Reliability prediction based on degradation measure distribution and wavelet neural network

To avoid the errors caused by pseudo life prediction in degradation testing, this paper proposes a reliability prediction method based on degradation measure distribution and wavelet neural network. The style of degradation measure distribution is assumed to be unchangeable during degradation procedure, while the character parameters, such as location and scale parameters, are time-dependent covariates. Therefore, the evaluations of character parameters are critical factors for prediction results. To predict the character parameters, different wavelet neural network prediction models are established. The learning algorithm of wavelet neural network is Levenberg-Marquardt algorithm combining the advantages of both Gauss-Newton algorithm and fast gradient descent algorithm. Practical degradation data are utilized to verify the proposed method. Considering that data may not be intact in engineering, reliability prediction of partial degradation data is also implemented and the result is acceptable.