skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Neural network burst pressure prediction in impact damaged Kevlar/epoxy bottles from acoustic emission amplitude data

Abstract

Acoustic emission (AE) signal analysis has been used to measure the effect of impact damage on the burst pressure of 5.75 inch diameter filament wound Kevlar/epoxy pressure vessels. A calibrated dead weight drop fixture, featuring both sharp and blunt hemispherical impact tups, generated impact damages with energies up to twenty ft-lb{sub f} in the mid hoop region of each vessel. Burst pressures were obtained by hydrostatically testing twenty-seven damaged and undamaged bottles, eleven of which were filled with inert propellant to simulate a rocket motor. Burst pressure prediction models were developed by correlating the differential AE amplitude distributions, Generated during the first pressure ramp to 25% of the expected burst pressure for the undamaged vessels, to known burst pressures using back propagation neural networks. Independent networks were created for the inert propellant filled vessels and the unfilled vessels using a small subset of each during the training phases. The remaining bottles served as the test sets. The eleven filled vessels had an average prediction error of 5.6%, while the unfilled bottles averaged 5.4%. Both of these results were within the 95% prediction interval, but a portion of the vessel burst pressure errors were greater than the {+-}5% worst case errormore » obtained in previous work. in conclusion, the AE amplitude distribution data collected at low proof loads provided a suitable input for neural network burst pressure prediction in damaged and undamaged Kevlar/epoxy bottles. This included pressure vessels both with and without propellant backing. Work is ongoing to decrease the magnitude of the prediction error through network restructuring.« less

Authors:
;  [1];  [2]
  1. Univ. of Alabama, Huntsville, AL (United States)
  2. National Aeronautics and Space Administration, AL (United States) [and others
Publication Date:
OSTI Identifier:
482048
Report Number(s):
CONF-960503-
TRN: 97:002904-0080
Resource Type:
Conference
Resource Relation:
Conference: 1. international conference on nonlinear problems in aviation and aerospace, Daytona Beach, FL (United States), 9-11 May 1996; Other Information: PBD: 1994; Related Information: Is Part Of First international conference on nonlinear problems in aviation & aerospace; Sivasundaram, S. [ed.]; PB: 729 p.
Country of Publication:
United States
Language:
English
Subject:
42 ENGINEERING NOT INCLUDED IN OTHER CATEGORIES; 99 MATHEMATICS, COMPUTERS, INFORMATION SCIENCE, MANAGEMENT, LAW, MISCELLANEOUS; EPOXIDES; FAILURE MODE ANALYSIS; IMPACT TESTS; NEURAL NETWORKS; PRESSURE VESSELS; ARAMIDS; PRESSURE DEPENDENCE; ACOUSTIC EMISSION TESTING

Citation Formats

Walker, J.L., Workman, G.L., and Russell, S.S. Neural network burst pressure prediction in impact damaged Kevlar/epoxy bottles from acoustic emission amplitude data. United States: N. p., 1994. Web.
Walker, J.L., Workman, G.L., & Russell, S.S. Neural network burst pressure prediction in impact damaged Kevlar/epoxy bottles from acoustic emission amplitude data. United States.
Walker, J.L., Workman, G.L., and Russell, S.S. Sat . "Neural network burst pressure prediction in impact damaged Kevlar/epoxy bottles from acoustic emission amplitude data". United States. doi:.
@article{osti_482048,
title = {Neural network burst pressure prediction in impact damaged Kevlar/epoxy bottles from acoustic emission amplitude data},
author = {Walker, J.L. and Workman, G.L. and Russell, S.S.},
abstractNote = {Acoustic emission (AE) signal analysis has been used to measure the effect of impact damage on the burst pressure of 5.75 inch diameter filament wound Kevlar/epoxy pressure vessels. A calibrated dead weight drop fixture, featuring both sharp and blunt hemispherical impact tups, generated impact damages with energies up to twenty ft-lb{sub f} in the mid hoop region of each vessel. Burst pressures were obtained by hydrostatically testing twenty-seven damaged and undamaged bottles, eleven of which were filled with inert propellant to simulate a rocket motor. Burst pressure prediction models were developed by correlating the differential AE amplitude distributions, Generated during the first pressure ramp to 25% of the expected burst pressure for the undamaged vessels, to known burst pressures using back propagation neural networks. Independent networks were created for the inert propellant filled vessels and the unfilled vessels using a small subset of each during the training phases. The remaining bottles served as the test sets. The eleven filled vessels had an average prediction error of 5.6%, while the unfilled bottles averaged 5.4%. Both of these results were within the 95% prediction interval, but a portion of the vessel burst pressure errors were greater than the {+-}5% worst case error obtained in previous work. in conclusion, the AE amplitude distribution data collected at low proof loads provided a suitable input for neural network burst pressure prediction in damaged and undamaged Kevlar/epoxy bottles. This included pressure vessels both with and without propellant backing. Work is ongoing to decrease the magnitude of the prediction error through network restructuring.},
doi = {},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Sat Dec 31 00:00:00 EST 1994},
month = {Sat Dec 31 00:00:00 EST 1994}
}

Conference:
Other availability
Please see Document Availability for additional information on obtaining the full-text document. Library patrons may search WorldCat to identify libraries that hold this conference proceeding.

