Geri Dön

Gerçek zaman-imge işleme temelli kumaş kalite kontrol sistemi

Real time-image processing based fabric quality control system

PDF İndir

  1. Tez No: 935752
  2. Yazar: MEHMET ERDOĞAN
  3. Danışmanlar: PROF. DR. MUSTAFA DOĞAN
  4. Tez Türü: Doktora
  5. Konular: Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Mekatronik Mühendisliği, Computer Engineering and Computer Science and Control, Mechatronics Engineering
  6. Anahtar Kelimeler: Belirtilmemiş.
  7. Yıl: 2025
  8. Dil: Türkçe
  9. Üniversite: İstanbul Teknik Üniversitesi
  10. Enstitü: Lisansüstü Eğitim Enstitüsü
  11. Ana Bilim Dalı: Mekatronik Mühendisliği Ana Bilim Dalı
  12. Bilim Dalı: Mekatronik Mühendisliği Bilim Dalı
  13. Sayfa Sayısı: 139

Özet

Tekstil sektöründe üretimin her aşamasında kalite kontrol, küresel pazarda rekabet edebilmek için hayati bir unsurdur. Manuel kumaş kusur incelemesinin sorunları, hassasiyet eksikliği ve zaman alıcı olmasıdır. Bu nedenle erken ve doğru kumaş hatası belirlenmesi, kalite kontrolün kritik bir aşamasıdır. Başarılı bir otomatik kumaş hata inceleme sistemi oluşturmak için iki ana zorluk vardır: kusur tespiti ve kusur sınıflandırılması. Geleneksel olarak, kumaş kusurlarının sınıflandırılması verimsiz ve emek yoğun bir süreç olan insan görsel muayenesiyle yapılmaktadır. Artan kumaş hata çeşitliliği nedeniyle, kumaş ürünlerinin kalitesini garanti altına almak için daha yüksek doğrulukla hataları sınıflandırabilen etkili yöntemlerin geliştirilmesi gerekmektedir. Tekstil kumaş malzemeleri ve ürünleri için otomatik kalite güvencesi, gerçek dünya uygulamalarındaki en karmaşık algoritmaların kullanıldığı yapay görme problemlerinden biridir [1-3]. Örme işlemi sırasında, kumaş ipliğinin kalitesindeki ve üretim ile çalışma koşullarındaki rastgele değişiklikler genellikle boyut, şekil, görünüm ve renk bakımından değişen dinamik hatalara yol açar. Tekstil ürün kalitesinde görsel denetimden kaynaklanan ekonomik faydalar çok büyüktür ve ürün kalite güvencesi için otomatik bilgisayarlı görüntü işleme çözümlerine yapılan yatırımları haklı çıkarmaktadır. En yetenekli denetçilerin bile kumaş hatalarının yalnızca yaklaşık %70'ini tespit edebildiği ve kumaş hatalarının üretilen kumaşların değerini yaklaşık %45-65 oranında azalttığı tahmin edilmektedir. Mevcut tespit sistemlerinin maliyetleri önemli ölçüde yüksektir ve tespit edebildikleri kusur türleri oldukça sınırlıdır. Düşük maliyetli yüksek hızlı bilgisayarlar, yüksek çözünürlüklü dijital kameralar ve düşük maliyetli depolamanın artan kullanılabilirliği, güçlü otomatik tekstil denetim çözümlerinin gelişeceğini ve yaygınlaşacağını göstermektedir [4-6]. Tez çalışmasında, histogram tabanlı yöntemler, renk tabanlı yaklaşımlar, görüntü segmentasyon tekniği, frekans dönüşümü ile algılama, doku tabanlı kusur algılama, görüntü morfolojisi işlemleri ve derin öğrenme yöntemlerine ilişkin kapsamlı bir genel çözüm üzerinde durulmaktadır. Araştırma, farklı kumaş kusurlarını tespit etmek için bilgisayarla görme ve dijital görüntü işleme uygulamaları kullanarak akan kumaşlarda hata tespiti için gerçek zamanlı ve yüksek performanslı çalışan algoritmaları gerektirmektedir. Bu nedenle, grafik kartı tabanlı geliştirmeler ile tespit ve sınıflandırma yapacak, ayrıca farklı yöntemler kullanarak kumaşın hız ve genişlik bilgilerini de hesaplayarak hata geometrisini kaydetmeyi hedeflemektedir.

