Gun Barrel Life Prediction Based On Deep Learnin

As a conventional weapon in modern warfare,artillery plays a very important role in the situation of war.As a key structure of artillery,the parameters of the artillery barrel have an important impact on the firing accuracy,service life and operational safety of artillery.The damage to the inner wall of the gun barrel caused by high pressure and high temperature and mechanical impact of gunpowder during use is the main factor affecting the life of the barrel.The damage generally manifests itself in the form of defects and impurities on the inner wall,i.e.,appearance phenomena such as protrusions,depressions and cracks.But the body tube failure mechanism is a complex process,can not be directly to the body tube life as a prediction object,through the main factors affecting the life-the degree of damage to the inner wall for life prediction is scientific and feasible.In this paper,a deep learning-based prediction model is designed to predict the gun tube lifetime.Firstly,the unique advantages of convolutional neural network(CNN)in multivariate data feature extraction and long short-term memory(LSTM)in sequence data analysis are utilized,and the CNN-LSTM(Convolutional Neural Network-Long Short Term Memory)formed by the direct fusion of CNN and LSTM is constructed.In the process of model training,the Nadam(Nesterov-accelerated Adaptive Moment Estimation)optimization algorithm is used to further In the process of model training,Nadam optimization algorithm is used to further optimize the network weight parameters,and Bayesian Optimization(BO),a global optimization algorithm,can make full use of the historical tuning information,reduce unnecessary objective function evaluation,and improve the efficiency of parameter search,so that the hyperparameters of the life prediction model of the gun barrel can be Optimization makes the prediction results more accurate.In this paper,a gun barrel inner wall detection system is built to collect a data set of barrel inner wall topography parameters,which is composed of a control subsystem,a crawling subsystem,a measurement subsystem,and an industrial camera measurement module.Laser displacement sensors and industrial cameras detect the inner wall of the pipe and collect data on the topography of the pipe wall.Through the collection and transmission,storage and reading of point cloud data,the data set is obtained as the initial data for barrel life prediction.Through the rifling separation algorithm,inner wall contour restoration algorithm,abnormal data screening algorithm and least square method proposed in this paper,the initial data is processed to obtain a high-precision barrel inner wall parameter data set for barrel life prediction.The experimental results show that the data acquisition platform based on laser displacement sensors and industrial cameras built in this paper realizes the high-precision and automated data acquisition.The data set is of high quality,provides high-quality input to the prediction model,and ensures the credibility of the results at the input level.Using the Bayesian optimized CNN-LSTM model,the prediction error has been significantly reduced.Compared with CNN-LSTM,LSTM-FC(Long Short Term Memory-Full Connected),GRU-FC(Gated Recurrent Unit-Full Connected)and other neural network prediction models,the prediction accuracy has been improved,which verifies the model used in this article.Feasibility and superiority,the network model finally obtained has higher accuracy and generalization ability for barrel life prediction.