Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.697458
Title: Indentation curve prediction and inverse material parameters identification of hyperfoam materials based on intelligent ANN method
Author: Su, X.
ISNI:       0000 0004 5992 9244
Awarding Body: Liverpool John Moores University
Current Institution: Liverpool John Moores University
Date of Award: 2014
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
In this work, an ANN program has been developed to predict the indentation P-h curves with known properties. An interactive parametric FE model and python programming based data extracting program has been developed and used to develop data for the ANN program. Two approaches have been proposed and evaluated to represent the P-h curve. One is using 2nd order polynomial trendline approach, the other is to use the forces at different indentation depth. The performance of the ANN based on the trendline approach is evaluated with MSE and relative error of the curve coefficient and the average error in forces over different depths. A frequency method is used to analyse the data, which provided important data/base to further enhanced the accuracy of the P-h curve based on averaging multiple ANN tests. This approach effectively taking use of the fact that ANN prediction is not continuous around any property point. The ANN program with the depth based approach showed similar accuracy in predicting P-h curves of hyperfoam materials. The program was validated in blind tests with numerical data and experimental data on two EVA foams with known properties. Comparison with other approaches (including surface mapping and direct date space fitting process) showed that the ANN program is accurate and much quicker than some other approaches and direct FE modeling. The feasibility of using ANN to directly predict the material properties is evaluated including assessing its capacity to predict trained data and untrained data. The use of single indenter approach and dual indenter approach is assessed. It was found that the approach with 2nd order polynomial fitting of the P-h curves is not able to predict the material parameters. Using 3rd order fitting showed much improvement and it is able to predict the trained data accurately but could not be used to predict untrained data. Works on dual indenter approach with R4 and R6 showed some improvement in predicting untrained data but could not produce data with reasonable accuracy of the full dataset. A new approach utilising the direct ANN program for P-h curve prediction is developed. A computerised program (with Web based interface) has been developed including data generation through ANN, data storage, interface for input and viewing results. A searching program is developed which enables the identification of any possible materials property sets that produce P-h curves matching the experiment data within a predefined error range. The approach is applied to analysis single and dual indenter methods through blind tests with model materials (with known material properties). A new approach using foams of different thickness is also proposed. The results showed that in a single indenter approach, there are multiple materials property sets that can produce similar P-h curves, thus the results are not unique. Dual indenter size approach showed a significant improvement in mapping out all potential material sets matching the tetsign data. The new program successfully identify addition material property sets that can produce P-h curve that match both R4 and R6 data, which was not identified previously with other inverse programs. The new approach proposed of using the tested data on samples of different thickness showed that the uniqueness of the prediction can be improved. The accuracy and validity of the program is firstly assessed with blind tests (using numerical data as input/target) then used to predict the properties of the EVA foam samples. Some key results of the real foam data is compared to the target and prediction results from other programs and data processing method, the comparison results showed that the new ANN base computer program has clear improvement in accuracy, robustness and efficiency in predicting the parameters of EVA foams. Future work is to transfer the program and methodology developed to other material system and testing conditions and further develop the computer program for material developments and research.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.697458  DOI: Not available
Keywords: TJ Mechanical engineering and machinery
Share: