Scientists in the US have developed an “electronic tongue” which can identify differences in liquid samples using AI, such as milk with varying water content, soda types and coffee blends. The technology can also spot the signs of spoilage in fruit juices and instances of food safety concerns.

The technology is inspired by the human tongue’s sensitivity to different foods and aims to address issues like food adulteration and authenticity assurance. It consists of a graphene-based ion-sensitive field-effect transistor or a conductive device that can detect chemical ions linked to an artificial neural network trained on various datasets.

The team combined graphene-based chemical sensors with AI to classify food products, by overcoming hardware challenges like sensor drift and variation.

Saptarshi Das, corresponding author and professor of engineering science and mechanics at the Pennsylvania State University, which led the study, tells Food Ingredients First: “Our research is related to neuromorphic computing and brain-inspired computing and we have developed various sensors in the past. When we thought about developing a sensor for chemicals, we considered the best natural chemosensor — the tongue, since it is very sensitive to different types of food products.”

The findings, published in Nature, highlight that the electronic tongue imitates the brain’s biological neural network, the gustatory cortex, which perceives and interprets different tastes beyond the five broad categories of sweet, sour, bitter, salty and savory. 

The researchers developed an artificial neural network — a machine learning algorithm that mimics the human brain in analyzing data and differentiating the subtlety of flavors.

Analyzing with AI

While examining liquid samples, the research team found that results were even more accurate when AI used its own assessment parameters to interpret the data generated by the electronic tongue.

Das visualizes the human tongue as the hardware piece and the brain as a software that runs the taste codification or taste detection in the brain.

“That is wher we wanted to bring in the AI, because at the end of the day, an artificial neural network is literally mimicking the connection between neurons and how the synaptic connections enrich the network’s capability to classify different types of chemical signatures.”

The scientists provided the neural network with 20 specific parameters to assess, all of which were related to how a sample liquid interacts with the sensor’s electrical properties. 

based on these parameters, the team says that AI could “accurately detect samples, including watered-down milks, different types of sodas, blends of coffee and multiple fruit juices at several levels of freshness.” It also measured their content with more than 80% accuracy in approximately one minute.

Addressing milk fraud

As milk counterfeiting methods become more complex, techniques to identify adulteration have also evolved alongside, notes a study conducted in Romania last year. The rising demand for milk and increased market competition and complexity of the supply chain are some of the reasons behind such food frauds, which the FDA says are “designed to avoid detection.” 

Experts estimate food fraud affects 1% of the global food industry at a cost of about US$10-15 billion a year, with more recent estimates putting the cost as high as US$40 billion a year, states the organization.

Das believes Penn State’s electronic tongue can help manufacturers address these concerns, but developing it was challenging. 

“When you mix water with milk, the chemistry does not change too much. So if you think about a chemical sensor designed for milk, it is not only very different from the one for milk that contains only 5% of water. Similarly, when you think about a hardware piece, it’s generally very difficult to see a visual difference, which we struggled with for a long time.”

Since subtle differences are difficult for the sensor to pick up because of the sensor drift, the team decided to take the help of AI to resolve it.

“We thought that if we, as human beings, cannot solve this problem by looking at the characteristics of the device, maybe AI could figure out subtle changes in the device characteristics, which can then help us to classify different types of food products,” Das explains. 

“This is wher we brought in the artificial neural networks, which helped us resolve these kinds of challenges. So despite having a lot of manufacturing variation from one chip to the other chip in the devices, we can mitigate it, because the network looks into something more subtle than what the humans can derive.”

He explains that the neural network can determine the milk’s varying water content and whether any degradation indicators are meaningful enough to be considered a food safety issue.

In October, another team of US scientists tapped into AI to detect milk anomalies like contamination and unauthorized additive addition. They applied their “explainable AI tool” to publicly available, genetically sequenced datasets from bulk milk samples to showcase the method’s robustness.

Beyond milk safety

Das explains that the electronic tongue tool can be applied to other products besides milk, such as grape juice, which often contains a toxin called Patulin.

“Many grape juice manufacturing companies don’t have an easy way of detecting Patulin and very often need to recall their products. Our sensor can figure out whether the grape juice has even minute quantities of the toxin in it.”

Additionally, the sensor can detect the aging of pineapple and orange juice with added preservatives and differentiate between different types of cola drinks, such as zero-sugar or caffeine-free variants. 

“The sensor can indicate mislabeling to help consumers determine the drink’s authenticity. This will in turn help the consumer, manufacturer and distributors of the product. Thus, the AI-powered graphene electronic tongue can transform the food industry.”

The electronic tongue concept was also used to detect microbial faults in white wines by scientists at the Washington State University, US, earlier this year. 

Eyeing solid food safety

When asked if the technology can also be applied to solid foods, Das says efforts are ongoing.

“Right now, the sensor can only detect liquid solutions, but we are trying to work toward making this sensor work also for solid products like spices which face various challenges like adulteration and lack of authenticity.” 

Adulterants in spices may be fillers to increase volume or weight, substituting all or a portion of ingredients with a similar inferior quality ingredient or chemicals to artificially enhance color, flags the American Spice Trade Association.

The team’s future plans involve developing a handheld, low-power device for consumer use and an app to connect their device to cloud servers to detect adulteration in foods, since machine learning algorithms “cannot run on chips,” Das concludes.