1. Context and Objectives of the Project

Venair Biotech faced the need to modernize its quality control system in the silicone tube manufacturing line, a process that until now was executed manually by operators, with severe limitations in terms of traceability, precision and efficiency.

The main objective of the project was to develop an automated visual inspection system based on Machine Vision (VA) And Artificial Intelligence (AI), capable of detecting surface and structural defects in tubes in real time during the post-extrusion phase.

Specific objectives:

• Reduce the percentage of defective products and the waste generated.
• Increase the speed, accuracy and consistency of the inspection process.
• Ensure full traceability of inspections through structured records.
• Improve decision-making through automatic alerts and visual interfaces that can be interpreted by operators.

2. Initial Situation and Problem Detected

The previous inspection system had a number of technical and operational limitations:

• Human subjectivity in the detection of defects.
• Lack of traceability and visual records.
• High operator dependence and low scalability.
• Risk of undetected defects moving to later stages.

This context motivated the need for an intelligent, automatic and traceable system, aligned with the principles of industrial digitalization and total quality.

3. Implemented Technical Solution

3.1. System Architecture

A modular system was designed with the following capabilities:

• Real-time image capture, using high-resolution industrial cameras (IDS GV-5890CP), arranged in vertical and horizontal orientation.
• Deep Learning Image Processing, using algorithms such as Autoencoders and PADim (Anomalib), trained with real production images.
• Visual anomaly maps, interpretable by quality operators and technicians.
• Structured storage, using a PostgreSQL relational database that manages images, metadata, configurations and traceability by batch and reference.
• Dynamic parameter management, automatically adjusting binning, ROI, framerate and exposure depending on the type of tube.

3.2. Hardware and Infrastructure

• 12MP industrial cameras with Sony IMX226 sensors.
• Homogeneous lighting calibrated to minimize reflections and shadows.
• Secure and scalable storage infrastructure integrated with the production system.
• Adjustable mechanical support adaptable to the post-extrusion area.

3.3. Detection Algorithms

Two main approaches were evaluated:

1. Autoencoder (TensorFlow): based on defect-free image reconstruction.
2. PADim (Anomalib): statistical modeling of local patches, with better stability and performance in a real environment.

The PADim model was finally selected for its greater precision, lower latency (<100 ms), robustness to variations and ease of operational integration.

4. System Validation

Laboratory Phase

• Simulation of real conditions with real tubes and representative defects.
• Quantitative evaluation using metrics such as accuracy, recall, F1-score, IoU and AUC-ROC.
• Stability tests and extended operation with batch changes and references.

Pilot Production Phase

• Physical installation of the system on an active in-line post-extrusion machine.
• Technical, operational and usability validation by the plant team.
• Partial integration with production systems and real impact assessment.

5. Achieved Results

• Functional system operating in a productive environment capable of continuous and accurate inspection.
• Eliminating the need for manual visual inspection and improving operational efficiency.
• Robust detection of complex defects such as bubbles, cracks or surface irregularities.
• Full traceability by lot, reference and technical configuration.
• Demonstrated scalability for implementation in other lines and products.

6. Conclusion

The project made it possible to transform a manual, limited and poorly traceable process into a robust, interpretive automated inspection system aligned with the most demanding industrial quality standards. The combination of artificial vision, explainable AI, automation and integration with plant systems consolidates Venair Biotech's ability to compete in the biomedical sector with higher quality products and more efficient processes.

‍