An underwater robot thruster fault feature extraction method based on thrust deviation and cross-correlation algorithm

Abstract

This invention provides a method for extracting fault features of underwater robot thrusters based on thrust deviation and cross-correlation algorithms, belonging to the field of underwater robot fault diagnosis technology. The method first obtains two thrust deviation curves based on the correlation signals of the longitudinal and bow degrees of freedom. The fault occurrence time is determined based on the step point position of the curves to improve the positioning accuracy of the fault occurrence time. Next, to enhance the fault features, a multi-time-window Fourier transform and fusion method is designed to extract the fault features from the two thrust deviation curves separately, and the cross-correlation fault features of the two fault features are obtained. Finally, the three fault features are fused to obtain the final fault features. This project, by extracting fault features from multiple signals, can achieve relatively accurate positioning of the fault occurrence time and obtain a high ratio of fault feature values ​​to interference feature values. It is particularly suitable for fault feature extraction in ocean current environments and under conditions of weak fault severity.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of underwater robot fault diagnosis, in particular to an underwater robot propeller fault feature extraction method suitable for a weak fault degree. BACKGROUND

[0002] Due to the ability to freely navigate in complex ocean environments, underwater robots have a wide range of applications, such as underwater exploration, underwater search, and autonomous operation. Reliable fault diagnosis is an important part of ensuring that underwater robots can successfully complete their tasks, and the foundation of fault diagnosis is fault feature extraction. In existing fault feature extraction methods, scholars extract multiple signals for fault feature extraction to fully utilize the fault information in each signal, strengthen the fault feature, and improve the reliability of fault diagnosis.

[0003] Existing fault feature extraction methods mostly extract fault features of different signals separately and then fuse them to obtain the final fault feature. A typical method is to first extract fault features of different signals using a modified Bayesian method, and then fuse them using the D-S evidence theory method. However, this method has two problems due to the different signal changes after the occurrence of different signals, one is that the fault detection time lags behind the occurrence time, and the other is that it cannot extract effective fault features when the fault degree is low (the fault feature value is smaller than the interference feature value). Third, it cannot extract fault features of longitudinal and bow degree of freedom signals simultaneously. To solve the above problems, the applicant studies a method of first converting underwater robot control and state signals into thrust deviation signals, and then extracting fault features, which solves the three problems of the typical method, so this research is meaningful. SUMMARY

[0004] The purpose of the present application is to provide an underwater robot propeller fault feature extraction method based on thrust deviation and cross-correlation algorithm. The present application can extract fault features of multiple signals when the fault degree is low in the current environment, the fault detection time has high accuracy, and the ratio of fault feature to interference feature is high, which helps to improve the accuracy of fault detection.

[0005] The object of the application is achieved by the technical solutions below: the starting point of the fault feature extraction is to extract the fault features of multiple signals respectively, and then to further strengthen the fault features through a fusion method. Most of the existing fault feature extraction methods first extract the fault features of each signal, and then perform fusion. However, due to the different change times and trends of each signal after the fault occurs, the obtained fault feature wave peak positions are different, and part of the signal fault features cannot be used because the fault features are cancelled after fusion. Therefore, the application relates to an underwater robot thruster fault feature extraction method based on thrust deviation and cross-correlation algorithm, and the implementation of the regional tracking control method of the application includes the following three steps:

[0006] Step (1): based on the longitudinal velocity and the longitudinal thruster control voltage signal, the longitudinal signal thrust deviation curve is obtained by combining the underwater robot dynamics model and the thruster dynamics model, and based on the heading angle and the lateral thruster control voltage signal, the heading signal thrust deviation curve is obtained;

[0007] Step (2): based on the longitudinal and heading signal thrust deviation curves, the longitudinal and heading signal fault features are obtained by using the multi-time window Fourier transform and fusion method combined with the results of step (1);

[0008] Step (3): based on the results of step (2), the cross-correlation fault feature of the longitudinal and heading signal fault features is obtained by using the sliding time window cross-correlation method, and the final fault feature is obtained by fusing the two fault features.

