Deconvolution algorithm (decon 3.0)

Program features | Usage | Download and Installation | About the program | Acknowledgements | Access count | Last update Feb 23, 2010

Program features

• This program is dedicated to the analysis of individual properties of several material phases extracted from the results measured on all phases (overall experimental PDF).
• The program solves the problem of the deconvolution of experimentally observed property histogram (in the form of overall probability density function) into the given number of individual distributions (of Gauss type).
• It computes theoretical probability density function (PDF) that best fits the experimental results.
• The theoretical PDF is decomposed into M individual PDFs (i.e. Gauss distributions).
• At the end, statistical parameters of individual distributions (e.g. mean, st. dev., fraction) are given.

Usage

1. Select Input filenames. It is supposed "The list of all measurements" contains an experimentally measured property in the form of one column txt (ascii) format (see Test data in the Download section).
Further, "Experimental PDF" must be supplied in the two column format [category(i);PDF(i)]. It must correspond to "The list of all measurements" and must be computed before running the program.
It is possible to use "exp2pdf" to do the job.
2. Select Output filename. The output file will contain main results after the deconvolution in txt form.
3. Set maximum number of iterations and precision (limit norm).  The program stops if any of these conditions are reached.
4. Set "Number of expected material phases", M. In doubts, you can run the program several times with different M and see the norm. The lower norm, the better fit.
5. Press "Calculate"  button and see the result of individual statistical descriptors (maximum 6) and the graph.
it is possible to hold the iteration and see running results.
At the end (after reaching maximum number of iterations or the precision), results are written to the output file.

Running the program using the Test data (in.txt, in_pdf.txt with M=3, maxiter=500, prec=1e-6) produces output_decon.txt file and the result shown in Fig. 2.
Note, that the result can be slightly different in the next run since the solution uses random trials and the algorithm can converge to different local minimum.

Therefore, repeating of the computations (running the program several times) or a large number of iterations is recommended to find the best fit.

The theoretical algorithm is described here in [decon.pdf].
 

Fig.1 Screenshot of the program window at the start	Fig. 2 Example of the usage on the Test data

About the program

The program was written in MATLAB 704.
 It can be run under Matlab environment or installed as a standalone application under Windows XP.

Author: Jiří Němeček, Czech Technical University in Prague, Civil Engineering Faculty

Version 3.0 released on Feb 23, 2010

Download and Installation

Installation guide for the standalone application on Windows XP:

1. make the new folder named e.g. "decon"
2. download the zip file into this folder and unzip
3. run "decon.exe" (takes some time to install libraries for the first time)

Uninstallation: Delete the "decon" folder.

Acknowledgements

This program was developed in the frame of research project "GA103/09/1748 - Propojení experimentální nanoindentace s výpočetními nástroji pro homogenizaci nanomechanických vlastností heterogenních materiálů (2009-2011)", i.e. "Integration of experimental nanoindentation with numerical tools for upscaling of nanomechanical properties of heterogeneous materials" .

The software solution was also supported by Jan Mareth.