International Journal of Advance Engineering and Research Development

Transcription

1 Scientific Journal of Impact Factor (SJIF): 5.71 International Journal of Advance Engineering and Research Development Volume 5, Issue 04, April e-issn (O): p-issn (P): AUTOMATIC DEFECT MONITORING OF SPUNBOND NONWOVEN FABRICS & IT S ANALYSIS & GRADING USING IMAGE PROCESSING TECHNIQUES Prof. (Dr.) P. A. Khatwani 1, Prof.(Dr.) K. S. Desai 2 1 Sr. Professor, Dept. of Textile Technology, Sarvajanik College of Engg. Tech., Surat. 2 Associate Professor, Dept. of Textile Technology, Sarvajanik College of Engg. Tech., Surat. Abstract Achieving zero defect is the most important objective of any manufacturing industry in today s world, thus catering a need for an effective quality control system. With the increasing applications of technical textiles, the use of spunbond nonwovens has increased tremendously. The functional properties are of prime importance in technical textiles. Any defect in the product deteriorates the functional properties and thus the monitoring of the defects and the analysis of the same has become necessary. This paper deals with the designing and testing of an automatic inspection system for monitoring and analysis of defects in spunbond fabrics. Keywords- Fabric Defects, Fabric Inspection, Spunbond Nonwovens, Quality Monitoring, Image Analysis, Texture Analysis I.INTRODUCTION Visual Assessment of fabric quality in the Textile Sector by human inspectors is being done since long, and includes the detection of fabric defects for grading of fabrics. This method has a number of limitations like missing out of defects, high labour cost, time involved in process, offline inspection leading to wastage of defective fabric, etc. Therefore, an automatic fabric defect inspection is required to reduce the cost and time waste caused by defects. A lot of literature is found pertaining to automatic fabric defect detection in woven fabrics, but very little is found in the area of defect detection in spunbond fabrics. Since, the application areas of spunbond fabrics has increased tremendously ranging from agriculture, construction to medical and industrial applications in the last few years, a system to monitor and control the quality of these fabrics has become necessary. A cost and quality effective quality control system for spunbond fabrics has been designed and developed for defect detection and it s analysis, which has been described in this paper. II.BACKGROUND Any imperfection responsible for decreasing the worth and utility of the fabric may be termed as fabric defects [1]. Use of poor quality of fibres/yarns, error in manufacturing process of the fabric may result in fabric defects [2]. The grading of fabrics is normally influenced by the frequency and nature of the defects in fabrics. This fabric grading is the main factor in determining the price of the fabric and is often subjective. The profit margin varies with the grading, decreasing from first quality to the last quality. With the development and improvement in the production process, new materials and technology, the quality levels have increased to a great extent. Thus, the expectation of getting minimum defects is getting higher and higher. In this scenario, it becomes very necessary to maintain standards of quality in the fabrics. Implementing an efficient quality control system is the best solution to achieve defect free fabric. At present, the nature and frequency of defects are examined by human inspectors in majority of the small scale industries. The process involves high labour cost and is very time consuming. Also, the inspection may vary from individual to individual; errors in the inspection may result due to tiredness, boredom, fatigue of the inspectors. The method of inspection plays a significant role in detection of objectionable faults and hence proper grading of fabrics. Therefore an automatic inspection system may be highly desirable as it gives possibly the best objective and consistent evaluation [3]. An automated inspection system consists of a computer based vision system which may be offline or online. It mainly includes a fabric monitoring system for identification of defects and normally uses a high resolution camera for capturing images of the fabric. Image analysis is then used for determination of nature and frequency of defects [4,5,6,7,8] A variety of offline and online fabric inspection for woven fabrics has served as an important tool to cater to the needs of good quality control systems especially for the fabrics to be used by the garment industry [4,5,6,7,8]. In case of woven and knitted fabrics, which are largely used for apparel, defects lead to loss of aesthetic value of the fabrics [7]. All rights Reserved 2231

