam captures images and then analyzes them by machine learning based on a convolutional neural network (CNN), showing outstanding performance in both image classification and the recognition of objects. Sample images were trained based on a modified Visual Geometry Group Network (VGGNet) model and the trained model was evaluated, resulting in 97% accuracy. The pre-trained model was tested on a 3D printer monitoring system for its ability to recognize the “spaghetti-shape-error” and was able to detect 96% of abnormal deposition processes. The proposed method can analyze the ME process in real time and informs the user or halts the process when abnormal printing is detected.

show abstract

“…GAN was used to generate more data for the deep learning network to overcome the problem of small amount of data. Ferguson et al [28] used a defect detection system based on the mask region-based CNN (mask R-CNN) architecture to detect the casting defects on the GDXray dataset. Excellent performance is achieved based on transfer learning, using weights pre-trained on the ImageNet dataset, and then trained the defect detection system on the COCO (Common Objects in Context) dataset.…”

mentioning

confidence: 99%

Apparatus and Method of Defect Detection for Resin Films

et al. 2020

View full text Add to dashboard Cite

A defect inspection of resin films involves processes of detecting defects, size measuring, type classification and reflective action planning. It is not only a process requiring heavy investment in workforce, but also a tension between quality assurance with a 50-micrometer tolerance and visibility of the naked eye. To solve the difficulties of the workforce and time consumption processes of defect inspection, an apparatus is designed to collect high-quality images in one shot by leveraging a large field-of-view microscope at 2K resolution. Based on the image dataset, a two-step method is used to first locate possible defects and predict their types by a defect-shape-based deep learning model using the LeNet-5-adjusted network. The experimental results show that the proposed method can precisely locate the position and accurately inspect the fine-grained defects of resin films.Appl. Sci. 2020, 10, 1206 2 of 23 3.In some cases, the manufacturer may want to tolerate no defect larger than 25 µm. However, human vision is limited to objects larger than approximately 50 µm; this limit is not acceptable for high-quality products. 4.Defect detection is not immediate. Using human vision for the inspection process is considerably time consuming and inefficient. The average number of plastic resin films that can be checked by a person in a day is approximately 12 to 15.Automatic procedures, including machine vision-based methods of defect detection, have gained considerable popularity for their high speed, high precision, and real-time performance. Several strategies for defect detection have been explored in numerous studies. These machine vision-based methods are used in the inspection of many industrial products, such as metal [1], fabric [2], steel [3], car surfaces [4], and light-emitting diode chips [5]. The methods of surface defect detection can be divided into three types, as depicted in Figure 1, namely traditional image processing, feature extraction with machine learning, and deep learning methods. Appl. Sci. 2020, 10, 1206 2 of 23 miss some defects. Furthermore, it is nearly impossible to standardize human vision; each inspector's eyesight will have its own peculiarities. 3. In some cases, the manufacturer may want to tolerate no defect larger than 25 μm. However, human vision is limited to objects larger than approximately 50 μm; this limit is not acceptable for high-quality products. 4. Defect detection is not immediate. Using human vision for the inspection process is considerably time consuming and inefficient. The average number of plastic resin films that can be checked by a person in a day is approximately 12 to 15.Automatic procedures, including machine vision-based methods of defect detection, have gained considerable popularity for their high speed, high precision, and real-time performance. Several strategies for defect detection have been explored in numerous studies. These machine vision-based methods are used in the inspection of many industrial products, such as metal [1], fabric [2], steel [3],...

show abstract

Detection and Segmentation of Manufacturing Defects with Convolutional Neural Networks and Transfer Learning

Cited by 188 publications

References 48 publications

A Weakly Supervised Surface Defect Detection Based on Convolutional Neural Network

A Weakly Supervised Surface Defect Detection Based on Convolutional Neural Network

Image-based failure detection for material extrusion process using a convolutional neural network

Apparatus and Method of Defect Detection for Resin Films

Contact Info

Product

Resources

About