Storage Life Prediction Methods Research Of Weapons System

Storage life is an important quality characteristic of weapons and equipments. The purpose of it's correct prediction is to ensure weapons and equipments readiness, improve it's availability and reduce it's logistical support costs. Thus, research on prediction method has a vital theoretical and practical significance, and also has an important value in military and economy. The reliability prediction models are summarized in this paper. Prediction methods based on the models, existing puzzles and their solutions are respectively discussed. Following the discussion, new methods are presented.First, according to the actual reliability physical model of system, the failure rate and change law of storage which the regularly detected data contain, a reliability prediction model that is more practical is established. The storage reliability of weapons and equipments can be acquired after a storage period of time based on the reliability prediction model, and then the storage life is also got according to the relationship among failure rate, storage reliability and storage life. By reviewing many possible related literatures and references home and abroad, six practical reliability prediction models are analyzed and compared in some aspects of modeling method of reliability prediction model, data analysis of storage life and it's prediction method based on reliability prediction models. Some examples are respectively applied to verify the feasibility of getting the storage life on the basis of the reliability prediction models and their precisions are also compared. As a result, the reliability prediction models are relatively simple in form and adopted methods, namely parameter estimate methods with higher precision, which demonstrates that the prediction methods of storage life are applicable to engineering. But prediction effect depends on system model, store environment and human factors. Therefore more reasonable is the notion in practice that several methods are used to predict storage life with the comparison of their results for a more appropriate prediction model.Secondly, different methods, i.e. neural network, respectively, gray theory and combination prediction are adopted independently. By the properly improvement and combination of storage information, a corresponding prediction method of storage life is obtained based on neural network prediction model, gray prediction model and combination prediction model. The commonly used methods that are not applicable to complicated, small-sample weapon system with different stress are improved. Examples demonstrate that these methods are valid, and neural network prediction model can predict the storage life of complex and indefinite system by choosing optimal structure design and algorithm with mass data. The prediction method is simple and precise based on gray prediction model. Gray model makes a prediction by means of data processing, so its physical model is unreliable. Further, collected data are distorted to some extent due to the factors, such as human factor, etc. Hence it is very important for getting higher precision that how to restore data. Compared with single model, combination one can make the best of storage information and improve precision sharply.Finally, considering the advantage that gray and combination model have in the way of improving precision and inherent characteristics, these models can be applied to deal with data in accelerated storage test to propose the method of data processing based on an improved accelerated life test. The method is adopted to build gray or combination model, regarding life characteristic as original series under the circumstance of high stress, so that life characteristic of normal stress is predicted. Additionally, the method is aimed at a storage failure mode with failure mechanism and it is difficult to select reasonable accelerated stress and strongly depends on mass sample data in conventional accelerated life tests. The adopted method overcomes these shortcomings and is proven effective by examples. Some suggestions are made, namely accelerated stress should be sampled with uniform interval and the interval between normal stress and accelerated stress is a whole-number multiple of that of accelerated stress, and 1 time is the best.