Neural Network Robustness Analysis Using Sensor Simulations for a Graphene-Based Semiconductor Gas Sensor

Abstract

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.

Machine Learning Assisted Enhancement in a Two-Dimensional Material’s Sensing Performance

“…They concluded that the most significant information is provided by the electron field-effect mobility. Additionally, based on a sensor simulation model, Schober et al 170 provided a characterization technique to assess the resilience of several ML regression models (Support Vector Regression(SVR), Multilayer Perceptron (MLP), RNN using Gated Recurrent Units (GRU) cells, RNN using Long−Short-Term Memory (LSTM) cells) for a chemo-resistant gas sensor. By using a cutting-edge feature ranking technique known as SHapley Additive exPlanations (SHAP), the explainability of the machine learning models is examined.…”

Review–Modern Data Analysis in Gas Sensors

“…In many gas sensing applications, supervised learning methods showed tremendous success in improving the performance, robustness, and device reliability. 105 Different supervised algorithms such as support vector machine (SVM), random forest, XGBoost, Knearest neighbor (KNN), different neural networks are widely being implemented to address challenges like drifting, fault detection, calibration, and classification etc. [106][107][108][109][110] Models like SVM and KNN provide expected performance in online active learning applications even when encountering sensor drifting challenges.…”

Temperature Compensation for Semiconductor Gas Sensors Based on Whale Optimization Algorithm–Least-Squares Support Vector Machine

“…(14) However, this approach has high design costs, low reliability, high power consumption, and other defects, making it inconducive to practical application in engineering. (15,16) Software compensation usually uses information fusion techniques to build algorithm models to eliminate the effect of temperature on the sensor, which is more accurate and easier to implement than hardware compensation methods and is more commonly used for sensor temperature correction. (17) Li et al (18) addressed the problem of the serious degradation of the measurement accuracy of a six-axis force/torque sensor in extreme environments.…”