A prediction method of life and reliability for CSALT using Grey RBF neural networks

There are two problems of the traditional life and reliability estimation methods of Accelerated Life Test (ALT): one is the difficulty to establish the accelerated model and another is the complex computing of multiple likelihood equations. In this paper, we proposed a new prediction method of life and reliability for the constant stress accelerated life test using Grey RBF Neural Network. The accelerated stress levels and reliability are used as the training input vectors, while well-regulated failure data operated by Grey Accumulated Generate Operation (AGO) principle as training target vectors. Then RBF neural net is established and trained. Eventually, the failure data under normal stress can be predicted by putting the normal stress levels and the reliability into the model, and reliability curves can be drawn if life distribution is known. A simulation case is conducted and results are compared to that of BP algorithm, which demonstrates the validation of this model.