Abstract

X-ray imaging experiments shed light on internal material structures. The success of an experiment depends on the properly selected experimental conditions, mechanics and the behavior of the sample or process under study. Up to now, there is no autonomous data acquisition scheme which would enable us to conduct a broad range of X-ray imaging experiments driven by image-based feedback. This thesis aims to close this gap by solving problems related to the selection of experimental parameters, fast data processing and automatic feedback to the experiment based on image metrics applied to the processed data.

In order to determine the best initial experimental conditions, we study the X-ray image formation principles and develop a framework for their simulation. It enables us to conduct a broad range of X-ray imaging experiments by taking into account many physical principles of the full light path from the X-ray source to the detector. Moreover, we focus on various sample geometry models and motion, which allows simulations of experiments such as 4D time-resolved tomography.

We further develop an autonomous data acquisition scheme which is able to fine-tune the initial conditions and control the experiment based on fast image analysis. We focus on high-speed experiments which require significant data processing speed, especially when the control is based on compute-intensive algorithms. We employ a highly parallelized framework to implement an efficient 3D reconstruction algorithm whose output is plugged into various image metrics which provide information about the acquired data. Such metrics are connected to a decision-making scheme which controls the data acquisition hardware in a closed loop.

We demonstrate the simulation framework accuracy by comparing virtual and real grating interferometry experiments. We also look into the impact of imaging conditions on the accuracy of the filtered back projection algorithm and how it can guide the optimization of experimental conditions. We also show how simulation together with ground truth can help to choose data processing parameters for motion estimation by a high-speed experiment.

We demonstrate the autonomous data acquisition system on an in-situ tomographic experiment, where it optimizes the camera frame rate based on tomographic reconstruction. We also use our system to conduct a high-throughput tomography experiment, where it scans many similar biological samples, finds the tomographic rotation axis for every sample and reconstructs a full 3D volume on-the-fly for quality assurance. Furthermore, we conduct an in-situ laminography experiment studying crack formation in a material. Our system performs the data acquisition and reconstructs a central slice of the sample to check its alignment and data quality.

Our work enables selection of the optimal initial experimental conditions based on high-fidelity simulations, their fine-tuning during a real experiment and its automatic control based on fast data analysis. Such a data acquisition scheme enables novel high-speed and in-situ experiments which cannot be controlled by a human operator due to high data rates.

First assessor: Prof. Dr.-Ing. R. Dillmann
Second assessor: Prof. Dr. Tilo Baumbach

Abstract

Moore’s law stays the driving force behind higher chip integration density and an ever- increasing number of transistors. However, the adoption of massively parallel hardware architectures widens the gap between the potentially available microprocessor performance and the performance a developer can make use of. is thesis tries to close this gap by solving the problems that arise from the challenges of achieving optimal performance on parallel compute systems, allowing developers and end-users to use this compute performance in a transparent manner and using the compute performance to enable data-driven processes.

A general solution cannot realistically achieve optimal operation which is why we will focus on streamed data processing in this thesis. Data streams lend themselves to describe high-throughput data processing tasks such as audio and video processing. With this specific data stream use case, we can systematically improve the existing designs and optimize the execution from the instruction-level parallelism up to node-level task parallelism. In particular, we want to focus on X-ray imaging applications used at synchrotron light sources. These large-scale facilities provide an X-ray beam that enables scanning samples at much higher spatial and temporal resolution compared to conventional X-ray sources. The increased data rate inevitably requires highly parallel processing systems as well as an optimized data acquisition and control environment.

To solve the problem of high-throughput streamed data processing we developed, modeled and evaluated system architectures to acquire and process data streams on parallel and heterogeneous compute systems. We developed a method to map general task descriptions onto heterogeneous compute systems and execute them with optimizations for local multi-machines and clusters of multi-user compute nodes. We also proposed an source-to-source translation system to simplify the development of task descriptions.

We have shown that it is possible to acquire and compute tomographic reconstructions on a heterogeneous compute system consisting of CPUs and GPUs in soft real-time. The end-user’s only responsibility is to describe the problem correctly. With the proposed system architectures, we paved the way for novel in-situ and in-vivo experiments and a much smarter experiment setup in general. Where existing experiments depend on a static environment and process sequence, we established the possibility to control the experiment setup in a closed feedback loop.

First assessor: Prof. Dr. Achim Streit
Second assessor: Prof. Dr. Marc Weber