Abstract

Proposal for classification of wood members as bearing elements in wooden constructions using neural network is presented here. Experiment is based on testing 100 samples of Slavonian Oak. Using nondestructive methods (transient excitation by hammer impact, transient excitation putting the element out of state of equilibrium and ultrasound excitation) dynamic modulus of elasticity (MOE) is obtained from the dynamic parameters.

Equipment for measuring, registration and analysis were spectrum analyzer HP 3560 A and impulse hammer ICIP, type 086C03. Moisture was continuously measured by electronic device for measuring wood moisture Viva 12. After nondestructive testing, on the same samples, classical destructive probes are made to get all strength values (MOR, UTS,o, UTS,90, UCS,0, UCS,90, SHEAR). Also, the samples are modeled by FEM program package COSMOSM to get theoretical prove for the results obtained in laboratory.

Dynamic parameters (resonant frequency, decrement of vibrational decay and propagation speed of ultrasound wave), moisture, density, boundary conditions and geometry of the samples were input variables while strength values were output variables for teaching neural network. As a result RUNTIME option of neural network is obtained as main part of classification system which contains n-dimensional correlation between all input and output variables. It means that just with very simple hammer percussion on wood sample or by acoustic emission, using such neural network, we can get all strength values and MOE without any laboratory destructive testing.

Keywords: Ultrasonic Testing

Source: Conference Proceedings of the 4th International Conference of the Slovenian Society for Nondestructive Testing, 24 - 25 April 1997, Ljubljana, Slovenia.
Pages:59 - 66

Publisher:
Slovenian Society for Nondestructive Testing, Faculty of Mechanical Engineering, Askerceva 6, SLO 1000 Ljubljana, Slovenia. Tel +386 61 177 12 03, Fax +386 61 218 567, Email: janez.grum@fs.uni-lj.si.