Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

AUGMENTED NUMERICAL PROPULSION SYSTEM SIMULATION (NPSS) SURROGATE MODEL

IP.com Disclosure Number: IPCOM000248347D
Publication Date: 2016-Nov-17
Document File: 3 page(s) / 197K

Publishing Venue

The IP.com Prior Art Database

Abstract

A technique for building augmented versions of such surrogate models by using a processed version of the outcomes of the original model for training is proposed. Using processed version of the outcomes of the original model for training improves performance, for example, reduced dispersion.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 53% of the total text.

AUGMENTED NUMERICAL PROPULSION SYSTEM SIMULATION (NPSS) SURROGATE MODEL

BACKGROUND

 

The present disclosure relates generally to surrogate models and more particularly to augmented numerical propulsion system simulation (NPSS) surrogate model.

Generally, surrogate models are built for achieving improved computational performance. The technique of surrogate models has been employed in various types of models, including numerical propulsion system simulation (NPSS). These various conventional surrogate models provide improved computational performance. However, there is ample scope of improvement in the quality of the output compared to the original model.

It would be desirable to have a surrogate model that provides improved quality of output compared to the original model.

BRIEF DESCRIPTION OF DRAWINGS

Figure 1 depicts a comparison of the original model, a conventional surrogate model and a augmented surrogate model described herein.

Figure 2 depicts application of the augmented surrogate model on a dataset of ~160000 civilian turbofan engine takeoff snapshots, obtained from 149 different engine serial numbers.

DETAILED DESCRIPTION

A technique for building augmented versions of such surrogate models by using a processed version of the outcomes of the original model for training is proposed. Using processed version of the outcomes of the original model for training improves performance, for example, reduced dispersion.

A standard surrogate model is, in the best case, equivalent to the original model, at least for some region in the input space. The technique described herein includes post-processing the output of the original model to build the augmented surrogate model for and use this post-processed output to train the augmented surrogate model. The rationale is that post-processing, for example, smoothing with a running window mean or median, the output reduces or eliminates influence of some factors which were not considered in the original model.

According to one embodiment, new predictors may be added. By adding new predictors and/or using a non-linear model capable of capturing new interactions of the already considered predictors, the augmented surrogate model tends to learn these previously un-modeled aspects and thereby tends to provide outputs that are better than those of the original model. For example, the output has improved precision.

Figure...