2 nonwoven fabrics are largely used for technical applications; the influence of defects on the functional fabrics is of more concern [8]. With the increasing applications of nonwovens, an effective inspection system has become a prime necessity for maintaining the standards of the functional fabrics as well as the nonwoven fabrics. Image analysis involves capturing of images of the fabric and then the surface characteristics of the fabric are analysed to identify any deformity or variability in it[4,5,8,9]. Various approaches and algorithms using computer vision and image processing have been tried and implemented for woven and knitted fabrics. [5-8, 10-20]. The studies relating to it suggests, that each method had its own advantages and disadvantages. It was also found that very limited studies have been done in the area of inspection systems for nonwoven fabrics. Since, nonwoven have a fibrous structure, it s texture is completely different from woven and knitted fabric, there is a need to find out the most suitable method for texture analysis of nonwovens. The structure of nonwovens varies according to it s method of production. In this paper, we have restricted the study to spunbond nonwovens. III. MATERIALS & METHODS A fabric quality monitoring device with a fabric unwinding section, image acquisition section, fabric rewinding section and a monitoring and analysis section, which was developed has been used for capturing of images of fabric in roll form. The device included motorised driving arrangement, vertical top illumination, frame for camera mounting, replacement of the fabric take up and delivery roll spindles by package holders, covers for the device along with the switches for operation of the device. Figure 1 shows the main parts and the driving arrangement of the fabric delivery and take up roll. An AC motor with 20rpm, 100mA & 7 kg cm torque is mounted on the machine frame at the front. The drive from the motor is positively transferred to the fabric take-up roll by means of a chain and sprocket. The fabric delivery roll can be driven both positively and negatively. Sprockets are provided on each side of the package holder, which acts as dead weight for facilitating negative driving of the fabric delivery roll. Positive driving arrangement of the fabric take up roll was facilitated by providing side chain on the sprockets. Figure 1: Drive and main parts of the Device Figure 2 shows the photograph of the developed device. The switch box on the top machine cover consists of 4 main switches; responsible for power supply to the device, starting of the motor, using Top Illumination & using Bottom Illumination. A power and an USB cable are provided for getting the power supply to the machine and inter-phasing with a PC / Laptop All rights Reserved 2232

3 Figure 2: Final System Polypropylene spunbond fabrics of different GSM ranging from 40 to 135 were manufactured on Chinese make spunbond machine, 1.6m wide and having a capacity of 5 tonnes/day at M/S. Wovlene Tecfab India, Surat. About 11 different varieties of fabrics were produced. They were of full width and had to be cut to obtain a fabric width of 18 cms as per the requirement of the device developed. After the cutting of the fabric width, the fabrics of varying specifications were stitched to obtain a roll. The roll so obtained was ready to be monitored and analysed by the developed device. Defects occurring commonly in the spunbond fabrics were studied and 6 types of commonly occurring defects were considered in the study. The defects considered were the defects which were obtained in the manufactured fabric as a result of the fabric manufacturing process. The identified defects, it s definition, principal causes and remedy has been tabulated in Table 1. Table 1: Identified Defects in Spunbond Fabrics Sr. No. Fabric Defect Definition Principal Causes Remedy 1. Drops / bond point fusion Fused fibres on surface Breaking of bundle of filaments during the process. Proper setting of draw ratio. 2. Pinholes Very small holes in fabric 3. Wrinkles Wrinkle formation 4. Hard filaments Fused filaments on surface 5. Hole Holes in fabric / web Damaged surface of delivery roller. Improper tension across the width of fabric. Breaking of filaments during the process. Improper supply of polymeric material across the width of fabric, blockage of spinnerette holes. Filing of surface of roller. Maintaining uniform tension. Proper setting of draw ratio. Maintaining proper supply of polymeric material across the width of fabric, cleaning of spinnerrete. 6. Calendar cut Cut marks due to calendaring Rough surface of calendar roll. Polishing of surface of roller. The identified defects were not found in all varieties of spunbond fabrics. The actual images of the identified defects have been shown in Figure All rights Reserved 2233

4 Figure 3: Actual images of the 6 identified defects More than 400 images from the prepared fabric roll were captured for studying the basic surface characteristics of the different varieties of fabric in the prepared roll. The images were processed and the image parameters like mean intensity of the images, study of histogram and its properties of each of the fabric were studied using gray level conversion, contrast adjustment and studying their histogram. These values were used to determine the image processing parameters for the defect detection. The next step involved taking the images of the defective regions containing the 6 identified types of defects and processing them to learn the pattern of each type of defect. Histogram analysis of the images indicated the presence of major defects easily, however the minor defects especially the one whose intensity levels are almost same as the structure itself or the local defects could not be identified. A processing algorithm was designed by optimising various image parameters and also incorporating further processing of the images was developed for the defect extraction. The main steps involved in the processing can be seen in Figure 4. Figure 4: Steps involved in processing of Images for Defect Detection and Classification The images acquired were in true colour and they store the information of Red, Green and Blue (RGB) levels for each pixel. Owing to high computation and high memory involved with the RGB images they were converted into gray images. A gray image has only gray level values for each pixel (0-255). The grayscale images were enhanced for getting more information about the images, especially to find any regions different from the background. This was done using adaptive contrast stretch, which was found suitable for spunbond fabrics. The images so processed still had a lot of noise, which is normally removed using any filtering technique in the woven fabrics. But the spunbond fabrics have a random fibrous structure and therefore use of any filtering technique resulted in loss of important structural information All rights Reserved 2234

