Real-Time Fabric Defect Detection Using Machine Vision and AI

Abstract

In spite of the modern weaving machines in the industry many defects that occur in fabric are in weaving process and those defects are still defected from human eye. A machine vision can be adapted which detects defects in weaving machines is proposed in this paper.

Introduction

The disfigured areas which harm the appearance and execution of a fabric might be known as 'a fabric defect'. In modern weaving trchnology, numerous sorts of fabric defect still happen, The reasons of which are many, for example, the yarn quality utilized in weaving, its r weaving mechanics.

The defects experienced inside production must be distinguished and remedied at the beginning phases of the production cycle. Most defects emerging in the production cycle of a fabric are as yet identified by human inspection. A woven texture bar is put on the texture examination machine and twisted from the back to the front pillar while disregarding an enlightened surface. The quality control staff needs to filter almost 2 meters width of texture (Figure 1) and must identify little subtleties that can be situated in a wide zone that is traveling through their visual field.

Get quality help now

Prof. Finch

Verified writer

Proficient in: Technology

4.7 (346)

“ This writer never make an mistake for me always deliver long before due date. Am telling you man this writer is absolutely the best. ”

+84 relevant experts are online

Hire writer

The identification and classification of these defects are time consuming and tiring procedures.

In the best case, a quality control individual can recognize close to 60-70% of the defects present, and can't manage a texture more extensive than 2 meters. It is seen that the review speed of a texture woven with a productivity of even 97% is 30 m/min, and just about 60% of the defects are recognized

The issue is to plan a machine vision framework for fabric inspection machines and build up a calculation for fabric defect discovery naturally.

The framework will be convenient, and effectively versatile to a wide range of fabric examination machines. Fabric producers can adjust this framework by utilizing their current fabric assessment machine. The automated defect detection system will upgrade item quality and result in improved efficiency to fulfill both client demands and decrease cost

Related with off-quality. The employment cost will be decreased by replacing human vision identification with programmed visual recognition, where texture imperfections will be assessed equitably. An information base of the imperfections can be framed and they would then be able to be dissected all things considered. Thusly measurable assessment of denim quality can be accomplished. The input information obtained from the programmed review framework may gives the creation of great fabric with less defects in a shorter time.

Survey on image processing techniques numerous attempts are made to replace traditional human inspection by automated visual systems using camera nodes image processing routines. various kind of image processing methods such as gabor filters, wavelet analysis, and the fourier Transform have been applied for fabric defect detection. artificial in- telligence (AI) methods such as the neu- ral Network (NN), fuzzy Logic (FL) and fenetic Algorithm (GA) are also used for defect detection and classification studies. Studies are commonly performed off-line using digital images . Sample images are obtained by using different types of Image capturing devices or from a defective fabric image database. There are fewer studies on real-time defect detection than offline ones. Celik et al. suggested two different de- fect detection algorithms based on the wavelet transform and linear filters using the same system presented in this study.

Karayiannis et al. suggested a pilot system for fabric defect detection and classification in real-time. Double thresh-olding, binary filtering, binary labelling, multiresolution rotting via the wavelet transform, and statistical fabric feature removal methods were used for image analysis and defect segmentation. Eight types of classification were performed: no error, black vertical error, white vertical error, wrinkle, black hori- zontal error, white horizontal error, black spot, and white spot. This system was able to differentiate the eight types of defcts with an accuracy of 94%.

Goswami and Datta used morpho- logical operations such as corrosion and opening operations as well as the Fou- rier Transform for fabric image analysis. Fabric images that contain a knot and thick yarn were detected by morphologi- cal corrosion functioning Maketall proposed a model of a real-time computer vision system for detecting defects of textile fabrics. The Gabor function was used for defect division. Both online and offline tests were to be performed using the model sys tem.

