Quick detection of shifts under specific faults in multivariate statistical process control has been of interest. Process knowledge can be exploited to design a control chart to be sensitive to more specific mean shifts. Out-of-control observations are simulated representing the shifts resulted from the specific faults and thus the detection problem is converted to a supervised learning task. A control region can be learned through the classifier. The effectiveness of this approach is shown here through graphical illustrations in comparison with the results from normal theory and error rate tables.