Automatic Fabric Defects Inspection Machine

M A I M. Abhayarathne, I U Atthanayake

(Department of Mechanical Engineering, The Ope university of Sri Lannka,Nawala, Nugegoda, Sri Lanka)

Abstract: The textile industry is one of the most important industries in Sri Lanka. In most of the textile garment factories the defects of the fabrics are detected manually. The manual textile quality control usually depends on eye inspection. Famously, human visual assessment is drawn-out, tiring, and an exhausting errand, including perception, consideration and experience to recognize the fault occurrence. The precision of human visual assessment declines with dull positions and vast schedules. Some of the time slow, costly, and sporadic review is the outcome. In this manner, the programmed automatic visual review safeguards both the fabric quality inspector and the quality. This examination has exhibited that Textile Defect Recognition System is fit for distinguishing fabrics' imperfections with endorsed exactness with viability. With some products 100% inspection is important to ensure the stipulated quality or standard. The classifications for the automated fabric inspection approaches are expanding as the work is vast and complex. According to the algorithm used, the texture analysis problem is classified into different approaches. They are Structural, spectral, model-based methods, Unfortunately, the optimal plan does not yet exist for these vast numbers of applied methods, as each of them has some advantages and disadvantages.

Keywords: Fabric Inspection, Convolution Neural Network, Fabric Defects, Automation

1 Introduction

The textile industry is one of the most important industries in Sri Lanka. Substance defects have a significant impact on the price of fabrics that represents a major threat to the Garment industry. Defect detection system detectes small percentage of defects with well trained inspectors. The isnpection of different fabrics for defects can either be done manually or manually by human inspectors or automatically by computer vision techniques.

An automatic real time error detection system developed with computer vision techniques can increase the percentage of detection of defects. Such techniques with best method of inspection of fabrics being employed for quality control the textile manufacturing costs are reduced and the quality of the products increases. This research presents a new algorithm for detecting fabric defects in Textile Industry. The proposed algorithm uses a morphological process for automatic detection of defects in fabrics resulting the status of accepted or not accepted. The proposed algorithm produces promising results and eliminates several disadvantages arising from the existing algorithms. The final product of the textile industry is seriously affected by yarn quality and / or weaving error. It is estimated that the price of fabrics can be reduced from 45% to 65% in the presence of defects[1]. Human inspection system results many types of errors. Its outcome depends on experience and the skill levels of the human observers, and the inspection process involves leading to human monotomy, because the inspection process requires a lot of concentration, continuity and repeatability.Furthermore the human inspection is done in an offline workstations in a separate inspection phase of the production line. This process will definitely affect production speed. Computer-vision based inspection systems have been increasingly applied to replace the human-based systems which eliminate the disadvantages due to offline and human inspection methods by reducing human fatigue, bordom,repeatability, and inadvertancy, and improves the speed of production flow.

2 Literature Review

Heavy competition prevails in the textile industry.The productivity and quality of the fabrics are the two main parameters that gives a momentum to the competition to progressevily increase. There has been an increase in losses in the textile sector due to faulty fabrics. Most of the defects in the manufacturing process of a textile material are still detected by human inspection in the less developed countries. Inspectors work is very tedious and time consuming.Identification of minute area of a fabric and displaying it in wide area of vision is important in the fabric inspection. The defect identification rate using conventional systems is approximately 70%. In addition, the effectiveness through visual inspection methods decreases with human fatigue. Computerized image handling procedures have been progressively applied to finished samples investigation in the course of the most recent decade[2]. Quality has been a topical issue in the textile sector, which levers the competitive edge of the producer The modern weaving industry uses high-speed looms to produce the fabrics with highest quality stansadrds in the shortest time. In addition, quality assurance systems have been developed with the aim of providing a quality product to the consumer with the maintenance of trust and loyality to produce stocks for a considerable time conforming to the product specifications, manufacturing standards and original technical design[4].

The highest priority of all weaving mills is therefore to reduce the presence of weaving defects in the final product at the early stages of the production process in order to ensure optimized economic viability.Some false positives (rejecting good products) are more forgivable for manufacturers than false negatives(lack of defective products). However, it is worth remembering that fabric defects are loosely divided into two types[5]; one is global colour (shade) deviation and the other local textural irregularities, that is going to be adressedin the present study. The process of detecting these defects is normally called fabric inspection.

