University of Twente Student Theses
As of Friday, 8 August 2025, the current Student Theses repository is no longer available for thesis uploads. A new Student Theses repository will be available starting Friday, 15 August 2025.
Fluid properties estimation and pressure drop sensing based on Coriolis flow sensor using machine learning
Zubavičius, R. (2023) Fluid properties estimation and pressure drop sensing based on Coriolis flow sensor using machine learning.
|
PDF
6MB |
| Abstract: | All forms of sensing devices have optimal performance conditions to achieve accuracy and reliability. Environmental conditions such as temperature affect the sensing elements of sensors, which results in a drop in accuracy due to non-ideal effects. Their effects are difficult to model due to the simplification of physics. In this research, we focus on Coriolis Mass Flow Sensors (CMFS), which are microfabricated mass flow sensors. They are able to measure the true mass flow by detecting the inertial effects caused by a fluid flowing through a vibrating channel. The aim is to investigate the viability of machine learning to combine various sensing elements of a device to counteract the non-ideal effects. We present statistical machine learning methods applied on a CMFS, to estimate mass flow, temperature, pressure drop, viscosity, and density as well as predict the measured fluid based on its electrical readings. The chip (1.2 cm2 ) has been exposed to different combinations of temperatures, flows, and pressure for four different liquids (ethanol, water, isopropanol, acetone), and one gas (nitrogen). The sensing elements were sampled using a Data acquisition (DAQ) system at 250 kHz, and features were computed using a Fast Fourier Transform (FFT) based on its resonance frequency. This thesis covers the classical machine learning and deep learning approaches. The classical machine learning approach covers linear regression, support vector regression with a non-linear kernel, and Gaussian process regression to estimate the fluid-state properties. Additionally, Naïve Bayes, Gaussian process and deep learning approach classifiers are employed to predict which fluid was measured based on its electrical readings. The results for density and viscosity estimations show less fluid-state dependence using GPR on trained fluids, achieving an average error of 0.132 kg/m3 (0.01%) and 0.62% for density and viscosity respectively, similarly, mass flow estimation combining sensing element features by polynomial features using linear regression achieves 2% accuracy, which is better by a factor of > 2 compared to its counterpart by directly estimating using phase difference, while even further improved by a non-linear method, improving performance by a factor of > 4. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 33 physics, 54 computer science |
| Programme: | Computer Science MSc (60300) |
| Link to this item: | https://purl.utwente.nl/essays/97809 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page