Access to master’s theses is restricted to current ERAU students, faculty, and staff.

Date of Award

Fall 2010

Document Type

Thesis - Open Access

Degree Name

Master of Aerospace Engineering


Graduate Studies

Committee Chair

Dr. Eric v. K. Hill

Committee Member

Dr. Ibrahima K. Kaba

Committee Member

Dr. Yi Zhao


The purpose of this research was to predict burst pressures in composite overwrapped pressure vessels (COPV) using mathematically modeled acoustic emission (AE) data. The AE data were collected during hydrostatic burst (hydroburst) testing often 15-inch (380 mm) diameter COP Vs. The data gathered during these tests were filtered in order to remove long duration hits, multiple hit data, and obvious outliers. Based on the duration, energy, and amplitude of the AE hits, the filtered data were classified into the various failure mechanisms of composites using a MATLAB based self-organizing map (SOM) neural network. Previous research has demonstrated that the parameters from mathematically modeled AE failure mechanism data can be used to predict burst pressures from low proof load data. Thus, amplitude histograms from classified matrix cracking data were mathematically modeled herein using bounded Johnson distributions.

A backpropagation neural network (BPNN) was then generated using MATLAB. This BPNN was able to predict the burst pressures of the COPVs using (1) filtered AE data, (2) filtered and classified (matrix cracking only) AE data, and (3) mathematically modeled, filtered and classified (matrix cracking only) AE data. Using five bottles for network training and four for testing, the worst case BPNN errors obtained for these three data sets were -9.039 percent, -6.874 percent, and 1.997 percent, respectively. Multiple linear regression (MLR) analysis subsequently required all nine bottles to obtain a -1.666 percent worst case error. From these results, it was concluded that mathematically modeled AE data can be analyzed with either a BPNN or MLR to provide extremely accurate burst pressure predictions for COPVs. However, where a BPNN is capable of processing both noisy and sparse data sets with high accuracy, MLR requires both noise free data and significantly more of it to obtain comparable results.