Save / Share:
  • Acoustic emission signal analysis has been used to measure the effect impact damage has on the burst pressure of 146 mm (5.75 in.) diameter graphite/epoxy and the organic polymer, Kevlar/epoxy filament wound pressure vessels. Burst pressure prediction models were developed by correlating the differential acoustic emission amplitude distribution collected during low level hydroproof tests to known burst pressures using backpropagation artificial neural networks. Impact damage conditions ranging from barely visible to obvious fiber breakage, matrix cracking, and delamination were included in this work. A simulated (inert) propellant was also cast into a series of the vessels from each material class,more » before impact loading, to provide boundary conditions during impact that would simulate those found on solid rocket motors. The results of this research effort demonstrate that a quantitative assessment of the effects that impact damage has on burst pressure can be made for both organic polymer/epoxy and graphite/epoxy pressure vessels. Here, an artificial neural network analysis of the acoustic emission parametric data recorded during low pressure hydroproof testing is used to relate burst pressure to the vessel`s acoustic signature. Burst pressure predictions within 6.0% of the actual failure pressure are demonstrated for a series of vessels.« less
  • A burst pressure prediction model was generated from the acoustic emission amplitude distribution data taken during hydroproof for three sets of ASTM standard 145 mm (5.75 in.) diameter filament wound graphite/epoxy bottles. The three sets of bottles featured the same design parameters and were wound from the same graphite fiber, the only difference being in the epoxies used. Therefore, the three resin types were categorized using dummy variables, which allowed the prediction of burst pressures in all three sets of bottles using a single back-propagation neural network. Three bottles from each set were used to train the network. The resinmore » category and the acoustic emission amplitude distribution data taken up to 25 percent of the expect burst pressure were used as network inputs. The actual burst pressures were supplied as target values for the supervised training phase. Architecturally, the network consisted of a 48 neuron input layer (a categorical variable defining the resin type, plus 47 integer variables for the acoustic emission amplitude distribution frequencies), a 15 neural hidden layer for mapping, and a single output neuron for burst pressure prediction. The network, trained on three bottles from each resin type, was able to predict burst pressures in the remaining bottles with a worst case error of {minus}3.89 percent, well within the desired goal of {+-}5 percent.« less
  • A burst pressure prediction model was generated from the acoustic emission (AE) amplitude distribution data taken during hydroproof testing for a set of eleven ASTM standard 5.75 inch diameter filament wound, fiberglass/epoxy bottles. The bottles were tested at three different temperatures - 32{degrees}F, 70{degrees}F, and 110{degrees}F - which were input as categorical variables, allowing the prediction of burst pressures in all three sets of bottles using a single backpropagation neural network. Two of the bottles contained simulated manufacturing defects which lowered their burst pressures. Architecturally, the neural network consisted of 42 input neurons (one categorical variable for temperature plus forty-onemore » amplitude frequencies), a 15 neuron hidden layer for mapping, and a single output neuron for the predicted burst pressure. Seven of the eleven bottles were used to train the network. The AE amplitude distribution data taken up to 25% of the expected burst pressure were used as network inputs, and the actual burst pressures were used as target values for the supervised training phase. The trained network was not able to predict burst pressures accurately on the two defective bottles with such a small test group; they were therefore not considered. The network then used six of the nine bottles for training and blind predicted on the remaining three bottles. The network was then able to predict burst pressures with a worst case error within the desired goal of {+-} 5%.« less
  • Acoustic emission (AE) flaw growth activity was monitored in aluminum-lithium weld specimens from the onset of tensile loading to failure. Data on actual ultimate strengths together with AE data from the beginning of loading up to 25 percent of the expected ultimate strength were used to train a backpropagation neural network to predict ultimate strengths. Architecturally, the fully interconnected network consisted of an input layer for the AE amplitude data, a hidden layer to accommodate failure mechanism mapping, and an output layer for ultimate strength prediction. The trained network was then applied to the prediction of ultimate strengths in themore » remaining six specimens. The worst case prediction error was found to be +2.6 percent.« less
  • Research is continuing on the applications of acoustic emission testing for predicting burst pressure of filament-wound Kevlar 49/epoxy pressure vessels. This study has focused on three specific areas. The first area involves development of an experimental technique and the proper instrumentation to measure the energy given off by the acoustic emission transducer per acoustic emission burst. The second area concerns the design of a test fixture in which to mount the composite vessel so that the acoustic emission transducers are held against the outer surface of the composite. Included in this study area is the calibration of the entire testmore » setup including couplant, transducer, electronics, and the instrument measuring the energy per burst. In the third and final area of this study, the number, location, and sensitivity of the acoustic emission transducers used for proof testing composite pressure vessels are considered.« less