A machine-learning powered electronic tongue – or ‘e-tongue’ – developed by researchers at Penn State University and NASA Goddard Space Flight Center is the latest electronic sensing device with promising commercial applications in the food industry.
Rapid advancements in AI and machine learning technologies have accelerated the development of similar e-tongues and electronic noses (e-noses) in recent years, providing food and beverage manufacturers with powerful tools for food safety, quality control, and even new product development.
The Penn State scientists published their findings on the machine learning-powered e-tongue in Nature earlier this month. Using a ‘biological neural network’ to mimic the human gustatory cortex – the brain region that interprets sensations coming from the taste receptors – the tongue and its system can detect subtle chemical differences in liquids and solids.
The sensors, graphene-based ion-sensitive field-effect transistors, can identify differences between similar liquids, such as various types of milk, sodas, or coffee blends, while also detecting signs of spoilage in beverages like fruit juices.
This system stands out from traditional methods because it uses non-functionalised sensors. These sensors can detect a wide range of substances without needing a specific sensor for each chemical.
AI plays a critical role in the e-tongue’s ability to deliver highly accurate results. Initially, the AI was trained using human-defined parameters, achieving over 80% accuracy in detecting chemical differences. However, when allowed to develop its own parameters based on raw sensor data, the AI’s accuracy improved to more than 95%.
This precision is especially valuable in applications such as identifying spoilage, contamination, or adulteration early in the production process, helping manufacturers ensure food safety long before products reach consumers.
To provide deeper insight into the AI's decision-making process, the researchers employed a method called Shapley additive explanations. This technique reveals how the AI weighs various data points when making its assessments.
While e-tongue technologies are making headlines today, the history of electronic sensing systems spans several decades. The idea of replicating human taste dates back to the 1980s, when the first e-tongue systems emerged. Early models were basic, designed primarily to identify fundamental tastes like sweet, sour, and salty.
At the same time, electronic noses (e-noses) were also being developed, with the aim of mimicking human olfaction. The first e-nose was developed in 1982 at Warwick University, and the technology has since evolved dramatically. Early systems were bulky, energy-intensive, and expensive, limiting their commercial applications. Over the past few decades, these devices have become more compact, affordable, and sensitive, thanks to innovations in sensor arrays, gas sensing systems, and AI-powered pattern recognition.
Most of today’s e-noses, such as the Cyranose 320 – a popular lightweight quality control device used in both food and chemical industries – use nano-composite sensors to detect volatile compounds and gases with a high level of precision. These sensors can identify the ‘smellprint’ of complex chemical mixtures and have proven essential in sectors like food, beverage, and even healthcare.
A good example of e-tongue application is in beverage production, where they can ensure consistency in flavour profiles across large batches. Whether it’s distinguishing between different blends of coffee or ensuring the right balance of sweetness in fruit juices, the e-tongue provides precise, real-time feedback, allowing producers to fine-tune their recipes. E-tongues are also proving effective in monitoring water quality during beverage production, detecting contaminants that could affect both safety and taste.
E-noses mostly serve similar functions today. One example is Sensifi, a system developed by Israeli scientists that uses carbon-coated electrodes to detect gases released by bacteria such as E. coli and salmonella. These gases produce distinct chemical signals that are analysed by AI, allowing the system to rapidly identify contamination.
Sensifi’s e-nose can provide results in less than an hour, offering a faster alternative to traditional laboratory tests, which can take days. This real-time detection capability allows food producers to identify and address contamination risks on site, reducing the chance of unsafe products entering the market.
E-noses are also being used to assess food freshness. For example, German company NTT Data has developed an AI system trained to recognise the “reference value” of specific odours, such as the smell of fresh coffee. This allows the system to assess when food products are no longer at their peak freshness, helping producers and retailers better manage inventory and reduce waste.
The future of electronic sensing might be the replication of the entire human sensory system, by combining electronic tongues, noses, and even eyes to predict and evaluate product quality.
In studies of apple juice and olive oil, combining these tools with traditional sensory analysis allowed researchers to track key compounds and attributes linked to consumer-defined quality. The multi-sensor approach provided better results than using any single system alone, correlating sensory panel data with chemical measurements more effectively.
Such tools are especially useful for addressing complex sensory traits like bitterness, which can vary significantly among individuals. While sensory panels remain a standard for product formulation, electronic tongues and noses offer a faster, more consistent alternative. Calibrated to mimic human responses, these devices can streamline development processes, helping manufacturers create products that meet specific flavour profiles.
In some cases, sensory technologies can be applied to products or treatments not yet approved for human testing. For example, an electronic nose was used to evaluate treatments to prevent browning in fresh-cut apples, including one treatment not cleared for consumption. The technology provided detailed insights into aroma profiles, correlating these with sensory data from approved treatments to form a complete picture of the effects on flavour.
As these technologies advance, their role in product development will likely expand. AI-powered systems that combine taste, smell, and visual data could allow manufacturers to fine-tune formulations and create products that align more closely with consumer expectations. The integration of electronic tongues, noses, and eyes could offer manufacturers a powerful tool for developing new products and refining existing ones.
