Virtual Engine Devel­op­ment Using AI Methods

The research and devel­op­ment project VIT-VI (short for German: ​“VIrtuelle Trieb­w­erk­sen­twick­lung mit Verfahren der kün­stlichen Intel­li­genz” | English: ​“Virtual Engine Devel­op­ment using arti­fi­cial intel­li­gence methods”) focuses on arti­fi­cial intel­li­gence (AI) methods and their use in the context of virtual, sus­tain­able aero engine development.

The focus is on building and strength­en­ing AI com­pe­ten­cies and increas­ing the use of arti­fi­cial intel­li­gence methods to enhance pro­duc­tiv­ity in data- and sim­u­la­tion-driven design.

CAESES offers the pos­si­bil­ity to automate design explo­ration and opti­miza­tion processes for complex flow-exposed geome­tries. Pre­dom­i­nantly, CAESES is applied to generate robust para­met­ric models and acts as a frame­work for auto­mated design studies by coupling to fluid-dynamic solvers, whose results are used to optimize the shape — all within a single envi­ron­ment. CAESES does already provide machine learning (ML)-based solu­tions, such as response surface models (RSM) or prin­ci­pal com­po­nent analysis (PCA), which are sched­uled to be expanded within the VIT-VI research project. Here, the focus will be to imple­ment new AI‑, ML- and Deep Learning-based methods to reduce time and com­plex­ity in the modeling process and increase the effi­ciency of con­ven­tional opti­miza­tion processes.

Adjoint Opti­miza­tion

Adjoint methods have been imple­mented in CAESES with the goal of improv­ing the effi­ciency of the opti­miza­tion process by inte­grat­ing adjoint sensitivity/​gradient data into the workflow. Based on earlier studies, the existing func­tion­al­ity to deter­mine so-called para­met­ric sen­si­tiv­i­ties was improved and tested to ensure accuracy during the opti­miza­tion. These ​“para­met­ric sen­si­tiv­i­ties” connect the shape sen­si­tiv­i­ties from the adjoint CFD solver with the geo­met­ri­cal design veloc­i­ties in CAESES, which in turn describe the model surface dis­place­ment gra­di­ents for each design parameter.

New features were created to provide cus­tomers a quick and easy setup when intro­duc­ing adjoint methods in their opti­miza­tion process and a func­tion­al­ity to weigh the para­met­ric sen­si­tiv­i­ties to focus on critical regions of the geometry was imple­mented, further reducing the number of iter­a­tions and thus com­pu­ta­tional resources.

Prin­ci­pal Com­po­nent Analysis

Another way of reducing the need for expen­sive com­pu­ta­tional resources in an opti­miza­tion is to reduce the com­plex­ity of the geometry that needs to be opti­mized. In the scope of the VIT-VI project, research on how to expand the existing prin­ci­pal com­po­nent analysis (PCA) tool in CAESES is carried out in col­lab­o­ra­tion with the BTU (Bran­den­bur­gis­che Tech­nis­che Uni­ver­sität) and Rolls-Royce by imple­ment­ing and testing new algo­rithms to deter­mine the prin­ci­pal para­me­ters, as well as alter­na­tive methods to generate the back-trans­for­ma­tion from the prin­ci­pal com­po­nent space to the original CAD para­me­ter space.

Bal­anc­ing Process

Focusing on the accel­er­a­tion and improve­ment of modeling processes, research was con­ducted in col­lab­o­ra­tion with Rolls-Royce to incor­po­rate man­u­fac­tur­ing devi­a­tions into the design process of complex geometry, with the specific goal of dig­i­tal­iz­ing the rotor bal­anc­ing process.

Work­flows have been devel­oped to easily import mea­sure­ment data from man­u­fac­tured com­po­nents and directly link it to morphing-based mod­i­fi­ca­tions of the CAD geometry model. Future research will involve further inves­ti­ga­tions on more complex mor­phable shapes, con­tain­ing larger data sets, and uti­liz­ing the morphed geome­tries for estab­lish­ing a workflow to vir­tu­ally run the bal­anc­ing process. This aims at sig­nif­i­cantly reducing time and costs during the assembly of aircraft turbofan engines.

In close exchange with Rolls-Royce, data on the existing morphing methods can be col­lected and applied to further optimize Radial-Basis-Function (RBF)-morphing tech­niques and estab­lish new work­flows to connect large data sets with morphing methods in CAESES. The measured man­u­fac­tur­ing devi­a­tions are utilized to train a machine learning model and expand the data set to derive a sta­tis­ti­cal pattern, from which another ML model will be trained to optimize the bal­anc­ing process.

AI-based Chat Assistant

The devel­op­ment of an AI-based chat assis­tant in CAESES will help nav­i­gat­ing through the GUI and offering opti­mized search algo­rithms, so that users can quickly find case-specific infor­ma­tion inside the CAESES doc­u­men­ta­tion. Future research will include imple­ment­ing, eval­u­at­ing, and testing dif­fer­ent methods to intro­duce AI-based nav­i­ga­tion and assis­tance in the daily routine of engi­neers working with CAESES. In addition, further research will include auto­mated feature code gen­er­a­tion to help users write complex scripts without a deep back­ground knowl­edge of the feature pro­gram­ming language. Another part of this research will consist in col­lect­ing and uti­liz­ing data from typical geometry modeling processes to train ML models to suggest appro­pri­ate modeling steps and oper­a­tions during the creation of para­met­ric models in CAESES.

Outlook

In the later stages of VIT-VI, research on intro­duc­ing inter­faces to imple­ment reduced order models (ROMs) in the opti­miza­tion process in CAESES is planned, as well as expand­ing the applic­a­bil­ity of sur­ro­gate models for larger data sets. Addi­tion­ally, inter­faces will be provided for users to add custom sur­ro­gate models and thus offer even more freedom in devel­op­ing custom opti­miza­tion workflows.

Further research will include offering AI-based real-time flow field visu­al­iza­tion to quickly analyze shape vari­a­tions without the need to run addi­tional CFD calculations.

We are excited to be part of the VIT-VI project and are looking forward to a fruitful and inter­est­ing col­lab­o­ra­tion with Rolls-Royce Deutsch­land, BTU, TU Berlin (Tech­nis­che Uni­ver­sität Berlin) and THB (Tech­nis­che Hochschule Bran­den­burg), as well as diving deep into the pos­si­bil­i­ties to imple­ment AI-based methods to support engineer’s daily routines.

Funding

The project is funded by the program for the pro­mo­tion of research, inno­va­tions, and tech­nolo­gies of the state of Bran­den­burg (ProFIT Bran­den­burg) through the Ministry of Economic Affairs, Labor, and Energy of the state of Bran­den­burg (MWAE) and sup­ported by the Invest­ment Bank of the state of Bran­den­burg (ILB). The funding comes from the European Regional Devel­op­ment Fund (EFRE) and the state of Bran­den­burg. The VIT-VI research project is co-funded by the European Union.