Combination forecasting method for storage reliability parameters of aerospace relays based on grey-artificial neural networks

Based on a Grey-Artificial Neural Networks (G-ANN) model, this study proposes a combination intelligent forecasting method, which is applied for contact resistances of aerospace relays' prediction in long term storage. The storage reliability of aerospace relays is subject to many nonlinear factors, while the time series forecasting in essence aims to realize a nonlinear mapping. The G-ANN combination forecasting model has been built up. It has the adaptability of neural networks in the nonlinear environment and the trait of Grey theory to weaken the fluctuation of data sequences. There are three phases in the modeling process, including data initialization phase, Grey prediction phase and ANN prediction phase. This approach forecasts the contact resistance degradation of aerospace relays in the accelerated storage test by various Grey system models and G-ANN method. In order to obtain accurate accelerated test data, a testing system of relay storage parameters was designed and developed. The testing system can automatically measure the contact resistances of 40 aerospace relays at the same time when the relays are under different temperature stress. Compared with other forecasting methods, the G-ANN method shows the highest accuracy. In addition, this study also provides the basis and reference for contact life prediction of aerospace relays in accelerated degradation storage test.