A Fault Diagnosis and Fault Tolerance Method for Integrated Navigation System Based on Dynamic Threshold

Abstract

The present invention proposes a combined navigation system fault diagnosis and fault tolerance method based on dynamic thresholds. The method first uses Mahalanobis distance to construct dynamic anomaly factors for navigation system fault diagnosis; then uses the T-S model to evaluate the credibility of navigation information, and calculates the degree of compliance of each state according to the constructed membership function; then constructs an adaptive gradient descent algorithm according to the change rate of navigation information at the current moment, thereby performing optimal estimation and finally obtaining fault-tolerant information. This effectively solves the problem that slowly changing faults are difficult to detect and the navigation system accuracy decreases after a fault occurs.

Description

Technical Field

[0001] The present invention is a fault diagnosis and fault tolerance method for low-cost, miniaturized integrated navigation systems. The traditional method uses residual chi-square test for fault diagnosis, but it is difficult to detect slowly varying faults. The present invention uses Mahalanobis distance to establish a dynamic threshold evaluation method to achieve fault detection, and constructs fuzzy judgment logic according to the TS model, and uses an improved adaptive gradient descent algorithm to solve, thereby improving the accuracy of fault diagnosis of the integrated navigation system. Background Art

[0002] Navigation instruments are increasingly used in military and civilian fields, and can provide key navigation information for a variety of vehicles (such as vehicles, ships, and aircraft), helping the vehicles to accurately obtain the current key navigation information, namely position, speed, and attitude, during navigation. In order to improve navigation accuracy, people install multiple sensors on the same vehicle, and then use the optimal estimation algorithm to fuse the information of multiple sensors to obtain the final navigation result.

[0003] Miniaturized combined navigation systems are usually composed of an inertial navigation system (INS) and a global navigation satellite system (GNSS). INS can perform independent navigation without the help of external measurement information, but its error will accumulate over time; GNSS has high positioning accuracy, but is greatly affected by the environment. The combination of the two for navigation has become a mature navigation technology. However, in practical applications, navigation instruments often need to work in complex and harsh environments. When a vehicle passes through a complex environment such as a canyon, the GNSS signal is likely to be distorted due to occlusion, and INS and other navigation instruments may also malfunction due to vibration and electromagnetic interference, resulting in abnormal values. This will cause erroneous information to participate in the filtering operation, which will in turn affect the navigation accuracy.

[0004] In current engineering applications, fault diagnosis of integrated navigation uses chi-square test or residual chi-square test. According to the threshold set by the chi-square test, the status of the sub-sensor will be judged as "faulty" or "no fault". The size of the threshold is determined based on experience. If it is set too high, slow and subtle faults will not be detected, and if it is set too low, the false alarm rate may increase. If the incorrect navigation information is incorporated into the filter for calculation, it will lead to the output of incorrect navigation results, and the accuracy of the navigation system will be greatly reduced. If the threshold is lowered in order to increase the probability of detecting a fault, the system will report errors too frequently, causing the system to oscillate and the navigation results to be unstable.

[0005] In recent years, in order to further improve the efficiency of fault diagnosis, Zhang et al. proposed a two-state chi-square test method based on the residual chi-square test method to improve the reliability of fault detection. However, this method requires isolating the fault subsystem after detecting the fault, and then using the one-step prediction of the Kalman filter for state estimation, which will reduce the filtering accuracy. Xiong et al. proposed a fault-tolerant federated filtering method that uses fault detection functions to construct time-varying variable measurement noise. This method can significantly improve the accuracy of the fault subfilter and the global estimation accuracy, but the architecture based on the federated filter determines that this method requires a variety of navigation instruments, and its application is limited in miniaturized, low-cost integrated navigation systems.

