Case Study of HTST Pasteurization Process

To implement actuator faults, the controllers are turned off for a specific time period which results in a constant signal being sent to the actuators. During the implementation of this class of faults, when the controlled variables deviate from the set-points, the controller and the actuator cannot respond until the implementation is over. 

the performance of HMMs in representing HTST data is assessed using the model residuals and the correlation coefficients of observed and estimated values of process variables. This is done by checking the normality of residuals of some important process variables (e.g., holding tube inlet temperature and steam valve setting, variables 3 and 5, respectively). It appears that the residuals are autocorrelated most of the time. The normality property is affected by the extreme values of faulty signals since the model may not perform well to estimate the measurements at times of fault implementation. If the observation sequence contains many outliers, the residuals will likely not belong to Normal distribution. 

The HMM strategy can be modified by considering a moving window to detect changes with low magnitudes and duration. It has been shown 

E ault Fault Time (sec) Steam Valve Signal (mA) Duration (sec) 

---