Statistical process monitoring and fault diagnosis in a continuous HTST pasteurization process

An integrated statistical process monitoring method and a fault detection and diagnosis method are used to monitor and diagnose causes of abnormal operation of a h&barbelow;igh t&barbelow;emperature s&barbelow;hort t&barbelow;ime (HTST) pasteurization pilot plant. In this study, the parity space technique as a fault diagnosis method is merged with contribution plots and multivariate statistical process monitoring (MSPM) methods for the first time. Another aspect of the thesis is the use of real plant data of a pilot scale food pasteurization process. Use of multivariate monitoring and diagnosis techniques in food processes have not been considered except a few studies.; An empirical model of the process is developed by system identification methods by using process data collected under normal operating conditions (NOC). The data collected under the influence of different magnitude and duration of faults in the variables are used to validate the multivariate statistical monitoring methods and fault diagnosis method.; Hotelling's T2 chart and squared prediction error (SPE_N) chart are used as MSPM charts that use the information coming from all process variables. Since multivariate charts do not indicate the variable(s) causing the deviation in the process, parity space technique, which is a stochastic dynamic method, is used as a fault diagnosis (FD) method. The parity space technique is based on the residuals (parity residuals) computed by using the past and present observations of process variables and their predictions based on the stochastic model of the process. The parity residuals are then treated with a statistical test, the generalized likelihood ratio (GLR) test to diagnose the source cause of the abnormal behavior.; The results showed that the multivariate monitoring and diagnosis techniques can be used together to ensure the product quality and process safety in food processes. Multivariate statistical monitoring techniques detect the abnormal process behavior. The parity residuals of the measured variables and dynamic trends in contribution plots of T2 and SPE_N charts are used to assist in determining the source cause of a disturbance.