[0009] The application also includes the following structural features:

[0010] Further, the step (1) is specifically:

[0011] The longitudinal and heading dynamics models of the multi-thruster driven underwater robot can be generally described as equations (1) and (2):

[0012]

[0013]

[0014] Wherein, m is the mass of the underwater robot; u is the longitudinal velocity of the underwater robot; T is the longitudinal thrust received by the underwater robot, r is the heading angle of the underwater robot; τ is the heading torque received by the underwater robot; X u 、X u|u| 、 N r 、N r|r| is the correlation coefficient.

[0015] According to formulas (1) and (2), the actual thrust T generated by the thruster is obtained according to the longitudinal velocityz , the yawing torque τ of the underwater robot is obtained according to the yaw angle, and the actual thrust T of the propeller is obtained according to the torque and the length of the propeller force arm s .

[0016] The dynamics model of the underwater robot propeller can be generally described as the following equation:

[0017] T' = cV + d (3)

[0018] wherein T' is the theoretical thrust, V is the propeller control voltage; c, d are relevant coefficients.

[0019] The actual thrust T of the propeller is obtained z , T s The difference between the theoretical thrust T' and the actual thrust T is obtained, and the thrust deviation based on the longitudinal signal and the yaw signal is obtained, and the formula is as follows:

[0020] T" z = T' - T z (4)

[0021]

[0022] wherein T" z is the longitudinal signal thrust deviation, and T" s is the yaw signal thrust deviation.

[0023] Further, the step (2) is specifically:

[0024] According to the sliding time window Fourier transform formula:

[0025]

[0026] wherein X n is the Fourier transform result, k is the sequence number of the transform result, indicating the arrangement order of the frequency from low to high in the frequency spectrum, k = 1 indicating the lowest frequency transform result, and M is the number of signals in the time window, W M = e j2 π / M .

[0027] The lowest frequency amplitude in the sliding time window Fourier transform, that is, the amplitude when k = 1, is extracted, and T z " is substituted into formula (6), and the lowest frequency amplitude is calculated, and the formula is as follows:

[0028] The curve of the longitudinal signal thrust deviation T z " formed by the sliding time window is intercepted, and then the Fourier transform is performed on all the intercepted parts, and the lowest frequency amplitude of the transform result, that is, the amplitude when k = 1, is extracted, and T zSubstitute into formula (5) and calculate the lowest frequency amplitude value, as shown in the formula below:

[0029]

[0030] Where f is the lowest frequency amplitude.

[0031] By using multiple time windows, i.e. taking multiple M values, multiple lowest frequency amplitude curves are obtained.

[0032] The lowest frequency curve is shifted forward by half a time window along the time axis and used as a fault characteristic, as shown in the following formula:

[0033] d z (n)=f(nM / 2) (8)

[0034] Where, d z For longitudinal signal fault characteristics, n is the sequence number of the signal.

[0035] According to formula (8), the curves formed by the lowest frequency amplitudes under multiple time windows are shifted forward by half a time window length along the time axis to obtain multiple fault characteristic curves. These curves are then fused using the DS evidence theory method to obtain the fused longitudinal signal fault characteristic m. FZ .

[0036] The above method is used to determine the thrust deviation T based on the bow signal. s "Obtain the fault characteristics m of the fused heading signal" FS .

[0037] Furthermore, step (3) specifically includes:

[0038] A sliding time window is used to extract the two fault characteristics, as shown in the following formula.

[0039] u(n) = m FZ (r), r∈[nM / 2,n+M / 2] (9)

[0040] v(n) = m FS (r), r∈[nM / 2,n+M / 2] (10)

[0041] Where u(n) and v(n) are the longitudinal and forward signal fault features captured at time n, n and r are the signal sequence numbers, and m FZ m FS The fault characteristics are represented by longitudinal and bow signals, and M is the number of signals within the time window.

