In this paper, we evaluate the efficiency and accuracy of a method of detecting fabric defects that have been classified into different categories by a neural network. Four kinds of fabric defects most likely to be found during weaving were learned by the network. Based on the principle of the back-propagation algorithm of learning rule, fabric defects could be detected and classified exactly. The method used for processing image feature extraction is a co-occurrence-based method, by which six feature parameters are obtained. All of them consist of contrast measurements, which involve three spatial displacements (i.e., 1, 12, 16) and four directions (0, 45, 90, 135 degrees) of fabric defects' images used for classification. The results show that fabric defects inspected by means of image recognition in accordance with the artificial neural network agree approximately with initial expectations.It has become more and more important for textile engineers to use automatic techniques in production processes and management procedures. At present, fabric inspection still depends on human sight, and inspection results are greatly influenced by the mental and physical condition of an inspector. To economize on personnel and to increase the competitive ability of products it is necessary to automate the inspection of fabric defects during weaving.We have used an artificial neural network to construct a pattern recognition system for inspecting fabric defects in this study. Because of the learning capacity and fault tolerant nature of an artificial neural network, it can be used for this purpose.In this study, we use a neural network topology known as a &dquo;multi-layer perceptron,&dquo; which has not been used in the past because of the lack of effective training algorithms for it. Recently this has changed due to the development of an iterative gradient procedure known as the back-propagation algorithm [ 7 ] . Through this supervised learning algorithm, the neural network can become a classifier of fabric defects. Because of the highly parallel operation and quick response nature of a neural network, it is easy to apply this method to on-line monitoring of fabric defects in weaving.

Artificial neural networks (ANN), with capabilities of fault tolerance and learning, can be used to detect fabric defects. Because the back propagation algorithm, has higher learning accuracy and successful applications, we have used it in this study to identify missing ends, missing picks, oily fabric, and broken fabric, all often found as defects in fabrics. The correct selection of characteristic parameters for the input layer in an ANN plays a great role in the recognition rate. The spatial periodicity of a fabric image can be transferred into spatial frequency by fast Fourier transform owing to the fabric's periodicity. Once a defect occurs in the fabric, its periodicity is changed so that the corresponding intensities at the specific positions of the spectrum obviously change. These intensities can act as characteristic parameters and can be substituted in the ANN for learning. Altogether, nine parameters derived from the spectrum have been selected by the ordinary method, which provides the characteristic parameters without any extra modification, and by the statistical method, which modifies the characteristic parameters with variations between the defective and normal fabrics.Of the two plain fabrics used (with densities of 70 X 60 and 65 X 45), for each fabric, the results show that the total classification rates each above 96%. The total classification rate is 88% with the statistical method while the ordinary method is 24% if only one fabric is selected and the learned mode is applied for a new, unlearned fabric. The statistical method can be used for fabric defect recognition, and any inconvenience caused by various specifications of warp and weft densities can be minimized.Though most fabric defects can be eliminated or abated by means of on-line monitoring systems and automatic devices in a modem weaving mill, the technology is still limited. A weaving machine cannot yet recognize fabric defects. Developing an automatic system for fabric defect recognition is relatively difficult. Shimizu et al.[7] used an expert system to recognize fabric defects in 1989, but the method required a complex process for image interpretation and extensive image data, and had difficulty achieving high recognition rates. In 1991, Cohen et al.[2] used a Gaussian Markov random field to inspect fabric defects. Many complicated computations and timings were required for pattern recognition, even though this method improved the recognition effect. In 1993, Ribolzi et al. [6] used an optic-electric method to inspect fabric defects. This method tremendously reduced working time but could not recognize fabrics with higher warp and weft densities. Vangheluwe et aL [8] used the ANN to assess set marks but parameter selection was inadequate and the results were unsatisfactory. The ANN is widely recognized as a useful classification technique for pattern recognition [4].In this study, we have adopted the back propagation (BP) algorithm of an ANN to recognize fabric defects as a learning rule. The characteristic parameters of input...

