When implementing a control chart, the in-control parameters of the process are usually unknown and need to be estimated from the in-control data obtained from phase I analysis, and then used to construct the control limits for phase II online monitoring. Assume the in-control covariance matrix is unknown, we establish a one-sided-test-based control chart and two two-sided-test-based control charts based on the likelihood ratio test (LRT) statistics for testing the covariance matrix of the quality characteristic vector of the current process. Considering the randomness of the estimated covariance matrix, we construct the control limits by controlling the expected false alarm rate at a prescribed level. Algorithms that are computationally feasible are developed for constructing such control limits via Monte Carlo simulation. The performance of these control charts are evaluated in terms of the detecting power of various changes in the covariance matrix through a simulation study. The applicability and effectiveness of the three control charts are illustrated via a semiconductor example and two simulated examples.