[0042] Calculated using the cross-correlation function formula:

[0043]

[0044] In the formula, F represents the cross-correlation function values ​​of functions u and v. s The frequency used is m, which is an integer related to the delay, and the delay value is m / F. s .

[0045] The fault characteristic curves of longitudinal and bow signals are extracted using a sliding time window. According to formula (11), the cross-correlation operation is performed on the extracted curves to obtain the cross-correlation curve at time n. The formula is as follows:

[0046]

[0047] Based on the obtained cross-correlation curve, extract the maximum value of the cross-correlation curve using the following formula.

[0048]

[0049] Where, m FH (n) represents the maximum value of the cross-correlation curve between u(n) and v(n).

[0050] Based on formula (13), the maximum cross-correlation curves of longitudinal and bow signal fault characteristics can be obtained, which can be used as cross-correlation fault characteristic curves.

[0051] The fault characteristic curves are obtained by fusing longitudinal, heading, and cross-correlation fault characteristic curves using the DS evidence theory method.

[0052] Compared with existing technologies, the beneficial effects of this invention are mainly reflected in the following: Existing fault feature extraction methods extract fault features from different signals separately and then fuse them. However, since different signals may have different trends after a fault occurs, the extracted fault feature peak positions are different, resulting in poor direct fusion effects. To address this research background, this invention proposes a fault feature extraction method for underwater robot thrusters based on thrust deviation and cross-correlation algorithms. To ensure that the peak positions of fault features from different signals are the same, a thrust deviation signal is obtained based on control and state signals, and then fault features are extracted from this signal. To enhance the fused fault features, a fault feature extraction method based on cross-correlation algorithms is designed. Cross-correlation fault features are obtained based on the correlation of different fault features, and then fused, improving the ratio of fault features to interference features and achieving reliable thruster fault diagnosis. Attached Figure Description

[0053] Figure 1 This is a flowchart of the fault feature extraction process of the present invention.

[0054] Figure 2 This is a comparison of the fault characteristics obtained by the method of the present invention and the conventional method under a 20% fault condition.

[0055] Figure 3 This is a comparison of the fault characteristics obtained by the method of the present invention and the conventional method under a 10% fault condition. Detailed Implementation

[0056] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0057] Figure 1 This is a flowchart illustrating the thruster fault feature extraction method of the present invention. (In conjunction with...) Figure 1 The specific implementation steps of a method for extracting fault features of underwater robot thrusters based on thrust deviation and cross-correlation algorithms are as follows:

[0058] Step (1): Combining the underwater robot dynamics model and the thruster dynamics model, the longitudinal signal thrust deviation curve is obtained based on the longitudinal velocity and the longitudinal thruster control voltage signal, and the bow signal thrust deviation curve is obtained based on the bow angle and the lateral thruster control voltage.

[0059] The longitudinal and bow dynamics models of multi-thrust underwater robots can generally be described by equations (1) and (2):

[0060]

[0061]

[0062] Where m is the mass of the underwater robot; u is the longitudinal velocity of the underwater robot; T is the longitudinal thrust of the underwater robot; r is the heading angle of the underwater robot; and τ is the heading torque of the underwater robot. X u X u|u| , N r N r|r| It is the correlation coefficient.

[0063] Based on formulas (1) and (2), the actual thrust T generated by the propeller can be obtained from the longitudinal velocity. z The bow torque τ acting on the underwater robot is calculated based on the bow angle, and the actual thrust T generated by the thruster is calculated based on the torque and the thruster lever arm length. s .

[0064] The dynamic model of an underwater robot's thruster can generally be described by the following equations:

[0065] T″ z =T′-T z (3)

[0066] Where T′ is the theoretical thrust, V is the thruster control voltage, and c and d are correlation coefficients.

