Deep Learning Assessment of Wine Fermentation Processes Using IoT Nose and Tongue Probes

This paper presents a new IoT sensory system called SmartBarrel. The system can record wine fermentation attributes and parameters, acting as an electronic nose and tongue. The IoT devices that pertain to this smartBarrel capabilities are the probing nose and tongue devices, which can be easily attached to stainless steel wine barrels. These devices can periodically monitor fermentation parameters, including gas emissions from the nose and acidity, residual sugar, and color changes in the tongue. Nose and tongue IoT devices utilize low-power, low-cost IoT sensors for this purpose and have been validated in small wine fermentation tanks. Apart from the end node devices, the proposed system features a distributed cloud architecture incorporating open-source, industry-ready application services and tools, specifically the Thingsboard platform supported by a NoSQL Cassandra database for data storage and visualization. The authors developed and experimented with a deep learning auto-calibrating variable layer-length, variable cell LSTM recurrent neural network predictor, called V-LSTM, and a fuzzy controller model to analyze wine fermentation processes, provide fuzzy encoded attribute responses, and infer predictions of the fermenting wine alcoholic content concerning other parameters. The results of this approach were compared with existing machine learning techniques in the literature of shallow MLP classifiers, demonstrating an improvement in minimizing RMSE loss of at least 45%. The authors also indicate that their model can be easily adapted into a service capable of issuing breakpoint alerts and adapting to changes to the non-linear characteristic curves of wine fermentations.