| The intelligent development of future aircraft requires it to be able to sense the current state online,and make decisions autonomously in service.One of the key technologies to realize this vision is the online prediction technology of the load carrying capacity of the structure.However,there are some problems in the online prediction of the load carrying capacity of composite structures.On the one hand,due to the complicated process and anisotropy characteristics of composite structures,there are a lot of cognitive uncertainties in the properties of composite structures.On the other hand,due to the dynamic evolution of the structure state,it is difficult to quickly identify the damage caused by sudden loads such as impact and overload,and the traditional nondestructive testing methods have poor accessibility such as infrared and ultrasonic monitoring,while various prior health management models are difficult to consider the state and damage evolution of the structure under sudden loading.With the background above,this paper develops an online prediction method for the load carrying capacity of composite structures based on the Bayesian approach,combining an offline constructed database with online dynamic sensor data.Considering that delamination and compression load are widely present in typical composite structures of aircraft such as fuselage skins,the proposed method is illustrated and verified by the online prediction of the load carrying capacity(i.e.critical load for initial damage at delamination boundaries or critical buckling load of structure)of the composite laminate with delamination under compression load.The specific content is divided into four parts:(1)The loading behavior,finite element modeling and key parameter determination of the laminate with delamination under compression load were carried out.Firstly,the compression test of the laminate with delamination was carried out.The local buckling,delamination growth behavior and load-strain curve characteristics caused by initial delamination were analyzed.Subsequently,the finite element model of the laminate with delamination was built based on the cohesive zone theory in the commercial software ABAQUS,the key point is to ensure that the load carrying capacity of the model and the sensor data at the given measuring points of the model are consistent with the test,so as to ensure that the model can use the sensor data of the test to predict the load carrying capacity of the laminate.lastly,considering the anisotropic characteristics and damage behavior of the laminate with delamination,which makes the number of input parameters of the model large,and the distribution characteristics of partial input parameters are nonuniform.However,the online prediction of load carrying capacity needs to control the number of input parameters to reduce the calculation cost,so it is necessary to introduce sensitivity analysis to determine the key input parameters of the model under such multiparameter and non-uniform distribution.The random balanced design method for the Fourier amplitude sensitivity test was improved by introducing the Halton low-difference sequence,so that it had better parameter identification accuracy when dealing the case with multiple and non-uniform distribution of input parameters.Then,the improved method was used to complete the determination of the key parameters affecting the load carrying capacity of the structure.(2)The evaluation of load carrying capacity and identification of the key parameters of the laminate with multi-source cognitive uncertainty was carried out.Firstly,the compression test of the laminate with delamination was completed by using the designed fixture.Subsequently,a finite element model of the laminate with delamination damage was established,and the results of the finite element method were compared with the results of the test.It was found that due to the existence of multi-source cognitive uncertainties such as the discreteness of the processing technology,the non ideality of the loading of the structure,and the limitations of the finite element model,the deviation in the ultimate load carrying capacity between the two was as high as 32%.It is necessary to identify the required parameters of the structure based on the current model,so that the model is consistent with the response of the actual structure.Lastly,considering the large number of parameters to be identified and the time-order characteristics of sensor data,and the dynamic Bayesian network can better deal with the problem of time-series data.Therefore,a multi-parameter identification method based on dynamic Bayesian network was developed.In this method,the parent node of dynamic Bayesian network was sparse by using sensitivity analysis,which reduced the number of single identification parameters,so that the method had better accuracy in multi-parameter identification.Then,multiple key parameters of the structure were identified simultaneously by using this method.(3)An online prediction method of structural load carrying capacity based on offline/online combination was developed.In the offline stage,a database containing a large number of samples(the number of samples in the example was 15730)was established according to the range of variation of the key properties of the structure and possible damage forms to provide support for online fast prediction.In the online stage,based on the acquired sensor information and the prior information of the structural characteristics,the Bayes factor method was used to realize the rapid mapping between the current information and the samples in database,so as to achieve the purpose of online structural load carrying capacity prediction,and the proposed method was verified by numerical examples.Then,the sensor layout was optimized in order to further save the calculation time of online prediction and reduce the additional cost associated with sensor layout.The optimization algorithm adopted genetic algorithm,and the fitness function was composed of the highest Bayes factor value of the identification sample,the identification accuracy,the relative entropy between the distribution of material properties after identification and the initial distribution.The results of the example showed that the online prediction time was reduced from 6.77 s to 2.76 s without reducing the accuracy,which mean that the calculation efficiency could be increased 59%.(4)On the basis of the above research,the verification tests of the online load carrying capacity prediction of the laminate with delamination were carried out.Firstly,the key mechanical properties of the structure were identified by combining the estabilished model and the sensor data obtained from the compression test of the laminate with small load magnitude,and the offline database was established based on the identification results.Then,two compression tests were conducted using the identified laminate in turn,and the strain array information on the surface of the structure was obtained online to infer the load carrying capacity of the structure and compare it with the actual load carrying capacity obtained from the test.The results showed that after obtaining certain observation information,the errors between the predicted value of the ultimate load carrying capacity of the two tests given by the prediction method and the actual load carrying capacity were 4.6% and 10.8% respectively.Moreover,the increase of identification error after sensor optimization was not more than 1%.At the same time,the material properties and the delamination size of the structure could be accurately identified though the sensor data,which fully demonstrated the effectiveness of the proposed method. |
