Ultimate Load Prediction in Fiberglass/Epoxy Beams from Acoustic Emission Data using Neural Network and Statistical Analyses

作者：Hill Eric v K^*; Dorfman Michele D; Zhao Yi

来源：Materials Evaluation, 2013, 71(8): 977-986.

• acoustic emission
• neural network
• multiple linear regression
• ultimate load prediction

The research presented herein demonstrates the feasibility of predicting ultimate loads in composite beams subjected to three-point bending using a neural network analysis of acoustic emission testing amplitude distribution data. Fifteen unidirectional fiberglass/epoxy beams were loaded to failure in a hydraulically-actuated tension-compression load frame using a three-point bend test fixture. Acoustic emission amplitude distribution data taken during loading up to 80% of the average ultimate load were used as inputs for a backpropagation neural network (BPNN). The network was trained on seven beams, and tested on the remaining eight. The worst-case prediction error for the BPNN was 4.34%. A second analysis was then performed using a Kohonen self-organizing map (SOM) neural network and multiple linear regression (MLR) analysis. Here, the Kohonen SOM was utilized to classify the acoustic emission data into failure mechanisms. Then MLR analysis was performed using the number of acoustic emission hits associated with each failure mechanism to develop a prediction equation. The worst-case prediction error for the statistical analysis was approximately the same as that of the BPNN at 4.13%, suggesting that the BPNN and MLR analyses are equivalent techniques for accurately predicting ultimate loads in beams. However, it should be noted that because the BPNN is an iterative optimization algorithm, it required the data from only seven beams to train on, whereas the MLR result required all 15 beams to achieve comparable results. Finally, BPNNs can handle noisy data, whereas outliers typically invalidate MLR results.

• 出版日期2013-8