Real Time Yarn Characterization and Data Compression Using Wavelets. INVESTIGATORS : Moon W. Suh, Warren Jasper and Jae L.

Transcription

1 TITLE : CODE : Real Time Yarn Characterization and Data Compression Using Wavelets I97-S1 INVESTIGATORS : Moon W. Suh, Warren Jasper and Jae L. Woo (NCSU) STUDENTS : Jooyong Kim and Sugjoon Lee (NCSU) GOAL: To design a holistic system for measurement, and quality maximization in textile fiber processing by combining on-line measurement (data fusion) and analyzing the data using wavelet theories, neural networks and time-series models. The system to be developed will extract, retain, and synthesize only the essential information required for characterizing the salient qualities of yarns and fabrics, without having to process and store vast amounts of data acquired on-line. This task, if successful, will provide the textile industry with a meaningful and cost-effective quality monitoring and measurement system which is non-existent at present. ABSTRACT: Much of our initial effort for the project was focused on developing an integrated, simultaneous yarn measurement system which combines optical and capacitor sensors. Using the system, we will study the relationship between the mass density and diameter measurements obtained from the two types of sensors. We also analyzed and compared density profiles obtained from several spun yarns. We applied wavelet transform in order to represent the totality of the yarn characteristics by retaining a minimal amount of data without a significant loss of information. Using a wavelet transform, the original density profile of yarns were decomposed into different sub-frequency scales so as to detect and model the variation of the signals which could not be analyzed otherwise. The decomposed sub-signals were further analyzed and interpreted. While we are in the process of developing a reliable system for measuring the mass density and diameter of yarns simultaneously, we also are continuing to develop a most efficient wavelet algorithm for signal analysis. The progress made to date is most encouraging in that both the measurement system and the wavelet-based analysis system met our initial expectation for achieving the project objectives. RELEVANCE TO NTC GOALS: To compete internationally, the U.S textile industry must capitalize on its strength: information and computer technologies, and a high-technology infrastructure. These strengths facilitate an environment for high quality and customer-driven QR networking. The project incorporates these strengths through the PI s diverse backgrounds (Tex. Eng., Mech. Eng., Stat.), educating students and transferring the technology to the industry.

2 OBJECTIVES: 1. To design and construct a new system for measurement, control, and quality maximization in spun yarn manufacturing by combining the diameter and mass density signals of yarns, which in turn would provide a superior prediction of fabric qualities. 2. To develop a data compression algorithm based on wavelet theories by which a large amount of data can be compressed and stored without a significant loss of the original information. 3. To characterize the yarn s mass density and diameter profiles using a wavelet-based multi-resolution analysis. 4. To predict and visualize fabric qualities directly from yarn density profiles through the measurement and analysis systems developed. TECHNICAL APPROACH: A. Development of a Signal Capturing and Processing System Variation in mass per unit length and diameter of yarns has long been the most important characteristic in textile processing and quality control. The existing commercial systems, such as Zweigle and Uster-3, however, are inadequate for measuring the real features of yarns as each measures either the mass density or diameter, but not both. Consequently, the signals obtained from these are markedly different from the visual qualities of the yarns and the resulting fabrics. In addition, the true and complete features of the raw signals are not fully revealed by these conventional systems. For overcoming the limitation, we are developing a multi-sensor, simultaneous yarn measurement system by combining optical and capacitor sensors. In the new system, both the mass density and diameter are measured and combined in such a way that two quantities are matched in spatial domain and transformed into a new physical quantity we would define as bulkiness. The new system consists of 1) optical and capacitor type sensors, 2) a tension control device, 3) a yarn guide for dampening tension variation, and 4) a signal processor. Figure 1 shows the schematics of the new measurement system. B. Yarn Characterization and Data Compression Using Wavelet Analysis With faster sensors and modern data-acquisition systems, data flow rates have increased exponentially over the last few years. For a yarn length of only 600m, the size of data file required has to be at least 1.2 MB. Yet, the presently available analytical methods either compress the data into a single CV% [1], or convert the entire data set into a spectrogram [2], deleting the important spatial information completely. In practice, because of the redundancy embedded in the 1.2MB signals, the entire data can be compressed to less than 5% of the original 1.2 Mb based on our initial efforts. This can be shown in figure 2 where the simulated fabric image from the compressed data is shown to be almost identical to that of the original data. The key question for our project

