Evaluation and prediction methods for launch safety of propellant charge based on support vector regression

Abstract In view of the consensus that the oversized initial combustion surface area of fractured propellant charge is the main cause of breech blow, the evaluation method for launch safety of propellant charge with initial dynamic vivacity ratio as the characterization parameter has drawn extensive attentions and inspired many variants. However, the characterization of propellant charge stacking configuration and the prediction of initial dynamic vivacity ratio are still two unsolved problems. To fill the gaps, this study provides a novel evaluation method for the launch safety of propellant charge through applying the machine learning and the error analysis methods. A new description with three parameters is proposed for dynamic compression and fracture process, and verified more accurate than the previous method with maximum compression stress as the single parameter. In addition, it identifies that the support vector regression is more suitable than back propagation neural network and least squares support vector machine in small sample training. And the corresponding model has been demonstrated by experiments to be of convinced accuracy and superior generalization capability. Through determining the model error distribution, this study makes it feasible to predict initial dynamic vivacity ratio and give an upper limit value with high confidence.