A dual redundant position sensor fault diagnosis method based on dynamic time warping

Abstract

The present invention proposes a dual redundant position sensor fault diagnosis method based on dynamic time warping, which is suitable for integration into a lightweight and miniaturized servo system to realize the detection and location of position sensor faults. The present invention is designed to calculate the consistency distance of the position sensor signal after low-pass filtering based on a dynamic time warping algorithm, use a random algorithm to set the inconsistency detection threshold, realize the consistency discrimination of the position sensor, and use the rotary transformer angle to construct a virtual position sensor to complete the positioning of the faulty position sensor and the identification of the extreme value and zero value fault mode. The method has certain diagnostic robustness and good fault compatibility.

Description

Technical Field

[0001] The present invention relates to the technical field of servo system fault diagnosis, and in particular to a dual redundant position sensor fault diagnosis method based on dynamic time warping. Background Art

[0002] As the key actuator of the aircraft, the servo system can quickly follow the position command signal to adjust the position of the follow-up control surface through the "position-speed-current" three-loop control, and adjust the three-axis motion attitude of the aircraft in real time and stably.

[0003] As an important part of the servo system, a healthy and reliable position sensor is crucial to the safe operation of the servo system. However, the position sensor is a relatively fragile measuring device, which is easily affected by various environmental factors such as short-term overheating, electromagnetic interference, abnormal shock, etc. during long-duration flight and fails. In order to improve the reliability of the servo system, the position sensor usually adopts a redundant design method, aiming to use the increased redundancy to assist fault diagnosis and fault-tolerant control. Traditional position sensor redundancy generally adopts a "three-redundancy" design method, and adopts a "three-out-of-two" (3oo2) voting decision mode to determine the faulty position sensor. At present, airborne servo systems are developing towards lightweight and miniaturized applications. From the perspective of reasonably reducing design redundancy, the design of dual redundant position sensors has received widespread attention. How to perform fault diagnosis on dual redundant position sensors to ensure the healthy operation of the servo system is an urgent problem to be solved. Summary of the invention

[0004] The technical problem solved by the present invention is: to overcome the deficiencies of the prior art and to provide a dual redundant position sensor fault diagnosis method based on dynamic time warping.

[0005] The technical solution of the present invention is:

[0006] A dual redundant position sensor fault diagnosis method based on dynamic time warping, characterized by comprising:

[0007] The consistency distance r of the two position sensor outputs after low-pass filtering is measured based on the dynamic time warping algorithm;

[0008] The inconsistency detection threshold is set by using a random algorithm to achieve consistency judgment of the position sensor;

[0009] On the basis of consistency judgment, a virtual position sensor is constructed using the angle of the rotary transformer to assist in the detection, location and fault mode identification of the position sensor fault.

[0010] Preferably, the consistency distance of the outputs of the two position sensors after low-pass filtering is measured based on a dynamic time warping algorithm, and the implementation method is as follows:

[0011] Step A1: Real-time acquisition of two-way position sensor signals pos under the running state of the servo system 1 ,pos 2 , dynamically intercept two position sensor signals with a length of l and in k is the current sampling time;

[0012] Step A2: Initialize the cumulative distance matrix D, D∈R l×l ;

[0013] Step A3: Set the matching window length w, and use the dynamic time warping algorithm to update the cumulative distance matrix, where the update formula of the i-th row and j-th column element D(i, j) of the cumulative distance matrix D is as follows:

[0014] D(i,j)=d(i,j)+min(D(i-1,j), D(i-1,j-1), D(i,j-1))

[0015] Among them, |ij|≤w,

[0016] Step A4: Calculate the consistency distance r of the two position sensor signals.

[0017] Preferably, the consistency distance r of the two position sensor signals satisfies:

[0018]

[0019] Among them, dist() is the consistency distance function of the two position sensor signals.

[0020] Preferably, a random algorithm is used to set the inconsistency detection threshold to achieve consistency judgment of the position sensor. The specific implementation method is as follows:

[0021] Step B1: Collect position sensor signals in a fault-free state and construct a consistent distance training set Among them, Nsamp is a sufficiently large constant;

[0022] Step B2: Given ε∈(0,1), δ∈(0,1), select a random sample number N that satisfies

[0023] Step B3: setting the allowed false alarm rate FAR and the allowed increment Δ of the single-step iteration;

[0024] Step B4: Initialize the inconsistency detection threshold make Among them, J0 is a sufficiently small real number;

[0025] Step B5: From the collection Randomly extract N groups of training samples from the random training set

[0026] Step B6: Traverse the random training set Count the elements in the collection that are greater than the current The number of samples n, return the proportion

[0027] Step B7: Iteratively calculate the inconsistency detection threshold according to the following criteria:

[0028] like Returns the current Go to step B8;

[0029] like make Return to step B5;

[0030] Step B8: Determine the consistency of the two position sensor signals. Determine that the signals of the two position sensors are consistent; if Determine whether the signals of the two position sensors are inconsistent.

