Kinematic Reliability And Fault Diagnosis Of Automatic Ammunition Loading Subsystems

Automatic ammunition loading system is a complex system with stringent demand of performance but poor working environment.The fault rate is high and faults are hard to detect,isolate and repair without fault diagnosis system.Reliability and maintainability have become the performance bottle-neck.This paper takes the ammunition loading systems as research projects,for action reliability assessment,reliability-based optimization and fault diagnosis.Using some typical auto-loading subsystems as examples,the simulation models are built by RecurDyn and Simulink.The uncertainty parameters are created according to the known fault symptoms.Portable data acquisition devices and multi sensors are used to measure the important action signals in the whole auto loading procedure of a standard-state howitzer.The models are modified using the measured data to improve the simulation accuracy.Uncertainty model is used for sampling and simulation.RBF neural network agent model,with uncertainty parameters as inputs,and positioning errors as outputs,is built and Monte Carlo simulation is used to calculate probability densities of positioning errors and estimate the action reliability.To avoid repeated action reliability calculations in the optimization process and reduce the amount of computation,an optimization design method indirectly improve the reliability is proposed.The weighted sum of positioning errors as a performance index is optimized using particle swarm algorithm.The whole errors are reduced and the relatively large ones stay far from the mean state.In this paper,action fault diagnosis is a process of multi abstracts,abstract from real equipment to simulation model,abstract from simulation model to response curves,abstract from response curves to features,and abstract from feature parameters to fault information.Considering the smoothness of time series signals,functional data analysis is used to make the sample response data functional,and features are abstracted using principal components analysis for functional data.BP neural network is built and trained to be the basic fault diagnosis machine.To reduce the potential diagnosis inaccuracy and avoid repeat of wrong diagnosis,a neural network learning method using samples with different confidence levels is proposed as the self-correction diagnosis algorithm.This algorithm can use the maintenance result of real mechanism after the diagnosis system is equiped,the diagnosis accuracy will become higher and higher by taking profit from the feedback samples.The diagnosis system is developed including data acquisition devices and diagnosis software.