A Machine Learning-Driven Multimodal Sensing System for Advanced Mold Detection

Traditional methods for monitoring mold contamination in food and beverage products are costly, labor-intensive, and heavily reliant on human intervention. These approaches increase production expenses, delay detection times, and elevate risks of product spoilage and inefficiencies in quality control. Existing techniques lack the capability for effective, real-time detection of mold contamination. In response, we develop an innovative real-time smart mold detection system that leverages machine learning-driven multimodal sensing. Our system integrates data from four gas sensors (alcohol, carbon monoxide, hydrogen, and air quality), a pH sensor, and a miniature microbial electrochemical sensor. Mold contamination leads to fermentation processes that produce alcohol, hydrogen ions, and carbon monoxide as sugars are metabolized. Additionally, as mold converts alcohol into acetic acid, the solution's acidity increases, and mold growth alters electrochemical activity-both detectable by our sensors. Powered by artificial intelligence and a neural network algorithm, our system demonstrates 100% accuracy in mold detection using Kombucha as a model product. Integrating this system into production lines can significantly reduce the risk of mold contamination, enhance product quality, and improve overall safety and reliability.