Radiographic Simulator artist: Version 2

Transcription

1 18th World Conference on Nondestructive Testing, April 2012, Durban, South Africa Radiographic Simulator artist: Version 2 Carsten Bellon 1, Andreas Deresch 1, Christian Gollwitzer 1, Gerd-Rüdiger Jaenisch 1 1 BAM Federal Institute for Materials Research and Testing; Berlin, Germany Phone: , Fax: Carsten.Bellon@bam.de, Andreas.Deresch@bam.de, Christian.Gollwitzer@bam.de, Gerd-Rüdiger.Jaenisch@bam.de Abstract Computer simulation of radiography can be used for different purposes in NDT, such as qualification of NDT systems, optimization of radiographic parameters, feasibility analysis, model-based data interpretation, and training of NDT/NDE personnel. BAM has been working on modeling in the field of radiographic testing for many years. With the gathered theoretical background and the familiarity with practical requirements of industrial application the simulation software artist has been developed. This analytical simulator includes a description of the radiation source, the interaction of radiation with test pieces and flaws, and the detection process with special focus on film and digital industrial radiology. It features high processing speed with nearinteractive frame rates and a high level of realism. Here we focus on the recent developments of the simulator, notably the release of artist version 2. Extended functionality regarding automated virtual computed tomography now allows for arbitrary scan paths. Another program extension supports reliability investigations and provides a user interface for planning automatic simulations with varying parameters and defects. Keywords X-ray Imaging, Radiographic Testing, Simulation, Computed Tomography 1 Introduction The artist program is a software tool for computer simulation of film and digital radiography developed by BAM. It generates synthetic radiographs based on a three-dimensional virtual setup with respect to a wide range of radiographic parameters. Over the years the BAM radiographic simulator has been developed from a laboratory application, running on dedicated hardware, to a user-friendly PC program which fulfills practical industrial requirements and can be applied for inspection planning as well [1]. Here an update to [1] is presented with focus on recent developments of the artist program. Within the last year the program has been revised entirely which resulted in a major step forward in terms of usability, stability, and speed. In the following section an update on the program features is given. Finally, program extensions facilitating simulation scenarios like virtual CT and reliability investigations are presented. 2 The artist simulator and its recent update Modeling of radiographic testing procedures requires the consideration of various theoretical and practical aspects. The theoretical aspect focuses on the choice of appropriate physical approximations in order to implement a fast simulation program with sufficient and problem adjusted accuracy. Particularly important for industrial application are suitable interfaces for handling of virtual part representations. The presented simulation program uses an analytical process model, efficient ray tracing algorithms, industry standard boundary representations, and a graphical user interface with 3D visualization to meet these requirements (see Fig. 1).

2 Figure 1: User interface of the radiographic simulator artist. 2.1 Geometrical model The computer simulation models the real inspection scenario by defining a virtual setup. Besides radiation source and detector, one or more geometrical part representations can be arranged freely in virtual space. Parts are modeled by triangulated boundary representations, which separate areas of homogeneous material. An overlap of the parts defined in the scene is resolved according to their priority. This allows for flexible defect descriptions that are independent from the surrounding material. To determine the attenuation of radiation, the penetrated lengths of each material have to be calculated along the linear connections between source points and detector pixels. Making use of the systematic linear spreading of the primary radiation an efficient ray tracer [2] has been developed, allowing for the handling of realistic scenes (huge number of triangles) at practical (short) run times. Within the new implementation the ray tracer was optimized even further for higher efficiency. In addition multi-core support and SIMD instruction sets were enabled to profit from latest processor technology, and a graphics processing unit can be utilized as well. Consequently, the calculation speed is several times higher in version 2 than it was before. This allows for working with scenes carrying several ten millions of triangles while at the same time generating images of several ten millions of pixels in less than a minute, assuming that higher ranked 64bit hardware is utilized. Another extension of the geometrical model is the introduction of curved detectors developed for simulation of pipe inspections. Last but not least the ray tracer has a new multisampling feature for detector pixels, which overcomes aliasing effects of edge projection. This can be combined with the optional focal spot sampling, in order to reduce the amount of additional computing expense. 2.2 Model of X-ray penetration process The radiographic process can be described in its three independent parts, consisting of radiation generation, interaction, and detection. For radiographic simulation the radiation generation model basically needs to describe the source intensity, spectrum, and focal spot