Özet (Çeviri)

Quality control at every stage of production in the textile industry is a vital factor for competing in the global market. The textile industry's reliance on high-quality production demands a robust inspection system to ensure minimal defects in fabric products. Traditionally, manual inspection methods have been the primary approach to fabric defect detection. However, these methods suffer from critical limitations, such as a lack of precision and high time consumption. These shortcomings often lead to missed defects or inconsistencies in inspection outcomes, highlighting the need for more reliable and efficient solutions. Consequently, early and accurate fabric defect identification has become a critical phase of quality control in the textile production process, ensuring consistent product quality and meeting customer expectations [1-3]. The primary challenges in developing a successful automatic fabric defect inspection system are defect detection and defect classification. Defect detection involves identifying anomalies on the fabric surface, while defect classification requires categorizing these anomalies into predefined types based on their characteristics. Manual methods, which involve human visual inspection, are inherently inefficient and labor-intensive. These methods also pose challenges related to inspector fatigue and subjective judgment, which can result in variability and errors in defect detection. With the increasing diversity and complexity of fabric defects, such traditional approaches are no longer sufficient. Thus, the development of automatic methods capable of achieving higher accuracy in defect detection and classification has become imperative for maintaining product quality and meeting global market demands. Automatic quality assurance for textile materials and products represents one of the most complex artificial vision problems. This complexity arises from the need to analyze dynamic and variable defect patterns in real time. The implementation of advanced algorithms in real-world applications has shown promising results in addressing these challenges. Several state-of-the-art approaches leverage developments in artificial intelligence, digital image processing, and computer vision technologies to create innovative solutions for automatic defect inspection systems. These technologies provide the capability to process large volumes of data, identify intricate defect patterns, and adapt to various production conditions. By integrating these advanced methodologies, the industry can achieve significant improvements in precision, reliability, and speed compared to manual inspection. During the knitting process, random variations in yarn quality and operational conditions often result in dynamic defects. These defects differ significantly in their size, shape, appearance, and color, making their detection a complex task. For instance, defects such as holes, slubs, stains, or misalignments may vary across different fabric types and production conditions. Such variability necessitates the use of adaptive algorithms capable of handling a wide range of defect types and severities. Moreover, the economic implications of these defects are substantial, as fabric defects can reduce the value of the produced materials by approximately 45-65%. This loss impacts not only the profitability of manufacturers but also the perception of quality by customers. Furthermore, it is estimated that even the most skilled inspectors can detect only about 70% of fabric defects during manual inspections. These limitations underscore the necessity of investing in automated solutions for fabric quality assurance, which can address both economic and operational challenges in the industry [4-6]. The adoption of automated computer vision-based systems for textile quality control has several economic and operational advantages. These systems enable the detection and classification of a wider range of defects while significantly reducing the time and labor involved. Additionally, advancements in low-cost, high-speed computing, high resolution digital cameras, and efficient data storage solutions have made automated inspection systems increasingly accessible and practical. With the reduction in costs associated with these technologies, small- and medium-sized enterprises are now able to implement advanced inspection systems that were previously cost-prohibitive. As these technologies continue to evolve, the adoption of powerful automated textile inspection solutions is expected to become more widespread, leading to higher standards of quality control across the industry. Furthermore, automated systems ensure consistent inspection quality, reducing the likelihood of customer complaints and returns, thereby strengthening brand reputation and customer loyalty. The methodologies explored in this thesis are designed to address these multifaceted challenges. By presenting a comprehensive general solution for automatic fabric defect detection and classification, the research aims to improve existing systems significantly. The proposed methodologies incorporate a variety of techniques, including histogram-based methods, color-based approaches, image segmentation techniques, frequency transformation-based detection, texture-based defect detection, image morphology operations, and deep learning algorithms. Each of these approaches contributes to enhancing the system's ability to identify and classify defects with greater accuracy and efficiency. For instance, histogram-based methods provide a statistical foundation for identifying anomalies, while frequency transformation techniques, such as Fourier and wavelet analysis, allow for the detection of periodic defects. Deep learning methods, on the other hand, introduce adaptability and precision by leveraging large datasets and training neural networks to recognize intricate patterns. Texture-based detection, combined with image morphology operations, further refines the accuracy by capturing microstructural differences in fabric surfaces. To ensure real-time performance, the research focuses on developing high performance algorithms capable of processing flowing fabric data. These algorithms leverage computer vision and digital image processing technologies to analyze the fabric surface dynamically. Additionally, the thesis explores the use of graphics card based enhancements to accelerate computation and improve overall system performance. By integrating these techniques, the system not only detects defects but also records their geometry by calculating the speed and width information of the fabric. This detailed analysis enables manufacturers to trace defects back to specific stages in the production process, facilitating targeted interventions and process improvements. Moreover, real-time monitoring systems allow for immediate feedback and corrective actions, minimizing waste and downtime in production lines. In conclusion, the economic and operational advantages of automated fabric defect inspection systems make them an indispensable tool for modern textile production. By addressing the limitations of manual inspection and leveraging advanced technologies, this thesis aims to contribute to the development of efficient, reliable, and cost effective solutions for ensuring high-quality textile products. The integration of innovative methodologies and cutting-edge technologies underscores the potential for transformative advancements in fabric quality assurance, paving the way for a more competitive and sustainable textile industry. The outcomes of this research are expected to have a far-reaching impact, not only in improving defect detection rates but also in enhancing the overall efficiency and sustainability of textile manufacturing processes. Furthermore, the successful implementation of such systems will encourage ongoing innovation, setting new benchmarks for quality and reliability in the textile industry.

