Intelligent belt conveyor carrier roller fault monitoring method based on DAS

Abstract

The invention discloses an intelligent belt conveyor carrier roller fault monitoring method based on a DAS. The method comprises the steps that a distributed optical fiber sensing system DAS is used for collecting whole-line carrier roller vibration signals; decomposing the vibration signal into a series of components by using wavelet decomposition, and calculating a correlation coefficient of each component and an autocorrelation function of the original vibration signal; effective components are selected for denoising processing and superposition to obtain denoised signals; calculating a time-frequency domain characteristic parameter and a KL distance between a denoising fault and a normal signal sample, screening out six characteristic parameters with the highest resolution capability, and training a built machine learning model; and inputting carrier roller vibration signals collected by the whole line of the belt conveyor to obtain six characteristic parameters, and sending the six characteristic parameters to the trained machine learning model to obtain a carrier roller fault prediction result at each position. According to the method, the limitation of single-domain fault analysis is overcome, the robustness to noise and variable working conditions is effectively improved, and intelligent positioning and monitoring of the whole-line carrier roller faults of the belt conveyor are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of belt conveyor idler fault monitoring, and particularly relates to an intelligent belt conveyor idler fault monitoring method based on DAS. BACKGROUND

[0002] The conveyor is the core conveying equipment in the coal, electric power and other industries, and the idler thereof has a high failure rate (the failure rate is more than 60%) due to long-term bearing of high load and strong dust environment. The traditional detection methods (such as manual inspection and point-type accelerometer) have three major defects, i.e., limited coverage (a single-point sensor is difficult to monitor a conveying line of several kilometers, and intensive arrangement leads to high cost), poor real-time performance (manual inspection has a long cycle and cannot capture sudden faults), and weak anti-interference (industrial electromagnetic noise easily overwhelms the vibration signal, and the false alarm rate is high).

[0003] The distributed optical fiber sensing (DOFS) technology has the unique advantage of realizing long-distance continuous monitoring with a single optical fiber, and has characteristics such as anti-electromagnetic interference and corrosion resistance, and thus has great potential in the field of industrial equipment state monitoring. However, the existing technology still has significant limitations in the application of belt conveyor idler fault monitoring. First, the Chinese patent application for invention with the publication number CN116484196A mainly focuses on optical fiber vibration area identification and fails to optimize the idler fault characteristics. Second, the Chinese patent for invention with the publication number CN114039656B involves fault positioning, but only relies on a single monitoring parameter and does not introduce an artificial intelligence algorithm, resulting in insufficient positioning accuracy (less than 90%). In addition, the Chinese patent for invention with the publication number CN112033669B only uses the frequency domain feature comparison with a reference signal for fault judgment, and this method has three obvious defects: first, the influence of environmental noise on fault judgment is not considered, second, single frequency domain analysis loses other key features of the fault signal, and third, the lack of intelligent algorithm support easily leads to misjudgment.

[0004] In view of the above technical bottlenecks, in order to fully realize the practical application value of the DOFS technology in the belt conveyor idler fault monitoring, an innovative solution is urgently needed: the solution should be able to effectively overcome the interference of environmental noise, realize multi-domain feature fusion extraction, integrate artificial intelligence diagnosis technology, and finally realize accurate positioning and intelligent identification of the idler fault. SUMMARY

[0005] The problem to be solved by the present application is to provide a DAS-based intelligent belt conveyor roller fault monitoring method, which overcomes the problem that point sensors are difficult to network monitor the belt conveyor by using distributed optical fiber sensing technology, overcomes the problem that noise overwhelms the fault characteristics by using multi-algorithm joint denoising, overcomes the limitation of single domain feature representation by using multi-domain feature extraction, accurately captures the transient impact and periodic characteristics of the fault, and finally fuses artificial intelligence technology, uses the high discrimination characteristics of multi-domain features, and constructs an intelligent diagnosis model with strong generalization ability. Not only does it overcome the limitations of single domain analysis of faults, but also effectively improves the robustness to noise and variable working conditions, and ultimately realizes intelligent positioning and monitoring of belt conveyor roller faults.

