Fault diagnosis method and diagnosis system for underwater propeller based on predicted tact dynamic adjustment

Abstract

The present invention discloses a method for diagnosing faults of an underwater propeller and a diagnosing system for the underwater propeller based on predictive tact dynamic adjustment, which collects dynamic signals of an underwater robot, establishes a grey prediction model, obtains the time series of the dynamic signals at the current time, performs grey prediction on the time series to obtain a prediction trajectory, extracts fault features from the prediction trajectory based on a modified Bayesian algorithm to obtain a fault feature prediction sequence, selects one characteristic value of a predictive tact that changes based on the slope of the fault feature prediction sequence from the fault feature prediction sequence as the fault characteristic value at the current time, substitutes the selected fault characteristic value into a fault identification model to perform underwater propeller fault identification, and obtains the fault degree of the underwater propeller at the current time. By dynamically adjusting the predictive tact when diagnosing faults of the propeller, the fault identification accuracy can be improved, and the mean absolute error, mean relative error, and root mean square error between the identification result and the actual fault degree can be reduced.

Description

Technical Field

[0001] The present invention relates to the fault detection of an underwater robot propeller, and specifically to a fault diagnosis method and system for an underwater propeller based on predictive tact dynamic adjustment.

Background Art

[0002] With the development of science and technology, underwater robots have unique advantages as the main equipment for ocean resource exploration. For an underwater robot, the propeller, as an important component, plays an important role in the process of ocean exploration. To determine whether the underwater robot is operating normally, it is necessary to monitor the state of the propeller.

[0003] Regarding the state monitoring of the propeller, in the prior art, for example, the Chinese patent application with the publication number CN107132760A discloses an underwater robot state monitoring method based on fuzzy support vector region description. By constructing a monitoring model and substituting underwater robot propeller data samples into the monitoring model, it is possible not only to determine the presence or absence of propeller faults but also to determine the degree of propeller faults. However, since the underwater robot is a large-inertia system, there is a transition stage from when the propeller fails to when the target state is re-tracked by closed-loop control. In this transition stage, the identification accuracy of the fault degree is limited in this manner.

[0004] For example, in the paper "Fault degree identification method for thruster of autonomous underwater vehicle using homomorphic membership function and low frequency trend prediction" published in the Proceedings of The Institution of Mechanical Engineers Part C-Journal of Mechanical Engineering Science in the prior art, a method for predicting and compensating the identification result of propeller fault degree is proposed, and the fault identification accuracy is improved by means of single-pass least squares prediction and fixed prediction tact. However, in this aspect, the improvement of fault identification accuracy is limited.

Summary of the Invention

Problems to be Solved by the Invention

[0005] Object of the Invention: Regarding the above-mentioned drawbacks, the present invention provides a fault diagnosis method for an underwater propeller based on dynamic adjustment of prediction tact to improve fault identification accuracy.

[0006] The present invention further provides a fault diagnosis system for an underwater propeller based on dynamic adjustment of prediction tact.

Means for Solving the Problems

[0007] Technical Solution: To solve the above problems, the present invention adopts a fault diagnosis method for an underwater propeller based on dynamic adjustment of prediction tact, and this method includes: Step (1) of collecting dynamic signals of an underwater robot; Step (2) of establishing a grey prediction model, obtaining the time series of the dynamic signal at the current time, performing grey prediction on the time series, and obtaining a prediction trajectory; Step (3) of extracting fault features from the prediction trajectory based on the modified Bayes algorithm to obtain a fault feature prediction sequence; Obtain the slope K of the fault feature prediction sequence, select the feature value of the Nth prediction tact in the fault feature prediction sequence, and the calculation formula of N is as follows: N(K)=aK b However, a and b are constants obtained through experiments. Step (4) of selecting the feature value of one prediction tact from the fault feature prediction sequence as the fault feature value at the current time, Substitute the selected fault feature value into the fault identification model established by the underwater propeller fault test to perform underwater propeller fault identification and obtain the fault degree of the underwater propeller at the current time in step (5), Obtain the time series of the next time tact, repeat steps (2) to (5), and include step (6) of obtaining the fault degree of the propeller at the next time tact.

