Study Of Fault Diagnosis Algorithm For Inertial Measurement Unit And Integrated Navigation System

Inertial navigation system(INS)is strong at autonomy,concealment and immunity,and has been widely used in both civil and military fields.Some other navigation systems are often combined assistly with INS to suppress the drift and improve the availability of navigation system.In order to ensure the accuracy and reliability of inertial-based integrated navigation system,fault diagnosis method and fault tolerant structure which can detect,separate and deal with the fault in time are necessary for the navigation system.this paper studies the fault diagnosis algorithms and fault tolerant strategies for redundant inertial measurement unit and auxiliary navigation system respectively,improving the navigation accuracy and fault tolerance of the inertial-based integrated navigation system under the condition of sensor fault.The main contents of this paper are as follows.The traditional parity space-based method cannot isolate faulty sensor in a quadruplet of the inertial measurement unit(IMU).Aiming at this problem,a fault diagnosis method assisted by support vector machine(SVM)is proposed.The generalized likelihood ratio method is employed to do a real-time fault detection for the quadruplet.The outputs of fault classifiers which combine the relative wavelet packet energy of the sensor output with the linear support vector machine(LSVM)are auxiliaries to diagnose the fault components.And a fault diagnosis identification is set up to realize a continuous monitoring of the quadruplet.Simulation results show that the proposed algorithm can quickly and accurately identify the faulty sensor in both stationary process hard and short-time dynamic process,and can also realize the soft fault detection in a long-time dynamic process.Working at the problem that there is an effective time limitation of signal feature-assisted fault diagnosis algorithm on hard fault diagnosis,a fault isolation method based on slope mutation detection is proposed for the quadruplet of IMU.The data simple size is set according to the statistics of fault detection delay.The slope of inertial sensor output is fitted by the least square method.And the fault isolation threshold is constructed by the variance of sensor noise.Simulation results demonstrate that the proposed algorithm has lower false alarm rate and is more sensitive to hard fault than other algorithms,which further improves the reliability of the quadruplet in the long-time dynamic process.Although lots of fault diagnosis algorithms for the redundant inertial measurement unit(RIMU)have been presented in previous decades,the performance of these two-stage algorithms is restricted by the fault detection accuracy as far as to small fault.Therefore,a data driven-based fault diagnosis method which combines the self-organizing incremental neural network(SOINN)with the principal component analysis(PCA)in parity space is proposed for a RIMU with six-axis redundancy configuration.The topology structure of parity residuals is extracted by fast-SOINN algorithm,and a rough fault detection plane is designed based on the primary-neuron set of the network.At the same time,the Q contribution plot of sensor residuals is calculated by PCA to ensure the accuracy of fault isolation algorithm.Quantitative evaluations show that the fault identification accuracy of proposed algorithm is 7.89%higher than other related algorithms in a small fault case with an acceptable elapsed time which can meet a real-time requirements of RIMU at a 300Hz output frequency.The non-reset scheme of federated filter is badly affected by fault tracking while the filter variance in full-reset scheme will continuously amplify in the case of subsystem fault.Thus,a conditional reset fault-tolerant scheme is proposed for fully constrained federated filter in which auxiliary navigation system can provide complete state constrains to INS.Reset the state estimation of sub-filter to reduce the influence of fault tracking.And set the information sharing coefficients of the faulty filters according to the fault diagnosis results of all subsystems to prevent variance amplification.Simulation results and hardware experiment results show that the proposed scheme outperformed the traditional federated structure when the observation information of the subsystem is completely faulty or the fault detection rate is low.Fault detection ability is a key to the fault tolerance ability of integrated navigation system.Therefore,an energy dictionary-based fault diagnosis algorithm is proposed.The wavelet packet analysis method is adopted to acquire the energy distributions of filter prediction residual.And the dictionary learning theory is used to design a fault classifier based on the minimum error of data reconstruction,which improves the fault detection ability in integrated navigation filter.Simulation results and hardware experiment results show that the proposed algorithm is more sensitive than the clasical residual chi-square test method,and the fault tolerance strategy combined with conditional reset scheme has has a strong robustness for faults.