Sensors are a common way of collecting health data on any sophisticated machinery. Utilizing the real time machine sensor data in order to predict problems ahead in time to mitigate the risk associated with unplanned failures has been of great interest to statisticians and reliability engineers as there are direct cost saving benefits. In effort to reduce downtime and improve overall reliability of the system, a robust, scalable modelling technique is desired to understand and forecast the future dynamics of the system as whole. We propose and analyze a framework for stochastic modeling of multidimensional machine sensor data. The system developed is self-adaptive and can be used with live data feed to provide real time predictions. We utilize the concept of low rank matrix approximation for efficient storage, retrieval and faster computation of real time data. Markov chain is used to model the process dynamics and to calculate short and long range probabilities of the system which can be used to identify potential failure introduction points in the system.