Supervised and Unsupervised Soft Sensors for Capsule Recognition in Espresso Coffee Machines

Soft sensing technologies are of fundamental importance in production systems and in intelligent devices, especially in the Internet of Things scenario. Soft sensing approaches can be leveraged to create new functionalities or to reduce costs. In this work, we present one of the first soft sensor applications for coffee machines, the first one in the literature for capsule recognition. Coffee maker users expect the machine to deliver the product in the shortest time possible, without sacrificing the quality of the brewing. Traditional heated water tanks (boilers) have been replaced with electrical instantaneous water heaters, allowing the water to be heated as it flows from the tank to the brewing unit. Furthermore, instead of manually loading coffee powder, sealed capsules are now commonly adopted. To maintain the desired water temperature during brewing, it is crucial to modulate the electrical power applied to the heater in a predictive manner. However, distinguishing between different capsule types becomes a challenge in a multi-vendor scenario where capsules may adhere to standard shape constraints but originate from different companies. To address this problem, we propose an approach that utilizes Artificial Intelligence techniques to infer information related to the capsule characteristics and detect anomalies. The soft sensor estimations can be applied to adjust the brewing process or inform users about lower-quality capsules or improper usage of the machine. Experimental results support the effectiveness of the procedure, which analyzes the water flow during the pre-infusion phase.