An Integrated Phenomenological and Physics-based framework for Multi-Sensor Vibration Analysis

Management Sciences has developed a layered algorithmic framework for high dimensional data analysis and fault detection focusing on multi-sensor vibration data. Processing modes include signal processing, factor analysis, source separation, feature extraction, classification, and Bayesian contextualization. These are applied dynamically to support online health monitoring and prognostic health management. Our existing approach, called “Data Driven Diagnostics and Prognostics”, is based on a physics-free “phenomenological” methodology for characterizing the vibration signatures associated with normal and fault states, for detecting and classifying unknown states through their vibration and related sensor signatures, and for tracking progressions to failure. This phenomenological approach has been successfully applied and tested on a workbench including power supply, motor, drive shaft, gears, couplings, pump, and valves. This proposed research project would develop and extend this technology in two ways. First, we would integrate our current phenomenological approach with a physics (kinematics and geometry)–based approach for correlating vibration frequencies, magnitudes, and energies with the relevant excitatory phenomena including fault modes. Second, we would extend our existing application domain to high speed turbo-machinery, making the necessary adaptations and modifications to address the challenges associated with a multiplicity of operational and transitional states and the potential for short event horizons.