[0031] Preferably, Nsamp satisfies Nsamp>N.

[0032] Preferably, on the basis of consistency determination, a virtual position sensor is constructed using the rotary transformer angle to assist in the detection, location and fault mode identification of the position sensor fault, and the implementation method is as follows:

[0033] Step C1: constructing a virtual position sensor using the angle signal of the rotary transformer according to the transmission relationship between the rotary motion and the linear motion;

[0034] Step C2: Dynamically intercept the virtual position sensor signal with a length of l

[0035] Step C3: Determine the two position sensor signals With virtual position sensor signal Inconsistency detection threshold between and

[0036] Step C4: After the virtual position sensor is introduced, the consistency of the position sensor signals is compared in pairs. If the signals of the two position sensors are consistent, both position sensors are not faulty, and the one with the smaller consistency distance with the virtual position sensor is selected for position feedback; if the signals of the two position sensors are inconsistent, and the signal of one position sensor is consistent with the signal of the virtual position sensor, it is determined that the other position sensor is faulty; if the comparison results of the two are both inconsistent, it is determined that both position sensors are faulty, and the virtual position sensor is selected for position feedback.

[0037] Step C5: Set the extreme fault detection threshold pos max , compare the maximum absolute value of the fault position sensor signal within the time window of length l with the extreme fault detection threshold to identify the extreme fault mode;

[0038] Step C6: Set the zero fault detection threshold pos zero , compare the maximum absolute value of the fault position sensor signal within the time window of length l with the zero-value fault detection threshold to perform zero-value fault identification.

[0039] Preferably, in step C1, pos 3 =θ×ratio, where pos 3 is the reconstructed virtual position sensor signal, θ is the angle of the resolver output, and ratio is the transmission ratio.

[0040] Preferably, the implementation of step C4 is as follows:

[0041] like If both position sensors are fault-free, the position feedback should be the one with the smaller consistency distance with the virtual position sensor;

[0042] like It is determined that the first position sensor fails, and the second position sensor is used for position feedback;

[0043] like It is determined that the second position sensor fails, and the first position sensor is used for position feedback;

[0044] like It is determined that both position sensors are faulty and a virtual position sensor is selected for position feedback.

[0045] Preferably, in step C5, for the faulty position sensor, if i={1, 2}, it is determined that an extreme fault occurs in the j-th position sensor.

[0046] Preferably, in step C5, for the faulty position sensor, if i={1, 2}, it is determined that the j-th position sensor has a zero-value fault.

[0047] Compared with the prior art, the present invention has the following advantages:

[0048] ① When judging the consistency of two-way position signals, the present invention uses a dynamic time warping algorithm to realize the consistency measurement of the position signal sequence within the time window. Compared with the point-to-point consistency measurement method, it has a certain robustness to signal wild values ​​and asynchronous communication delays.

[0049] ② The present invention adopts a random algorithm to design the non-uniform detection threshold, which has certain guiding significance compared with the artificial experience selection, and can avoid the problem of missed detection caused by the selection of too large a threshold to a certain extent.

[0050] ③ The present invention introduces a soft redundancy design of the position sensor and constructs a virtual position sensor using the angle signal output by the rotary transformer, so that the dual redundant diagnosis problem can be converted into a "triple redundant" diagnosis problem. The diagnostic logic is simpler. When both position sensors fail, there is still fault tolerance, good fault compatibility and high reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is the fault diagnosis schematic diagram of the dual redundant position sensor of the servo system;

[0052] Figure 2 It is a simplified schematic diagram of a specific implementation technology;

[0053] Figure 3 It is the virtual sensor output effect diagram;

[0054] Figure 4 This is the effect diagram of position sensor fault detection;

[0055] Figure 5 This is the effect diagram of the zero-value fault identification of the second position sensor. DETAILED DESCRIPTION

[0056] The present invention aims to solve the problem of dual redundant position sensor fault diagnosis. Based on the dynamic time warping algorithm, the consistency distance of the two position sensor outputs after low-pass filtering is measured, and the inconsistency detection threshold is set by a random algorithm to achieve consistency judgment of the position sensor. On the basis of consistency judgment, the virtual position sensor constructed by the rotary transformer angle is combined to realize the detection, location and fault mode identification of the position sensor fault. Figure 1 As shown in the figure, the key links of fault diagnosis design include: ① consistency distance measurement; ② inconsistency detection threshold design; ③ fault diagnosis logic.