Han and Zhang derived one even uniform Gabor filter mask and one odd uniform Gabor filter mask from an efficent Gabor filter. The limit of the efficient Gabor filter were accquired by Genetic Algorithm (GA). It was conclud- ed that even the symmetric Gabor filter was good at detecting blob-shaped fabric defects like knot, and the odd symmetric Gabor filter was adept at detecting edge- shaped fabric defects like miss pick. The efficency of the system expressed is of 95.24% over all the detection rate.

Cho et al. suggested a model system for fabric defect detection. The defects observed by the system were warp float, broken pick, hole and oil spot. It was stated that a nearly optimal recognition rate was obtained for an oily spot and spot. However, the recognition rate for warp and pick float was approximately 80%.

Conci & Proença had compared the results of three image-processing methods such as fractal dimension, thresholding, and edge detection. In this study, the relationship between the optimality of each method and the type of fabric fault were presented. Ten types of faults were observed, all of which were plain weave. Tests were implemented using 100 images of fabrics. The results showed that fractal dimension was the reliable way as it detected correctly all the types of defects. By using edge detection, faults were detected 98% of time while the threshold approach correctly 82% of these defects.

Hu and Tsai observed the best packet wavelet transform to detect fabric defects. The three major effects on the classification rate of fabric defect inspection comprising wavelets with various maximum vanishing moments, different numbers of resolution levels, and different scaled fabric images were observed. Basically Four kinds of fabric defects such as missing ends, missing picks, broken fabrics, and oil stains were used. A back propagation neural network (BPNN) was used to classify the fabric defects.

Methodology

1. Image Processing Techniques: Utilization of Gabor filters, wavelet analysis, and Fourier Transform to process fabric images and detect defects. These techniques form the foundation for analyzing the texture and identifying irregularities in fabric samples.
2. AI Methods for Defect Detection: Incorporation of Neural Networks (NN), Fuzzy Logic (FL), and Genetic Algorithms (GA) to classify and accurately identify different types of fabric defects. These methods allow for the automation of defect detection and classification processes.
3. Machine Vision System Configuration: The system comprises an industrial fabric inspection machine, illumination unit, CCD line-scan camera, frame grabber, and rotary encoder. This configuration is crucial for capturing high-quality images of the fabric as it undergoes inspection.
4. Real-time Fabric Detection System: The implementation of a real-time system capable of identifying various fabric defects, such as warp lacking, weft lacking, holes, soiled yarn, yarn flow, and water stains, with high accuracy. The system's performance is evaluated based on the detection rate, showcasing its effectiveness in identifying defects.
5. Algorithm Development: The development of an algorithm utilizing convolutional neural networks and noise reduction techniques to process images and detect defects. The algorithm consists of four basic steps: convolution, max pooling, flattening, and the dense layer, each contributing to the system's ability to accurately identify fabric defects.

Conclusions

There are many different vision system alternatives which are proposed for real-time defect detection a new fabric defct detection machine design is present. A vision detection system which is easily adaptable to any fabric defect detection machine was developed in our study. Now textile industries can easily afford this system and set it up on their current fabric defect detection machines. By using this system the fabric inspection process can be done in a shorter time and less fabric quality inspection persons are to be needed. Fabric inspection processes can be achieved briskly with elimination of any disputes over the quality of the fabric. The performance of the system is tested on fabric. The defect detection rate can be increased by using high performance computers. Also the size of the defects detected can be decreased by higher resolution cameras and specially manufactured illumination.

