Stanford University engineers have developed a new AI model for food texture testing to improve the sensory experience of plant-based meat and encourage wider acceptance, especially among meat lovers.
The researchers say the method could accelerate the development of appealing vegan alternatives by replacing trial-and-error approaches with a systematic, data-driven strategy that can tailor specific features to match consumer expectations.
“Instead of using a trial-and-error approach to improve the texture of plant-based meat, we could envision using generative artificial intelligence to scientifically generate recipes for plant-based meat products with precisely desired properties,” the authors state.
Testing the “mechanical signature”
In the study published in Science of Food, the researchers show how they used a combination of mechanical testing and machine learning to measure and replicate the texture of plant-based and animal meat, similar to human taste testers.
The researchers used tension, compression, and shear tests to evaluate the texture of eight products: five plant-based (tofurky, sausage, hot dog, extrafirm and firm tofu) and three animal products (spam turkey, animal sausage, and animal hotdog).
“What’s really cool is that the ranking of the people was almost identical to the ranking of the machine”
Additionally, they used constitutive neural networks to find if plant-based meat mimics the “mechanical signature” of animal meat across the entire three-dimensional spectrum model instead of the usual one-dimension. The data was processed using machine learning while human testers ate samples to rate the products on a 5-point scale for 12 categories: soft, hard, brittle, chewy, gummy, viscous, springy, sticky, fibrous, fatty, moist, and meat-like.
After 157 mechanical tests and 288 neural network simulations, the researchers found that some plant-based meats already mimic the texture of animal meats well. Mechanical tests showed that plant-based hotdogs and sausages had textures similar to their animal counterparts. Meanwhile, plant-based turkey was stiffer, and tofu was softer than animal products. Human and mechanical texture assessments ranked similarly, validating the mechanical test accuracy.
“We were surprised to find that today’s plant-based products can reproduce the whole texture spectrum of animal meats,” said Ellen Kuhl, professor of mechanical engineering and senior author of the study shared with Stanford University.
“What’s really cool is that the ranking of the people was almost identical to the ranking of the machine. That’s great because now we can use the machine to have a quantitative, very reproducible test,” she added.
Exploring ingredients
Other findings showed that animal meats exhibited higher tension-compression asymmetry than plant-based meats, showing stiffer tension than compression. In contrast, most plant-based products showed the opposite pattern or were symmetric, which means they could break more easily.
Additionally, according to the researchers, the mechanical properties of plant-based meat depend heavily on the source of the plants and processing technique, with the most successful combinations featuring wheat, pea, or potato proteins. These findings were possible due to the AI-driven neural networks that map relationships between ingredients and textures. The researchers suggest exploring first which ingredients deliver the correct textures to reverse-engineer formulas and create more precisely engineered products.
The team is sharing their data online to promote collaboration and innovation in plant-based meat development, as artificial intelligence needs lots of data. The researchers will continue testing various food items, including engineered fungi, and build a public database to support further research since traditional testing methods are not standardized and results are not usually shared.
“Our study shows that the one-dimensional stiffness of plant-based and animal sausage and hotdog is nearly identical, but their three-dimensional characteristics are not.
“Our approach to automatically discover the mechanics of plant-based and animal meat with constitutive neural networks could be a starting point towards using generative artificial intelligence to reverse-engineer formulas for plant-based meat products with customer-friendly tunable properties,” said the authors.
