Using AI to characterise and measure ingredients

By SHERMAN HO, Chief Scientific and Technology Officer, ProfilePrint 

In the dynamic world of food manufacturing, ensuring consistency, quality, and safety of ingredients and end products is crucial. The advent of artificial intelligence (AI) is transforming these processes, spearheaded by companies like ProfilePrint. This technological leap is enhancing efficiencies, enabling faster decision-making, and significantly reducing costs, particularly those related to quality assurance. 

The need for AI in ingredient characterisation 

Traditionally, material characterisation and measurement in the food industry have relied on specific but time-consuming methods like laboratory testing or potentially inconsistent and subjective methods such as human tasting panels. At the basic end, there is reliance on Certificates of Analysis, which often lack detailed material functionality data crucial for end-users. Moreover, in-line near-infrared (NIR) spectrometers deployed for monitoring raw material quality or process completion often prove insensitive and costly to customise and maintain.

AI, particularly machine learning, offers a unique opportunity to address these challenges. Machine learning enables IT systems to learn and define quality or grade levels and verify material consistency across batches. ProfilePrint, a Singapore-based AI-driven ingredient quality platform, exemplifies this transformation. Their patented digital food identity-as-a-service technology provides rapid decision-making capabilities by leveraging digital fingerprints captured by analytical devices to build models that facilitate efficient decision-making. 

Analysing cocoa nibs with ProfilePrint

How AI transforms ingredient characterisation 

1. Digital fingerprinting  ProfilePrint's technology involves creating digital fingerprints of food ingredients. Using spectroscopy, it captures the unique molecular signature of each sample. Like human fingerprints, these digital fingerprints are unique to each ingredient. This approach fundamentally differentiates itself from conventional methods by focusing on overall patterns in the fingerprints rather than only specific signals like peaks at certain wavelengths. 

2. Keeping the human in the loop  A successful model requires correlating a set of input data with expected output data, akin to how humans learn. When trained sensory panels taste a food sample, their senses receive inputs processed by the brain to produce an output, such as a score or a pass/fail decision. ProfilePrint operates similarly but relies on electromagnetic spectral rather than taste and smell. The analyser sees digital fingerprints as inputs, and ProfilePrint's proprietary “Toolbox” analyses these patterns to predict various quality metrics like sensory scores, flavour notes, defects, blend component ratios, and more. Initially, the system learns these metrics by correlating fingerprint inputs to known scores. 

3. Rapid sample analysis, unlimited potential  Once these relationships are established, simple sample preparation, coupled with fast scanning and prediction speeds enhance throughput, allowing manufacturers to make informed decisions swiftly. This approach also mitigates issues like environmental changes or sensory panellist fatigue, adding a layer of consistency. For example, at the Global Let’s Talk Coffee Conference  2023 in Honduras as well as the China International Coffee Fair in Kunshan 2024, ProfilePrint predicted coffee scores aligning with expert tasters’ evaluations, without the need to roast and brew the beans. 

The potential applications of AI in ingredient characterisation are vast. For instance, estimating the ratio of soy and pea proteins in a pancake mix with single-digit percentage errors can facilitate quality checks on blend homogeneity. Similarly, raw materials can be quickly analysed for "Pass" or "Fail" before acceptance, ensuring only suitable materials proceed to downstream processing. Anomalous batches of processed materials like modified starches or cocoa powder can be flagged before release, ensuring high quality for downstream buyers or consumers. 

AI can also optimise blending recommendations. Different lots of raw coffee beans can undergo combinatorial optimisation to produce recommended proportions for blending, approximating a target blend. This shortens the product development cycle and helps find substitutes when ingredient stocks fluctuate. 

The future of AI in food manufacturing

The integration of AI in ingredient characterisation and measurement is poised to redefine food manufacturing. By providing rapid, accurate, and comprehensive analysis, AI enables manufacturers to enhance product quality, streamline operations, and reduce costs. Companies like ProfilePrint are pioneering this transformation, offering solutions for a wide range of ingredients and applications. 

As AI technology continues to evolve, its applications in food manufacturing will likely expand, offering even more sophisticated tools for quality control and product development. Embracing these innovations will be crucial for manufacturers aiming to stay competitive in an increasingly demanding market. 

AI’s role in characterising and measuring ingredients marks a significant leap forward for the food manufacturing industry. By enhancing the efficiency and reliability of quality control processes and opening new possibilities for product innovation, AI ensures the highest standards of quality and consistency in our food supply.