Conventional preventive maintenance strategies and condition monitoring methods are often deficient in accounting for complex behavior of most industrial systems. These methods, capable of assessing health of components fall short on reliable system level health assessment. This integration problem often requires fusion of different measurable process parameters to tackle data disparity. Physical failure-based degradation modelling of single fault modes has limited account for interdependencies between subsystems and their effect on systems overall health. For this purpose, predictive maintenance is established to optimize maintenance efforts with prognostics and estimations of Remaining Useful Life. To meet the goal of the industry to provide reliable failure predictions and solutions for real-time derivation of prognostic parameters for predictive maintenance, thesis describes methods for deployment of predictive maintenance methods and provides workflow for implementation of prognostic algorithm that can be useful for applications across the industry attempting to utilize sensor data for maintenance management. The results show that multivariate multidimensional data can be useful for estimating predictions and utilized for maintenance purposes by means of data-driven methods such as dimensionality reducing for anomaly detection, residual similarity degradation modeling and use of machine learning classification algorithms.Methods described in the thesis are considered to suite as an extension of traditional condition-based maintenance management with use of novel technologies and prognostic algorithms.Thesis is intended to be useful for students interested in applicable methods of implementing predictive maintenance and all those with interest in RAMS oriented fields.