We developed a computer-controlled gas mixing system that provides automated test procedures for the characterization of gas sensors. The focus is the generation of trace gases (e.g. VOCs like benzene or naphthalene) using permeation furnaces and pre-dilution of test gases. With these methods, the sensor reaction can be analyzed at very low gas concentrations in the ppb range (parts per billion) and even lower. The pre-dilution setup enables to cover a high concentration range (1:62 500) within one test procedure. Up to six test gases, humidity, oxygen content, total flow and their variation over time can be controlled via a LabVIEW-based user-interface.

In this study, methods from the field of deep learning are used to calibrate a metal oxide semiconductor (MOS) gas sensor in a complex environment in order to be able to predict a specific gas concentration. Specifically, we want to tackle the problem of long calibration times and the problem of transferring calibrations between sensors, which is a severe challenge for the widespread use of MOS gas sensor systems. Therefore, this contribution aims to significantly diminish those problems by applying transfer learning from the field of deep learning. Within the field of deep learning, transfer learning has become more and more popular. Nowadays, building a model (calibrating a sensor) based on pre-trained models instead of training from scratch is a standard routine. This allows the model to train with inherent information and reach a suitable solution much faster or more accurately. For predicting the gas concentration with a MOS gas sensor operated dynamically using temperature cycling, the calibration time can be significantly reduced for all nine target gases at the ppb level (seven volatile organic compounds plus carbon monoxide and hydrogen). It was possible to reduce the calibration time by up to 93% and still obtain root-mean-squared error (RMSE) values only double the best achieved RMSEs. In order to obtain the best possible transferability, different transfer methods and the influence of different transfer data sets for training were investigated. Finally, transfer learning based on neural networks is compared to a global calibration model based on feature extraction, selection, and regression to place the results in the context of already existing work.

Based on a diode coupled silicon carbide field effect transistor (FET) with platinum as catalytic gate material, the influence of dynamic temperature modulation on the selectivity of gas analysis sensors FETs has been investigated. This operating mode, studied intensively for semiconductor gas sensors, has only recently been applied to FETs. A suitable temperature cycle for detection of typical exhaust gases (CO, NO, C₃H₆ , H₂, NH₃) was developed and combined with appropriate signal processing. The sensor data were evaluated using multivariate statistics, e.g., linear discriminant analysis. Measurements have proven that typical exhaust gases can be discriminated in backgrounds with 0, 10, and 20% oxygen. Furthermore, we are able to quantify the mentioned gases and to determine unknown concentrations based on training data. Very low levels of relative humidity below a few percent influence the sensor response considerably but for higher levels the cross interference of humidity is negligible. In addition, experiments regarding stability and reproducibility were performed.