[0057] The flow chart of the above invention is as follows Figure 2 As shown, it is mainly achieved through the following technical solutions:

[0058] ①Consistency distance measurement

[0059] According to the collected signals of the two position sensors, the consistency distance measurement is performed based on the dynamic time warping algorithm. The specific steps include:

[0060] Step A1: Real-time acquisition of the position sensor signal pos under the running state of the servo system 1 ,pos 2 , dynamically intercept two position sensor signals with a length of l and

[0061]

[0062] Among them, k is the current sampling time.

[0063] Step A2: Initialize the cumulative distance matrix D∈R l×l , the initial value of the matrix element D(i, j) is set to infinity.

[0064] Step A3: Set the matching window length w and use the dynamic time warping algorithm to update the cumulative distance matrix.

[0065] D(i,j)=d(i,j)+min(D(i-1,j), D(i-1,j-1), D(i,j-1))

[0066] Among them, |ij|≤w,

[0067] Step A4: Output the consistency distance metric r of the position sensor signal,

[0068]

[0069] Where D(l,l) is and The minimum matching distance between two position signal sequences.

[0070] ② Design of non-uniform detection threshold

[0071] Based on the above consistency distance metric, a random algorithm is used to design the inconsistency detection threshold of the two position sensor signals. The specific steps include:

[0072] Step B1: Collect enough position sensor signals in a fault-free state and construct a consistent distance measurement training set

[0073]

[0074] Wherein, Nsamp is a sufficiently large constant.

[0075] Step B2: Given ε∈(0,1), δ∈(0,1), select a random sample number N that satisfies

[0076]

[0077] Step B3: Set the allowable false alarm rate FAR and the allowable increment Δ of the single-step iteration.

[0078] Step B4: Initialize the inconsistency detection threshold make

[0079]

[0080] Wherein, J0 is a sufficiently small real number.

[0081] Step B5: From the collection Randomly extract N groups of training samples from the random training set

[0082] Step B6: Traverse the random training set Calculate the number of elements in the collection that are greater than the current J th The number of samples n, return the proportion

[0083] Step B7: Iteratively calculate the inconsistency detection threshold Do the following:

[0084] like Returns the current Exit algorithm;

[0085] like make Return to step B5.

[0086] Step B8: Determine the consistency of the dual position sensor signals. If Determine that the signals of the two position sensors are consistent; if Determine whether the signals of the two position sensors are inconsistent.

[0087] ③Fault diagnosis

[0088] On the basis of consistency judgment, the occurrence of position sensor failure is detected and the faulty position sensor is located. The specific steps include:

[0089] Step C1: Based on the transmission relationship between rotational motion and linear motion, a virtual position sensor is designed using the angle signal of the rotary transformer.

[0090] POS 3=θ×ratio

[0091] Among them, pos 3 is the reconstructed virtual position signal, θ is the angle of the resolver output, and ratio is the transmission ratio.

[0092] Step C2: Dynamically intercept the position sensor signal with a length of l in the same manner as in step A1

[0093] Step C3: According to steps B1 to B7, determine the two position sensor signals With virtual position sensor signal Inconsistency detection threshold between and

[0094] Step C4: Position sensor fault detection. After the virtual position sensor is introduced, if If both position sensors are fault-free, the position feedback should be the one with the smaller consistency distance with the virtual position sensor; It is determined that the first position sensor is faulty, and the second position sensor is used for position feedback; if It is determined that the second position sensor is faulty, and the first position sensor is used for position feedback; if It is determined that both position sensors are faulty and a virtual position sensor is selected for position feedback.

[0095] Step C5: Extreme fault mode identification. Set the extreme fault detection threshold pos max , compare the maximum absolute value of the fault location sensor signal within the time window of length l with the extreme fault detection threshold. It is determined that the i-th position sensor has an extreme fault.

