AI-Based Visual Defect Detection for Textile Production

Problem and Context

A machinery manufacturer serving textile production needed to detect and classify material defects directly during live production to reduce quality risk and avoid relying on manual downstream inspection. The challenge was especially relevant because defects occurred in different forms and frequencies, making consistent recognition difficult in real time. In addition, classification quality depended heavily on image labeling quality, while the solution also needed to run close to the machines and support continuous improvement across the installed customer base.

28

defect types detected and classified in production

98%

classification accuracy after model and labeling improvements

+7 pp

accuracy improvement versus the initial ML phase

Approach and Solution

The project combined close collaboration with the client’s domain experts and an iterative improvement process to build a production-ready visual inspection solution. High-speed cameras were installed on the machines to capture fabric images during operation, and the resulting solution was trained to recognize defect patterns and classify them into distinct categories. As performance improved, deeper reviews showed that inconsistent labeling was a major constraint. To address this, a dedicated support app was introduced to help experts label images more consistently and improve data quality. The solution was embedded into operations and connected to a cloud-based feedback loop for continuous enhancement.

Results and Impact

The initiative enabled automated in-line quality recognition during live production and significantly improved classification performance over time. The solution moved from a strong initial level to near-industrial accuracy by combining process refinement with better labeling support. It also created a scalable learning loop in which image data and classifications could be shared through the cloud, allowing ongoing improvement and giving customers access to additional value-added functions beyond the initial defect detection use case.

Want to learn more about this case?

Contact:

Role:

Email:

Steffen Illig

Partner, Project Manager and Expert for Data Analytics

steffen.illig@5v-strategy.com

⬅ Back to Case Studies