3 size. While for the typical isotopes tables of gamma spectra are provided, artist comes with a quantitative model for X-ray spectra that is configurable with common tube settings. With the consideration of multiple source points at the same time the extended source focus is described. A detailed solution for the radiation-material interaction with its stochastic processes of absorption and scattering (and pair production for higher energies) is difficult and computing time intensive and thus not suitable in this case. The analytic model used here is based on the exponential attenuation law. The influence of scattered photons can be considered by a build- Transmission functions like the characteristic film curves for different types of film classes up factor or through analyticall scattering models for single application cases. are used in order to describe the properties of different detectors. The inner unsharpness is simulated by Gaussian filtering. Noise is added to the synthetic image taking into account its gray value dependency. One innovation of this artist version is a new, more general model for the spectra of X-ray tubes, based on fundamental interaction cross sections and including a parameter-free description of the characteristic radiation [3]. Also, the spectral sensitivity of the detector can now be considered within artist. To improve the consideration of scattering effects, an optional integration of a Monte-Carlo program has been realized. 2.3 User interface The graphical user interface allows for interactive simulation of radiographic scenarios with the artist program. The program window presents a parameter panel and a 3D scene view of the radiographic setup in the central area. Additional modules like the image viewer open in a separate window but can be attached to the main window. The user interface is configurable in many aspects, in order to serve different screen conditions and personal preferences. Figure 2: Direct scene manipulation: A selected part is translated by three mouse clicks; 1. Select transformation mode, 2. Select direction, 3. Drag the part.

4 The user can specify the radiographic setup, including position and orientation of the source and detector, in an interactive fashion utilizing direct manipulation in a 3D view (see Fig. 2). During the interactive manipulation a live preview is generated, which simplifies the construction of the virtual setting. Simple objects can be constructed directly in artist from primitives like cuboids, spheres, and cylinders by using Boolean operations, and complex objects can be read from triangulated CAD exchange files. All user-specified parameters are related in a straight-forward way to the real setup; for example, the material of objects is chosen from a predefined list of frequently used materials and new material definitions can be added by entering the respective chemical composition. Similarly, the source spectrum and brightness can be computed for an X-ray tube by entering the voltage, current, and filtering applied. The detector can be chosen from a predefined list of radiographic films, imaging plates, and digital detectors. With the new user interface of artist version 2 the usability has been improved, and it has become quite easy to learn how to simulate radiographic scenarios with artist. One supporting feature is the above mentioned preview of the synthetic radiograph, providing instant visual feedback of the interactions. The undo/redo functionality helps to correct unintended changes to the setup, or allows for easy switching between settings for comparison. The arranged scene can be saved with all its parameters to a single project file. This file can be opened by artist to continue the same work, even on a different computer. 3 Advanced simulation scenarios Modular extensions have been developed to facilitate specific simulation scenarios. They serve to configure repeated simulation runs and provide additional functionality, e.g. for the processing of simulation results. With additional program windows these modules integrate into the interactive program handling. 3.1 Virtual CT Virtual CT (computed tomography) describes the generation of data sets for 3D reconstruction. Repeated simulations will gather projections of a virtual scene with relative sample movement, e.g. rotation around one of its axes for conventional computed tomography. The appropriately simulated data can be processed by the algorithms developed for reconstruction of real projections. Knowing the exact geometrical definition, virtual CT data can be used to test algorithms for tomographic reconstruction and to investigate influencing parameters. Two independent modules provide virtual CT functionality in the current version of artist, namely CtScan and TomoSynth. The CtScan module has been added to simulate scenarios of conventional CT (see Fig. 3). It is primarily controlled by one parameter: the number of steps of the sample rotation to generate a projection image. The images will be saved as separate TIFF files. A sample reconstruction algorithm for small datasets is also included, as is the functionality to visualize the reconstructed volume in artist s scene view (see Fig. 4). The second virtual CT module can be used to realize arbitrary scan paths and additional parameter variations. The TomoSynth module allows for defining and executing a series of simulation runs with a programmed parameter variation. Besides source and detector position other parameters can be activated for virtual manipulation.

