null

AI Texplorer™ Project

Exploring the Potential of AI in Automated Fabric Inspection


Drunen, March 2025, rev003

The rapid rise of artificial intelligence (AI) has opened new doors for automated defect detection in industrial vision applications, such as automated fabric defect detection, classification and real-time conditioning. Thanks to financial support from the Province of North Brabant under the Stimulating Innovation in the Region program, COMVIS has successfully taken the first important steps towards the final developed and implementation of a flexible AI-based vision plugin for both surface inspection, defect classification and -conditioning in high-speed production environments. Key Insights from COMVIS’ Innovation Project:

From Research to Real-World Application

This innovation marks a significant step forward for COMVIS, which began exploring AI technologies over five years ago. At that time, various limitations — including immature tools, insufficient computing power, and a lack of data — made practical implementation unfeasible. Today, these barriers have largely been overcome.

The primary objective of the project was to create a robust and reliable AI solution capable of detecting and classifying surface defects in printed and woven materials — all within a tight 20 ms processing window. After two years of development, testing, and optimization, this goal is now closer than ever. Key developments that contributed to this success include:

  • Increased AI Expertise
    Over the past few years, COMVIS has built strong internal knowledge of AI and machine learning. This allowed the team to select the right models, fine-tune them effectively, and integrate them into production systems with confidence.

  • Powerful New Hardware
    Advances in GPU technology — driven largely by the gaming industry — have significantly improved processing power. These new capabilities made it possible to deploy AI models that operate within strict real-time requirements.

  • Availability of AI Tools and Pre-Trained Models
    Thanks to some of the industry’s leading suppliers’ latest developments in AI technology, including user-friendly tools and pre-trained networks, COMVIS was able to speed up training processes and reduce development time significantly.

  • Improved Access to Training Data
    With increased trust and cooperation from customers, COMVIS now has access to extensive image datasets from real production environments. This allowed the training of highly specialized AI models tailored to specific defect types and fabric styles.

Key Outcomes of the Project

The project delivered a modular Vision Plugin (VP), leveraging API-based architecture, that integrates seamlessly into the existing Texplorer Software Suite and provides measurable improvements in various domains:

  • Improved Detection Accuracy in highly texturized fabrics
    The AI vision plugin successfully detect defects that were previously challenging using conventional rule-based systems.

  • Higher Flexibility
    AI models can generalize across product variations and adapt to new defect types with limited retraining.

  • Real-Time Performance
    Optimized inference pipelines ensure that defect classification occurs within the required low timeframes, meeting the strict demands of customer processes.

  • Customer Adoption
    Pilot customers using the new AI vision plugin in live production environment, with positive results and promising feedback.

Looking Ahead

This successful project is an important milestone in COMVIS’ ongoing innovation journey. The company will continue to refine its AI tools, broaden its applications across other product lines, and explore hybrid approaches that combine traditional and AI-based vision techniques to meet the ever growing requirements of our customers.

This project was made possible through financial support from the Province of North Brabant under the “Stimulating Innovation in the Region” program.