[0008] Furthermore, perform multi-path grey prediction on the dynamic signal in step (2), establish a multi-path grey fusion prediction model, obtain the time series of the dynamic signal at the current time, perform overlapping sampling on the time series to obtain multiple single-path sequences, perform grey prediction on each single-path sequence to obtain multiple single-path prediction results, and fuse the multiple single-path prediction results into the prediction trajectory of one path based on K-means clustering analysis.

[0009] Furthermore, the specific steps of obtaining the time series of the dynamic signal at the current time in step (2) and performing overlapping sampling on the time series to obtain multiple single-path sequences are as follows: (2.1) Cut out the dynamic signal using a time window with a length of L, sequentially extract L data from the current time to the left, and obtain a time series with a length of L: {u(n)} = [u(1) u(2) ··· u(L)].[[]]END]] (2.2) Perform overlapping sampling on the time series {u(n)} = [u(1) u(2) ··· u(L)] to obtain multiple single-path sequences. The data length of each single-path sequence is l, and there are a total of L - l + 1 single-path sequences, that is, {u 1(n)} = [u(1) u(2) … u(l)] {u 2 (n)} = [u(2) u(3) … u(l + 1)],... {u (L-l+1) (n)} = [u(L - l + 1) u(L - l + 2) … u(L)].

[0010] (2.3) For each single - path sequence {u i (n)}, perform wavelet decomposition to obtain the wavelet scale component {u iA (n)} = [u iA (i) u iA (i + 1) … u iA (l + i - 1)] and the wavelet detail component {u iD (n)} = [u iD (i) u iD (i + 1) … u iD (l + i - 1)], where i = 1, 2, …, L - l + 1, (2.4) Based on the grey prediction theory, perform forward prediction on the wavelet scale component {u iA (n)} for Q × l - tact data to obtain the predicted data sequence {u iAP (n)}.

Number

[0011] The present invention further provides a fault diagnosis system for an underwater propeller based on predictive tact dynamic adjustment. This system includes a collection module for collecting the dynamic signals of an underwater robot, a model - establishment module for establishing a grey prediction model, obtaining the time - series of the dynamic signal at the current time, performing grey prediction on the time - series, and obtaining a predicted trajectory, Extract fault features from the predicted trajectory based on the modified Bayesian algorithm to obtain a fault feature prediction sequence, obtain the slope K of the fault feature prediction sequence, select the feature value of the Nth prediction tact in the fault feature prediction sequence, and the calculation formula of N is as follows: N(K)=aK b However, a and b are constants obtained through experiments. A fault feature value acquisition module for selecting the feature value of one prediction tact from the fault feature prediction sequence as the fault feature value at the current time, and a fault identification module that substitutes the selected fault feature value into a fault identification model established by an underwater propeller fault test to perform underwater propeller fault identification and obtain the fault degree of the underwater propeller.

Advantages of the Invention

[0012] Beneficial effects: Compared with the prior art, the remarkable advantage of the present invention is that by dynamically adjusting the prediction tact during the fault diagnosis of the propeller, the fault identification accuracy can be improved, and the mean absolute error, mean relative error, and root mean square error between the identification result and the actual fault degree can be reduced.

Brief Description of the Drawings

[0013]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Embodiment for Carrying out the Invention

[0014] As shown in FIG. 1, it is a method for diagnosing the failure of an underwater propeller based on single-pass grey prediction and tact dynamic adjustment in this embodiment. As shown in FIG. 2, it is a method for diagnosing the failure of an underwater propeller based on predictive tact dynamic adjustment and multi-pass grey fusion prediction. In this embodiment, the multi-pass grey fusion prediction in FIG. 2 will be taken as an example for explanation.

