Research On Fault Prediction And Health Management Technology Of Fire Control System Based On Machine Learning

The fire control system is the core unit of armored vehicle operations.It has the characteristics of complex structure and many parameters,and the signal measurement is not easy.When a fire control system fails,it is difficult to determine the location and type of the failure,which greatly affects the combat performance of armored equipment.In the past,the maintenance methods were mostly based on expert experience,and human failure diagnosis was performed through post-failure maintenance.The accuracy of human diagnosis was not high,and the maintenance cost was also very large.Now,the condition-based maintenance method has entered people's vision.With the increasing intelligence of large and complex equipment,the application of prognostics and health management in weaponry is becoming more and more important.In order to reduce the manpower and material consumption of fire control system maintenance and improve the reliability of armored vehicle field operations,this topic builds on the basis of artificial intelligence fault diagnosis and combines machine learning algorithms to construct three intelligent models for fire control system performance parameter trend prediction,multi-fault prediction and health status assessment:(1)For the characteristics of insufficient failure information,nonlinearity signal and high dimensionality of fire control system,the features of performance parameters are first extracted by ensemble empirical mode decomposition,and then the least squares support vector regression machine performs parameter trend prediction.The parameters of the regression machine are selected by the whale optimization algorithm,which improves the prediction accuracy;(2)Establish an improved multi-fault prediction model to make the fire control system's failure prediction not only stay recognition of "normal" or"fault" status,but to distinguish between different types of fault status.The decision directed acyclic graph is used to complete the multi-class conversion of the support vector classifier,and the improved separation measure calculation is introduced to improve the decision directed acyclic graph to reduce the error caused by the improper order of the initial forms.The parameters of the support vector multi-classifier are optimized by the whale optimization algorithm,which improves the accuracy of multi-classification;(3)Establish a model for the health status evaluation of the fire control system.For the characteristics of continuous signals,use the neighborhood rough set for performance attributes.It is reduced to find the important attribute set,which is convenient for the calculation of status assessment.Aiming at the difference between the number of heterogeneous samples and positive samples,the method of support vector data description describes the hypersphere model of positive samples.The parameters that affect the boundary of the hypersphere are determined by the whale optimization algorithm.The distance between the sample and the center of the hypersphere is used to determine the severity of component performance degradation and remaining service life,which is used as a reference for maintenance decision-makers.Taking the power module of the fire control computer and the sensor sub-system of the fire control system as the experimental objects,the measured historical samples verify the above three prognostics and health management models.The experiment proves that the proposed performance parameter trend prediction method has a high accuracy to guarantee the accuracy of the fault prediction;the proposed multi-fault prediction method realizes the possibility of identification of multiple types of fault,making the fault prediction more refined;the proposed health status assessment method has is a breakthrough in judgment of the health degree of the fire control system and has a high reliability.The three methods provide powerful technical support for prognostics and health management of weaponry.