Digital Twin Implementation for Coffee Machine Health Monitoring

As coffee machines become increasingly intelligent and data-driven, manufacturers are leveraging IoT and AI technologies to enhance brewing performance, optimize maintenance, and improve the user experience. This thesis, developed in collaboration with Zoppas Industries (ZI), presents a machine learning–based approach for assessing the health of thermoblocks, critical heating elements vulnerable to limescale buildup. Limescale reduces thermal efficiency and degrades brewing performance. A digital twin of the coffee machine was developed to simulate the behavior of its core components, including the thermoblock, water pump, and water pathway. Due to confidentiality concerns, the specific model and brand of the commercial coffee machine cannot be disclosed. Because of experimental constraints, such as limited time and resources, sub-models for pump temperature, efficiency, and water dynamics were created using simplified behavioral trends and interpolation methods. These simulated parameters, assuming a healthy thermoblock, were fed into a sequence-based LSTM neural network with a learned neural initializer to predict the flow of brewed coffee. During real-time deployment, the predicted parameters (flow, pressure, temperature) were compared to actual sensor readings. The resulting residuals, reflecting deviations from healthy behavior, were used to estimate the degree of limescale accumulation. Eleven statistical features were extracted from these residuals, followed by dimensionality reduction via PCA. A Support Vector Machine (SVM) classifier then categorized the thermoblock’s condition into four levels: healthy, mild, medium, and critical. The best-performing model achieved 81% overall classification accuracy. Notably, it identified healthy thermoblocks with 100% accuracy, making it a reliable early-warning system. Although distinguishing medium from critical levels was more challenging, the method proved highly effective for detecting early-stage limescale, supporting predictive maintenance in smart coffee systems.