A neural network approach is developed to determine the crack opening load from differential displacement signal curves. A backpropagation neural network of three layers was employed. In order to examine the measurement accuracy and precision of the neural network method, computer simulation was extensively performed for various combinations of crack opening levels and signal-to-noise (S/N) ratios. For measuring the crack opening load within a small error of 3% in both the accuracy and the precision, the S/N ratio of differential displacement signal is recommended to be beyond 15 dB.
The proposed method was applied to constant amplitude loading tests on CCT and SEB specimens of aluminum alloy. The effective stress intensity factor range based on measurements by the neural network can describe well fatigue crack growth rates. As the neural network approach does not need any special assumption, the method is expected to give consistent and unbiased results.
In addition, the neural network approach is applied to determine automatically the crack opening load under random loading. The crack opening results obtained are compared with the results of visual measurements by previous researchers.