Effectiveness improvement of electronic nose by modifying flow pattern of gas carrier combined with transient data analysis

Abstract

The objective of this research is to design the gas flow pattern entering an array of gas sensors to achieve the highest signal response from the sensor group or the maximum efficiency. Additionally, the study aims to develop methods for analyzing signal data from the gas sensor group during periods of maximum slope changes (transient features). The researchers designed and constructed an electronic nose chamber capable of installing flow pattern adjustment models, along with designing gas flow pattern adjustment devices for the electronic nose chamber in 4 formats. These formats include 1 laminar flow pattern and 3 turbulent flow patterns. Subsequently, all 4 pattern models were utilized to perform computational fluid dynamics simulations of the airflow entering the electronic nose chamber at a flow rate of 0.4 l/min. using the ANSYS program. The obtained results align with the research objectives. Afterward, physical models were fabricated using three-dimensional printing techniques and subjected to testing with real devices. Smoke tests were conducted to observe the behavior of the airflow entering the electronic nose chamber. The observed behavior closely matched the characteristics of the developed computational fluid dynamics model, thus enhancing the credibility and reliability of the simulation model. Subsequently, samples of rancid odor in brown rice were prepared by subjecting the rice to light-induced aging and accelerating the reaction with oxygen gas at a temperature of 40 oC over a period of 1 month. The non-rancid odor in brown rice samples were then tested using the electronic nose equipped with the flow pattern adjustment device. The obtained samples were sent to a laboratory to analyze the chemical components using the SPME-GC-MS method to confirm the quality of the samples used for testing. Following this, the researchers initiated olfactory tests using the electronic nose, conducting a total of 800 tests. The resulting data were processed and refined to ensure smooth signal values. This processing was done to extract the response signal values from the sensor group during both the transient state and steady state. As a result of this processing, the dataset expanded by a factor of 2, resulting in a total of 2,400 data points. Subsequently, the gathered data were analyzed to establish relationships, and prediction equations were created using Partial Least Square Regression (PLSR) technique in The Unscrambler software. Upon careful consideration and comparison of the accuracy of the predictive equations, it was found that the model adjusted for flow pattern and signal data range had the most accurate and high-precision predictions. Specifically, Turbulent Flow Model 1 at 50% transient state based on the maximum signal slope demonstrated the highest accuracy. With a decision coefficient (R-squared) value of 0.9749 and a minimal Bias value of 0.000181, it excelled in capturing the dynamics of the signal. However, it's important to note that using signals from the steady-state range also yielded results with similar efficiency, providing good outcomes for all the remaining flow pattern models, which approached an R-squared value of nearly 1, in general.