5 therefore only contrast enhancement of the images was done. As the images could not be processed for noise removal, deciding of common threshold for defect extraction was found very difficult. The other typical characteristic of spunbond nonwoven fabric structure is random thick and thin places or we can say minor mass variations is common in these fabrics and cannot be accounted for a defect. A solution to this problem was found by taking different thresholds for extracting the extreme light and extreme dark regions. Morphological operations were done on the binary images obtained after thresholding to correct any errors in the images. Feature extraction of the defects using blob analysis was done to extract the dimensions of the defects. The connected components in the binary images were found out, which are known as objects or blob. Each blob has its own statistics which were found out. Stepwise elimination of these blobs was done to identify the defective regions. The blobs with minimum area which cannot be termed as defect were identified and eliminated in the first step. In the second step the features of the remaining blobs were obtained and the general parameters for the defects were calculated. The same were compared with the statistics obtained by manual - visual examination for validation and classification. The designed algorithm for the optimum defect detection was tested and validated by three means: The area, shape and size of the defects were extracted using the algorithm which was compared with the visual examination of the defects. Based on the defect statistics, the grading of the defects was done. Multiple images of the same defect regions were captured and processed to check for any variation in the definition and grading of defects. Regions of the fabric other than those studied for defect detection were captured and processed to grade the same. On the basis of the defect parameters obtained as result of the processing of images of the fabric lot & considering the proposed classification of defect, a fabric grading system was developed. The Defect Classification is shown in Table 2. Table 2: Defect Classification Defect Name (DN) Drops / bond point fusion, Pinholes, Wrinkles, Hard filaments, Holes, Calendar cut Defect Size (DS) Mendable- 10 % Defective Area; Permissible- 30% Defective Area; Critical - 60% Defective Area; Rejected - 80% Defective Area Defect Frequency (DF) Frequency of occurrence of defects Defect Orientation (DO) Machine Direction / Warp Way; Cross Direction / Weft Way The Proposed Grading System is shown below: Grade of Fabric A (Best) B (Good) C (Poor) D (Rejected) Proposed Performance of Fabric The defect has no or very negligible influence, the fabric can thus be used for suggested applications. DN - All DS Mendable (10% Defective Area / upto 3 length of defect) DF 1- within an image frame, 10% of the total images for the fabric roll. DO - any Substandard applications of suggested areas are possible with this grade of fabrics. DN - All DS Permissible (30% Defective Area / 3-6 length of defect) DF 1- within an image frame, 30% of the total images for the fabric roll. DO - any Can be considered after repairing or taking preventive measures for suggested areas of applications. DN - All DS Critical (60% Defective Area / upto 6-9 length of defect) DF more than 1 within an image frame, 60% of the total images for the fabric roll. DO any Not to be considered for any suggested applications. DN - All DS Rejected (80% Defective Area more than 9 length of defect) DF more than 1 within an image frame, 80% of the total images for the fabric roll. DO - All rights Reserved 2235

6 IV. RESULTS AND DISCUSSION The images of defect free regions of spunbond nonwovens were processed. It was found that there was some difference between the intensity values of each type of fabric, and the same might influence the thresholding value. The contrast enhancement in the images increases the range of the high and low level intensity values giving a considerable improvement in the visual appearance of the images. The higher GSM fabric had maximum mean intensity while the lower GSM had minimum mean intensity, but the other fabrics exhibited almost same mean intensity value. The variation in the intensity levels in the images could be accounted due to the basic fibrous structure of the nonwovens. Again here also no significant difference was found between the images for different GSM fabrics. A comparison of the mean intensity values of unprocessed and enhanced defective images and the general set of images have been shown in Figure 5. Figure 5: Comparison of the Mean Intensity Values between the unprocessed & enhanced Images of the Defects & general Fabric in Spunbond No significant difference was found from the basic image parameters between the images of defect free and defective regions. Difference in the Mean Intensity Values can be seen between major and minor defects. Therefore, histograms of a large amount of images for all varieties of spunbond fabrics were studied. The histograms of fabrics void of defects show a single peak. Also it was found that the shape and pattern of intensity distribution of each sample varies. The histograms were not symmetric around the mean intensity values. The defect wise histogram of the images of the fabrics with defects and with its corresponding defect free fabric image for each defect have been shown in Figure 6. The difference in the shape and intensity distribution can be accounted due to the variation in the texture and appearance of the fabrics. Figure 6: Histogram of Images of defect free region & various All rights Reserved 2236

