An improved method of nonintrusive load monitoring (NILM) using hidden Markov models is proposed. The proposed method is intended for electrical systems with a closed and known set of devices. Systems that meet this constraint are more commonly found in industrial or commercial contexts than the residential settings in which NILM is traditionally studied. The proposed method is designed to support applications relevant to these contexts, such as fault monitoring and system health assessments. To this end, the energy predictions of the disaggregation algorithm offer accurate descriptions of device behavior in time. The accuracy of the proposed method is validated using standard NILM performance metrics and data from public databases.