[0067] The actual thrust T of the thruster was obtained. z T s The difference between the thrust and the theoretical thrust T′ yields the thrust deviation based on the longitudinal and bow signals, as shown in the following formula:

[0068] T″ z =T′-T z (4)

[0069] T″ s =T′-T s (5)

[0070] Among them, T″ z It is the longitudinal signal thrust deviation, T″ s It is the forward thrust deviation.

[0071] Step (2): Combining the results of step (1), the fault characteristics of the longitudinal and bow signals are obtained based on the thrust deviation curves of the longitudinal and bow signals using the multi-time window Fourier transform and fusion method;

[0072] According to the sliding time window Fourier transform formula:

[0073]

[0074] Among them, X n The result is the Fourier transform, k is the sequence number of the transform result, representing the order of frequencies in the spectrum from low to high. When k=1, it represents the lowest frequency transform result. M is the number of signals within the time window. W M =e j2 π / M .

[0075] Extract the lowest frequency amplitude from the sliding time window Fourier transform, i.e., the amplitude when k=1, and then... z Substitute into formula (6) and calculate the lowest frequency amplitude value, as shown in the formula below:

[0076] Using a sliding time window to measure the longitudinal signal thrust deviation T″ z The resulting curve is truncated, and then a Fourier transform is performed on the truncated portion. The lowest frequency amplitude of the transform result is extracted, that is, the amplitude when k=1. T″ z Substitute into formula (5) and calculate the lowest frequency amplitude value, as shown in the formula below:

[0077]

[0078] Where f is the lowest frequency amplitude.

[0079] By using multiple time windows, i.e. taking multiple M values, multiple lowest frequency amplitude curves are obtained.

[0080] The lowest frequency curve is shifted forward by half a time window along the time axis and used as a fault characteristic, as shown in the following formula:

[0081] d z (n)=f(nM / 2) (8)

[0082] Where, d z For longitudinal signal fault characteristics, n is the sequence number of the signal.

[0083] According to formula (8), the curves formed by the lowest frequency amplitudes under multiple time windows are shifted forward by half a time window length along the time axis to obtain multiple fault characteristic curves. These curves are then fused using the DS evidence theory method to obtain the fused longitudinal signal fault characteristic m. FZ .

[0084] The above method is used to determine the thrust deviation T based on the bow signal. s "Obtain the fault characteristics m of the fused heading signal" FS .

[0085] Step (3): Combining the results of step (2), the cross-correlation fault features of longitudinal and heading signal fault features are obtained by using the sliding time window cross-correlation method, and then fused with the two fault features to obtain the final fault features.

[0086] A sliding time window is used to extract the two fault characteristics, as shown in the following formula.

[0087] u(n) = m FZ (r), r∈[nM / 2,n+M / 2] (9)

[0088] v(n) = m FS (r), r∈[nM / 2,n+M / 2] (10)

[0089] Where u(n) and v(n) are the longitudinal and forward signal fault features captured at time n, n and r are the signal sequence numbers, and m FZ m FS The fault characteristics are represented by longitudinal and bow signals, and M is the number of signals within the time window.

[0090] Calculated using the cross-correlation function formula:

[0091]

[0092] in, F represents the cross-correlation function values ​​of functions u and v. s The sampling frequency is m, where m is an integer related to the delay, and the delay value is m / F. s .

[0093] The fault characteristic curves of longitudinal and bow signals are extracted using a sliding time window. According to formula (11), the cross-correlation operation is performed on the extracted curves to obtain the cross-correlation curve at time n. The formula is as follows:

[0094]

[0095] Based on the obtained cross-correlation curve, extract the maximum value of the cross-correlation curve using the following formula.

[0096]

[0097] Where, m FH (n) represents the maximum value of the cross-correlation curve between u(n) and v(n).

[0098] Based on formula (13), the maximum cross-correlation curves of longitudinal and bow signal fault characteristics can be obtained, which can be used as cross-correlation fault characteristic curves.

[0099] The fault characteristic curves are obtained by fusing longitudinal, heading, and cross-correlation fault characteristic curves using the DS evidence theory method.