In this study, we use best wavelet packet bases and an artificial neural network (ANN) to inspect four kinds of fabric defects. Multiresolution representation of an image using wavelet transform is a new and effective approach for analyzing image information content. In this study, we find the values and positions for the smallest-six entropy in a wavelet packet best tree that acts as the feature parameters of the ANN for identifying fabric defects. We explore three basic considerations of the classification rate of fabric defect inspection comprising wavelets with various maximum vanishing moments, different numbers of resolution levels, and differently scaled fabric images. The results show that the total classification rate for a wavelet function with a maximum vanishing moment of four and three resolution levels can reach 100%, and differently scaled fabric images have no obvious effect on the classification rate.Applying automatic techniques to production processes and management procedures has become more and more important for the textile industry. At present, fabric inspection still depends on human vision, and inspection results are greatly influenced by the mental and physical condition of an inspector. It is desirable to detect and identify fabric defects, and to prevent these defects from reoccurring. Off-line or on-line inspection systems cannot prevent defects, but on-line inspection can enhance product quality to meet customer demand.Image-based inspection systems have been widely developed in the textile industry, for blend irregularity in yarns, evenness and pore size in nonwovens, nep and trash evaluation in webs, carpet texture evaluation, and fabric defect inspection. There are many recent studies on the inspection of fabric defects. Takato et al. developed a gray-level matching algorithm for fabric inspection [ 11 ]. The shortcomings of this approach are that all imaging conditions are always assumed constant and threshold choice is subjective. An expert system has been used to recognize fabric defects [ 10], but the method requires a complex process for image interpretation and extensive image data, and cannot achieve a higher recognition rate. In 1991, Cohen et al. used a Gaussian Markov random field to inspect fabric defects [2]. Many complicated computations and timings were required for pattern recognition, even though this method improved the recognition effect. Vangheluwe et al. used an ANN to assess set marks [ 13], but parameter selection was inadequate and the results were unsatisfactory. In 1996, we used a Fourier transform and an ANN to inspect fabric defects [12], but the method yielded less than optimal representations for analyzing fabric images containing localized components. Jasper et al. [7] used three different image analysis techniques, including a Sobel edge operator, fast Fourier transform, and discrete wavelet transform to analyze mispicks in woven fabrics, and they proved that wavelet transform gives better results than the other two methods. However, they only explor...

In this study, an intelligent diagnosis system capable of tracing possible breakdown causes of fabric defects is applied to fabric inspection processing. Apart from the basic structure ( e.g., user interface, inference engine, and knowledge base) of the expert system, there is also a diagnosis system developed using fuzzy set theory. This system, which uses the Turbo C program, can act as an expert consultant for operators tracing the causes of breakdowns at any time, and the expert system, developed in an artificial intelligence language PROLOG, can provide the operator with a knowledge base for further consultation in fabric inspection. The accuracy and efficiency of system implementation are also discussed. Implementing automation techniques for production processes and management procedures has become increasingly important in textile engineering. At present, however, fabric inspection processing and tracing breakdowns that occur during weaving still heavily depend on the experience of the technical operator. Results obtained from inspection and diagnosis are exclusively influenced by the mental and physical condition of the inspector.In addition to the time needed to train a new operator to become expert in the specific technical knowledge of weaving engineering, there is the problem of retraining once the trained operator leaves the job. This approach is not only time-consuming but also economically infeasible.To solve these problems, we have applied an intelligent diagnosis system developed using an artificial intelligence programming language PROLOG and Turbo C to fabric inspection processing and diagnosis of breakdowns during weaving.Expert systems are computer-based and use a large body of heuristic and logical rules for solving a specific class of problems [11. These systems have been evaluated for many years by many companies to guide plant operations [ 12,13,14 ] . They work in ways similar to those of very experienced operating and technical personnel and make evaluations to enhance operations or diagnose problems. These systems are potentially important for early detection and diagnosis of mechanical or process problems in the plant.Apart from the basic structure [ 1 ] ( e.g., user interface, inference engine, and knowledge base) of a conventional expert system, the intelligent diagnosis system also has a diagnosis component based on fuzzy set theory. By combining these two components, we can construct an intelligent diagnosis system that has the ability to function as an expert consultant to help a new operator trace breakdowns, to provide the operator with a knowledge base and a data base for various weaving problems, and to apply these bases to fabric inspection. . Theory . AUTOMATIC DIAGNOSIS USING FUZZY SET THEORY ' ,The automatic diagnosis procedure can be demonstrated by an electric circuit, as shown in Figure 1 a, in which switches xi -xm represent m kinds of breakdown causes and lamps yi -yn represent n kinds of symptoms. respectively. Assuming that the breakdown is caused by a switching e...

The aim of this research is to construct an appropriate back-propagation neural network topology to automatically recognize neps and trash in a web by color image processing. After studying the ideal background color under moderate conditions of brightness and contrast to overcome the translucent problem of fibers in a web, specimens are reproduced in a color BMP image file format. Assuming that neps and trash can be distinguished without difficulty from the color image, the image-taking device in the system can be easily altered as long as the optical conditions for other color image resources (i.e., CCD) are considered to ensure image quality. With a back-propagation neural network, the RGB (red, green, and blue) values corresponding with the image pixels are used to perform the recognition, and three categories (i.e., normal web, nep, and trash) can be recognized. The numbers and areas of both neps and trash can also be determined. According to experi mental analysis, the recognition rate can reach 99.63% under circumstances in which the neural network topology is 3-3-3. Both contrast and brightness are set at 60% with an azure background color. The results show that both neps and trash can be recognized well, and the method is suitable not only for cotton and man-made fibers of different lengths, but also for different web thicknesses as to a limit of 32.9 g/m2. Since neps and trash in a web can be recognized, yarn quality not only can be assessed but also improved using a reference for adjusting manufacturing parameters.