Inspection suppliers weigh in on AI’s real-world applications in food and beverage, its current limitations, and what’s next for smarter inspection systems.
AI can allow for improvements in food and beverage inspection, but it's accuracy depends on robust training data.
sorbetto via Getty Images
Artificial intelligence (AI) has become one of the most talked-about, and often misunderstood, developments in CPG manufacturing.
While the technology is still evolving, it’s clearly influencing the future of product inspection systems, with suppliers already using it for more accurate contaminant detection and adaptive learning for new products. That said, it’s important to recognize that AI is not an end-all solution, and it requires careful application to be truly effective.
AI-powered inspection took center stage at PACK EXPO Las Vegas, where suppliers demonstrated their latest advancements and practical applications. Just a few examples from the show include Oxipital AI’s V-CORTX no-code AI vision platform and Eriez’s SenseGuard Systems with AI-enhanced analysis. Stay tuned for our Inspection/Detection Innovations Report coming soon, compiling all the latest advancements in inspection and detection our editors found at the show. Sign up here for first access once the report is published.
To better understand the current role, limitations, and outlook of AI in food inspection, ProFood World reached out to inspection system suppliers Eagle Product Inspection, METTLER TOLEDO, and ProSpection Solutions. We’ve compiled their insights below, offering a focused look at AI in food inspection today.
(Responses have been edited for length and clarity. Repeated definitions and concepts between suppliers have been trimmed.)
How would you define the roles of AI, machine learning (ML), and generative AI in food inspection? Why is it important to distinguish between them?
Norbert Hartwig, Head of R&D, Eagle Product Inspection:
AI is best understood as the broad field of computer systems that learn to interpret information and make decisions in ways similar to human intelligence.
ML is a subset of AI focused on using data to automatically adapt and improve performance, and it’s already applied in Eagle systems. For instance, our x-ray platforms use ML to analyze absorption patterns and adjust mass measurement calculations in real time.
Generative AI, by contrast, is designed to create new ideas or designs based on prior data. In one test, it suggested an X-ray system concept that included features like shielding curtains but placed critical components in unrealistic locations. That experiment showed both the promise and the limitations of generative AI.
Jeff Youngs, President and CEO, ProSpection Solutions:
In food inspection, AI-powered systems are smarter, faster, and more consistent at quality and safety checks.
It is important to understand which technology is at work to set realistic expectations, guide investment decisions, and ensure compliance with regulatory and safety standards. At Prospection, our systems leverage these distinctions, using ML and AI to optimize detection and generative AI to strengthen model training, delivering smarter, more reliable inspections.
What does AI realistically mean in food and beverage product inspection today? Do you see any limitations in its current form?
Ian Scott-Mance, Digital Marketing Manager at METTLER TOLEDO:
Today, AI in food and beverage inspection mainly involves machine learning models trained to detect contaminants and defects and ensure product quality by analyzing images and sensor data.METTLER TOLEDO metal detectorMETTLER TOLEDO
Limitations include reliance on quality training data, potential challenges in detecting novel defects and the need for ongoing model updates to handle changing product variations.
Youngs (ProSpection Solutions):
AI in food inspection powers X-ray, visual, and sensor-based inspections to detect contaminants, defects, and packaging issues faster and more consistently than manual checks.
Its limitations include that AI depends on large, high-quality datasets, so rare defects may be underrepresented. It lacks deep context understanding, so some subtleties require human judgement. False positives and negatives may occur, especially with new products. High-performance hardware may be costly and complex to implement. Regulatory validation is also still required to ensure compliance.
Does your company currently use AI or machine learning in inspection systems? Can you provide examples?
Youngs (ProSpection Solutions):
Yes. Prospection Solutions partners with System Square to integrate AI, ML, and deep learning into X-ray and visual inspection systems. These onboard AI systems adapt in real time, learning from product variations to improve safety and quality. Together, Prospection and System Square deliver inspection solutions that optimize safety, maintain product quality, and give manufacturers confidence from grind to final package.
One example is the Dual Energy X-Ray “Bone Hunter,” which uses System Square’s deep learning neural networks to detect bone fragments in meat, poultry, and seafood, adapting to size, shape, and density variations to reduce false rejects and protect brand integrity.
Hartwig (Eagle Product Inspection):
Eagle X-Ray inspection machine equipped with PXT AI-powered photon counting technology.Eagle Product InspectionYes, Eagle systems already incorporate machine learning. One example is in mass measurement; by analyzing X-ray absorption patterns, the system detects changes in product composition and adjusts its weight calculations automatically. This ensures consistent portion control, reduces giveaway, and improves overall quality. We also combine these capabilities with advanced technology such as dual-energy photon counting, which adds pixel-level material data for even more precise inspection.
What’s a common misconception about AI in inspection systems that you think the industry should better understand?
Hartwig (Eagle Product Inspection):
The biggest misconception is that inspection systems are now “fully AI-driven” and can manage themselves without human involvement. In practice, most so-called AI systems depend on significant upfront training and curated datasets. Results often take months to reach full effectiveness.
