Despite their advantages regarding production costs and flexibility, chemiresistive gas sensors often show drawbacks in reproducibility, signal drift and ageing. As pattern recognition algorithms, such as neural networks, are operating on top of raw sensor signals, assessing the impact of these technological drawbacks on the prediction performance is essential for ensuring a suitable measuring accuracy. In this work, we propose a characterization scheme to analyze the robustness of different machine learning models for a chemiresistive gas sensor based on a sensor simulation model. Our investigations are structured into four separate studies: in three studies, the impact of different sensor instabilities on the concentration prediction performance of the algorithms is investigated, including sensor-to-sensor variations, sensor drift and sensor ageing. In a further study, the explainability of the machine learning models is analyzed by applying a state-of-the-art feature ranking method called SHAP. Our results show the feasibility of model-based algorithm testing and substantiate the need for the thorough characterization of chemiresistive sensor algorithms before sensor deployment in order to ensure robust measurement performance.

No abstract

Chemiresistive gas sensors are a crucial tool for monitoring gases on a large scale. For the estimation of gas concentrations based on the signals provided by such sensors, pattern recognition tools, such as neural networks, are widely used after training them on data measured by sample sensors and reference devices. However, in the production process of low-cost sensor technologies, small variations in their physical properties can occur, which can alter the measuring conditions of the devices and make them less comparable to the sample sensors, leading to less adapted algorithms. In this work, we study the influence of such variations with a focus on changes in the operating and heating temperature of graphene-based gas sensors in particular. To this end, we trained machine learning models on synthetic data provided by a sensor simulation model. By varying the operation temperatures between −15% and +15% from the original values, we could observe a steady decline in algorithm performance, if the temperature deviation exceeds 10%. Furthermore, we were able to substantiate the effectiveness of training the neural networks with several temperature parameters by conducting a second, comparative experiment. A well-balanced training set has shown to improve the prediction accuracy metrics significantly in the scope of our measurement setup. Overall, our results provide insights into the influence of different operating temperatures on the algorithm performance and how the choice of training data can increase the robustness of the prediction algorithms.