[0015] As shown in FIG. 2, it is a method for diagnosing the failure of an underwater propeller based on multi-pass grey fusion prediction in this embodiment, and the specific steps are as follows.

[0016] In the first step, conduct an underwater robot propeller failure test and record the dynamic signals of the underwater robot including dynamic signals such as the propeller control voltage signal and the vertical direction speed signal of the underwater robot.

[0017] In the second step, based on the modified Bayesian algorithm, extract the failure characteristics from the obtained dynamic signals, and establish a failure identification model using the failure characteristics based on the fuzzy support vector region description algorithm.

[0018] In the third step, collect the dynamic signals of the underwater robot in real time, adopt a time window with a length of L, cut out the collected dynamic signals, extract L data sequentially from the current time to the left, and obtain a time series {u(n)} = [u(1) u(2) … u(L)] with a length of L.

[0019] In the fourth step, overlapping sampling is performed on the extracted time series {u(n)} = [u(1) u(2) … u(L)], obtaining a plurality of single-path sequences. The data length of each single-path sequence is l, and there are a total of L - l + 1 single-paths, that is, as follows: {u 1 (n)} = [u(1) u(2) … u(l)], {u 2 (n)} = [u(2) u(3) … u(l + 1)],..., {u (L-l+1) (n)} = [u(L - l + 1) u(L - l + 2) … u(L)].

[0020] As shown in Figure 1, in the fault diagnosis method of the underwater propeller with single-path gray prediction, without performing overlapping sampling on the time series in this step, wavelet decomposition in the fifth step is directly performed on the directly extracted time series.

[0021] In the fifth step, wavelet decomposition is performed on the first single-path sequence {u 1 (n)}, obtaining the wavelet scale component {u 1A (n)} = [u 1A (1) u 1A (2) … u 1A (L)] and the wavelet detail component {u 1D (n)} = [u 1D (1) u 1D (2) … u 1D (L)].

[0022] In the sixth step, based on the gray prediction theory, Q × l tact data of the wavelet scale component {u 1A (n)} is predicted forward, obtaining the predicted data sequence {u 1AP (n)}.

Number

[0023] The specific content of the grey prediction theory is as follows. The original sequence x (0) (k) is successively added to generate a new sequence x (1) (k), and adjacent values are generated for the sequence x (1) (k) obtained during the addition generation process, z (1) (k)=αx 1 (k)-(1-α)x (1) (k - 1) where α, also called the generation coefficient, satisfies α∈[0,1], and z (1) (k) is the whitening background value.

[0024] The differential equation x (0) (k)+βz (1) (k)=θ is established, where β is the development coefficient and θ is the grey action quantity.

Number

[0025] The grey prediction model can also be expressed as Y = Bu. The minimum prediction values of b and q are obtained by the least squares method

Number

Number

Number

[0026] In the seventh step, the wavelet detail components {u 1AP (n)} of the original data are superimposed on the prediction data sequence {u 1D (n)} to obtain a prediction trajectory {u 1APR (n)} with the amplitude value reconstructed.

Number

[0027] In the 9th step, based on the K-means clustering analysis, fuse the multi-path prediction results {u 1APR (n)}, {u 2APR (n)}, {u 3APR (n)}, ……, {u (L-l+1)APR (n)} into one path {u F (n)}.

[0028] In the 10th step, extract the fault characteristics from the fused path {u F (n)} based on the modified Bayesian algorithm to obtain the fault feature prediction sequence {F F (n)} = [F F (1) F F (2) … F F (Q×l)].

[0029] In the 11th step, calculate the slope K of the fault feature prediction sequence {F F (n)}.

[0030] In the 12th step, select the feature value F F (n) of the Nth prediction cycle in the fault feature prediction sequence {F F (n)} as the fault feature value at the current time. The selection of N is related to the slope K of the fault feature prediction sequence {F F (n)}. The larger the slope K, the smaller N is, and the smaller the slope K, the larger N is. The relationship between N and K is shown in Equation (12). N(K) = aK b (12)

[0031] Where K is the slope, and a and b are constants, and specific numerical values are obtained through multiple tests. In the 13th step, the fault characteristic value F F (N) is substituted into the fault identification model to perform propeller fault identification and obtain the degree of the propeller fault at the current time.