[0096] Step C6: Zero value fault mode identification. Set the zero value fault detection threshold pos zero , compare the maximum absolute value of the fault position sensor signal within the time window of length l with the zero-value fault detection threshold. It is determined that the i-th position sensor has a zero value fault.

[0097] Example:

[0098] The embodiment of the present invention is a differential speed integrated dual redundant servo system, the position adjustable range is [-35mm, 35mm], and the transmission ratio ratio=0.0783.

[0099] Given a position command signal with an amplitude of 35 mm and 0.05 Hz, the consistency distance measurement parameter settings are: the dynamically intercepted window length l = 50, the matching window length w = 2; the inconsistency detection threshold design parameter settings are: ε = 0.01, δ = 0.01, the allowable fault false alarm rate FAR = 0.02, the number of random training set samples N = 20000, the initial inconsistency detection threshold J0 = 0.0001, and the allowable increment of single-step iteration Δ = 0.0001; the fault mode identification parameter settings are: the extreme fault detection threshold pos max =35mm, zero value fault detection threshold pos zero =0.08mm.

[0100] Collect enough position sensor data in a fault-free state, calculate the consistency distance between the three groups of position signals (two position sensor signals and one virtual position sensor signal) in pairs according to the consistency distance measurement method, and construct three groups of consistency distance measurement training sets According to the above-mentioned non-consistent detection threshold design method, the three sets of non-consistent detection threshold designs are calculated as follows: According to the fault diagnosis logic, the position sensor fault is detected, located and identified in mode. It should be pointed out that due to the use of differential speed integrated dual redundant structure design, the dual-channel rotary transformer angles θ1 and θ2 need to be added to obtain the angle θ used to calculate the virtual position signal.

[0101] Figure 3 The position output of the virtual sensor is shown. To facilitate observation of the effect, a partial enlarged view of the position output is captured. In the fault-free state, the output of the virtual position sensor is well consistent with the output of the two position sensors, verifying the effectiveness of the designed virtual position sensor.

[0102] Figure 4 The fault diagnosis results of dual redundant position sensors are shown. In the experiment, a zero-value fault is injected into the second position sensor at the 50th second. The consistency distance between the first and second sensors exceeds the threshold at the time of the fault. The two signals are inconsistent; the consistency distance between the first signal and the virtual sensor does not exceed the threshold The two signals are consistent; the consistency distance between the second signal and the virtual sensor exceeds the threshold at the time of the fault The two signals are inconsistent. Therefore, it can be determined that a fault occurred at the 50th second and the second position sensor failed. It should be noted that in the fault-free state, the phenomenon of occasionally exceeding the threshold is because the threshold is designed under the allowable fault false alarm rate. This threshold design method can avoid missed detection due to excessive threshold selection. In this example, if 10 consecutive sampling points do not exceed the threshold, it can be determined that there is no fault. In addition, the fault position sensor of the second path like Figure 5 As shown, it can be determined that a zero value fault has occurred.

[0103] The present invention is suitable for integration into a lightweight and miniaturized servo system to detect and locate position sensor faults. The present invention is designed to calculate the consistency distance of the position sensor signal after low-pass filtering based on a dynamic time warping algorithm, use a random algorithm to set the inconsistency detection threshold, realize the consistency discrimination of the position sensor, and use the rotary transformer angle to construct a virtual position sensor to complete the positioning of the faulty position sensor and the identification of the extreme value and zero value fault mode. The method has certain diagnostic robustness and good fault compatibility.

[0104] The contents not described in detail in the specification of the present invention belong to the common knowledge of the professionals in this field.

