Use of Multivariate Statistical Techniques in Haccp Programs

Abstract

Food safety is the major concern for the food industry and Hazard Analysis and Critical Control Points (HACCP) is an effective safety management system. Data analysis is an important ingredient of this system. The use of Statistical Process Monitoring (SPM) methods in critical control point monitoring step can further improve a HACCP system,since SPM and HACCP have a common goal which is to prevent failures before they occur.Food production processes include many variables and generally they are not independent of each other. The use of multivariate statistical methods is more appropriate than that of univariate statistical methods for food processes and provides comprehensive analysis of the data. The aim of this study was to display the benefits of the use of multivariate SPM techniques in HACCP system.In this study, data were taken from a food processing plant, which uses HACCP program in the production. They were collected in a frozen vegetable production line and composed of raw material properties, process conditions, microbiological counts and end product analyses. The data were analyzed by using multivariate statistical techniques such as Principal Component Analysis (PCA), Multiple Linear Regression (MLR), Principle Component Regression (PCR) and Partial Least Square Regression (PLSR). In the monitoring step, multivariate statistical tools such as Hotelling's T2, Squared Prediction Error (SPE) and contribution plots were utilized. Cause and effect diagrams were also employed as a problem analysis tool to improve the process.Uncorrelated score variables of PCA of process data and quality data successfully analyzed out of control observations on time basis in T2 and SPE plots. Contribution plots displayed the responsible variables, which alarmed at particular time instant. Contribution percentages of variables obtained from these out of control points displayed that blanching temperature and microbial counts are very important contributing factors. Blanching temperature is a variable of the first critical control point (CCP-1) and microbial counts are the verification of that CCP. This result indicates that CCP-1 is the point which extra care should be taken.PCR and PLSR techniques were successful in analyzing the process and product data individually. T2 and SPE plots of these models were nearly the same with the PCA of process data and product data. The regression models (MLR, PCR and PLSR) were not able to explain the correlation structure between process and product data, completely. The in-control data set used in this study was insufficient to construct regression models since it failed to explain the normal operating conditions exactly.It was stated that the proper data collection in the production line would cause an enhancement in the application of multivariate statistical techniques, in both monitoring and prediction of critical control point measurements.