1. EXPERIMENTAL NOISE INJECTION IN SIMULATED MODEL SIGNALS.
- Author
-
Khan, Tariq, Udpa, Lalita, and Udpa, Satish
- Subjects
ELECTRONIC noise ,SIMULATION methods & models ,SIGNAL processing ,NONDESTRUCTIVE testing ,MATHEMATICAL optimization ,ALGORITHMS ,DISTRIBUTION (Probability theory) ,MAXIMUM likelihood statistics ,EXPECTATION-maximization algorithms - Abstract
Nondestructive testing methods have been routinely, designed, evaluated and optimized using simulation models developed using various computational techniques. The simulated signal using computational model differs from the true signal in that the signal does not simulated experimental noise. In order to use the computational models more effectively for signal processing algorithm development, experimental noise should be injected in the simulated signals. Experimental noise PDF (probability density function) can be numerically calculated from measured noise. The experimental signal PDF can then be generated by combining the simulation signal and measurement noise PDF. Sampling from experimental signal distribution is not a straight forward task as the distribution is generally not a standard parametric distribution. This paper presents a method that approximates experimental signal PDF as a mixture of Gaussian densities. Maximum-likelihood estimate of the parameters of Gaussian distributions from a given data set are computed using expectation-maximization (EM) technique. A sampling scheme from the mixture of Gaussian densities is also discussed. The overall algorithm is implemented on eddy current inspection data from steam generator (SG) tubing. [ABSTRACT FROM AUTHOR]
- Published
- 2011
- Full Text
- View/download PDF