INCREASING THE SELECTIVITY OF COMMERCIALLY AVAILABLE TIN OXIDE-BASED GAS SENSORS FOR MONITORING COMBUSTIBLE GASES IN-PROCESS ENVIRONMENTS

Tin oxide (SnO2) is one of the most popular metal oxide semiconductors used as a gas sensor. Commercial sensors that use TiN oxide are inexpensive and readily available. Computer controlled thermal cycling techniques were used with only one sensor to collect a series of conductance surfaces for various compounds versus concentration. These patterns were treated as inputs to a neural network program that was trained to associate particular patterns to a specific compound. When the neural network was successfully trained to recognize pure patterns, the network was presented with a conductance pattern from a combination of compounds to test if the neural network could identify the constituents. This analysis method allowed the qualitative recognition of the different gases in a mixture or in one single gas. The results using a one sensor approach for the neural network identification of methanol, isopropanol, and hexane are presented and discussed. This, to our knowledge, is the first attempt to use data from a single sensor and process it via a neural network program to identify gases found in process environments. © 1995, American Vacuum Society. © 1995, American Vacuum Society. All rights reserved.