Irregularly sampled data in the design of a soft sensor system: some preliminary results

Abstract

In modern industrial applications, sensors are an expensive part of installed systems. Nevertheless, many system variables cannot be measured sufficiently frequently or accurately. Thus, soft sensors have been developed to estimate those variables without the expense of additional hardware. The use of a soft sensor with a bias update term has shown to perform well for disturbed systems with time delays and multirate sampling times. In industrial application, the time delay and sampling times often vary. Yet, the case of variation of the time delay and sampling time in the bias update term has not been considered in previous publications. This thesis tests a soft sensor with bias update term in simulation and gives a modification yielding better performance. It is shown that the tested method gives unstable results. Hence, a more general method with a bias update term that considers all possible sampling times in each step is proposed, giving stable results in simulation. Furthermore, the stability of the general method is proven mathematically by building a state space representation and applying