Abstracts Engineering

Estimation of physical parameters in mechanical systems for predictive monitoring and diagnosis

by Thomas Nickel

Monitoring, diagnosis and prediction of failures play key roles in automatic supervision of machine tools. They have received much attention because of the potential for reduced maintenance expenses, down time, and an increase in the equipment utilization level. At present, signal analysis techniques are predominantly used. But methods involving system analysis are capable of providing more reliable information, especially for predictive applications of supervision. System analysis involves comprehensive analytical models combined with techniques developed in control theory, and experimental modal analysis. The primary objective of this research is to develop a methodology to monitor critical physical parameters of mechanical systems, which are difficult to measure directly. These parameters are inherent features of constitutive rigid body models. A method for computer aided model generation developed in this thesis leads to a gray box model structure by which physical parameters can be estimated from experimental data. Lagrange's energy formalism, linear algebra and homogenous transformations are used to promote parsimonious three-dimensional model building. A software environment allowing symbolic and arbitrary precision computations facilitates efficient mapping of physical properties of the actual system into specific quantities of the analytical model. Six different methods are postulated and analyzed in this thesis to estimate physical parameters such as masses, stiffnesses and damping coefficients. Implementation of this methodology is a prerequisite for the design of an on-line monitoring and diagnosis system, which can detect and predict process faults. Two mechanical systems are used to validate the proposed methods: (1) A simple multi degree-of-freedom (MDOF) system and (2) a machine tool spindle assembly. A practical application of physical parameter estimation is proposed for preload monitoring in high-speed spindles. Preload variations in the bearing can lead to thermal instability and bearing seizure. The feasibility of using accelerometers located on the spindle housing to estimate bearing preload is evaluated. The optimal environment for continuation of this research is collaboration with machine tool companies to incorporate the proposed methodology (or parts of it) into current design practices.