AI Quality Control in Glass Wool Production with Computer Vision

Surface Defect and Color Detection

Glass wool travels through the line in long, continuous sheets. Cameras mounted under or beside the conveyor capture a constant stream of images, frame by frame, as the product passes. Image processing algorithms look for surface defects such as holes, scratches, fiber irregularities, or patches of discoloration that may indicate chemical inconsistencies.

• Unlike human inspection, which can only check a sample, this system reviews 100% of the product.
• Detected defects are flagged immediately, which helps operators to react before an entire batch is compromised.

Lamella Inspection

In many cases, glass wool is cut into lamellae (narrow strips used for insulation boards or pipe coverings). Here, overhead cameras are positioned above the cutting station. As each lamella moves forward, the system checks for structural issues like delamination, missing fibers, or uneven cuts. If a faulty piece is detected, it can be automatically diverted from the line, without halting production. This way only defect-free lamellae continue to packaging.

Thermal Anomaly Detection

Some defects in glass wool are not visible on the surface but hidden inside the material, such as overheated fragments or density variations. Thermal cameras, positioned along the line, capture temperature maps of the product in real time.

• AI models analyze these thermal signatures to spot anomalies that could weaken insulation performance or create safety risks later.
• While this approach requires feasibility testing to fine-tune thresholds, it has the potential to eliminate high-risk defects that manual inspection would almost certainly miss.

3. The Best Way to Start Are AI Workshops

Before cameras and algorithms can be introduced, it’s important to map out the manufacturer’s needs and constraints. That’s the role of AI workshops that are at the beginning of every custom AI solution development project that we do.

In practical terms, a workshop is where we sit down with the client to walk through the process step by step. We look at the factory, understand where defects appear, and discuss how an automated system could realistically fit in.

Workshops clarify:

• Business goals – What success looks like (e.g., reduce scrap by 15% or cut complaints from downstream installers).
• Process mapping – Identifying points where inspection cameras should be mounted.
• Defect taxonomy – Agreeing on what faults actually matter (holes, discoloration, delamination, overheated spots) and how they are defined for AI model training.
• Technical setup – Choosing between camera types (RGB or thermal), planning for dust protection, and ensuring lighting and data throughput work under factory conditions.
• Integration – How the vision system will connect to MES or ERP, so defects are logged automatically.
• Operational requirements – Defining reaction times (should the line stop? should defective pieces be diverted automatically?), alarm handling, and what information operators need to see on the floor.

The AI workshop outcome is a clear, factory-specific specification document that shows which problems to address first, what equipment is needed, and what we need to proceed to the next stage: Proof of Concept.

4. How We Build an AI Quality Control Proof of Concept for Glass Wool Production

Building a Proof of Concept (PoC) the solution in real production conditions.

Step-by-step Process of Building an AI Pilot Project

To test if AI-powered quality control can work in your factory, you need to run a structured Proof of Concept (PoC) directly in your production environment.

The PoC follows these steps:

1. Data collection
2. Defect annotation
3. Pilot algorithm
4. Live testing
5. Operator feedback loop
6. Evaluation and next steps

Step 1: Data Collection

Temporary cameras are installed to capture product flow under normal conditions. Both defect-free and faulty samples are recorded.

Step 2: Defect Annotation

Captured images are labeled by experts. Defects are marked and categorized (delamination, discoloration, density problems). Quality controllers’ input here is critical.

Step 3: Pilot Algorithms

Early machine learning models are trained to distinguish acceptable from faulty products. The goal isn’t perfection but proof: can obvious defects be detected reliably?

Step 4: Live Testing

Models are connected to real-time camera feeds. The system flags potential defects, but final approval remains with human operators. This way, the PoC supports decisions without risking production quality.

Step 5: Operator Feedback Loop

Operators validate or correct system detections. Their feedback is added to the dataset, making the models more accurate with each cycle.

Step 6: Evaluation and Next Steps

The PoC concludes with answers to key questions:

• How many defects were caught that manual checks missed?
• Did the system slow or speed up the line?
• What is the potential cost saving if scaled to full production?

With these insights, management and quality teams can decide confidently on scaling up.

5. Why This Matters for Quality Controllers

Computer vision in quality control is about giving inspectors sharper, more reliable tools for quality control. Unlike manual checks that can vary from person to person, AI ensures consistency, with every product inspected against the same objective standard. It also brings speed, detecting and flagging defects in real time without slowing or stopping production. At the same time, the system provides traceability, recording every inspection result to support audits, compliance, and continuous improvement efforts. Finally, it offers scalability, making it possible to expand the same approach to additional lines or plants as production grows.

6. Ready to Invest in an AI Glass Wool Quality Inspection System?

Glass wool is a high-performance material, and its value depends on maintaining consistent quality. Computer vision offers manufacturers a practical path forward: start with focused workshops to define requirements, then move into a structured Proof of Concept that tests AI directly on the production line. Each step builds on the last, ensuring progress is measurable and low-risk.

For quality controllers, it’s about gaining sharper tools. AI brings consistency, inspecting every product against the same standard. It adds speed, flagging defects in real time without slowing production. It provides traceability, recording every inspection for audits and continuous improvement. And it enables scalability, making it possible to extend the system to new lines or facilities as production grows.

By combining expert knowledge with AI-based automation, manufacturers can move from manual spot checks to a collaborative system that works alongside their teams.