[0032] In the 14th step, for the collected dynamic signal, the time window with a length of L is moved one time tick to the right, and the 3rd step to the 13th step are repeated to identify the degree of the propeller fault at the next time tick.

[0033] As shown in FIGS. 3 and 4, in this embodiment, an underwater robot propeller fault test is performed to obtain dynamic signals such as the propeller control voltage signal and the longitudinal velocity signal of the underwater robot. During the test process, the underwater robot target speed of 0.4 m / s is set by closed-loop control, starting from rest, the speed of the underwater robot gradually increases, and at the 100th tick, the underwater robot reaches 0.4 m / s, and then starts operating at a constant speed of 0.4 m / s. At the 300th tick, the propeller generates a slow-changing fault until the test ends.

[0034] As shown in FIG. 5, multi-pass gray fusion and predictive tick dynamic adjustment are performed on the collected dynamic signal. As shown in FIG. 5(a), a time window with a length of L = 180 is adopted to cut out the data from the 251st tick to the 430th tick in FIG. 3 as the sample data {u(n)}.

[0035] As shown in FIG. 5(b), overlapping sampling is performed on the data {u(n)} in FIG. 5(a) to obtain a plurality of single paths. The data length of each single path is set to l = 160, and there are a total of L - l + 1 = 21 single paths {u 1 (n)}, {u 2 (n)}, ……, {u 2l (n)}.

[0036] As shown in FIG. 5(c), the first single-path sequence {u 1Perform wavelet decomposition on {u(n)} to obtain the wavelet scale component {u 1A (n)} and the wavelet detail component {u 1D (n)}.

[0037] As shown in Figure 5(d), based on the grey prediction theory, perform forward prediction on the wavelet scale component {u 1A}(n) with Q×l = 9×160 = 1440 clock cycles of data to obtain the predicted data sequence {u 1AP}(n).

[0038] As shown in Figure 5(e), superimpose the wavelet detail component {u 1AP}(n) of the original data on the predicted data sequence {u 1D}(n) to obtain the predicted trajectory {u 1APR}(n) whose amplitude value is reconstructed.

[0039] As shown in Figure 5(f), use the same method to perform prediction and amplitude value reconstruction on single paths {u 2}(n), {u 3}(n), ……, {u (L-l+1)}(n) respectively to obtain multiple single path prediction results, namely {u 2APR}(n), {u 3APR}(n), ……, {u (L-l+1)APR}(n).

[0040] As shown in Figure 5(g), based on the K - means clustering analysis, fuse the multi - path prediction results {u 1APR}(n), {u 2APR}(n), {u 3APR}(n), ……, {u (L-l+1)APR}(n) into one path {u F}(n).

[0041] As shown in Figure 5(h), extract the fault features from the fused path {u F}(n) based on the modified Bayesian algorithm to obtain the fault feature prediction sequence {F F}(n), and the fault feature prediction sequence {F FCalculate the slope K = 4.438 of (n), and calculate N(K)=2902.2K -0.827 = 846 is obtained, and 846 is the rounded data. The fault feature prediction sequence {F F (n)} The feature value F of the 846th prediction cycle F (N)=144.857 is selected, and the fault feature value F F (N)=144.857 is substituted into the fault identification model to perform propeller fault identification, and a fault degree of 22.374% is obtained, and this fault degree of 22.374% is used as the propeller fault degree identification result at the current actual time (the 430th cycle).

[0042] As shown in FIG. 6, it is a fault degree diagram of a fault diagnosis method for an underwater propeller based on single-pass gray prediction and prediction cycle dynamic adjustment, and it is closer to the actual fault degree than the known method 1 and the known method 2 provided in the background art.

