Monitoring the performance of a closed-loop by detecting whether the mean mu and the standard deviation sigma of the control error ( or the control signal) belong to a fixed triangular region in the (mu, sigma)-plane is shown to be a sensible approach. This triangular region is approximated by a set of circles and for each of them, a CUSUM algorithm for local linear hypotheses is performed. A systematic approach to adjust the parameters of the monitoring algorithm is described.