7 Histogram analysis shows that presence of major defects could be easily detected, however the minor defects especially the one whose intensity levels are almost same as the structure itself could not be identified. The deviation of skewness in histograms of defective images from the defect free images is useful in determining the threshold for further processing of the image. It was seen that the defects could be broadly categorised as the ones having more skewness deviation towards dark intensity (drop / bond point fusion, stains) and the ones having more skewness deviation towards light intensity (pin hole, wrinkles, hard filaments, holes, calendar cut). However, a common threshold for all fabric images and defects could not be obtained. Since the nonwovens are characterised by thin and thick regions, different thresholds were used to obtain regions with the extreme light and dark regions. Threshold Intensity for dark regions was 100 and for light regions was 180 used. All pixels with intensity below 100 were eliminated to obtain the extreme dark regions and all pixels with intensity above 180 were eliminated to obtain the extreme light regions. The visual results of the grayscale image, binary image after thresholding and the image with highlighted defective regions after the further processing of the images with defective regions of each defect is shown the Figure 7. Figure 7: Grayscale image, binary image, highlighted defective regions of each defect As compared to the woven fabrics, the extraction of the defects is quite difficult due to the presence of thin - thick regions in the nonwoven fabrics. The detection rate of the local defects i.e. the one tending to merge with the basic structure was found to be low. The area of the identified defects was calculated and the same has been tabulated in Table 3. The values were compared with the values obtained by manual - visual examination of the defects and can be seen in Table All rights Reserved 2237

8 TABLE 3: Defect Statistics obtained from the System Sr. No. Defect Name GSM Defect Statistics obtained from the System Defect Statistics obtained from Manual - visual Examination Defect Area Defect Type Defect Area Defect Type (sq. cm) (sq. cm) 1 Drop / Bond Pt. Fusion Irregular(dark) 1.54 Spots 2 Pin Hole Small circle 0.33 Small hole (light) 3 Wrinkle Irregular lines 2 lines (light) 4 Hard Filament Irregular lines 9.17 Patch 5 Hole Irregular (light) Holes 6 Calender Cut Very Fine line / curve 10 Fine line / curves It can be seen from Table 3 that the defect statistics obtained from the system are quite similar to the ones obtained from manual - visual examination. The detection rate of Calender Cut was found to be less as compared to the other defects, the reason being very less difference between the intensity levels of the defect and defect free area. Also the defect was very fine, however some information about the defect could be gathered to mark presence of little variation. The fabric images had multiple defective regions and so the grading of the spunbond images was done. The grading was based on the total defective area, objectionable defective area and the number of defective areas. The total defective area present in the image, percentage defective area, objectionable defective area, percentage defective area, total number of objectionable defects were extracted from the images using the algorithm developed and translated into grades and the same have been tabulated in column 9 of Table 4. Column 10 shows comparison of these grades with the manual - visual grading of the fabrics. Table 4: Grading of Spunbond Images using the System Main Defect Present Drop / Bond Pt. Fusion Fabric Sample GS M Total Defec t-ive Area % Total Defecti ve Area Number of Objectio n-able Defects Objectio n-able Area % Objectio n-able Defectiv e Area GRAD E OF FABRI C Manua l - visual Gradi ng NS B Minor Pin Hole NS A Minor Wrinkle NS B Major Hard Filament NS B Minor Hole NS D Major 8 Calender Cut NS A Minor The results obtained as above were validated by capturing multiple images of same fabric regions and have been summarized in Figure 8. The results show the variability of the order of only 5-10 % in the defect statistics which is considered to be negligible, while no variation was seen in the grading of the images thus giving a consistent result. Figure 8: Comparison of Objectionable Area Multiple Images of Regions with same All rights Reserved 2238