[0006] The present application adopts the following technical scheme: a DAS-based intelligent belt conveyor roller fault monitoring method, comprising the following steps:

[0007] S1, the sensing optical cable is arranged on the side support of the belt conveyor roller group, a distributed optical fiber sensing system DAS is used to inject an optical pulse into the sensing optical cable, a backscattering Rayleigh scattering signal carrying the vibration information of the roller is collected, and a full-line roller vibration signal S(x) is demodulated by a phase demodulation algorithm;

[0008] S2, in combination with the fault characteristic frequency distribution and the system sampling rate, a suitable wavelet basis function and decomposition layer number are selected, the vibration signal S(x) is decomposed into a series of components IMF(x) by using a wavelet decomposition algorithm, the correlation coefficients of the autocorrelation functions of each component IMF(x) and the original vibration signal S(x) are calculated, and effective components IMF'(x) are selected according to a preset coefficient threshold;

[0009] S3, the effective components IMF'(x) are subjected to Bayesian shrinkage wavelet threshold denoising processing to obtain denoised components, and a denoised signal D(x) is obtained by superposition;

[0010] S4, 11 time-frequency domain feature parameters of the denoised signal D(x) are extracted, the KL distance between the denoised fault signal sample and the denoised normal signal sample is calculated, and the first six feature parameters with the largest distance are taken as the six most sensitive feature parameters F(m) to the fault;

[0011] S5, the six selected feature parameters F(m) are taken as a single sample to construct a sample set and train a machine learning classification model;

[0012] S6, the belt conveyor full-line real-time collected roller vibration signal is input, the feature parameters processed by steps S2 to S4 are input into the trained machine learning classification model, and the roller fault prediction results at each position are output.

[0013] Preferably, step S2 further comprises:

[0014] S21, confirming the fault characteristic frequency distribution interval according to the time-frequency characteristics of the typical bearing fault idler and broken idler in the sample library.

[0015] S22, selecting Daubechies 4 (db4) wavelet for 6-layer decomposition according to the fault characteristic frequency distribution and combining the system sampling rate, to obtain a series of scale components IMF(x) containing high-frequency transients, low-frequency trends and mechanical vibrations.

[0016] S23, calculating the autocorrelation function of the multi-scale component IMF(x) and the autocorrelation function of the original signal S(x), and the autocorrelation function is calculated as follows:

[0017]

[0018] Wherein, x(t) is a multi-scale component or an original signal, x(t-τ) is obtained by reversing x(t) along the time dimension, R x (τ) is the autocorrelation function calculated.

[0019] S24, calculating the correlation coefficient ξ(j) of the autocorrelation function of the jth component IMF(x) and the autocorrelation function R x of the original signal:

[0020]

[0021] Wherein, N is the number of signal points, and the correlation coefficient ξ(j) is compared with a preset threshold to obtain the first n effective components IMF'(x).

[0022] Preferably, step S3 further comprises:

[0023] S31, calculating the wavelet threshold T using the prior information of the generalized Gaussian distribution of the effective component IMF'(x):

[0024]

[0025] Wherein, σ 2 is the noise variance, σ x 2 is the signal variance.

[0026] The noise variance σ 2 can be calculated by the following formula:

[0027]

[0028] Wherein, IMF max (x) is the effective component under the maximum decomposition scale.

[0029] The signal variance σ x2 can be obtained by the following formula:

[0030]

[0031] wherein IMF'(x) is the effective component, is the corresponding mean value.

[0032] S32, according to the obtained wavelet threshold T corresponding to the effective component, the effective component IMF'(x) is contracted by using the threshold function, and the calculation process is as follows:

[0033]

[0034] wherein IMF'(x) k is the kth effective component, is the de-noised component processed by the threshold function.

[0035] S33, the de-noised components are superimposed to obtain the de-noised signal D(x):

[0036]

[0037] Preferably, step S4 further comprises:

[0038] S41, the samples in the carrier roller fault sample set are processed according to steps S2-S3 to obtain the de-noised signal D(x), and 6 time domain parameters are calculated: the mean value reflects the average level of the signal, and embodies the overall offset; the standard deviation measures the fluctuation intensity of the signal; the kurtosis is used to describe the peak degree or abnormality of the signal; the peak-to-peak value represents the difference between the maximum value and the minimum value of the signal, reflecting the dynamic range; the root mean square measures the energy or power level of the signal; and the peak factor is the ratio of the maximum absolute value to the RMS.

[0039] S42, 5 frequency domain parameters of the de-noised signal D(x) are extracted: the barycenter frequency reflects the weighted average position of the energy distribution of the signal, which can be used to identify the dominant vibration frequency; the spectral bandwidth is calculated by the standard deviation of the spectral energy, which quantifies the dispersion degree of the energy distribution, and the smaller the value is, the more concentrated the energy is around the barycenter frequency; the mean square frequency reflects the energy weighted root mean square value of the frequency component, which is sensitive to high frequency components and can effectively capture the high frequency energy changes caused by impact faults; the 300Hz-700Hz frequency band energy ratio and the 1.8kHz-2.2kHz frequency band energy ratio are calculated by calculating the proportion of the energy of the specific frequency band to the total energy, which accurately locates the energy aggregation characteristics of the fault characteristic frequency band.