[0100] Application examples of this invention are as follows:

[0101] To verify the effectiveness of the underwater robot thruster fault feature extraction method based on thrust deviation and cross-correlation algorithm designed in this invention, the following comparative experiment was designed:

[0102] 1) In a simulated underwater robot operating in a dry environment, the left thruster experiences a 20% thrust loss failure after 40 seconds. The traditional method is used to fuse the longitudinal velocity and longitudinal thruster control voltage signals to obtain the final fault features. These features are then compared and verified with the fault feature fusion method based on cross-correlation algorithm designed in this invention, which extracts fault features from four signals: longitudinal velocity, longitudinal thruster control voltage, heading angle, and lateral thruster control voltage.

[0103] 2) In a simulated underwater robot operating in a flowing environment, the left thruster experiences a 10% thrust loss failure after 40 seconds. The traditional method is used to fuse the longitudinal velocity and longitudinal thruster control voltage signals to obtain the final fault features. These features are then compared and verified with the fault feature fusion method based on cross-correlation algorithm designed in this invention, which extracts fault features from four signals: longitudinal velocity, longitudinal thruster control voltage, heading angle, and lateral thruster control voltage.

[0104] In the simulation experiment verification process, the same original data were used in both the traditional method and the method designed in this invention, and the verification carrier was the "Beaver II" underwater robot.

[0105] The comparison results obtained by the conventional method and the method designed in this invention are as follows: Figure 2 , 3 As shown.

[0106] from Figure 2 Based on the given thruster fault characteristics, both the traditional method and the method of this invention produce obvious peaks after the fault. The fault occurrence lag time, determined by the peak position, is 1.2 seconds, lower than the 11.8 seconds obtained by the traditional method. The ratio of the peak value of the fault characteristic wave to the peak value of the interference wave is 8.51 × 10⁻⁶. 5 The accuracy is also significantly higher than the traditional method's 13.07, verifying that the method of the present invention has significant improvements in terms of the accuracy of fault detection and the ratio of fault characteristics to interference characteristics.

[0107] from Figure 3 Based on the given thruster fault characteristics, the fault characteristics obtained using the method of this invention produce obvious peaks after the fault. The time lag between the fault occurrence and the peak position is determined to be 3.0 seconds. The ratio of the peak value of the fault characteristic wave to the peak value of the interference wave is 12.38, which can reliably detect the fault. In contrast, the ratio of the peak value of the fault characteristic wave to the peak value of the interference wave in the traditional method is 0.44, which cannot detect the fault. This verifies the effectiveness of the method of this invention when the fault level is low.

[0108] In summary, this invention studies a method for extracting fault features from underwater robot thrusters. It achieves fault feature extraction from multiple signals even when the fault severity of the underwater robot thruster is relatively weak in ocean current environments. Compared with traditional methods, it has significant improvements in terms of positioning accuracy at the time of fault occurrence and the ratio of fault features to interference features. Moreover, it achieves fault feature extraction from signals related to different degrees of freedom, which is of great significance for underwater robot fault diagnosis.

