Dipl.-Math. Robert Seidl M.Sc Research Associate   Room: 3157 Email: This email address is being protected from spambots. You need JavaScript enabled to view it. Tel: +49 89 289-23204 Fax: +49 89 289-25051

Research

• Inverse Problems
• Full Waveform Inversion
• Time Reversal and Adjoint Methods
• Simulation in applied mechanics
• High-order Finite Elements

Publications



• Seidl, R.; Rank, E. Show abstract Full waveform inversion for ultrasonic flaw identification
  In: AIP Conference Proceedings 1806, 090013 (2017), , 2017
  
  
  
  Ultrasonic Nondestructive Testing is concerned with detecting flaws inside components without causing physical damage. It is possible to detect flaws using ultrasound measurements but usually no additional details about the flaw like position, dimension or orientation are available. The information about these details is hidden in the recorded experimental signals. The idea of full waveform inversion is to adapt the parameters of an initial simulation model of the undamaged specimen by minimizing the discrepancy between these simulated signals and experimentally measured signals of the flawed specimen. Flaws in the structure are characterized by a change or deterioration in the material properties. Commonly, full waveform inversion is mostly applied in seismology on a larger scale to infer mechanical properties of the earth. We propose to use acoustic full waveform inversion for structural parameters to visualize the interior of the component. The method is adapted to US NDT by combining multiple similar experiments on the test component as the typical small amount of sensors is not sufficient for a successful imaging. It is shown that the combination of simulations and multiple experiments can be used to detect flaws and their position, dimension and orientation in emulated simulation cases.
  
  
  
• Seidl, R.; Rank, E. Show abstract Iterative time reversal based flaw identification
  Computers & Mathematics with Applications 72 (4), pp. 879-892, 2016
  DOI: 10.1016/j.camwa.2016.05.036
  
  
  
  In the field of ultrasonic non-destructive testing, ultrasonic impulses are used to detect flaws in components without causing damage. Based on performing experiments alone, it is possible to infer the state of the component–but this usually provides only limited details about the interior damage such as its position, dimension or orientation. Furthermore, the amount of sensors that can be used to record the signals is restricted to only a few because of the shape and the dimensions of a typical specimen and considerations such as costs, additional mass and structural integrity. Further information about the flaw is hidden in the recorded experimental signals. The idea of the proposed method is to use the experimental data together with a wave speed model of a healthy component and to try to adapt the model to generate these experimental measurements. Formally, the problem is posed as a nonlinear optimization, and the wave speed model is adapted in such a way that the discrepancy between the experimental measurements and the model output is minimized. Moreover, to overcome the problem of only a few available sensor measurements, a combination of multiple simulations corresponding to generally performed multiple experiments in SHM practice is used to improve the accuracy. Following this approach, the position, dimension and orientation of a flaw can be detected for an emulated damaged aluminum plate. For cases when there are only few sensors available, it is shown how a combination of similar experiments can be used to improve the inversion results.
  
  
  

Conference Talks

• Kopp, P.; Korshunova, N.; Seidl, R.; Zander, N.; Kollmannsberger, S.; Rank, E. Finite Cell discretizations for Full Waveform Inversion
  In: European Conference on Computational Mechanics, 2018
  
  
  
  
  
  
  
• Seidl, R.; Rank, E. Full waveform inversion for ultrasonic non-destructive testing
  In: QNDE 2016, Atlanta, USA, 2016
  
  
  
  
  
  
  
• Seidl, R.; Wille, H.; Fichtner, A.; Kollmannsberger, S.; Rank, E. Show abstract Monitoring healing processes in bones using the adjoint method
  In: AEWG-56, Salt Lake City, USA, 2014
  
  
  
  One goal of non-destructive testing procedures is the identification of changes of interior details of a structure over time. More general, the deviation of the present state of a structure from a measured or simulated model state has to be identified. These and similar problems occur in different areas of natural sciences, engineering and technology, where geophysics, biomechanical engineering and Structural Health Monitoring are only a few examples. Over the last decade newly developed numerical methods in geophysics allow to identify the interior of the earth in much more detail than ever before [1]. Their dramatic progress is based on smart combinations of detailed seismic measurements, advanced mathematical approaches and efficient algorithms for high performance computing [2]. We present our attempts to transfer and further develop these methods to problems in biomechanics with the goal to monitor healing processes of fractured bone. To this end we investigate the so-called adjoint method, using a numerical bone model which stems from a CT scan of the damaged bone, an acoustic simulation based on this model and synthetic measurements obtained during the healing process. Because the wave equation is symmetric in time, the propagation of sound is time-reversible [3]. Therefore, the differences of simulated and generated signals at the receiver locations can be time-reversed and played backward into the model. Within the backward simulation sensors become sources and the wave fronts focus on the location of differences in the bone density, being the origin of the difference in the signals. Using the results of this simulation, we construct a sensitivity kernel highlighting the regions of maximal change of material parameters. We illustrate the potential of this approach on numerical experiments with a synthetic model and show that it could open the possibility to reduce cost and radiation exposure by reducing the number of necessary CT scans without compromising the quality of monitoring of the healing process.
  
  
  

Theses

• Seidl, R.
  Preconditioning for Hessian-free Optimization
  Master's thesis, Department of Informatics, Technische Universität München, 2012
  
  
• Seidl, R.
  Analysis and implementation of a step-size restriction for a SQP method with constraint regularization
  Diploma thesis, Department of Mathematics, LMU München, 2008

Supervised Student Projects

• Kopp, P.; Korshunova. N. Crack detection by full waveform inversion using multi-level hp-FEM
  Honours project, Technische Universität München, 2016
  
  
  
  
  
  
  

• Jupyter Notebooks in Teaching: Interactive exercises for HOFEM lecture, Software Lab, 2016
• Kernel based detection of material perturbations, Software Lab, 2015
• Adjoint Method for Localizing Soil Perturbations, Software Lab, 2014