[0040] S43, after obtaining 11 time-frequency domain feature parameters of the de-noised signal D(x), the Kullback-Leibler divergence (KL divergence) corresponding to each feature parameter between the carrier roller fault samples is calculated:

[0041]

[0042] Wherein, P1 and P2 are respectively the probability distribution of two different samples, for the sample composed of discrete signals, the probability distribution is calculated as follows:

[0043] P(v i ) = c(v i ) / n i = 1,...k

[0044] Wherein, P(v i ) is the probability distribution of the sample, c(v i ) represents the number of occurrences of each v i in the sample, v i is the possible value in the sample; the greater the KL distance, the stronger the discrimination of the feature parameter.

[0045] S44, the six feature parameters with the largest KL distance: kurtosis, root mean square, peak factor, center frequency, spectral bandwidth and 1.8kHz-2.2kHz frequency band energy ratio are taken as the feature parameters F(m) of the roller fault sample set.

[0046] The machine learning model of step S5 is a lightweight integrated model, including any one of the following: gradient boosting decision tree (GBDT) combined with random forest; XGBoost combined with support vector machine (SVM); five-fold cross-validation is used to optimize the hyperparameters during training, and the output layer uses softmax function processing.

[0047] Preferably, step S6 further comprises:

[0048] S61, real-time acquisition of vibration signals V n (t) at each place on the line, n represents the sensing position; after the V n (t) is processed by steps S2-S3, the denoising signal D n (t) is obtained.

[0049] S62, according to step S4, the six feature parameters to be calculated are: kurtosis, root mean square, peak factor, center frequency, spectral bandwidth and 1.8kHz-2.2kHz frequency band energy ratio; the denoising signal D n (t) obtained by S61 is calculated for feature parameters, and then a feature vector F n (m) is formed.

[0050] S63, the feature vector F n (m) corresponding to V n (t) is sent into the machine learning model trained in step S6 to obtain the fault classification result P fault (n), at this time P fault(n) can reflect the whole line roller fault position and fault type.

[0051] Compared with the prior art, the technical scheme has the following technical effects:

[0052] 1. The DAS-based intelligent belt conveyor roller fault monitoring method uses a distributed optical fiber vibration sensing system to realize high-precision, multi-node real-time sensing of the running state of each roller in the conveying line.

[0053] 2. The DAS-based intelligent belt conveyor roller fault monitoring method uses wavelet decomposition and wavelet threshold denoising to process the collected roller vibration signals, effectively removing noise and improving the signal fault feature retention capability.

[0054] 3. The DAS-based intelligent belt conveyor roller fault monitoring method extracts multi-domain effective features such as time domain, frequency domain, time-frequency domain and nonlinear features, which can fully characterize the transient impact and periodic characteristics of the fault, overcome the limitations of single-domain analysis, effectively improve the robustness to noise and variable working conditions, provide more distinguishable feature input for AI classification algorithm, and significantly improve the classification accuracy and model generalization ability.

[0055] 4. The DAS-based intelligent belt conveyor roller fault monitoring method significantly improves the speed and accuracy of belt conveyor roller fault positioning, and has significant practical application value for improving the operation and maintenance level of the belt conveyor. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The DAS-based intelligent belt conveyor roller fault monitoring method flowchart of the present application;

[0057] Figure 2 The distributed optical fiber monitoring system architecture diagram of the present application;

[0058] Figure 3 The time-frequency feature map of the fault roller and the normal roller of the present application;

[0059] Figure 4 The time-frequency feature map of the effective component superimposed signal after wavelet decomposition and the signal after wavelet threshold denoising of the present application;

[0060] Figure 5 The normalized KL distance of the 11 feature parameters in the sample set;

[0061] Figure 6 The whole line roller fault prediction and positioning information output by the model. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical scheme and advantages of the present application more clear, the technical scheme of the application is further described in detail below in combination with the drawings, and the described embodiments are only a part of the embodiments involved in the present application. All non-innovative embodiments of other researchers in the art on the embodiments belong to the protection scope of the present application. Meanwhile, the step numbers in the embodiments of the present application are only set for the convenience of description and explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiments can be adaptively adjusted according to the understanding of the person skilled in the art.

[0063] The distributed optical fiber sensing system DAS is to use the optical fiber as the transmission and sensing medium at the same time, to obtain the vibration information along the whole optical fiber, and to realize the continuous monitoring of the distance of thousands of meters or even tens of thousands of meters. The vibration wave generated by the roller vibration affects the surrounding arranged sensing optical fiber, and the vibration signal can be obtained by demodulating the back Rayleigh scattering light generated in the optical fiber. The optical fiber will inevitably perceive environmental noise such as wind and material spilling, and these noises will be mixed into the vibration information of the roller, covering the fault characteristics.

