Optimal rapid multidisciplinary response networks: RAPIDDISK |
| |
Authors: | S Nagendra JB Staubach AJ Suydam SJ Ghunakikar VR Akula |
| |
Affiliation: | (1) Pratt and Whitney Engines Division, United Technologies, AT&PD – System Optimization, 400 Main St., M/S 165-16, E. Hartford, CT 06108, USA;(2) Software Units Layout, InfoTech Enterprises Limited, Plot No. 11, Hyderabad, AP, India |
| |
Abstract: | The role of uncertainty in information rich design systems is critical to the development of advanced propulsion systems. Future turbine engines would have lower lifetime operating costs similar to current evolving automotive systems. Detailed multi-physics models (thermo-fluid, structural and mechanical systems) and an operational environment are enablers of rapid correction and model-based predictive analyses. Bayesian machine learning paradigms are developed to identify the behavior of turbo-machinery components for preliminary design. The embedded models approach enables systematic evolution from individual components level to the advanced engine. The embedded models approach enables systematic evolution from individual components level to the advanced engine. A rapid response strategy is proposed, for design of turbine disks by using multidisciplinary optimization and neural networks. iSIGHT optimization software is interfaced with ANSYS to find optimum designs for a given set of design boundary conditions (rpm, live rim load, thermals, etc.). The optimum designs obtained from iSIGHT for different set of design conditions are used for machine learning and design knowledge recognition using the neural network technique. The trained network is used to predict responses for design boundary conditions. Responses predicted by the neural network are validated using ANSYS. Discrete design points are chosen from the wide design space of turbine disks. A hierarchical neural network approach provides an ability to quickly train the network and predict responses (weight, stresses, burst margin, etc.) for applied design conditions. This basic building process involves four steps starting from identifying design boundary conditions to the prediction of design shape for the disk. Sensitivity-based scaling rules are developed, to accommodate different materials for the disk. The technique is developed in RAPIDDISK, which provides an optimal preliminary shape and design attributes for a turbine disk. |
| |
Keywords: | multidisciplinary optimization (MDO) iSIGHT neural networks rapid response strategy rapiddisk basic building block approach and disk optimization |
本文献已被 SpringerLink 等数据库收录! |
|