AI Based Fabric Inspection System

Abstract

Textile industry continues to face challenges in ensuring consistent fabric quality due to its reliance on manual inspection processes. This study introduces an AI-driven automated fabric inspection system designed to detect surface defects in real time using deep learning and computer vision techniques. The system is built around a Raspberry Pi 5 platform integrated with TensorFlow and a Pi Camera module. A convolutional neural network (CNN) model processes captured images to identify defects, prompting the system to halt fabric movement and activate a pump-based liquid applicator for marking the flawed regions. The proposed solution is portable, cost-effective, and offers high detection accuracy, making it particularly advantageous for small and medium-sized textile enterprises seeking to modernize their quality control operations. This approach reduces inspection time and improves accuracy in textile manufacturing environments. Experimental validation achieved 92% detection accuracy across multiple fabric types.

Introduction

I. Introduction: Importance of Fabric Quality

• Fabric quality directly affects product value, customer satisfaction, and brand reputation.

• Defects like holes, stains, and weaving issues can cause major economic losses.

• Manual inspection, though common, is slow, subjective, and prone to error, especially due to human fatigue.

II. The Need for Automation

• Manual methods lack consistency, speed, and scalability.

• AI and computer vision provide a more reliable, consistent, and efficient alternative for real-time fabric inspection.

• AI systems (especially CNNs) can detect, classify, and localize defects continuously and with high precision.

III. Literature Review

• Traditional image processing techniques (e.g., edge detection, filtering) are simple but lack robustness.

• AI-based methods:

  • CNNs: Strong performance in defect detection and classification.

  • SVMs + feature extraction: Useful, but rely heavily on high-quality features.

• Recent trends: Use of low-cost, real-time systems (e.g., Raspberry Pi) for on-site defect detection.

IV. Key Issues in Current Systems

• Subjectivity and inconsistency in human inspection.

• Fatigue-related errors in long shifts.

• Lack of scalability and real-time data logging.

• High cost and rigidity of existing automated systems make them unaffordable for small to mid-scale manufacturers.

V. Comparison Table: Manual vs AI System

VI. Proposed AI-Based System Architecture

A. Hardware Overview

• Core Controller: Raspberry Pi 5

• Camera: Pi Camera V2 (8MP)

• Fabric Motion: Stepper motor + TB6600 driver

• Defect Marking: 12V diaphragm pump + L298N motor driver

• Power Supply: Dual-voltage SMPS (5V for logic, 12V for motors/pumps)

• Other Components: Marking nozzle, aluminum frame, bearings, wiring

B. Software Architecture

• Language: Python on Raspberry Pi OS

• Libraries: OpenCV (image processing), TensorFlow Lite (AI model), GPIO control

• Functions:

  • Real-time image acquisition and preprocessing

  • AI inference for defect detection

  • Synchronized motor stop and marking activation

  • Live GUI with logging and manual override

  • Error handling for disconnects, delays, and faults

C. Image Processing Pipeline

• Steps: Grayscale → Noise Reduction → Resize → Normalize → CNN Inference

• Output: Probability of defect, location marking

• Speed: ~0.3–0.5 seconds per frame using optimized TensorFlow Lite model

• Multithreaded processing ensures real-time performance.

VII. Implementation and Methodology

A. Physical Setup

• Mounted aluminum frame ensures stable camera and fabric alignment.

• Rollers support smooth fabric motion.

• Pump nozzle precisely marks defects upon detection.

• Camera height, lighting, and motor speed are calibrated for optimal accuracy.

B. System Layout

• Compact, production-ready design with all components modular and upgradeable.

• Marking unit is precisely aligned with camera’s view for accurate defect labeling.

• Enclosure includes space for ventilation, cable routing, and potential add-ons (e.g., touchscreen GUI, wireless modules).

C. Software Pipeline

• Handles:

  • Camera streaming and frame capture

  • Preprocessing (grayscale, blur, resize, normalization)

  • Real-time inference and decision logic

  • Motor and marking control synchronization

  • GUI display, logging, and future extensibility

Conclusion

The development of the AI-based automated fabric inspection system presented in this work represents a significant step forward in modernizing quality control processes in the textile industry. By integrating low-cost embedded computing hardware with advanced deep learning models and intelligent actuation, the system provides a robust and scalable alternative to traditional manual inspection techniques. The key innovation lies in the use of a Raspberry Pi 5 paired with a lightweight TensorFlow Lite CNN model, enabling real-time defect detection with over 92% accuracy — all without the need for external GPUs or cloud processing. Unlike conventional systems that are either fully manual or prohibitively expensive, this system bridges the affordability-accessibility gap, making it a viable solution for small to medium textile manufacturers. Its modular design and software-driven logic allow for rapid deployment, easy upgrades, and minimal operator training. The marking mechanism, which utilizes a diaphragm pump and precision nozzle, offers an effective way to visually identify defects on the fabric for further analysis or reprocessing. This not only minimizes waste but also enhances traceability across production batches. In practical terms, the system improves operational efficiency, reduces human fatigue-induced errors, and promotes data-driven quality control. The ability to log, store, and analyze defect data creates opportunities for long-term process optimization. Moreover, the system\'s reliance on open-source tools and off-the-shelf components ensures sustainability and ease of maintenance, even in resource-constrained environments. Overall, this research establishes a strong foundation for implementing intelligent, low-cost automation solutions within India\'s growing textile manufacturing ecosystem.

References

[1] Kim and D. Lee, “Fabric Defect Detection Based on Artificial Intelligence,” Int. J. Intell. Syst. Appl. Eng., vol. 11, no. 5s, pp. 01–07, 2024. [2] W. H. Cheung and Q. Yang, “Fabric Defect Detection Using AI and Machine Learning for Lean and Automated Manufacturing of Acoustic Panels,” Proc. IMechE Part B: J. Eng. Manuf., vol. 238, no. 12, pp. 1827–1836, 2024. [3] B. Mohankumar and R. Venkatesan, “An Intelligent System for Automatic Fabric Inspection,” Asian J. Inf. Technol., vol. 13, no. 3, pp. 308–312, 2014. [4] K. S. Prakash and R. N. Mohanty, “Automated Fabric Inspection Using Machine Vision,” in Proc. Int. Conf. on Computing, Communication and Automation, pp. 450–455, 2017. [5] Y. Wang, Q. Zhou, and X. Lin, “Fabric Defect Detection Using Deep Learning,” School of Computer Science and Technology, Tsinghua University, China, 2020. [6] R. Harinarayan and V. D. Kulkarni, “Automatic Textile Defect Detection and Classification Using Image Processing Techniques,” Dept. of Electronics Eng., BMS College of Eng., India, 2019. [7] H. Xu, Y. Wang, and T. Liu, “Fabric Defect Detection Using Convolutional Neural Networks (CNN),” Dept. of Computer Science Eng., China, 2020. [8] A. Patel and S. Joshi, “Real-Time Fabric Defect Detection System on Raspberry Pi Using Machine Learning,” Dept. of Computer Science, Nirma University, India, 2021. [9] M. Ghiasi and M. B. Menhaj, “Automated Visual Inspection of Textiles Based on Local Binary Patterns and Support Vector Machines,” Dept. of Electrical Eng., Amirkabir University of Technology, Iran, 2018. [10] P. Mahajan, S. R. Kolhe, and P. M. Patil, “A Review of Automatic Fabric Defect Detection Techniques,” Adv. Comput. Res., vol. 1, no. 2, pp. 18–29, 2009.

Copyright

Copyright © 2025 Sujatha D., Swaraj R. S., Atitya Ram S., Adhith S., Girishwaran S., Athiyavan Snovas. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.