[0006] In order to overcome the above problems, the present invention proposes an improved integrated navigation fault diagnosis and fault tolerance algorithm, which makes up for the deficiency that the traditional residual chi-square detection cannot detect, and achieves the purpose of detecting slowly varying faults as early as possible, so as to improve the accuracy of the integrated navigation system. The method first uses the Mahalanobis distance to establish a dynamic threshold evaluation method to realize fault detection, then constructs a fuzzy fault tolerance algorithm according to the TS model, and finally solves it through an improved adaptive gradient descent algorithm, so that the accuracy and stability of the navigation system are improved. Summary of the invention

[0007] The purpose of the present invention is to provide an improved navigation system fault diagnosis and fault tolerance method, after a sudden fault or a slowly changing fault occurs, the navigation system can quickly perform fault diagnosis and achieve fault tolerance to achieve high accuracy.

[0008] The object of the present invention is achieved by following steps:

[0009] Step 1: Design a vehicle-mounted INS / GNSS integrated navigation experiment, obtain simulation data or measured data through path simulation or sports car experiment, and add fault signals on this basis;

[0010] Step 2: According to the Kalman filter equation, the residual obtained by calculating the state parameters is obtained, and the fault diagnosis of the navigation system is performed using the Mahalanobis distance;

[0011] Step 3: Use the TS model to evaluate the credibility of the navigation information and calculate the degree of compliance of each state according to the constructed membership function;

[0012] Step 4: Construct an adaptive gradient descent algorithm based on the rate of change of the navigation information at the current moment, thereby performing optimal estimation and obtaining fault-tolerant information, thereby improving the accuracy of the navigation system.

[0013] In step 2 of the above method, the specific method for fault diagnosis based on Mahalanobis distance is:

[0014] 1) Assuming that no fault occurs within the first m moments, the Mahalanobis distance is defined as: where μ m is the mean of the data set at the previous m moments, P k / k-1 is the mean square error matrix for the state one-step prediction;

[0015] 2) Set the abnormal factor of the current data to in is the i-th element in the Mahalanobis distance vector at time k, Th(k,i) is the threshold of the i-th element in the Mahalanobis distance vector at time k, which is determined by the state of the abnormal factor at time k-2 and time k-1. The determination process is as follows: Figure 2 shown.

[0016] 3) Let ρ th is the warning value. According to the size of the abnormal factor, it can be judged whether the data point is normal: if The data is normal; if The data state is uncertain and needs to be determined based on the abnormal factor calculated at the next moment; if The data is abnormal.

[0017] In step 3 of the above method, the specific method of evaluating the credibility of navigation information using the TS model is:

[0018] 1) Construct the filter error factor according to the state mean square error matrix in the Kalman filter, combine it with the abnormal factor output by the fault diagnosis, and define the fuzzy rules of the fuzzy judgment system;

[0019] 2) Set the membership function of the fuzzy prediction system is the Gaussian membership function of the fuzzy antecedent parameter x, let γ m Indicates the degree of compliance with the mth rule, y m represents the actual output value of the navigation instrument under the mth rule, M is the total number of rules, and the output prediction value of the system is

[0020] 3) At time k+1, according to Calculate the fitting error.

[0021] In step 4 of the above method, the specific method of solving using the adaptive gradient descent algorithm is:

[0022] 1) The angular velocity of the load system measured by the gyroscope in the INS is The specific force under the load system measured by the accelerometer in the INS is This allows the introduction of an adaptive parameter ξ, which is defined as where λ 0 is the initial value, t 0 is the system sampling time, ζ1 , 2 is the regulating factor;

[0023] 2) Determine the fuzzy antecedent parameters so that the fitting error between the predicted output and the actual output data of the navigation instrument is minimized. Use the gradient descent method to solve the parameters of the fuzzy prediction system.

[0024] The advantages of the present invention are as follows: the method of the present invention can evaluate whether a fault occurs in the integrated navigation system through a dynamic threshold, and without increasing the system hardware redundancy, the fault-tolerant method established according to the TS model can effectively solve the problem that slowly varying faults are difficult to detect and the navigation system accuracy decreases after a fault occurs, and has great application prospects for the integrated navigation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the process of the present invention.

[0026] Figure 2 Schematic diagram of the process for dynamic threshold determination.

