Algorithms and software development.

From WIKIMEDIA TOMOGRAPHY_GROUP
Revision as of 14:08, 23 September 2021 by Admin (talk | contribs) (→‎Machine Learning)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search
Xrayct2.jpg


Objectives:
- Build AI algorithms for image processing applied in tomography
- 3D reconstruction of CiCC cables
- Software developing for investigation of CiCC-tomograms
- Database of CiCC tomograms
- Provide input for multi-physics simulations of CiCC cables (FEM model, void fraction, etc.)
Team:
Daniel Dumitru (daniel.dumitru@inflpr.ro), Mihail Lungu.

Introduction[edit]

The modelling of electrical (i.e. AC loss) and thermal-hydraulic performances relies on strands identification and their trajectory determination. The main algorithm used in the section's strands detection is Hough transformfeature extraction. The detection strategy of each centroid inside the tomographic volume is based on two approaches:

  1. strand-to-strand detection
  2. slice-to-slice detection

CiCC Software version 1.0[edit]

Cicc software1.0 slice detections.jpg

An integrated software was developed based on specific algorithms improved with Regression Machine Learning model to identify the strands and to give a 3D reconstruction of trajectory in the strands. Various information is needed to be extracted from the XCT reconstruction such as: strand detection and positioning coordinates, local and global void fractions in relation to a relevant length of the cable and twist pitch (TP) of individual strands. All these parameters have a substantial impact on CCiC performances, and the mentioned software is applicable to different CiCC configurations.


Steps in using the CiCC Software

1. Convert all the images in grayscale and 8-bit type. If it is higher than 8-bit than the detection process will be slower.

Machine Learning[edit]

Linear Regression Model[edit]

Regression models applied in CiCC cables

Polynomial Regression[edit]

Support Vector Regression[edit]

Convolutional Neural Network[edit]

Morphology Shape Reconstruction[edit]

Finite Element Discretization[edit]

After the 3D reconstruction of the strand's trajectory it is possible to buid a 3D FEM model of the strand using a finite-element mesh generator such as Gmsh.