Research On Quality Mapping Method For Multilevel Testing Of Inertial Navigation Product

Inertial navigation products is the core equipment in the airborne navigation system.at present,there are the three manufacturing stages A/G,IMU and INS in the the inertial navigation production.at every stage of manufacturing product assembly,product debugging and test project is multifarious and low production efficiency and need to be completed manually,which make unstable process and poor on time delivery rate.It need to enhance the level of intelligent production line and strengthen the quality control ability.To solve the above problems in the three-stage manufacturing of inertial navigation products,this paper extracted two business problems that combining with the business characteristics of threestage manufacturing,which are single-level internal quality prediction and three-stage quality mapping.we adopted machine learning and data analysis methods to realize the quality control of inertial navigation products production.For the above research purposes,the main work of this paper includes the following points:(1)Design the method framework of production data extraction and feature construction.the data of INS manufacturing process are combed out by the business process of INS manufacturing,and then the framework of data extraction is designed based on the business objectives.the data distribution and missing situation were statistically analyzed based on the extraction of production data.According to the sample differences,K-means clustering based on cosine distance was used to complete the missing data filling.the process knowledge was integrated into the data feature extraction method,and the complex feature extraction was accomplished by combining the process knowledge PCA method and the sparse autoencoder method.(2)Build DE optimized XGBoost and DNN manufacturing quality prediction models.The tree model can guarantee the physical meaning of features and make the process adjustment operable.The XGBOOST algorithm is used to predict the quality of inertial navigation products,and the differential evolution algorithm is used to complete the optimization of the super parameters in the XGBOOST model.The quality prediction of inertial navigation products was accomplished by using the excellent nonlinear fitting ability of DNN model,and the adaptive optimization of the number of nodes in each hidden layer was accomplished by using the differential evolution algorithm.(3)Construct a three-level quality mapping model of DE-optimized RNN and bidirectional RNN.According to the prediction accuracy requirements,the RNN model is used to mine the one-way mapping relationship,and the differential evolution algorithm is used to optimize the number of hidden layer nodes of the model;the complex two-way mapping characteristics of the three-level indicators are analyzed,and the bidirectional RNN three-level indicator mapping model is constructed,and combined The differential evolution algorithm completes the optimization of the number of hidden layer nodes.Based on the above work content and combined with the production line production data,the model results were analyzed and compared to verify the effectiveness of the method and model.