2.1 Fabric Inspection

Nowadays in manufacturing industries, such as electronics, automotive and medical industries, product inspection is inevitable. Inspection is a preventive process that can be broadly defined as the process of determining whether a product deviates from a specification set. There are two distinct possibilities for detecting fabric defects. The first one is the product or end (offline) inspection that requires inspection of the manufactured fabric through fabric inspection machines. The second possibility is the process inspection (online) in which the weaving process (or its parameters) for the occurrence of defects can be constantly monitored. In both cases computer vision techniques play an important role[6].

2.2 Types of Defects and Reasons

8-20 meters per minute. When a fault on the moving fabric is found by the inspector, he stops the machine,documents the fault and its position, and restarts the engine. The number of defects per meter length is measured for each inspected fabric sheet, and acordingly the fabric is graded. Early detection of repeated faults and unusual faults is left to the operators or so-called (roving inspectors)[8]. During the test, if an unusual defect rate or repeat defects are detected by the supervisor, these roving inspectors alert the production team so that corrective action can be taken to reduce the defect rate. Bowlinget al.suggested the use of two inspectors on the same computer as another method to decrease this level when inspecting the material[9].

Most fabric defects occur while being woven on the loom. Some of these defects in the fabric are visible while others are not. Nevertheless, when weaving and after weaving, certain fabric defects may be rectified where as other defects can not be rectified. The table 1 outline the most common fabric defects, as well as their causes and severity. The tables are constructed on the basis of the position of the defect or the region where the defects are spread[7]. The most common defects appear in fabric are explained in Table 1 with the respective reason for the defect.

In spite of the fact that people can show improvement over machines much of the time, the visual assessment experiences numerous downsides.Because these downsides drawbacks address the main arguments for the advent of another robust inspection method, they can be summarized and discriminated as follows: Human specialists for this task are challenging to track down or keep up with in an industry; human requires preparing and their abilities invest in some opportunity to create; Sometimes, visual investigation will in general be drawn-out or troublesome or tedious,in any event, for the best prepared specialists; Human is more slow than the machines which implies that examination is a tedious errand; Human monitors become exhaustion over the long run (get worn out rapidly). Consequently, visual texture review is very tiring errand, and inevitably the sight can't be engaged(the maximum period of concentration is 20-30 min).In any case, the administrator unavoidably misses little deformities and once in a while even enormous ones with the quantity of meters of the reviewed fabric roll.Human inspectors have to deal with an extensive variety of defects, and make mistakes because inspection is unreliable when the fabric of 1.6-2 meters’width is unfolded at a speed of 20 m/min. It is challenging for people to stay alert all the time of these hard circumstances. Due to the fact that their productivity depends on experience and, surprisingly,in an all around run activity, the reproducibility of a diffect examination will seldom be more than half while the most extreme location efficiency is around 70% - 80%. The reviewer can barely decide the level of issues that is satisfactory, while comparing such a level between several inspectors is almost impossible. It is an abstract strategy that finds it hard to duplicate outcome. The evaluating system is slow and changes from one fabric viewing mill to another. Generally,there is a shortfall of criticism to help processes for remedial measures.

2.4 Drawbacks of Visual Fabric Inspection

The inspection process is based strictly on the human eye and is performed after the process of fabric formation. According to the poet Alexander Pope,"error is human; divine forgiveness is human." In the ethical context, this may be the motto, but modern manufacturing is error-free. A key fact: in practice, the probability of human error cannot be reduced to zero even with the best-designed man-machine interface.Therefore, visual inspection performed well for many years due in part to the limited and manageable amount of data[10].

“Are you still cold,” she asked, as she kissed him on the forehead. The kiss was colder than ice; it went quite through to his heart, which was already almost a lump of ice; he felt as if he were going to die, but only for a moment; he soon seemed quite well again, and did not notice the cold around him.

In vain did her husband represent to her their extreme poverty: she would not consent to it; she was indeed poor, but she was their mother. However, having considered what a grief it would be to her to see them perish with hunger, she at last consented,8 and went to bed all in tears.

