Liquid level estimation and fluid identification using FBG temperature sensors via machine learning algorithms
Name: KATIUSKI PEREIRA NASCIMENTO
Publication date: 05/05/2021
Advisor:
Name |
Role |
|---|---|
| ANSELMO FRIZERA NETO | Co-advisor * |
| ARNALDO GOMES LEAL JÚNIOR | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| ANSELMO FRIZERA NETO | Co advisor * |
| ARNALDO GOMES LEAL JÚNIOR | Advisor * |
| MARIA JOSE PONTES | Internal Examiner * |
Summary: This dissertation proposes the use of the fiber Bragg grating (FBG) temperature sensors array to estimate the fluid level. A optical fiber sensor (OFS) level is ideal for evaluating oil tank level because it is a sensor that does not conduct electricity, is small size and resists corrosive areas. However, these sensors are complex to assemble, requiring several steps after fiber fabrication. Due to the temperature variation inside the tank, there is a need for a temperature sensor with no connection to these sensors, to measure the temperature. FBGs are intrinsically sensitive to temperature and strain. Therefore, level sensors also need a temperature sensor to reduce the temperature cross-sensitivity issues. To demonstrate the possibility of using the FBG temperature sensor for liquid level estimation, the temperature distribution of an oil storage tank, 200 cm height and 40 cm in diameter, receiving solar radiation at the top, is simulated. Then, the presence of a 200 cm long and 125 μm diameter fiber inside the tank with different amounts and distribution of FBGs along the fiber is simulated. In the simulation, due to the low variability of the classes, the Random Forest (RF) algorithm was chosen for classification. Starting with 200 FBG equidistant, decreasing to 6, with different distributions along the fiber. It was possible to classify the oil with an accuracy of 94.89% using 8 FBGs, using Tests for Two Proportions with a significance of 5%, the accuracy is equal to use 50 FBGs. Using the results obtained in the simulation, we utlized a 22.5 cm beaker, with 3 FBGs inside. In the beaker, 3 different fluids are identified: water, mineral oil, and kyro oil. Afterwards their levels are estimated from the temperature distribution along the beaker (using the 3 FBGs). Furthermore, we keep the fluid inside the beaker heated by a peltier at the bottom of the beaker to 318.15 K during the entire experiment. We followed the same principle for the beaker experiment, using RF for both level identification, obtaining 100% accuracy in fluid identification, and fluid level measurement the mid RMSE was 0.2603. After the simulation commented above, and the bench tests, using the beaker at constant temperature, we decided to expand the experiment. In this way we propose a full-scale experiment, using 9 FBGs distributed in this tank to estimate the liquid level. The tank is 100 cm in height and 30 cm in width, with 9 FBG sensors distributed along with the tank height. For the detection, we use the following Machine Learning (ML) algorithms: Logistic Regression (LogR), Decision Tree (DT) and Support Vector Machine (SVM). Initially the algorithm chosen was RF, but when using it we obtained RMSE of 16.32 cm. The algorithms chosen are Weighted Linear Regression (WLR), Support Vector Regression (SVR), SVR with kernel selection minimize cost (SVRmin). We propose the Mixed Model (MM), which selects the lowest Root Mean Square Error (RMSE) among the tested regression algorithms at each level, and associates it to it. The MM has RMSE of 3.56 cm, which is approximately four times smaller than when using WLR. The SVM and SVMmin have RMSE of 6.28 cm and 6.14 cm, respectively.
