The Project

We are building a novel Computed Laminography (CL) pipeline to allow for robust and reliable non-destructive testing of important and safety-critical objects, like airplane wings, printed circuit boards or fiber-reinforced lighweight construction structures that cannot be appropriately examined using 3D-CT. A typical industrial 3D-CT system consists of an X-ray tube, a digital flat panel detector and an object manipulation system. Rotating the object by 360 degrees while taking X-ray images results in a set of 2D digital radiographs that are used to obtain a 3D representation of the object using tomography reconstruction techniques.In practice one is however often faced with objects like large rotor blades, printed circuit boards (PCBs), fiber-reinforced lightweight construction structures and other extended planar objects that are too large or too heavy to fit in a 3D-CT scanner or it might be impossible to get radiographs with high resolution from a 360 degrees circle because of geometrical or physical reasons. This makes non-destructive testing of those objects impossible, leaving currently no viable inspection technique for them.To improve on this situation and deliver a practically usable technique, we develop an alternative non-destructive testing system for this kind of objects, based on Computed Laminography. We are addressing the problem comprehensively in all three key stages of the testing pipeline:


1. Data acquisition systems

In CL the object is required to be rotated only in its lateral plane, while still keeping the source and detector stationary, meaning the object can be kept “lying” in its most natural position. This significantly simplifies the scanning system, improves its accuracy and lowers its costs. Unlike stationary 3D-CT systems, the lateral plane-only rotation minimizes the unstable, unwieldy or downright impossible manipulations with the object that would be required for transversal plane manipulations.

Additionally, the high-absorbing directions of scanning are not required, allowing to tune the X-ray energy to better deal with materials like fibre-reinforced plastic or polymer composites often used in light weight construction, which have low X-ray contrast and a detailed internal structure.

Our experimental CL scanner “CLARA” and its schematic

2. Mathematical algorithms

The geometry of Computed Laminography does not allow us to use most common reconstruction techniques like Filtered Backprojection, as these fundamentally require the whole angular range of projections and are known not to cope well with other geometries.Instead, we are using iterative reconstruction methods that have been shown by several groups independently to perform better in non-circular geometry scenario. We are especially developing aspects related to the choice and optimization of basis functions, different variants of a priori information, and hierarchical iteration schemes.

3. High-performance software

The advanced iterative techniques exploiting a priori information however generally lead to significantly increased computational demands and more complex convergence criteria have to be devised to prevent biasing the solution. We develop our tools for first-class support of highly parallel, many-core hardware like GPUs or Intel Xeon Phi processors using modern software parallelization solutions like OpenCL or CUDA.

Example of a basic ray casting algorithm in CUDA:

__global__ void kernel(unsigned int* pixels){

 __shared__ float3 s_rayDir[BLOCK_SIZE];
 __shared__ float3 s_dst[BLOCK_SIZE];

 int sIndex = tIdx.y*blockDim.x + tIdx.x;
 s_rayDir[sIndex] = ComputeRayDirection();

 if( (tOut - tIn) > 0.0f)
  f = c_rayOrigin + tIn*s_rayDir[sIndex];
  old = text3D(datasetTex,f);
  while(tIn < tOut)
   tIn += c_rayStepSize;
   f = c_rayOrigin + tIn*s_rayDir[sIndex];

   f = text2D(preintTexture2D,old,next);
   old = next;