Human oversight remains essential, and processors should view AI as a powerful tool to enhance proven inspection methods, not as a replacement for them.
Scott-Mance (METTLER TOLEDO):
A common misconception is that AI can solve all inspection challenges autonomously. AI systems need well-curated data and integration with other technologies to perform effectively.
Are there inspection system features or capabilities that might be labeled as “AI” but predate current AI trends? How should processors think about these features in terms of value and performance?
Scott-Mance (METTLER TOLEDO):
Some features labeled as "AI" may be advanced image processing or rule-based systems developed before recent AI trends. Processors should assess these features based on their actual performance and value, not just the AI label.
Hartwig (Eagle Product Inspection):
Yes. Features like pattern recognition, image filtering, and intelligent automation are sometimes marketed as “AI,” but they’ve been available for years. Eagle’s SimulTask PRO, for example, delivers exceptional image clarity and resolution—up to 65,535 grayscale values—using advanced algorithms.
These capabilities demonstrate that processors don’t need an “AI” label to recognize real value. The key is to focus on measurable inspection performance and proven reliability, rather than on buzzwords.
Youngs (ProSpection Solutions):
Yes. Some older systems are called AI but rely on pre-programmed rules or statistical monitoring.
This includes basic metal or foreign material detection that identifies anomalies but doesn’t learn over time, rule-based vision systems that use programmed rules to spot deviations or imperfections, and statistical process monitoring that flags out-of-spec products using simple rules.
These systems can work well for consistent products but lack adaptability. Modern AI and ML systems, like those from Prospection and System Square, learn and improve over time, reducing waste, catching subtle defects, and handling product variability.
What kind of training data is most critical in training AI models in food inspection applications? How can we ensure that the AI is receiving “good” data, that’s accurate and representative of the tasks it’s designed to complete?
Youngs (ProSpection Solutions):
ProSpection's IP69K ai X-Ray SystemProSpection SolutionsTo ensure accurate data we need to accurately label images, maintain high-resolution imaging, and continuously update datasets as products or defect types evolve.
Balanced, high-quality data is key. Effective AI models require defect-free images to establish what a normal product looks like, examples of contaminants to accurately detect foreign material, defective product samples to identify quality issues, and variations in shape, packaging, and rare edge cases to handle real-world production.
Scott-Mance (METTLER TOLEDO):
Critical training data includes both defect-free images for baseline comparison and diverse examples of known contaminants and defects to teach the model. Ensuring good data quality involves representative sampling, accurate labeling and continuous validation.
How is your company preparing or investing in AI’s future role in inspection, beyond what’s commercially available today?
Hartwig (Eagle Product Inspection):
We’re investing in feeding richer data, such as pixel-level material information from dual-energy photon counting, into AI models to improve their ability to detect subtle differences. Our R&D efforts focus on enhancing deterministic algorithms with AI, rather than replacing them, to create adaptive systems that evolve with production conditions. This approach ensures that AI development is firmly grounded in practical, validated inspection technology, while gradually unlocking new capabilities that benefit processors.
Youngs (ProSpection Solutions):
Prospection Solutions is investing in next-generation AI and ML technologies to anticipate defects, adapt to new products, and enhance food safety beyond current standards. Our R&D focuses on predictive analytics, adaptive learning, and intelligent integration across X-ray, vision, and package inspection systems.
Where do you see AI and machine learning realistically taking inspection capabilities in the next five to 10 years?
Scott-Mance (METTLER TOLEDO):
In the next five to 10 years, AI and machine learning are expected to improve inspection speed, accuracy, and adaptability, potentially enabling real-time detection of complex defects and predictive quality control.
Youngs (ProSpection Solutions):
Looking ahead, as this technology keeps learning and adapting, inspection will become more of a command center for food safety: able to handle product variations, anticipate risks, and help processors run more efficiently while staying compliant.
Hartwig (Eagle Product Inspection):
Over the next five to 10 years, I see AI and ML enabling adaptive learning, predictive maintenance, and even dynamic recipe adjustments, helping systems fine-tune themselves as products vary. I see this as “flexible adaptation,” where AI enhances existing inspection systems with greater responsiveness and efficiency. Even as AI evolves, food safety will continue to depend on validation as much as innovation.
The future is not about replacing today’s high-performing machines, but about making them even smarter, more adaptable, and more reliable over time.
Looking for CPG-focused digital transformation solutions? Download our editor-curated list from PACK EXPO featuring top companies offering warehouse management, ERP, digital twin, and MES software with supply chain visibility and analytics capabilities—all tailored specifically for CPG operations.
Looking for engineering services? Our curated list features 100+ companies specializing in civil, process, structural, and electrical engineering. Many also offer construction, design, and architecture services. Download to access company names, markets served, key services, contact information, and more!
METTLER TOLEDO metal detectorMETTLER TOLEDO
