Smart density and viscosity sensing based on edge machine learning and piezo electric MEMS for edible oil monitoring
The new industry requires embedded sensor systems that can monitor processes and address issues quickly and intelligently. These devices must operate efficiently at any stage of the product’s lifecycle, using minimal resources and low energy. Integrating sensors with AI via microcontrollers is thus...
| Autores: | , , |
|---|---|
| Formato: | artículo |
| Fecha de publicación: | 2025 |
| País: | España |
| Recursos: | Consejo Superior de Investigaciones Científicas (CSIC) |
| Repositorio: | RUIdeRA. Repositorio Institucional de la UCLM |
| OAI Identifier: | oai:ruidera.uclm.es:10578/47694 |
| Acesso em linha: | https://doi.org/10.1016/j.sna.2025.116258 https://hdl.handle.net/10578/47694 |
| Access Level: | acceso abierto |
| Palavra-chave: | MEMS Deep learning Viscosity Density Olive oil |
| Resumo: | The new industry requires embedded sensor systems that can monitor processes and address issues quickly and intelligently. These devices must operate efficiently at any stage of the product’s lifecycle, using minimal resources and low energy. Integrating sensors with AI via microcontrollers is thus commercially significant and highly relevant. In this work, we introduce a compact, intelligent, and portable device based on a piezoelectric resonator microelectromechanical system (MEMS) and discrete electronic circuits managed by a microcontroller unit (MCU), capable of estimating the physical properties of fluids using deep learning algorithms deployed on the MCU, without relying on cloud-based computations. By analyzing multiple MEMS resonances in a training and calibration process with blends of various vegetable edible oils, we obtained valuable insights into the liquid properties. Various combinations of hyperparameters of a convolutional neural network were examined to optimize the model’s performance in terms of calibration and resolution errors. The system is employed to monitor oil mixtures, addressing a critical industrial issue such as olive oil adulteration. By measuring viscosity, our system can detect blends of pure olive oil with other vegetable oils in concentrations as low as 2%, with calibration and resolution errors of 0.47% and 0.14 mPa s for viscosity, and 0.0331% and 9.25 ⋅ 10−5 g/mL for density. These results are comparable to other commercial laboratory instruments, offering a low-cost, portable and precise alternative. |
|---|