Data driven approaches to characterizing cross-sensitive sensors and to improve calibration transer

Today, sensors can be found in many different applications, such as in home appliances, vehicles, medical equipments, mobile technology etc. Nearly all products with a reasonable technological height have a sensor integrated.Chemical Sensors are devices able to sense the chemical environment and are today an active area of research. A great potential has been predicted for these kinds of sensors, meeting future demands of e.g. environmental monitoring, health-care issues and safety. The potential lies in the fact that chemical sensors are unspecific in general, responding to several different species in the ambient. By using several different sensors simultaneously a great portion of the chemical environment can be sensed at one instant.The data generated when making measurements with multiple chemical sensors contain a lot of information, but t he information is not clearly visible and it is not easy to interpret the data as it is.Along with the development of complex sensor systems comes the need for advanced data analysis procedures, able to interpret and visualize the information provided by the sensors. In this work, various aspects of data analysis procedures are discussed. Techniques for exploring large dataset are treated and a general overview is given on algorithms able to learn characteristic patterns within data. Difficult ies caused by unwanted sensor phen6mena such as drift and noise are identified and it is discussed how to counteract for these.The thesis touches upon applications where chemical sensors are useful, and which requirements these applications put on the sensors and the data analysis procedures.Work will be presented in which attempts have been made to learn the composition of flue gases produced by boilers used for heat and power production. It will be shown that it is possible roughly estimate the concentration of carbon monoxides, oxygen, and hydrocarbons by using a set of relatively in-expensive chemical sensors and by applying data analysis procedures. Work will also be presented in which it is discussed how to counteract for unintentional differences between sensor elements, a problem causing trouble prior to commercialization and mass-market production of chemical sensor systems.