| Aerospace equipment refers to integrated systems with specific functions or important components within the system,such as aerospace power supplies,aerospace engines,and aerospace electronic units.Loose particles are objects that created and encapsulated inside during the production process of the aerospace equipment.The loose particle is an important factor affecting the reliable operation of aerospace equipment,and with the increase of equipment volume,the research on the loose particle localization has begun to receive extensive attention.The realization of loose particle lcalization can not only help inspectors to complete the cleaning of loose particles in a timely and efficient manner,but also guide researchers to prevent and control them.At the same time,the research on loose particle localization is of great significance to improve the reliability of the aerospace system,and is a supplement to the loose particle detection research.In this paper,an in-depth study is carried out on the loose particle localization method for aerospace equipment.Firstly,based on the Particle Impact Noise Detection method,with the help of the existing loose particle automatic detection system,the loose particle localization experimental system is established,and the conventional loose particle localization scheme and the transfer-learning-based loose particle localization scheme are determined,and the specific implementation steps are given.In addition,other related works in the research on loose particle localization is described,including the production of aerospace equipment models,the selection of loose particle samples,the setting of loose particle experimental conditions,the selection and layout of acoustic emission sensors,and the introduction of loose particle automatic detection system.Secondly,according to the signal synchronous collection circuit inside the existing loose particle automatic detection system,thus a single CPLD logic controller is used to control the collection and transmission of four-channel loose particle signals,which has the problem of asynchronous serial transmission,and the problem of limited application scenarios signal transmission through wired mode,a multi-channel signal synchronous collection system based on Pulse Per Second is designed.By introducing the Beidou Pulse Per Second to complete the synchronous triggering of the pin level,the synchronous collection of the four-channel loose particle signal is realized.By introducing UTC time to time-calibrate the signal data,the Wi Fi-based remote transmission of the loose particle signal is completed.Thirdly,the two-stage dual-threshold pulse extraction algorithm is used to deal with the continuous multi-pulse and a small amount of noise interference in the loose particle signal.Aiming at the problem that the pulses in the four-channel loose particle signal are not synchronized due to the different layout of the acoustic emission sensor,the zero-fill time difference pulse matching algorithm is designed to eliminate the asynchronous pulse signal in the four-channel loose particle signal,so as to ensure the synchronous matching of the stored signal data on the time scale.Fourth,based on the existing feature library of loose particle detection research,the Mel Frequency Cepstrum Coefficient features are newly extracted,thereby extracting fifty same type features.Combined with the four-channel characteristics of the loose particle signal,a multi-channel weighted threshold feature selection method is proposed.It is used to select and retain thirt-three same type features,that is,one hundred and thirtytwo single localization features,for building the loose particle localization data set.The Random-Forest-based classification learner that used for evaluation has achieved a 0.75%improvement in classification accuracy before and after feature selection.Then,for the problem of missing values and outliers in the loose particle localization data set,the incomplete data processing method based on the measurement of abnormal degree and missing rate is proposed,and the applicable incomplete data processing model is trained,and its rules in the outlier processing part and the missing value processing part are obtained.Aiming at the problem of uneven distribution of feature data in the loose particle localization data set,the optimal z-score normalization method is used to process it.The Random-Forest-based classification learner that used for evaluation has achieved an 8.77% improvement in classification accuracy before and after normalization.Furthermore,in view of the problem that the traditional acoustic emission(AE)source localization method is affected by the internal structure and composition material of the object,and the classification accuracy obtained by the conventional classification learner is limited,the loose particle localization method based on the XGBoost ensemble classifier is proposed,and the loose particle localization model is trained.The Simulated Annealing algorithm is used to complete the parameter optimization,and the optimal loose particle localization model is obtained,and the highest average classification accuracy is achieved as 96.80%.Combined with the requirements of engineering applications,the definition of the loose particle localization accuracy is proposed,and the average localization accuracy of 90.91% is achieved by the localization model.Finally,in order to improve the ubiquity of conventional loose particle localization methods to deal with unfamiliar aerospace equipment,the loose particle transfer localization method based on Instance Transfer Learning is proposed,and the loose particle transfer localization model is trained.The average classification accuracies higher than 82% are achieved by the transfer localization model,and the average localization accuracy of 83.33% is achieved by the transfer localization model as well.By summarizing the existing research results of the loose particle localization,the preliminary idea of loose particle auxiliary localization is put forward. |
PDF Full Text Request