3 is how to best represent the totality of yarn characteristics with a minimal amount of data. This involves measuring one or more physical characteristics (mass density, diameter, hairiness) on-line, combining the information (data-fusion), analyzing the data (filtering, data compression), and presenting them in a meaningful form for decision support and feedback control. It has been found that the wavelet transform offers many advantages over other methods in performing data compression, space-frequency analysis, and texture characterization [3]. Previous studies in other fields [4] suggest that the waveletbased multi-resolution analysis can be an excellent tool for characterizing yarn qualities. This new tool, in combination with soft computing algorithms such as neural networks and fuzzy logic, would provide a better data analysis and decision making method suitable for on-line control environments. CURRENT PROGRESS: A. EXPERIMENTAL We obtained 15 different kinds of 16/1 Ne open-end yarns produced from different spinning frames under identical process conditions. The samples ranging CV of 15-17% were mechanically conditioned under a standard atmosphere for a month. Finally, the samples for each yarn were tested by using Uster-3 and Keisoke Evenness Tester in order to compare the measures of yarn density profiles. The yarn density signals were analyzed by Daubechies wavelets (Fig. 3), and decomposed into sub-band frequency signals for a further analysis. B. YARN CHARACTERIZATION USING WAVELET TRANSFORM We analyzed and compared two yarn signals with similar visual qualities using waveletbased multi-resolution technique. The yarns showing no significant differences of CV% and spectrogram were decomposed into several sub-frequency signals in order to investigate the characteristics of those yarns in different frequencies. In spite of their similarity in their original density profiles, the MRA windows show quite different patterns at the mid-frequency levels. It is not yet clear how the differences will affect the visual qualities of the resulting fabrics as we are in the process of producing the fabrics. In view of the fact that the cloudiness or barre effect result mainly from midterm variation in yarn densities, the significant differences found in the mid-frequency levels are expected to be translated directly into visual qualities of the fabrics.

4 PROSPECTS: 1. Prospects are good for developing a multi-sensor, simultaneous yarn quality measurement system in the near future. The opto-capacitive system will measure the yarn diameter and mass density simultaneously and combine them in spatial domain to generate a new uniformity measure. The matching errors caused by tension variation during yarn transport will be minimized by a specially designed control mechanism. The system, if successfully built, will be the first of its kind with a builtin signal processor. 2. The density profiles obtained from the system will be analyzed by wavelets. The wavelet-based multi-resolution analysis will validate the usefulness of the new measurement system by linking the measured yarn characteristics to visual qualities of the resulting fabrics through the spatial decomposition of yarn density signals. 3. We will also develop an efficient algorithm for compressing the yarn signals by retaining the least amount of information. For an on-line process control in spun yarn production, the large amount of data must be compressed by discarding the redundant information while the data are being captured. Using a wavelet thresholding method, we can store only a portion of the original signals for either control purposes or graphical representation of the visual qualities of the yarns. Since the system will filter out the unnecessary information in real-time, it will also reduce the data storage requirement significantly. Spatial mapping of non-uniformities in slivers, rovings, yarns and fabrics is the ultimate motivation for this particular area of research. LITERATURE CITED: 1. Furter, R., Evenness Testing in Yarn Production : Part I, The Textile Institute and Zellweger Uster AG, Manchester (1982). 2. Wood, E. J., Applying Fourier and Associated Transforms to Pattern Characterization in Textiles, Textile Res. J. 60, P (1990). 3. Jasper, W. and Potlapalli, H., Image Analysis of Mispicks in Woven Fabric, Text. Res. J., 10, P (1995). 4. Mallat, S., Multiresolution approximations and wavelet orthonormal bases in L 2 (R), Trans. Amer. Math. Soc., 315, P69-87 (1989).

5 Yarn Package 2. Optical Sensor 3. Capacitance Sensor 4. Constant tension transport unit 5. Controller 6. Signal processing unit 7. Computer 7 Figure 1 The schematics of the new measurement system

6 Figure 2 Comparison of simulated fabric images from a) original yarn density signals and b) simulated yarn densities with 95% data compression

7 Figure 3 Representation of a Daubechies wavelet