[0043] As shown in FIG. 7, it is a fault degree diagram of a fault diagnosis method for an underwater propeller based on multi-pass gray fusion prediction and prediction cycle dynamic adjustment. For the problem that the fault degree identification results obtained by the known method 1 and the known method 2 lag behind the actual fault degree, through the experimental error index, the effectiveness of reducing the lag of the fault identification result of the method of this embodiment is compared. The error calculation formula is as follows.

[0044] Mean absolute error:

Number

Number

Number

[0045] The error index analysis table is as follows.

Table 1

[0046] In the known method 1, each of the mean absolute error, mean relative error, and root mean square error is 0.156, 0.759, and 0.091, respectively. In the known method 2, each of the errors is 0.069, 0.350, and 0.064, respectively. In the single-pass grey prediction method, they are 0.023, 0.165, and 0.050, respectively, and the effect is better than that of the known method 1 and the known method 2. In the single-pass grey prediction, the fault identification accuracy can be improved to a certain extent by the prediction cycle dynamic adjustment. In the multi-pass grey fusion prediction method, they reach 0.017, 0.129, and 0.043, respectively, and the effect is better than that of the known method 1, the known method 2, and the single-pass grey prediction method.

Claims

1. A fault diagnosis method for an underwater propeller based on predictive tact dynamic adjustment, comprising: (1) collecting dynamic signals of an underwater robot; (2) establishing a grey prediction model, obtaining a time series of dynamic signals at a current time, and performing grey prediction on the time series to obtain a predicted trajectory; (3) extracting fault features from the predicted trajectory based on a modified Bayes algorithm to obtain a fault feature prediction sequence; Obtain the slope K of the fault feature prediction sequence, and select the feature value of the Nth predicted tact in the fault feature prediction sequence, and the calculation formula of N is as follows: N(K)=aK ｂ where a and b are constants obtained by experiment, and a step (4) of selecting a feature value of one predicted tact from the fault feature prediction sequence as a fault feature value at a current time; (5) substituting the selected fault feature value into a fault identification model established by underwater propeller fault testing to perform underwater propeller fault identification and obtain the fault degree of the underwater propeller at the current time; and (6) obtaining a time series of the next time tact, repeating steps (2) to (5), and obtaining a fault degree of the next time tact of the propeller.

2. 2. The fault diagnosis method for an underwater propeller according to claim 1, wherein in step (2), multi-path grey prediction is performed on the dynamic signal, a multi-path grey fusion prediction model is established, a time series of the dynamic signal at the current time is obtained, convoluted sampling is performed on the time series to obtain a plurality of single-path sequences, grey prediction is performed on each single-path sequence to obtain a plurality of single-path prediction results, and the plurality of single-path prediction results are fused into a single-path prediction trajectory based on K-means clustering analysis.

3. The specific content of establishing the fault identification model is as follows: carry out an underwater propeller fault test, obtain the dynamic signal of the underwater robot, extract fault features from the dynamic signal based on the modified Bayes algorithm, and establish the fault identification model using the fault features based on the fuzzy support vector domain description algorithm. The fault diagnosis method for underwater propellers as described in claim 1.

4. The specific steps of obtaining a time series of a dynamic signal at a current time in step (2), performing overlap sampling on the time series, and obtaining a plurality of single-path sequences are as follows: (2.1) A time window of length L is used to extract the dynamic signal, and L pieces of data are extracted sequentially from the current time to the left, to obtain a time series of length L: {u(n)} = [u(1) u(2) ... u(L)]; (2.2) Perform superposition sampling on the time series {u(n)} = [u(1) u(2) ... u(L)] to obtain multiple single-pass sequences, each of which has a data length of l, and there are L-l+1 single-pass sequences in total, i.e., {u １ (n)}=[u(1) u(2) ... u(l)], {u ２ (n)}=[u(2) u(3) ... u(l+1)], {u （Ｌ－ｌ＋１） 3. The fault diagnosis method for an underwater propeller according to claim 2, wherein {u(L-l+1) u(L-l+2) ... u(L)].

