摘要
Turbomachinery blisks are paramount to the safe operation of gas turbines. These structures are found in the compressor and turbine sections of gas turbines, rotate at very high speeds, and are exposed to large aerodynamic forces and extreme temperatures. Furthermore, they are manufactured as one single piece, making them susceptible to stress concentration, which in time can result in high-cycle fatigue, causing failure of the structure. In addition, as-manufactured turbomachinery blisks are inherently mistuned, having small differences from their nominal properties at the blade level. This phenomenon breaks the system’s cyclic behavior and can result in enhanced vibration amplifications that are not trivial to predict. Additionally, wear formation at the blade level or blade damage is classified as large mistuning and can further enhance blisk vibration amplifications. Hence, it is of great importance to study turbomachinery blisk vibrations for the safe and efficient operation of gas turbines. To do so, the state-of-the art focused on developing methods to predict mistuning in blisks using as-manufactured blisk responses. Through physics-based, data-driven, and probabilistic approaches, these experimentally computed responses can be used to identify the blisk mistuning. However, full blisk experiments require very complex setups, long times to assemble, and can be financially expensive. Additionally, once mistuning is identified, it is critical to have computational tools to accurately and efficiently predict responses of blisks with both small and large mistuning. Physics-based methods have been developed using subsets of tuned modes to create reduced-order models (ROMs) to compute mistuned responses. However, a limited number of methods leverage data-driven approaches to predict blisk responses with both small and large mistuning, which could enhance accuracy by including experimental data in their training and could significantly reduce total computational cost. While mistuning identification and blisk response predictions are critical to gas turbine testing, the turbomachinery community has also focused on ways to mitigate blisk vibrations. Several studies have focused on introducing resonant vibration absorbers (RVAs). Other techniques enabled energy dissipation through friction or impact contacts to reduce vibrations. However, impact contacts have only been modeled using a restitution coefficient between rigid bodies. Additionally, no computational tool has been developed to study the joint effect of blisks with both friction and impact contacts modeled microscopically. To address these knowledge gaps, this dissertation proposes four main techniques: 1) a mistuning identification method using mass detuning, simple ping test data from isolated blades, and so-called influence coefficients to quantify the blade couplings through the disk. 2) A physics-informed neural network approach to compute ROM of blisks with small and large mistuning. Two neural networks with the same architecture are trained to compute matrix transfer functions that capture the blade couplings through the disk and the effect of the forcing on the blade root response. The root responses are predicted by solving a linear system of equations using the network output transfer functions. 3) A novel microscopic contact approach to model the impact contact interface using virtual springs and dampers, accounting for the elastic deformation and energy dissipation at contact. A previously developed RVA with friction contacts was adapted to include impact contacts. 4) A hybrid computational and experimental approach to predict the stiffness and damping values at the contact interface to accurately characterize the impact contacts in blisk-RVA application.