Claims

1. A method for underwater robot thruster fault feature extraction based on thrust deviation and cross-correlation algorithm, characterized in that, The steps are as follows: Step (1): based on the longitudinal velocity and the longitudinal thruster control voltage signal, the longitudinal signal thrust deviation curve is obtained by combining the underwater robot dynamics model and the thruster dynamics model, and based on the heading angle and the lateral thruster control voltage, the heading signal thrust deviation curve is obtained; The longitudinal and heading dynamics model of the multi-thruster driven underwater robot is: wherein m is the mass of the underwater robot; u is the longitudinal velocity of the underwater robot; T is the longitudinal thrust force received by the underwater robot, r is the bow angle of the underwater robot; τ is the bow torque received by the underwater robot; X u 、X u|u| 、 N r 、N r|r| is a correlation coefficient; The actual thrust T generated by the thruster is obtained from the longitudinal velocity z The actual thrust T generated by the thruster is obtained from the longitudinal velocity s ; The thruster dynamics model of the underwater robot is: T'=cV+d Wherein, T' is the theoretical thrust, V is the thruster control voltage; c, d are the relevant coefficients; Calculate the actual thrust T of the thruster z T s The difference between the thrust and the theoretical thrust T′ yields the thrust deviation based on the longitudinal and bow signals: T''=T'-T T''=T'-T where T z " is the longitudinal signal thrust deviation, T s " is the bow signal thrust deviation; Step (2): based on the results of step (1), the longitudinal and heading signal thrust deviation curves are obtained by using the multi-time window Fourier transform and fusion method based on the longitudinal and heading signal thrust deviation curves; Step (3): based on the results of step (2), the cross-correlation fault feature of the longitudinal and heading signal fault features is obtained by using the sliding time window cross-correlation method, and the final fault feature is obtained by fusing the two fault features.

2. The method according to claim 1, wherein the method is characterized in that: The step (2) is specifically: According to the sliding time window Fourier transform formula: wherein X n is the Fourier transform result, k is the sequence number of the transform result, represents the arrangement order of the frequency from low to high in the frequency spectrum, k = 1 represents the lowest frequency transform result, and M is the number of signals in the time window, W M = e j2π / M ; The lowest frequency amplitude value in the sliding time window Fourier transform, i.e. the amplitude value when k = 1, is extracted, and T z "Substitute into formula (6), and calculate the lowest frequency amplitude value. The longitudinal signal thrust deviation T z "Substitute into formula (6), and calculate the lowest frequency amplitude value. The longitudinal signal thrust deviation T z "Substitute into formula (6), and calculate the lowest frequency amplitude value. The longitudinal signal thrust deviation T Wherein, f is the lowest frequency amplitude; A plurality of time windows are used, that is, a plurality of M values are taken, and a plurality of lowest frequency amplitude curves are obtained; The lowest frequency curve is translated along the time axis by half the time window length, which is used as the fault feature, and the formula is as follows: d z (n) = f(n - M / 2) (8) wherein d z is a longitudinal signal fault feature, n is a sequence number of the signal; According to formula (8), the curve composed of the minimum frequency amplitudes under multiple time windows is translated forward along the time axis by half the length of the time window to obtain multiple fault characteristic curves, and the D-S evidence theory method is used for fusion to obtain the fused longitudinal signal fault characteristic m FZ ; the thrust deviation T s of the heading signal is obtained according to the heading signal FS .

3. The method according to claim 1, wherein the method is characterized in that: The step (3) is specifically: The two fault features are intercepted by using the sliding time window, and the formula is as follows: u(n) = m FZ (r), r e [n - M / 2, n + M / 2] (9) v(n) = m FS (r), r e [n - M / 2, n + M / 2] (10) Wherein, u(n), v(n) are longitudinal and bow direction signal fault features at n time, n, r are signal sequence numbers, m FZ , m FS are longitudinal and bow direction signal fault features, M is the number of signals in the time window; According to the cross-correlation function calculation formula: wherein is a cross-correlation function value of functions u and v, F s is a frequency of adoption, m is an integer related to a delay, and a delay size is m / F s ; The longitudinal and heading signal fault feature curves are extracted by using the sliding time window, and according to formula (11), the cross-correlation operation is performed on the intercepted curves, and the cross-correlation curve at time n is obtained as follows: According to the obtained cross-correlation curve, the maximum value of the cross-correlation curve is extracted as: where m FH (n) is the maximum value of the cross-correlation of u(n) and v(n) Based on formula (13), the maximum value curve of the longitudinal and heading signal fault feature cross-correlation is obtained, which is used as the cross-correlation fault feature curve; the longitudinal, heading signal and cross-correlation fault feature curves are fused by using the D-S evidence theory method, and the final fault feature curve is obtained.

Images