Estimation of uniform static regression model with abruptly varying parameters

A modular framework for monitoring complex systems contains blocks that evaluate condition of single signals, typically of sensors. The signals are modelled and their values must be found within the prescribed bounds. However, an abrupt change of the signal increases the estimated parameters´ variance, which raises uncertainty of the sensor condition although it operates correctly. This increase affects the whole system in evaluation of condition uncertainty. The solution must be fast and simple, because of runtime application requirements. The signal is modelled by a static model with uniform noise, variance increase is tested and if detected, the model memory is cleared. The fast and efficient algorithm is demonstrated on industrial rolling data. The method prevents the parameters´ variance from the artificial increase.