Real-time monitoring of bread baking in ovens by smart odour sensors: focus on calibration transfer

Smart products or digital manufacturing allowing industries to perform flexible operations based on a strong human-machine collaboration can be considered the core of Industry 4.0, which are currently leading to the implementation in household and industrial appliances of smart systems emphasizing process optimization and product innovation. In this context, the electronic nose (e-nose) technology, combining low-cost gas sensors and machine learning, can address novel market needs concerning the realtime monitoring of production processes. While immense progress was made in instrumental odour sensing and e-nose technology by the end of the 20th century, there still are important aspects related to lack of stability and reproducibility, which require substantial refinement and improvement to enable the scaling up on large-scale production, thereby including specific calibration transfer (CT) and drift compensation techniques [1]. Focusing on CT, this paper investigates the transferability among nominally identical devices of the calibration model implemented for the real-time monitoring of bread baking in high-end domestic ovens. More in detail, this research involved two nominally identical e-nose setups equipped with 6 MOS sensors array each, to be used for the real-time monitoring of odour changes associated with the ongoing of the bread baking process. The research evaluated the possibility of implementing specific CT approaches directly on real samples aimed at enabling the use of the calibration model implemented on a primary unit for processing responses by secondary units. Different CT methods (i.e., Direct Standardization (DS) and Model Update) were compared. Requiring the lowest number of transfer samples (i.e., 2,4), DS showed the best results for this case study. Moreover, the results obtained proved that increasing the number of sensors used in developing the calibration model enhances its robustness and transferability.