Abstract
Tilting-pad journal bearings (TPJBs) support the rotors of systems in many industries such as power generation, HVAC, and oil \& gas. Reliable operation with little to no unplanned downtime is paramount. Stability predictions using rotordynamic analysis tools ensure safe and reliable operation. As demands on rotating machinery push operating conditions towards higher speeds and more demanding loads, the accuracy of rotordynamic analysis becomes critical. Accurate rotordynamic analysis relies on having accurate component level models, especially for the bearings, such as TPJBs. The dynamic behavior of the TPJBs, expressed as stiffness and damping coefficients, are used in system-level analyses. The coefficients are predicted with a variety of TPJB models used in different bearing codes. Bearing codes must be validated by comparing predictions with experimental data. Validation experiments for TPJB bearing codes are often performed on dedicated test rigs.
Reliable validation for TPJB models requires data from test rig to have the lowest uncertainty possible. The state-of-the-art for performing and presenting uncertainty analysis is inadequate for confidently validating TPJB dynamic coefficients. Therefore, a framework for analyzing the uncertainty of TPJB coefficient identification that is more suitable to the problem is proposed in this dissertation.
First, the framework is described and applied to single-axis models. The framework is a simulation-based method that (1) defines a truth model to represent the physics of the test rig and expected measurement errors, (2) establishes an identification model that will process simulated measurements into dynamic coefficients, and (3) compares the coefficients identified in the simulation using the identification model with the true values defined in the truth model. Through the single-axis applications, important trends that affect the uncertainty of TPJB coefficient identification are identified. These trends are applicable for TPJBs and other components (such as seals) on rotating machines with behavior described by stiffness and damping coefficients. Furthermore, some common assumptions used in TPJB identification experiments are shown to be problematic, especially when identifying dynamic coefficients in high performance conditions (e.g. - high rotation speeds or high-frequency excitations).
The models are extended to higher-fidelity models in two-axes. First, uncertainty analysis is performed using models based on existing test rigs and compared with the single-axis models. The results in the higher-fidelity models support the results from the single-axis models and add further details into TPJB identification uncertainties. For example, identifying the cross-coupled coefficients of TPJBs is incredibly challenging due to the small magnitude of the cross-coupling. While at a system level these uncertainties may not change the dynamics significantly, they add challenges to bearing model validation. Second, the utility of the uncertainty analysis is demonstrated by updating the design of a new test rig for TPJB dynamic coefficient identification. The uncertainty analysis serves as a design tool to make changes that would reduce uncertainty.
Improving the analysis of uncertainty for TPJB coefficient identification will lead to test rig designs and experimental methods that reduce the identification uncertainty and allow more accurate model validation. This will ultimately improve TPJB modeling for rotordynamic analysis and increase rotating machine performance and reliability.
