NEURAL NETWORKS FOR BACKCALCULATION OF MODULI FROM SASW TEST

The spectral-analysis-of-surface-waves (SASW) testing procedure is a seismic technique for the in-situ assessment of elastic moduli and layer thicknesses for pavement and soil systems. Through an inversion or backcalculation procedure, experimental dispersion curves from SASW test results are determined. The complexity of the inversion process has led to the development of neural-network models. This paper describes prototype neural networks that perform the inversion procedure for SASW testing of asphalt concrete pavements. Three-, four-, and five-layer back-propagation models are employed; a general regression neural-network model is studied also. The models accept a description of the dispersion curve as input, and as output, provide ratios detailing the elastic moduli and layer thicknesses of the pavement. Results for all models were reasonably close to actual output, with the best results found with back-propagation neural networks using multiple hidden layers and jump connections.

Language

  • English

Media Info

  • Features: Appendices; Figures; References; Tables;
  • Pagination: p. 1-8
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 00672327
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Jan 16 1995 12:00AM