Alarm systems are of paramount importance for safe and efficient operations of industrial plants. This paper proposes a method to design dynamic alarm limits and adjust manipulated variables for multivariate systems. A hyper-ellipsoid model is built from historical normal data points to represent a normal operating zone (NOZ). Dynamic alarm limits of each involved process variable are mathematically defined and calculated based on the NOZ. Adjustment amplitudes of some manipulated variables are determined in order to remove occurring alarms, by exploiting reduced NOZs in parallel coordinates. The effectiveness of the proposed method is illustrated via numerical and industrial examples.