Research On Storage Reliability Modeling And Life Prediction Method Of Electromagnetic Relay For Missile Based On Accelerated Degradation Test

Electromagnetic relays for missiles possess functions such as control,safety protection,and signal transmission.Their reliability directly impacts the overall reliability of the weapon system,and their storage life greatly exceeds their operating life.Therefore,assessing the reliability of electromagnetic relay storage and predicting its storage life is of great practical significance.To address the issue of high reliability and limited failure data of electromagnetic relays for missiles,this thesis uses accelerated degradation testing to study its degradation parameters during storage degradation,analyze its failure mechanism and failure mode,and establish multiple storage life prediction models using the accelerated degradation data.The main environmental factor affecting degradation is determined to be temperature stress.The thesis designs an experimental plan for the accelerated degradation test system from aspects such as the type of the test,the type of stress,the level of stress,and the number of samples.To deal with noisy degradation test data,the thesis preprocesses the data using wavelet denoising method and determines a suitable method based on its basic functions,threshold function,and decomposition level.Contact resistance is identified as the sensitive parameter for relay storage failure based on graphical analysis of degradation data trends.To address the complexity of degradation data processing,the thesis establishes a missile electromagnetic relay storage life prediction model based on Extreme Learning Machine(ELM)due to its fast operation speed.Multiverse Optimization(MVO)and Sine Cosine Algorithm(SCA)are introduced to optimize ELM’s weight and threshold values.The MVOELM and SCA-ELM models are used to train and predict the degradation data,and the mean absolute error(MAE)and root mean square error(RMSE)of the predicted values are used as evaluation metrics.The results show that the MVO-ELM model has the lowest error and higher accuracy but slow convergence.To further improve the speed of convergence,Least Squares Support Vector Machine(LSSVM)is introduced to establish the prediction model.LSSVM converts inequality constraints into equality constraints based on SVM,reducing computational complexity.The performance of the LSSVM model is determined by regularization parameters and kernel function width.To improve the accuracy of the model,Harris’ s hawk algorithm and Ant Lion Optimizer algorithm are introduced to optimize the parameters.The corresponding prediction models are established.The results show that the prediction accuracy of LSSVM and its improved models is slightly lower than that of ELM and its improved models,but its convergence speed is faster,with only 10%-20% convergence iterations of the ELM improved model.In practical applications,different prediction models can be selected according to different requirements.ELM and its improved models are suitable for high-precision prediction scenarios,while LSSVM and its improved models are suitable for preliminary prediction of storage life or when fast convergence is required.