Probabilistic Virtual Sensor for On-line Viscosity Estimation

Virtual sensors or software sensors are computer programs that estimate the value of difficult variables, using a model of the process and the measurements of the related variables. These difficult variables can be measurements in inaccessible places for hardware sensors, expensive instruments or instrumentation difficult to calibrate and maintain. Examples of variables are temperature measurements in turbines, emissions in chimneys, or viscosity of oil fuel. The estimation of the value of a variable utilizes a model and real readings of related variables. Several types of sensors can be built according to the type of model implemented. This paper presents a probabilistic viscosity sensor that utilizes probabilistic relations between the variables of the process. The model is based on Bayesian networks that can be learned using historical data and some expert advice. Experiments are presented where two data set are used. One is used for training the model and the other is used for validating the model. Promising conclusions are also presented.