This paper proposes a covariate-regulated sparse subspace learning (CSSL) model to address the challenges of complex, time-varying cross-correlation in multivariate functional data, and demonstrates its application in process monitoring and fault isolation using SCADA data from wind turbines.