
The “Last Mile Problem” illustrates how achieving near-perfect accuracy (99.9%) in quality inspection dramatically increases in difficulty, underscoring the value of AI-powered synthetic data.
In 2025, sustainability is no longer optional—it’s a strategic imperative. Manufacturers, responsible for nearly 40% of global material waste, face rising demands to reduce emissions, cut waste, improve product consistency, and enhance efficiency.
ADVERTISEMENT |
Artificial intelligence (AI) is central to this transformation, with more than $200 billion invested in industrial AI last year alone. Yet one area is emerging as the true game-changer: quality control.
Effective quality control reduces material waste, energy use, and downtime—supporting not only operational efficiency but also corporate sustainability goals. However, achieving the necessary level of precision—especially the critical leap beyond the approximately 80% defect detection accuracy often reached relatively easily in proof-of-concepts or early deployments—has proven a major hurdle.
When manual inspection hits its limits
For manufacturers, the tipping point often arrives quietly—when manual processes that once sufficed start holding everything back.
Jaap Wiersema, managing director at Aviation Glass & Technology BV, recalls reaching that exact moment:
“Our inspection process was still largely manual. A single panel could take up to 30 minutes to inspect, and even then, there was a real risk of either missing small defects or rejecting good panels unnecessarily. Over time, the scrap losses and production delays started adding up—and we knew it wasn’t good enough anymore.”
Many manufacturers find themselves in the same position: too much variability, too much waste, and too little adaptability. In a world where customer expectations and regulatory demands keep rising, the cost of inaction grows higher every year.
Synthetic data: Turning a bottleneck into a breakthrough
Traditional AI systems trained solely on real-world production data have struggled to surpass the 80% accuracy ceiling. Manual data collection and labeling is slow, tedious, inconsistent, and expensive—and rare defects are, by definition, rare.
That’s where synthetic data offers a new path forward. Instead of the endless burden of collecting and labeling training data, manufacturers can now simulate defects under countless conditions—lighting, texture, material type—creating vast, perfectly labeled datasets in a fraction of the time and cost.
At Aviation Glass, this shift was transformational.
“We realized we couldn’t just digitize what we were doing manually. We had to rethink it altogether,” Wiersema says. “Synthetic data allowed us to build inspection models that understood our product variants immediately without months of retraining.”
Comparison of real and synthetic data: Synthetic data accelerates AI model training, offering consistency and efficiency beyond traditional methods.
Real-world results: Faster, smarter, more sustainable
For Aviation Glass, the move to synthetic data and AI-driven inspection delivered immediate, measurable improvements. Inspection times dropped to under 30 seconds. Over the course of deployment, the system scanned thousands of panels and assessed hundreds of thousands of potential defects—all while seamlessly managing more than 40 different product variants and more than 30 unique pass/fail criteria.
Crucially, the adoption of AI-enabled automated quality control allowed Aviation Glass to evaluate all defects on each panel—unlike manual inspection, which would stop at the first fail condition. With full defect statistics captured for every panel, AG could identify patterns and root causes, leading to targeted process improvements. As a result, they achieved an estimated 5% increase in yield while simultaneously reducing scrap and material waste by a similar margin. More importantly, this shift enabled Aviation Glass to move from reactive quality checks to a proactive, data-driven strategy—embedding continuous improvement and operational agility across its production line.
Human-in-the-loop technology: Working with people, not instead of them
While synthetic data and AI have transformed what’s possible, human expertise remains essential. The most effective systems integrate “human-in-the-loop” (HITL) frameworks where experienced operators validate, adjust, and teach AI models in real time.
At its best, it’s a true partnership. Technology doesn’t replace skilled people; it amplifies their capabilities, insights, and decision-making power.
Imagine not a robot taking over your line, but a supercharged magnifying glass handed to your best people, making them even sharper. That’s the kind of integration that doesn’t just speed up operations; it future-proofs them.
HITL integrates human expertise with AI, enhancing accuracy and decision-making in manufacturing quality control.
A new era for manufacturing leaders
For manufacturers striving to meet the demands of efficiency, sustainability, and agility, a new quality control paradigm is emerging.
By leveraging synthetic data, AI-driven inspection, and HITL collaboration, companies like Aviation Glass are proving it’s possible to cut waste, accelerate deployment, and build more resilient operations—all without compromising on quality.
As technology reshapes industry, quality control is no longer just about catching defects; it’s about defining standards, preserving knowledge, and building systems that are resilient, responsible, and ready for the future. Manufacturers who recognize this shift have the opportunity to lead not only in efficiency but in integrity and innovation.
In the end, quality control is not just about what we build—it’s about what we choose to protect. The future will belong to those who see in every inspection not a task to complete but a promise to keep.
Add new comment