References

1. Anagnostopoulos C, Vergados D, Kaya- fas E, Loumos V, Stassinopoulos G. A computer vision approach for textile quality control. J. of Vis. and Comp. Ani. 2001; 12: 31–44.
2. Bodnarova A, Bennamoun M, Latham S. Optimal Gabor filters for textile flaw de- tection. Pattern Recognit.2002; 35: 2973– 2991.
3. Çelik Hİ, Dülger LC, Topalbekiroğlu M. Development of a machine vision sys- tem: real-time fabric defect detection and classification with neural networks, J. Text. Inst. 2014; 105, 6: 575-585.
4. Çelik Hİ, Dülger LC, Topalbekiroğlu M. Fabric defect detection using linear fil- tering and morphological operations. Ind. J. Fib. Text. R. 2014; 39, Sept: 254- 259.
5. Karayiannis YA, Stojanovic R, Mitropou- los P. Defect detection and classification on web textile fabric using multiresolu- tion decomposition and neural networks. In: The 6th IEEE Int. Conf. on Elect. Cir- cuits and Systems. 1999; 2: 765 – 768.
6. Goswami MB, Datta KA. Detecting de- fects in fabric with laser-based morpho- logical image processing. Text. Res. J. 2000; 70(9): 758-762.
7. Mak KL, Peng P, Lau HYK. A real-time computer vision system for detecting de- fects in textile fabrics. In: IEEE Int. Conf. on Industrial Tech., Hong Kong, China, 2005, pp. 469-474.
8. Han R, Zhang L. Fabric defect detection method based on Gabor filter mask. In: Global Congress on Intell. Systems, Xia- men, China, 2009: 184-188.
9. Cho CS, Chung BM, Moo-Jin P. Devel- opment of real-time vision-based fabric inspection system. IEEE Trans. on In- dustrial Elect. 2005; 52: 1073-1079.
10. Conci A, Proença CB. A comparison be- tween image-processing approaches to textile inspection. J. Text. Inst. 2000; 91, 2: 317-323.
11. Hu MC, Tsai IS. The inspection of fabric defects by using wavelet transform. J. Text. Inst., 2000; 91, 3: 420-443.
12. Mak KL, Peng P, Yiu KFC. Fabric defect detection using morphological filters. Image Vision Comput. 2009; 27: 1585– 1592.
13. Furferi R, Governi L. Machine vision tool for real-time detection of defects on tex- tile raw fabrics. J. Text. Inst. 2008; 99, 1: 57-66.
14. Çelik HI. Development of an intelligent fabric defect inspection system. Univer- sity Of Gaziantep Graduate School Of Natural & Applied Sciences. Ph.D The- sis In Mechanical Engineering, 2013.
15. Çelik Hİ, Dülger LC, Topalbekiroğlu M. Developing an algorithm for defect de- tection of denim fabric: gabor filter meth- od. Teks ve Konf. 2013; 23(2): 255-260.
16. Jain AK, Farrokhnia F. Unsupervised texture segmentation using Gabor fil- ters. Pattern Recognit. 1991; 24(12): 1167–1186.
17. Porat M, Zeevi YY. The generalized Gabor scheme of image representation in biological and machine vision. IEEE Trans. on Pattern Analysis and Mach. Intell. 1988; 10: 452– 468.
18. Bovik AC, Clark M, Geisler WS. Multi- channel texture analysis using localized spatial filters. IEEE Trans. on Pattern Analysis and Mach. Intell. 1990; 12, 1: 55–73.
19. Kumar A, Pang G. Fabric defect seg- mentation using multichannel blob de- tectors. Opt. Eng. 2000; 39(12): 3176- 3190.
20. Grigorescu SE, Petkov N, Kruizinga P. Comparison of texture features based on Gabor filters. IEEE Transactions on Im- age Processing 2002; 11, 1: 1160 – 1167.
21. The MathWorks, Inc. Create prede- fined 2-D filter, http://www.mathworks. com/help/toolbox/images/ref/fspecial. html#bqkft1d (accessed 10 June 2012).
22. Sonka M, Hlavac V, Boyle R. Image pro- cessing, analysis and machine vision international student edition. 3rd. ed. Toronto, 2008: 661-665.
23. Solomon C, Breckon T. Fundamentals of digital image processing a practical approach with examples in MATLAB. Chichester, 2011: 200-202.
24. Mak KL, Peng P. Detecting defects in textile fabrics with optimal Gabor filters. World Aca. of Sci. Eng. and Tech. 2006; 13: 75-80.