5 Figure 3: CT simulation with CtScan module of artist: N projections are gathered while the sample revolves by one complete turn. A variety of functions is available to control the parameters to vary in a deterministic or random way. For random variations, a probability distribution with its associated parameters is chosen. For deterministic variation, several predefined functions can be selected and adjusted by the corresponding parameters, e.g. to describe linear or circularr variations. Figure 4: Visualisation of the reconstructed volume from virtual CT in its original scene. The red surface displays the original CAD part, the gray volume represents the reconstructed data. For better visual comparability the original part is moved out of its place in this view.

6 Finally, an arbitrary mathematical formula can be entered to define a deterministic distribution as done in the example presented in Figure 5. Here the source position (Sx, Sy, Sz) variation is defined as: Sx formula: 100*cos($step/($Nsteps-1.0)*2*$PI) Sy formula: 100*sin($step/($Nsteps-1.0)*2*$PI) Sz formula: 60*cos( ($step/($nsteps-1.0)*4*$pi), with step number $step, and the total number of steps $Nsteps. This describes a closed circular path with a sinusoidal out of plane distortion (with shape of the number eight in a side view). Reconstruction results for this example have been shown in [2]. Figure 5: Scene view of virtual setup and scan path (dotted curve) for Virtual CT in artist. 3.2 Programmable parameter variations and reliability investigations To facilitate an easy workfloww for the computation of POD (probability of detection) data, artist has been enhanced with the module SimuPOD [4], which adheres to the POD(a) model. At first, the user has to set up a radiographic scene that contains defects. Then, the defects are marked and a variation schedule is defined. This comprises the geometric transformation of the flaw size in the range of interest plus additional random fluctuations, and the desired number of images. This series of simulations is run in batch, and after its completion, an automatic image processing evaluates the visibility of the defects in the resulting images. An interactive analysis module performs the â-vs-a calculation, and the resulting POD(a) curve is displayed and the critical thresholds a 90 and a 90/95 are computed. Any intermediate results namely the images, the â-vs-a data, and the POD(a) curve can be exported in standard formats to enable further processing by other means. Figure 6 shows the interface of the SimuPOD module after finishing a POD simulation trail. Here the defect detectability is investigated with respect to the sample thickness. Additionally a random variation of the defect size was introduced. As all other parameters can be influenced in the same way, this is an example of how to consider real world uncertainties with deterministic modelling.

7 Figure 6: Screens of artist s SimuPOD module. Left: variation schedule controls, Right: graphical and interactively controllable presentation of resulting POD data. 4 Conclusion artist, an easy to use and practical simulation tool generating realistic radiographic images from virtual scenes has been developed. With the latest update, the physical models of source and detector have been improved, yielding quantitative results. A major speed up of the core ray-tracing engine leads to interactive frame rates for typical scenes, allowing a live preview. A complete interface redesign significantly improves usability and workflow. Additionally, modular extensions have been introduced in order to support special simulation scenarios. This includes a batch planning interface for arbitrarily complex parameter variations. Virtual CT and reliability investigations are two applications of the latter functionality. References 1. G-R Jaenisch, C Bellon, U Ewert, artist Analytical RT Inspection Simulation Tool for Industrial Application, Proceedings of the WCNDT2008, 2. C Bellon, C Gollwitzer, D Fratscher, U Ewert, G-R Jaenisch, Simulation of Complex Scan Paths for 3D Reconstruction, Proceedings of the DIR2011, 3. A Deresch, G-R Jaenisch, C Bellon, A Warrikhoff, Simulating X-ray Spectra: From Tube Parameters to Detector Output, accepted at the 18 th WCNDT, Durban, ZA, April C Gollwitzer, C Bellon, A Deresch, U Ewert, G-R Jaenisch, U Zscherpel, Q Mistral, Simulation supported POD for RT test case concept and modelling, accepted at Review of progress in Quantitative Nondestructive Evaluation, July 2011