The paper presents a risk-informed predictive maintenance strategy to achieve condition-based maintenance. The paper will present on data architecture that is used to collect heterogeneous data from vertical motor-driven pumps and how the collected data is used by the feature engineering module to extract salient features associated with different faults. Once fault signatures are developed, diagnostics models like eXtreme Gradient Boosting is used for automate the fault classification process. Given the diagnostic outcome, prognostic model like Auto Regression Integrated Moving Average is used to forecast the health condition of the motor-driven pump. Along with the prediction for 1 hour, 24 hour, and 48 hour prediction horizon, uncertainty bounds are also computed.