[0027] Figure 3(a) shows the result of fault detection using traditional residual chi-square using longitude information.

[0028] FIG. 3( b ) is a diagram showing the result of fault detection using the longitude information of the present invention.

[0029] Figure 4(a) shows the result of fault detection using traditional residual chi-square using latitude information.

[0030] FIG. 4( b ) is a diagram showing the result of fault detection using the latitude information of the present invention.

[0031] FIG5( a ) is a comparison diagram of the longitude positioning error using the traditional residual chi-square and the present invention.

[0032] FIG5( b ) is a comparison diagram of the latitude positioning error using the traditional residual chi-square and the present invention. DETAILED DESCRIPTION

[0033] The present invention is described in detail below in conjunction with specific implementation cases.

[0034] The present invention provides a fault diagnosis and fault tolerance method for an integrated navigation system based on Mahalanobis distance and TS model. Figure 1 The object of the present invention is achieved by the following steps:

[0035] 1. Design a vehicle-mounted INS / GNSS combined navigation experiment, obtain the output data of INS and GNSS through sports car experiments or path simulation, and add fault signals on this basis. The expression of the measurement information is Z k =H kX k +V k +N k , where N k is the fault information, and its expression is

[0036]

[0037] Set at t mb <t<t me During the period, the sensor has a sudden fault, and the fault is a constant value η 0 ; at t sb <t<t se During the period, the sensor has a slow-changing fault, t sf The failure rate reaches its peak at the moment, and the growth rate and slowdown rate are η 1 and η 2 ;

[0038] 2. Use the strapdown inertial navigation solution method to solve the problem, and get the residual error of the state parameter calculation in the Kalman filter as ε k =Z k -H k X k ;

[0039] 3. Assuming that no fault occurs within the first m moments, the Mahalanobis distance is defined as: where μ m is the mean of the data set at the previous m moments, P k / k-1 is the mean square error matrix for the state one-step prediction;

[0040] 4. Set the abnormal factor of the current data to in is the i-th element in the Mahalanobis distance vector at time k, Th(k,i) is the threshold of the i-th element in the Mahalanobis distance vector at time k, which is determined by the state of the abnormal factor at the previous two moments. The process is determined as follows: Figure 2 As shown;

[0041] 5. Let ρ th is the warning value. According to the size of the abnormal factor, it can be judged whether the data point is normal: if The data is normal. If The data state is uncertain and needs to be determined based on the abnormal factor calculated at the next moment. The data is abnormal;

[0042] 6. Construct the filter error factor according to the state mean square error matrix in the Kalman filter, combine it with the abnormal factor output by the fault diagnosis, define the fuzzy rules of the fuzzy judgment system, and use the TS model to evaluate the credibility of the navigation information;

[0043] 7. Set the membership function of the fuzzy prediction system is the Gaussian membership function of the fuzzy antecedent parameter x, let γ m Indicates the degree of compliance with the mth rule, y m represents the actual output value of the navigation instrument under the mth rule, M is the total number of rules, and the output prediction value of the navigation system is

[0044] 8. At time k+1, according to Calculate the fitting error;

[0045] 9. The angular velocity of the load system measured by the gyroscope in the INS is The specific force under the load system measured by the accelerometer in the INS is This allows the introduction of an adaptive parameter ξ, which is defined as where λ 0 is the initial value, t 0 is the system sampling time, ζ 1 , 2 is the regulating factor;

[0046] 10. Determine the fuzzy antecedent parameters so that the fitting error between the predicted output and the actual output data of the navigation instrument is minimized. Use the gradient descent method to solve the parameters of the fuzzy prediction system.

[0047] In this way, the fault diagnosis and fault tolerance of the combined navigation system based on the Mahalanobis distance and TS model can be completed, which can effectively solve the problem that slowly varying faults are difficult to detect and the navigation system accuracy decreases after a fault occurs.

