All sensors drift over time. For example, assume that a temperature sensor was calibrated within 0.1 % accuracy when installed. The sensor could report that the temperature of the fluid was 100,1°C while the real value was 100,0°C. After some years of operation, all though the true value might be unchanged at 100,0°C, the sensor would have drifted and consequently could report a value of 103,0°C instead. As these measurements are often used to calibrate (or as input to) a process model, it would result into a mismatch between the model and the true plant. Any optimization based on the model may therefore result in sub-optimal plant performance. For top side facilities, the solution to the drift problem is simply to recalibrate the sensor at regular service intervals. This is not possible when the sensor is installed on the seabed. The drift must instead be estimated from the process data and process model.
Incorporating information about the sensor drift into the modelling phase is one of the aims of this project. Furthermore, automatic model calibration and model selection routines which take sensor drift into account will also be investigated. The overall project goals are therefore to 
i.    Minimize the real plant-model mismatch, resulting in operation closer to the truly optimal point. 
ii.    Inform whether your model is trustworthy or not, by quantifying the uncertainty. This is important information if the model is to be used as a tool for decision making, e.g. related to condition based maintenance.