The dynamic derivatives are widely used in linear aerodynamic models in order to determine the flying qualities of an aircraft: the ability to predict them reliably, quickly and sufficiently early in ...the design process is vital in order to avoid late and costly component redesigns. This paper describes experimental and computational research dealing with the determination of dynamic derivatives carried out within the FP6 European project SimSAC. Numerical and experimental results are compared for two aircraft configurations: a generic civil transport aircraft, wing-fuselage-tail configuration called the DLR-F12 and a generic Transonic CRuiser, which is a canard configuration. Static and dynamic wind tunnel tests have been carried out for both configurations and are briefly described within this paper. The data generated for both the DLR-F12 and TCR configurations include force and pressure coefficients obtained during small amplitude pitch, roll and yaw oscillations while the data for the TCR configuration also include large amplitude oscillations, in order to investigate the dynamic effects on nonlinear aerodynamic characteristics. In addition, dynamic derivatives have been determined for both configurations with a large panel of tools, from linear aerodynamic (Vortex Lattice Methods) to CFD. This work confirms that an increase in fidelity level enables the dynamic derivatives to be calculated more accurately. Linear aerodynamics tools are shown to give satisfactory results but are very sensitive to the geometry/mesh input data. Although all the quasi-steady CFD approaches give comparable results (robustness) for steady dynamic derivatives, they do not allow the prediction of unsteady components for the dynamic derivatives (angular derivatives with respect to time): this can be done with either a fully unsteady approach i.e. with a time-marching scheme or with frequency domain solvers, both of which provide comparable results for the DLR-F12 test case. As far as the canard configuration is concerned, strong limitations for the linear aerodynamic tools are observed. A key aspect of this work are the acceleration techniques developed for CFD methods, which allow the computational time to be dramatically reduced while providing comparable results.
► The paper deals with experimental and numerical prediction of dynamic derivatives. ► Wind tunnel data for two civil transport aircraft configurations are analyzed. ► Results from vortex lattice methods, Euler and RANS solvers are compared. ► CFD methods are more accurate and robust than vortex lattice methods. ► Acceleration techniques such as frequency domain solvers are very efficient.
Deep generative adversarial networks (GANs) are the emerging technology in drug discovery and biomarker development. In our recent work, we demonstrated a proof-of-concept of implementing deep ...generative adversarial autoencoder (AAE) to identify new molecular fingerprints with predefined anticancer properties. Another popular generative model is the variational autoencoder (VAE), which is based on deep neural architectures. In this work, we developed an advanced AAE model for molecular feature extraction problems, and demonstrated its advantages compared to VAE in terms of (a) adjustability in generating molecular fingerprints; (b) capacity of processing very large molecular data sets; and (c) efficiency in unsupervised pretraining for regression model. Our results suggest that the proposed AAE model significantly enhances the capacity and efficiency of development of the new molecules with specific anticancer properties using the deep generative models.
Recent advances in deep learning and specifically in generative adversarial networks have demonstrated surprising results in generating new images and videos upon request even using natural language ...as input. In this paper we present the first application of generative adversarial autoencoders (AAE) for generating novel molecular fingerprints with a defined set of parameters. We developed a 7-layer AAE architecture with the latent middle layer serving as a discriminator. As an input and output the AAE uses a vector of binary fingerprints and concentration of the molecule. In the latent layer we also introduced a neuron responsible for growth inhibition percentage, which when negative indicates the reduction in the number of tumor cells after the treatment. To train the AAE we used the NCI-60 cell line assay data for 6252 compounds profiled on MCF-7 cell line. The output of the AAE was used to screen 72 million compounds in PubChem and select candidate molecules with potential anti-cancer properties. This approach is a proof of concept of an artificially-intelligent drug discovery engine, where AAEs are used to generate new molecular fingerprints with the desired molecular properties.