5. The specific steps of performing grey fusion prediction for each single-pass sequence as described above are as follows: (2.3) For each single-pass sequence {u ｉ (n)} is subjected to wavelet decomposition to obtain the wavelet scale component {u ｉＡ (n)}=[u ｉＡ (i) u ｉＡ (i+1) … u ｉＡ (l+i−1)] and wavelet detail components {u ｉＤ (n)}=[u ｉＤ (i) u ｉＤ (i+1) … u ｉＤ (l+i-1)], where i=1, 2..., L-l+1; (2.4) Based on the gray prediction theory, the wavelet scale component {u ｉＡ (n)}, and forward predicts the data of Q×1 tacts, and the predicted data sequence {u ｉＡＰ (n)}, ##EQU00011## (2.5) The predicted data sequence {u ｉＡＰ (n)} of the wavelet detail components of the original sequence {u ｉＤ (n)} is convolved to reconstruct the width value of the predicted trajectory {u ｉＡＰＲ (n)} is obtained.

6. The gray prediction theory is specifically as follows: Original sequence x （０） (k) to generate a new sequence x （１） (k) and generate the new sequence x （１） Generate adjacent values ​​for (k); z （１） (k)=αx （１） (k)-(1-α)x （１） (k-1) where α is a generation coefficient, α∈[0,1], and z （１） (k) is the whitened background value, Differential equation: x （０） (k) + βz （１） Establish k = θ, where β is the evolution coefficient and θ is the grey function; Establish a grey prediction model, Y=Bu, where: ##EQU00012## The minimum predicted value of b and q by the least squares method ##EQU00013## and hence the corresponding whitened differential equation ##EQU14## Then, we obtain the prediction equation by solving the whitening differential equation. ##EQU00015## 6. The fault diagnosis method for an underwater propeller according to claim 5, further comprising: performing cumulative reduction on the prediction equation to obtain a prediction sequence.

7. Based on the K-means clustering analysis, multiple single-pass prediction results {u １ＡＰＲ (n)}, {u ２ＡＰＲ (n)}, {u ３ＡＰＲ (n)}, ..., {u （Ｌ－１＋１）ＡＰＲ (n)} along one path {u Ｆ (n)}, Based on the modified Bayes algorithm, the fusion path {u Ｆ (n)}, and a fault feature prediction sequence {F Ｆ (n)}=[F Ｆ (1) F Ｆ (2) … F Ｆ 6. The fault diagnosis method for an underwater propeller according to claim 5, further comprising the step of obtaining a fault vector x of the underwater propeller.

8. a collection module for collecting dynamic signals of the underwater robot; a model establishment module for establishing a grey prediction model, obtaining a time series of dynamic signals at a current time, performing grey prediction on the time series, and obtaining a predicted trajectory; Extract fault features from the predicted trajectory according to the modified Bayes algorithm, obtain a fault feature prediction sequence, obtain the slope K of the fault feature prediction sequence, and select the feature value of the Nth predicted tact in the fault feature prediction sequence, where the calculation formula of N is as follows: N(K)=aK ｂ where a and b are constants obtained by experiment, and a fault feature value acquisition module for selecting a feature value of one predicted tact from the fault feature prediction sequence as a fault feature value at a current time; A diagnosis system for the underwater propeller fault diagnosis method described in any one of claims 1 to 7, characterized in that it includes a fault identification module for substituting the selected fault feature value into a fault identification model established by an underwater propeller fault test, performing underwater propeller fault identification, and obtaining the fault degree of the underwater propeller.

9. A computing device comprising a memory, a processor and a computer program stored in the memory and executable by the processor, the computing device implementing the steps of the method according to any one of claims 1 to 7 when the processor executes the computer program.

10. A computer-readable storage medium having a computer program stored thereon, the computer program implementing the steps of the method according to any one of claims 1 to 7 when executed by a processor.

Images