Claims

1. A dual redundant position sensor fault diagnosis method based on dynamic time warping, characterized in that: include: The consistency distance r of the two position sensor outputs after low-pass filtering is measured based on the dynamic time warping algorithm; The inconsistency detection threshold is set by using a random algorithm to achieve consistency judgment of the position sensor; On the basis of consistency judgment, a virtual position sensor is constructed using the angle of the rotary transformer to assist in the detection, location and fault mode identification of the position sensor fault. The consistency distance of the two position sensor outputs after low-pass filtering is measured based on the dynamic time warping algorithm. The implementation method is as follows: Step A1: Real-time acquisition of two-way position sensor signals pos under the running state of the servo system 1 ,pos 2 , dynamically intercept two position sensor signals with a length of l and in k is the current sampling time; Step A2: Initialize the cumulative distance matrix D, D∈R l×l ; Step A3: Set the matching window length w, and use the dynamic time warping algorithm to update the cumulative distance matrix, where the update formula of the i-th row and j-th column element D(i,j) of the cumulative distance matrix D is as follows: D(i,j)=d(i,j)+min(D(i-1,j),D(i-1,j-1),D(i,j-1)) Among them, |ij|≤w, Step A4: Calculate the consistency distance r of the two position sensor signals; The consistency distance r of the two position sensor signals satisfies: Wherein, dist() is the consistency distance function of the two position sensor signals; The inconsistency detection threshold is set by using a random algorithm to achieve consistency judgment of the position sensor. The specific implementation method is as follows: Step B1: Collect position sensor signals in a fault-free state and construct a consistent distance training set Among them, Nsamp is a sufficiently large constant; Step B2: Given ε∈(0,1), δ∈(0,1), select a random sample number N that satisfies Step B3: setting the allowed false alarm rate FAR and the allowed increment Δ of the single-step iteration; Step B4: Initialize the inconsistency detection threshold make Among them, J0 is a sufficiently small real number; Step B5: From the collection Randomly extract N groups of training samples from the random training set Step B6: Traverse the random training set Count the elements in the collection that are greater than the current The number of samples n, return the proportion Step B7: Iteratively calculate the inconsistency detection threshold according to the following criteria: like Returns the current Go to step B8; like make Return to step B5; Step B8: Determine the consistency of the two position sensor signals. Determine that the signals of the two position sensors are consistent; if Determine whether the signals of the two position sensors are inconsistent.

2. A dual redundant position sensor fault diagnosis method based on dynamic time warping according to claim 1, characterized in that: Nsamp satisfies Nsamp>N.

3. A dual redundant position sensor fault diagnosis method based on dynamic time warping according to claim 1, characterized in that: On the basis of consistency judgment, a virtual position sensor is constructed using the rotary transformer angle to assist in the detection, location and fault mode identification of the position sensor fault. The implementation method is as follows: Step C1: constructing a virtual position sensor using the angle signal of the rotary transformer according to the transmission relationship between the rotary motion and the linear motion; Step C2: Dynamically intercept the virtual position sensor signal with a length of l Step C3: Determine the two position sensor signals With virtual position sensor signal Inconsistency detection threshold between and Step C4: After the virtual position sensor is introduced, the consistency of the position sensor signals is compared in pairs. If the signals of the two position sensors are consistent, both position sensors are not faulty, and the position feedback uses the one with a smaller consistency distance with the virtual position sensor; if the signals of the two position sensors are inconsistent, and the signal of one position sensor is consistent with the signal of the virtual position sensor, it is determined that the other position sensor is faulty; if the comparison results of the two are inconsistent, it is determined that both position sensors are faulty, and the virtual position sensor is used for position feedback; Step C5: Set the extreme fault detection threshold pos max , compare the maximum absolute value of the fault position sensor signal within the time window of length l with the extreme fault detection threshold to identify the extreme fault mode; Step C6: Set the zero fault detection threshold pos zero , compare the maximum absolute value of the fault position sensor signal within the time window of length l with the zero-value fault detection threshold to perform zero-value fault identification.

4. A dual redundant position sensor fault diagnosis method based on dynamic time warping according to claim 3, characterized in that: In the step C1, pos 3 =θ×ratio, where pos 3 is the reconstructed virtual position sensor signal, θ is the angle of the resolver output, and ratio is the transmission ratio.

5. The dual redundant position sensor fault diagnosis method based on dynamic time warping according to claim 3 is characterized in that: The implementation of step C4 is as follows: like If both position sensors are fault-free, the position feedback should be the one with the smaller consistency distance with the virtual position sensor; like It is determined that the first position sensor fails, and the second position sensor is used for position feedback; like It is determined that the second position sensor fails, and the first position sensor is used for position feedback; like It is determined that both position sensors are faulty and a virtual position sensor is selected for position feedback.

6. A dual redundant position sensor fault diagnosis method based on dynamic time warping according to claim 3, characterized in that: In step C5, for the faulty position sensor, if It is determined that the i-th position sensor has an extreme fault.

7. The dual redundant position sensor fault diagnosis method based on dynamic time warping according to claim 3 is characterized in that: In step C5, for the faulty position sensor, if It is determined that the i-th position sensor has a zero value fault.

Images