A Data Driven Method for Model Based Diagnostics and Prognostics

This article’s model based diagnostics system has four modules. Diagnosis and fault location forms physics models of the machine, measures states off the real in-service machine, generates simulated machine states and simulated sensor outputs for the machine model with loads same as the real machine, and compares simulated sensor outputs to real sensor outputs. The parameter tuning module adjusts (tunes) the parameters of the model until the simulated sensor outputs closely mimic real sensor outputs. Tuning transfers information on the system’s health from the sensor data to the model’s parameters. Parameters changed from nominal values locate faults and bad parts. For the health assessment module to assess machine health, we view a machine as a “machine channel” that organizes power and information flow through the machine. Machines focus power via an organization inherent in its components and design. Broken or degraded components disrupt this organization and the power and information flows. Shannon’s information theory for communications channels can then be applied as a health metric to this “machine channel”. Ageing of components degrades machine functional health. To prognose future health, differential equations that model ageing of the machine’s components are formulated and solved. These equations predict component degradation, and update values of parameters in the model associated with component ageing. With these future parameter values, simulations of the machine operation model can then predict “future” machine behavior, and system health. This article demonstrates these methods on motors and a pump.