Fault Element Diagnosis Considering Array Errors Based On Multitask Neural Networks

Array antenna has been widely used in actual production and life.Compared with single antenna,it has better performance.Array antenna has many advantages,it can be used to improve the gain of the antenna,also can be used to resist the specific direction of interference.Generally speaking,the more antennas you have,the better it performs.But in practice,some elements may fail,which will affect the performance of the whole antenna array.There is no way to avoid such failure.These failure elements will have a great impact on the internal structure of the array,which will make the antenna array unable to be used normally.It is necessary to diagnose the array.There may also be a variety of array mismatches when there are failed elements.In recent years,deep learning and neural network have been widely used in practical production and life.In this thesis,multi-task neural network is used to diagnose array faults with array mismatch.There are two main aspects.1.Use multi-task neural network to diagnose uniform linear array and planar array with frequency errors,and use radiation compensation method to improve the accuracy of neural network prediction.The interval analysis(IA)method is proposed to train the auxiliary network to reduce the influence of frequency error on the neural network.The difference of the array far-field pattern with frequency error is compared.The simulation results show that the Location Refinement based on Radiation Compensation(LRRC)neural network with radiation compensation has a good effect on the diagnosis of the array with frequency error compared with the traditional machine learning algorithm.It also has a good diagnostic effect for different types of arrays.And the method is still robust in the case of increasing frequency error.When Gaussian noise is added with frequency error,the simulation results show that the algorithm has good robustness to noise.2.The far-field radiation physical model of uniform linear array and planar array with array position error is proposed,and the influence of directional graph with array position error is given.The IA method is used to obtain the analysis of the position error of the array,so as to further assist the training of the neural network.Finally,the multitask neural network with radiation compensation is used to diagnose the array with array position error,and the traditional machine learning algorithm is compared to prove the feasibility of the method.The far-field radiation physical models of uniform array and planar array with both frequency error and array position error are proposed.The effects of the two errors on the array pattern are analyzed.The performance of the algorithms is compared when there are both frequency error and array position error.3.In the case of frequency offset error,the power information of the array far-field is used to carry out array diagnosis,and the frequency offset error is optimized at the same time.The multi-label neural network is used to complete the diagnosis of the failure array,which verifies the diagnostic performance of the algorithm and the robustness of the added noise.