STP1077

    Solving AE Problems by a Neural Network

    Published: Jan 1991


      Format Pages Price  
    PDF Version (304K) 18 $25   ADD TO CART
    Complete Source PDF (7.7M) 18 $68   ADD TO CART


    Abstract

    This paper summarizes recent work utilizing an adaptive learning system to the characterization of acoustic emission phenomena. The processing system resembles a neural network including an associative memory. Data is input into the system as a vector composed of either AE signals or their spectra and encoded information about the source. The mapping of AE signals from the sensors to the descriptors of the source and vice versa is accomplished by learning in the system. This is performed by presenting experimental signals to the system and adaptively forming a memory whose output is an autoregression projection of the input. Discrepancies between the input and output are applied in a delta learning rule. Experiments are described which utilize a system running on a minicomputer to process signals from a localized, simulated source of discrete acoustic emission events in a block of material and to process the AE signals generated during a metal drilling operation. It is shown that the characteristics of the source can be estimated from the AE signals or vice versa by the auto-associative recall from the correlation memory.

    Keywords:

    Neural signal processing, neural networks, quantitative AE, inverse problems, adaptive systems, learning, source characterization, process monitoring


    Author Information:

    Grabec, I
    professor on the Faculty and graduate student of Mechanical Engineeringprofessor, E. K. University of LjubljanaCornell University, LjubljanaIthaca, New York

    Sachse, W
    professor on the Faculty and graduate student of Mechanical Engineeringprofessor, E. K. University of LjubljanaCornell University, LjubljanaIthaca, New York

    Govekar, E
    professor on the Faculty and graduate student of Mechanical Engineeringprofessor, E. K. University of LjubljanaCornell University, LjubljanaIthaca, New York


    Paper ID: STP19092S

    Committee/Subcommittee: E07.04

    DOI: 10.1520/STP19092S


    CrossRef ASTM International is a member of CrossRef.