Publikationen

A combined convolutional autoencoder-recurrent neural network machine learning model is presented to directly analyze and forecast the dynamics and low-order statistics of the local convective heat flux field in a two-dimensional turbulent Rayleigh-Bénard convection flow at Prandtl number Pr=7 and Rayleigh number Ra=10^7. Two recurrent neural networks are applied for the temporal advancement of turbulent heat transfer data in the reduced latent data space, an echo state network, and a recurrent gated unit. Thereby, our work exploits the modular combination of three different machine learning algorithms to build a fully data-driven and reduced model for the dynamics of the turbulent heat transfer in a complex thermally driven flow. The convolutional autoencoder with 12 hidden layers is able to reduce the dimensionality of the turbulence data to about 0.2% of their original size. Our results indicate a fairly good accuracy in the first- and second-order statistics of the convective heat flux. The algorithm is also able to reproduce the intermittent plume-mixing dynamics at the upper edges of the thermal boundary layers with some deviations. The same holds for the probability density function of the local convective heat flux with differences in the far tails. Furthermore, we demonstrate the noise resilience of the framework. This suggests that the present model might be applicable as a reduced dynamical model that delivers transport fluxes and their variations to coarse grids of larger-scale computational models, such as global circulation models for atmosphere and ocean.

Purpose To investigate the spectral characteristics of fundus autofluorescence (FAF) in AMD patients and controls. Methods Fundus autofluorescence spectral characteristics was described by the peak emission wavelength (PEW) of the spectra. Peak emission wavelength (PEW) was derived from the ratio of FAF recordings in two spectral channels at 500-560 nm and 560-720 nm by fluorescence lifetime imaging ophthalmoscopy. The ratio of FAF intensity in both channels was related to PEW by a calibration procedure. Peak emission wavelength (PEW) measurements were done in 44 young (mean age: 24.0 ± 3.8 years) and 18 elderly (mean age: 67.5 ± 10.2 years) healthy subjects as well as 63 patients with AMD (mean age: 74.0 ± 7.3 years) in each pixel of a 30˚ imaging field. The values were averaged over the central area, the inner and the outer ring of the ETDRS grid. Results There was no significant difference between PEW in young and elderly controls. However, PEW was significantly shorter in AMD patients (ETDRS grid centre: 571 ± 26 nm versus 599 ± 17 nm for elderly controls, inner ring: 596 ± 17 nm versus 611 ± 11 nm, outer ring: 602 ± 16 nm versus 614 ± 11 nm). After a mean follow-up time of 50.8 ± 10.8 months, the PEW in the patients decreased significantly by 9 ± 19 nm in the inner ring of the grid. Patients, showing progression to atrophic AMD in the follow up, had significantly (p ≤ 0.018) shorter PEW at baseline than non-progressing patients. Conclusions Peak emission wavelength (PEW) is related to AMD pathology and might be a diagnostic marker in AMD. Possibly, a short PEW can predict progression to retinal and/or pigment epithelium atrophy.

In this paper, a third-order sliding mode control (SMC) with a super-twisting finite time observer is proposed for the position control of a Stewart platform. The Stewart platform is a parallel robotic manipulator with a fixed base platform connected to a movable top platform with the help of six actuator legs. The proposed control makes the movable platform track a reference trajectory in finite time while rejecting matched disturbances. Conventional SMC offers robustness towards matched disturbances, but will induce chattering in the system. The proposed third-order SMC can ease out the chattering problem and achieve finite-time tracking. Furthermore, a robust super-twisting observer is designed, which converges to the actual states in finite time, such that the proposed control can be computed from the system output using the observer. A simulation study of the Stewart platform is done for the proposed robust continuous finite-time control with the finite-time observer. The results using the proposed control are compared with conventional SMC, PID control and twisting control.

Sepsis is a dysregulated host response of severe bloodstream infections, and given its frequency of occurrence and high mortality rate, therapeutic improvements are imperative. A reliable biomimetic strategy for the targeting and separation of bacterial pathogens in bloodstream infections involves the use of the broad-spectrum binding motif of human GP-340, a pattern-recognition receptor of the scavenger receptor cysteine rich (SRCR) superfamily that is expressed on epithelial surfaces but not found in blood. Here we show that these peptides, when conjugated to superparamagnetic iron oxide nanoparticles (SPIONs), can separate various bacterial endotoxins and intact microbes (E. coli, S. aureus, P. aeruginosa and S. marcescens) with high efficiency, especially at low and thus clinically relevant concentrations. This is accompanied by a subsequent strong depletion in cytokine release (TNF, IL-6, IL-1β, Il-10 and IFN-γ), which could have a direct therapeutic impact since escalating immune responses complicates severe bloodstream infections and sepsis courses. SPIONs are coated with aminoalkylsilane and capture peptides are orthogonally ligated to this surface. The particles behave fully cyto- and hemocompatible and do not interfere with host structures. Thus, this approach additionally aims to dramatically reduce diagnostic times for patients with suspected bloodstream infections and accelerate targeted antibiotic therapy. - Statement of significance - Sepsis is often associated with excessive release of cytokines. This aspect and slow diagnostic procedures are the major therapeutic obstacles. The use of magnetic particles conjugated with small peptides derived from the binding motif of a broad-spectrum mucosal pathogen recognition protein GP-340 provides a highly efficient scavenging platform. These peptides are not found in blood and therefore are not subject to inhibitory mechanisms like in other concepts (mannose binding lectine, aptamers, antibodies). In this work, data are shown on the broad bacterial binding spectrum, highly efficient toxin depletion, which directly reduces the release of cytokines. Host cells are not affected and antibiotics not adsorbed. The particle bound microbes can be recultured without restriction and thus be used directly for diagnostics.

The growth of data collection in industrial processes has led to a renewed emphasis on the development of data-driven soft sensors. A key step in building an accurate, reliable soft sensor is feature representation. Deep networks have shown great ability to learn hierarchical data features using unsupervised pretraining and supervised fine-tuning. For typical deep networks like stacked auto-encoder (SAE), the pretraining stage is unsupervised, in which some important information related to quality variables may be discarded. In this article, a new quality-driven regularization (QR) is proposed for deep networks to learn quality-related features from industrial process data. Specifically, a QR-based SAE (QR-SAE) is developed, which changes the loss function to control the weights of the different input variables. By choosing an appropriate inductive bias for the weight matrix, the model provides quality-relevant information for predictive modeling. Finally, the proposed QR-SAE is used to predict the quality of a real industrial hydrocracking process. Comparative experiments show that QR-SAE can extract quality-related features and achieve accurate prediction performance.