Then the King remembered Jack my Hedgehog, and he told his daughter how he had been obliged to give a written promise to bestow10 whatever he first met when he got home on an extraordinary creature which had shown him the way

And with great love the Wumpalump stretched forth3 his hand and with Himself painted the heavens in beautiful blues4 and sparkles of gold. With laughter and warmth he placed a brilliant light to rule the day and a softer glow to rule the night. Embracing his fullness he brought forth the earth and with tears of joy created the oceans. He fashioned whales and moving creatures of all kinds. He set birds in the sky to dance amongst the clouds.

Besides, the issue of the visual review doesn't compare just to the undetected deformities yet in addition, it changes the mechanical properties of the texture under assessment. For example, the texture aspects (longitudinally and width-wise) generally different because of the applied strain on texture roll during the investigation interaction. From the selling point of view the buyers are not happy and not ready to pay for defective materials. In addition, the shrinkage happens after the spreading of the fabric in cutting department that builds the likelihood of delivering second quality pieces of clothing either because of poor quality or mistaken size.

The bad quality control speed when contrasted with the production speed offers a significant bottleneck in the high speed production lines. It is extremely difficult to achieve 100% fabric inspection with this traditional method. There is a cost of manual investigation, which is basically a non-value-added movement. Traditional visual fabric inspection is cost-intensive. Even, through the incidence of serious weaving faults can be diminished or controlled by the use of modern weaving technology, the defect detection in many fabric viewing mills still continues by creating a considerable extra cost (which increases with the labour cost).

In view of these tremendous downsides and to expand exactness, endeavors are being made to supplant manual visual examination via mechanized framework that utilizes a camera and image processing to protect the best chance of even handed and reliable assessment for fabric quality.

3 Methodology

The fabric roll is automatically unwinded and sent through a table on top of which a camera is fixed to facilitate capturing images of the selected length of the fabric roll. After exposing to the camera the fabric roll is set to rewind. The images are analyesed to detect defects using Convolution Neural Network.

3.1 Design of Fabric Roll Unwinding and Rewinding Mechanism

The fabric roll has to be unwinded and spreaded in order to take snapshots of the total width and selected length. The mechanism for unwinding and rewinding of the fabric roll was designed and fabricated. The design concept drawn with SOLIDWORK software is shown in Fig.1. The technical drawing with the isometric view is shown in Fig.2. A DC motor is used to rewind the fabric roll and the rollers were designed to facilitate the smooth unwinding and rewinding of the fabric roll. The motor was selected so that when the rewinding starts at first after the inspection the fabric roll itself gets unwound itself due to the rewinding at the other end of the roll. The camera was set on top of the fabric segment to captutre the image by using a stand as shown in both Fig.1 and Fig.2. The lighting required for capturing the images were provided by LED stripes fitted on top of the fabric roll as shown in Fig.1. According to the length and width of the fabric role the field of view for one image was selected as square region of length 200 mm. Then considering the speed of the movement of the fabric roll, the frame rate of the camera was calculated. A Logitech C270 web camera with 1920 x 1080 pixels maximum resolution is selected to match the calculated minimum resolution requirement and frame rate.

Fig.3 Design Concept of the Automated Fabric Defects Inspection Machine

Fig.4 Technical Drawing of the Automated Fabric Defects Inspection Machine

3.2 Image Processing for Defect Identification

Convolutional Neural Network (CNN) is used to process images of fabric and identify the defects. CNN is a very common approach to solving pattern recognition and classification problems. A neural network is a mathematical model based on artificial neurons linked to biological neural networks via each other's neural units. Usually, neurons are organized into layers, and only adjacent layers create relations between neurons. The low-level feature vector input is positioned in the first layer and converted to the high-level feature vector, moving from layer to layer.The quantity of the output neuron layer is equal to the number of classification groups. Therefore, the output vector is the probability vector that demonstrates the probability that the input vector belongs to the corresponding class. The neural network was trained first using known defect and in this study there were only three types of defects were used. Fig.5 shows the flow chart of the CNN training process. According to the images with known defects, the features were extracted and the heural network was trained.

Fig.5 Flow Chart for Process of CNN Training Using Images with Known Defects