A systematic approach for the automatic signal processing, evaluation and optimization of T-cycled gas sensors

Several groups have shown the improved functionality, especially concerning selectivity, sensitivity and stability, of semiconductor gas sensors either evaluated dynamically, i.e. by impedance spectroscopy [U. Weimar, et al., 1995], or operated dynamically, i.e. by temperature cycling [A.P. Lee, et al., 1999; W.M. Sears, et al., 1989; E. Llobet, et al., 2001; A. Gramm, et al., 2003]. The T-cycle approach has also been successfully tested for pellistors [M. Engel et al., 2004]. For an efficient development of sensor systems based on these approaches an application specific tailoring of the systems parameters, i.e. relevant frequencies for impedance measurements, as well as signal processing and evaluation is required. For T-cycling this includes the identification of the most sensitive operating temperatures of the sensor for the required gases and furthermore the search for stable features in the sensor signals to be used for evaluation. For many applications training data are obtained in the lab through the variation of gas mixtures and concentrations using automatic systems based on mass flow controllers. The subsequent signal processing is then done manually leading to a very time consuming and labor intensive process. In addition, the variation and adaptation of the system parameters requires an experienced expert to avoid errors, unnecessary measurements and sub-optimal classification results, which could be caused by faulty feature extraction and/or signal interpretation. In this paper we present an automated system including a methodology to acquire, process and evaluate sensor signals to create a largely self-calibrating system. In two independent case studies we present the results of the proposed automatic system.