
1041 views  Technical Discussions   Sébastien Petit
 Sébastien Petit
 02:02 Sep201999 Spectres classification, NDT basics decision, etc... Hello, I'd like to know what are the principles and basics of spectral analysis, data analysis in simple NDT testing. I don't mean images recognition and complex 3D reconstruction. Actually, i'd like to analyse resonant spectres of pieces wich should be detected as "good" or "bad". What are the generally useful tools one may apply to numeric data to simply classify the pieces as "good" or "bad" ? May neural networks be applied?, or fuzzy logic ? Just signal treatment, with statistics ? I use now some statistic tools ( means, standard deviation, correlation factor) but they seems not to be sufficiant, i.e. i suppose i should extract more significant parameteres from my spectres... Let me know what signal treatment you use for numerical data as spectres, please ! Thanks in advance Sebastien Petit IEMN CNRS (France) sebastien.petit@iemn.univlille1.fr
 
   Robert (Rocky) A. Day Engineering Milky Way Jewels, USA, Joined Nov 1998, ^{40}   00:14 Sep201999 Re: Spectres classification, NDT basics decision, etc... To over simplify, spectral analysis is based on analyzing signals in the frequency domain rather than in the time domain. There are several ways in which this can be done. The traditional method is the use of a spectrum analyzer that produces the power spectral density plot in real number space. Spectrums are complex so the use of digital spectral methods has increased in recent decades. The common methods in use are fast fourier transform and wavelet transforms. There are in fact several variations on these and some other less popular but useful transforms as well.A good basic reference is Oppenheimer & Shafer 'Digital Signal Processing' (I'm not positive on the title). This covers the fourier methods very well and discusses the Welch power spectral density algorithm which is very popular. It also discusses Cepstrum and some others that are often useful for vibration and acoustics problems. G. Strang's 'Wavelet Processing' is also very valuable. Sorry I am moving and the books are in one office while the computer is in the new one. So I'm going from memory on the book titles. It is indeed possible to do neural nets with spectral information and this is an active area of research. A literature search should provide a lot of cites. I do not recall a good review but I'm sure one is out there. Regards, Robert (Rocky) A. Day Second Sound Ultrasonic Systems 220 Gates Street San Francisco, CA 94110 (415) 6470625 Fax: (415) 6414947
: Hello, : I'd like to know what are the principles and basics of spectral analysis, data analysis in simple NDT testing. : I don't mean images recognition and complex 3D reconstruction. : Actually, i'd like to analyse resonant spectres of pieces wich should be detected as "good" or "bad". : What are the generally useful tools one may apply to numeric data to simply classify the pieces as "good" or "bad" ? : May neural networks be applied?, or fuzzy logic ? Just signal treatment, with statistics ? : I use now somestatistic tools ( means, standard deviation, correlation factor) but they seems not to be sufficiant, i.e. i suppose i should extract more significant parameteres from my spectres... : Let me know what signal treatment you use for numerical data as spectres, please ! : Thanks in advance : Sebastien Petit : IEMN CNRS (France) : sebastien.petit@iemn.univlille1.fr
 
   Godfrey Hands Consultant, PRI Nadcap, United Kingdom, Joined Nov 1998, ^{307}   07:57 Oct021999 Re: Spectres classification, NDT basics decision, etc... : Hello, : I'd like to know what are the principles and basics of spectral analysis, data analysis in simple NDT testing. : I don't mean images recognition and complex 3D reconstruction. : Actually, i'd like to analyse resonant spectres of pieces wich should be detected as "good" or "bad". : What are the generally useful tools one may apply to numeric data to simply classify the pieces as "good" or "bad" ? : May neural networks be applied?, or fuzzy logic ? Just signal treatment, with statistics ? : I use now some statistic tools ( means, standard deviation, correlation factor) but they seems not to be sufficiant, i.e. i suppose i should extract more significant parameteres from my spectres... : Let me know what signal treatment you use for numerical data as spectres, please ! : Thanks in advance : Sebastien Petit : IEMN CNRS (France) : sebastien.petit@iemn.univlille1.fr Sebastien, I have tried to contact you about making a demonstration with our resonant inspection equipment in Lille for you, but have been unable to get a message to your address. Please contact me. I can demostrate how we detect the changes in resonances associated with defects. Regards, Godfrey Hands Phone +44 2476 320812 Fax +44 2476 320813
 
  
