Cost-effective equipment maintenance for electric power transmission systems requires ongoing integration of information from multiple, highly distributed, and heterogeneous data sources storing various information about equipment. This paper describes a federated, query-centric data integration and knowledge acquisition framework for condition monitoring and failure rate prediction of power transformers. Specifically, the system uses substation equipment condition data collected from distributed data resources, some of which may be local to the substation, to develop Hidden Markov Models (HMMs) which transform the condition data into failure probabilities. These probabilities provide the most current knowledge of equipment deterioration, which can be used in system-level simulation and decision tools. The system is illustrated using dissolved gas-in-oil field data for assessing the deterioration level of power transformer insulating oil.