[0048] The effect of the present invention can be verified by the following simulation test:

[0049] First, construct the vehicle driving trajectory and specify the initial parameters. The initial geographic location is set to 45°N and 126°E, the initial speed is 0, and the initial heading angle is 0. The gyro constant drift and random drift are both set to 0.1° / h, and the accelerometer constant drift and random drift are both set to 50μg. Vehicle movement includes straight driving, turning, climbing, acceleration, deceleration, etc. The GPS position error is set to 10m, and the speed error is set to 0.2m / s. The simulated driving time is 1000s, and the filter period is set to 1s. The false alarm rate and missed detection rate are set to 1% and 20%. It is set that the GNSS has a sudden fault in the period of 146s to 221s, and the fault is a constant value of 20m; in the period of 355s to 875s, the GNSS has a slow fault, and the fault reaches a peak of 15m at 795s.

[0050] Considering longitude and latitude respectively, the traditional residual chi-square detection method is used to compare with the fault detection method based on Mahalanobis distance proposed in this paper, and the results are shown in Figures 3, 4, and 5. The above figures clearly show that both the traditional method and the fault detection method based on Mahalanobis distance can obtain accurate results for sudden faults. However, for slow-changing faults, the method proposed in this paper can detect them faster. Compared with the traditional method, the longitude fault detection time of the method proposed in the present invention is 89s earlier, and the latitude fault detection time is 79s earlier. Therefore, the fault diagnosis method provided by the present invention is significantly better than the traditional method. The traditional fault isolation method is compared with the fuzzy fault-tolerant method based on the TS model, and the latitude error and longitude error are examined, as shown in the figure. It can be seen from the figure that when a slow-changing fault occurs, the traditional fault isolation method is not effective and will greatly reduce the navigation accuracy. Therefore, the fault-tolerant method provided by the present invention can better restore the real navigation information and improve accuracy.

Claims

1. A fault diagnosis and fault tolerance method for an integrated navigation system based on a dynamic threshold, characterized in that: Includes the following steps Steps: Step 1: Design a vehicle-mounted INS (Inertial navigation system) / GNSS (Global Navigation Satellite System) combined navigation experiment, obtain simulation data or measured data through path simulation or sports car experiment, and add fault signals on this basis; Step 2: According to the Kalman filter equation, the residual obtained by calculating the state parameters is obtained, and the fault diagnosis of the navigation system is performed using the Mahalanobis distance; Step 3: Use the TS model to evaluate the credibility of navigation information and calculate the degree of compliance of each state according to the constructed membership function; Step 4: Construct an adaptive gradient descent algorithm based on the change rate of the navigation information at the current moment, thereby performing optimal estimation and obtaining fault-tolerant information, thereby improving the accuracy of the navigation system; The specific method of using Mahalanobis distance to diagnose the fault of the navigation system in step 2 is as follows: 1) Assuming that no fault occurs within the first m moments, the Mahalanobis distance is defined as: Where μm is the mean of the data set at the previous m moments, Pk / k-1 is the mean square error matrix of the state one-step prediction, and ε k is the residual calculated at the current moment; 2) Set the abnormal factor of the current data to in is the i-th element in the Mahalanobis distance vector at time k, Th(k,i) is the threshold of the i-th element in the Mahalanobis distance vector at time k, which is determined by the state of the abnormal factor at time k-2 and time k-1; 3) Let ρth be the warning value. According to the size of the abnormal factor, we can judge whether the data point is normal: if The data is normal; if The data state is uncertain and needs to be determined based on the abnormal factor calculated at the next moment; if The data is abnormal; The adaptive gradient descent algorithm described in step 4 is used to solve the problem. The specific method is: 1) The angular velocity and specific force measured by INS are The adaptive parameters are defined as , where λ0 is the initial value, t0 is the system sampling time, ζ1 and ζ2 are adjustment factors; 2) Determine the fuzzy antecedent parameters so that the fitting error between the predicted output and the actual output data of the navigation instrument is minimized, and use the gradient Degree descent method to solve the parameters of fuzzy prediction system

Images