9 The developed algorithm was thus found reliable and was tested for a number of images of other fabric regions. Some of the images with highlighted defective regions and the grading so obtained have been shown in Figures 9. Figure 9: Highlighted Defective Regions in Test Images Table 5: Comparison between the Grading achieved with developed System as against Manual - visual grading Test Image Grading As Per Software Manual - visual Grading 1 C Major 2 C Major 3 B Minor 4 A No Faults 5 B Major 6 A No Faults 7 A No Faults 8 A No Faults 9 B Minor It can be seen from Table 5 that Test Image 1 & 2 have been graded as C quality which matches perfectly with the severity of the defective region. Also Test Image 4, 6, 7 & 8 which had no defective regions had been graded as A quality. The grading obtained for Test Image 3 & 9 was also found to be perfect considering the severity of the defect. Test Image 5 had little deviation from the actual severity as it was graded of B quality by the system. It can be concluded that 8 out of 9 Test Images gave perfect results and therefore it can be said that the algorithm is able to detect the defective regions and the results show the variability of the order of only 5-10 % which is considered to be negligible. IV. CONCLUSION It was found that the developed system was able to detect the defective regions quite well as well grade the fabrics according to the severity of the defects for the spunbond nonwoven fabrics. The detection rate of very minor defects like calender cut was very low as there was very less difference between the intensity levels of the defect and defect free area. It was found that the retention of information of very fine defects in spunbond fabric was difficult as the some amount of structural variation is a typical characteristic of nonwoven fabrics. The exact size of defects in case of very minor defects was mot obtained. However some information about the defect could be gathered to mark presence of little variation and which could be useful in final grading of the fabric. The results were very desirable in case of the major defects. Thus we could conclude that the designed & developed prototype of device well supported with the user friendly software module to help the users in selection of proper quality of nonwoven / functional fabrics for specific end use applications. It will also avoid unnecessary wastage of time and materials, which otherwise would be due to wrong selection of materials for any specific All rights Reserved 2239

10 REFERENCES [1] J. B. Goldberg. (1950). Fabric Defects-Case Histories of Imperfections in Woven Cotton and Rayon Fabrics. Mc Graw-Hill Book Company, INC., USA. [2] Kumar A.: Computer vision-based fabric defect detection: a survey, IEEE, Transactions on Industrial Electronics, Vol. 55, Issue 1, 2008, pp [3] My thesis [4] Guruprasad R & Behera B K (2009) Automatic Fabric Inspection Systems. The Indian Textile Journal, 1 5. [5] Mahajan P, Kolhe S & Patil P (2009) A review of automatic fabric defect detection techniques. Advances in Computational Research, 1(2), [6] Brad R & Brad R (2004) A Vision System for Textile Fabric Defect Detection. Proceedings of 2nd International Istanbul Textile Congress, Istanbul, Turkey, April, [7] Islam A, Akhter S & Mursalin T E (2006) Automated Textile Defect Recognition System Using Computer Vision and Artificial Neural Networks. World Academy of Science, Engineering and Technology, 1 6. [8] Ngan H Y T, Pang G K H & Yung N H C (2011) Automated fabric defect detection - A review. Image and Vision Computing, 29(7), [9] Chan C, Liu H, Kwan T, Pang G (1998) Automation Technology for Fabric Inspection System.[Online].Available: [10] Zhang J & Meng X (2010) A Fabric Defect Detection System Based on Image Recognition. In Intelligent Systems and Applications (ISA), Proceedings of nd International Workshop, 1 4. [11] Thilepa R & Thanikachalam M (2010) A Paper on Automatic Fabrics Fault Processing Using Image Processing Technique In MATLAB. Signal & Image Processing : An International Journal, 1(2), [12] Weickert J (1999) A Real-Time Algorithm for Assessing In homogeneities in Fabrics. Real Time Imaging, 5, [13] Loonkar M S & Mishra D (2015) A Survey-Defect Detection and Classification for Fabric Texture Defects in Textile Industry. International Journal of Computer Science and Information Security, 13(5), [14] Niles S N, Fernando S & Lanerolle W D G (2015) A System for Analysis, Categorisation and Grading of Fabric Defects using Computer Vision. RJTA, 19(1), [15] Singh U, Moitra T, Dubey N & Patil M V (2015) Automated Fabric Defect Detection Using MATLAB. International Journal of Advanced Research in Computer Engineering & Science, 03(06), [16] Brad R & Brad R (2004) Automated Fabric Defect Inspection for Quality Assurance Systems. The 83rd Textile Institute World Conference, Shanghai, (1996), [17] Karunamoorthy B, Somasundareswari D & Sethu S P (2015) Automated Patterned Fabric Fault Detection Using Image Processing Technique In MATLAB. International Journal of Advanced Research in Computer Engineering & Technology, 4(1), [18] Faisal R H, Rokonuzzaman M & Ahmed F (2014) Automated Fabric Defect Inspection : A Survey of Classifiers. International Journal in Foundations of Computer Science & Technology (IJFCST), 4(1), [19] Abouelela A, Abbas H M, Eldeeb H, Wahdan A A & Nassar S M (2005) Automated vision system for localizing structural defects in textile fabrics. Pattern Recognition Letters, 26(10), [20] Fazekas Z, Komuves J, Renyi I & Surjan L (1999) Automatic Visual Assessment of Fabric Quality. IEEE International Symposium on Industrial Electronics, 178 All rights Reserved 2240