Benzer Tezler

  1. Tez No

    541786

    Deep convolutional neural network based unconstrained ear recognition

    Derin evrişimsel sinir ağı tabanlı kısıtsız kulak tanıma

    FEVZİYE İREM EYİOKUR

    Yüksek Lisans

    İngilizce

    İngilizce

    2018

    Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Teknik Üniversitesi

    Bilgisayar Mühendisliği Ana Bilim Dalı

    DOÇ. DR. HAZIM KEMAL EKENEL

  2. Tez No

    427835

    Image processing algorithms for histopathological images

    Histopatolojik imgeler için imge işleme algoritmaları

    OĞUZHAN OĞUZ

    Yüksek Lisans

    İngilizce

    İngilizce

    2016

    Elektrik ve Elektronik Mühendisliğiİhsan Doğramacı Bilkent Üniversitesi

    Elektrik-Elektronik Mühendisliği Ana Bilim Dalı

    PROF. DR. AHMET ENİS ÇETİN

  3. Tez No

    75247

    Markov random fields and a multiscale implementation of markov random fields on Bayesian image segmentation

    Başlık çevirisi yok

    UĞUR SIVAKÇI

    Yüksek Lisans

    İngilizce

    İngilizce

    1998

    Elektrik ve Elektronik Mühendisliğiİstanbul Teknik Üniversitesi

    Elektrik-Elektronik Mühendisliği Ana Bilim Dalı

    DOÇ. DR. ERTUĞRUL ÇELEBİ

  4. Tez No

    68904

    Çok ölçekli raslantı alan modeli için görüntü bölütleme

    Multiscale random field model for image segmentation

    FEZA ÇAKIR

    Yüksek Lisans

    Türkçe

    Türkçe

    1997

    Elektrik ve Elektronik Mühendisliğiİstanbul Teknik Üniversitesi

    Elektronik ve Haberleşme Mühendisliği Ana Bilim Dalı

    PROF. DR. ERDAL PANAYIRCI

  5. Tez No

    70899

    Radiküler kist ve granülomların dijital histogram analizi ile ayırdedilmesi

    Differention of radicular cyst and granulomas with digital histogram analysis

    ÜLKEM AYDIN

    Doktora

    Türkçe

    Türkçe

    1998

    Diş HekimliğiGazi Üniversitesi

    Diş Hastalıkları ve Tedavisi Ana Bilim Dalı

    PROF. DR. TAYFUN ALAÇAM

Geri Dön