[0064] Therefore, the present application firstly uses wavelet decomposition and wavelet threshold denoising to denoise and reconstruct the original vibration signal, so as to suppress the environmental noise as much as possible; different domain features have different sensitivities to various faults, therefore, the time domain and frequency domain features representing the transient impact and periodic characteristics of the faults are calculated according to the sample set, the feature vectors are composed of the features with the largest discrimination selected according to the KL distance, and the sensitivity and limitation of single feature to noise are overcome; the feature vectors of the input whole-line roller vibration signal are composed into a sequence according to the position, and are input into the trained machine learning model for classification prediction, and the output result not only reflects the accurate fault type, but also has accurate position information.

[0065] In an embodiment of the present application, referring to Figure 1 The intelligent belt conveyor roller fault monitoring method based on DAS comprises the following steps:

[0066] S1, the sensing optical cable is arranged on the side support of the belt conveyor roller group, the distributed optical fiber sensing system DAS is used to inject an optical pulse into the sensing optical cable, the back Rayleigh scattering signal carrying the roller vibration information is collected, and the whole-line roller vibration signal is demodulated by a phase demodulation algorithm;

[0067] S2, combining the fault characteristic frequency distribution and the system sampling rate, selecting a suitable wavelet basis function and a decomposition layer number, decomposing the vibration signal S(x) into multi-scale components IMF(x) by using a wavelet decomposition algorithm, calculating the correlation coefficient of the autocorrelation function of each component IMF(x) and the autocorrelation function of the original vibration signal S(x), and selecting the effective component IMF'(x) according to a preset coefficient threshold;

[0068] S3, performing Bayesian shrinkage wavelet threshold denoising processing on the effective IMF'(x) to obtain a denoised component, and then superimposing to obtain a denoised signal D(x);

[0069] S4, extracting 11 time-frequency domain feature parameters of the denoised signal D(x), and then screening six feature parameters F(m) sensitive to faults through KL distance;

[0070] S5, constructing a sample set based on the screened feature parameters F(m), and training a machine learning classification model;

[0071] S6, inputting the feature parameters of the real-time collected vibration signal processed through steps S2-S6 into the trained model, and outputting the fault type and position of the carrier roller.

[0072] Specifically, as shown in Figure 2 , the sensing optical cable is arranged on the side support of the belt conveyor carrier roller group, the distributed optical fiber sensing system DAS is used to inject optical pulses into the sensing optical cable, the backscattering Rayleigh scattering signals carrying the vibration information of the carrier roller are collected, and the vibration signals of the carrier roller are demodulated through the phase demodulation algorithm.

[0073] Further, according to the fault characteristic frequency distribution and in combination with the system sampling rate, a suitable wavelet basis function and decomposition layer number are selected, including:

[0074] In this embodiment, the vibration signal in the fault sample is subjected to short-time Fourier transform, and the time-frequency diagram obtained is as shown in Figure 3 As can be seen from the diagram, compared with the normal carrier roller, the characteristic of the fault carrier roller is a periodic impact signal in the frequency range of 500Hz-7000Hz. At a sampling rate of 20kHz, the Daubechies 4(db4) wavelet is used for 6-layer decomposition, which is suitable for analyzing the medium-high frequency transient signal and the low frequency trend, and the frequency band division covers 10kHz(D1) to 156Hz(A6).

[0075] Among them, D1(10k-5kHz) is suitable for ultrahigh frequency noise and transient impact detection; D2(5k-2.5kHz) is suitable for high frequency vibration; D3(2.5kHz-1.5kHz) is suitable for mechanical fault characteristic frequency, and D4 to D6(1.25kHz-156Hz) is suitable for rotating machinery low frequency characteristic frequency, and A6(0Hz-156Hz) is suitable for analyzing slow changing trend or ultralow frequency event.

[0076] D5(625Hz-312.5Hz) is particularly suitable for mechanical fault diagnosis (such as 1-5kHz impact component of bearings / gears).

[0077] In summary, the db4 wavelet balances the time-frequency resolution and can effectively capture the medium-high frequency transient characteristics of the fault carrier roller.

[0078] Further, the correlation coefficient of the autocorrelation function of each component IMF(x) and the autocorrelation function of the original vibration signal S(x) is calculated, and the effective component IMF'(x) is screened according to a preset coefficient threshold, including the following steps:

[0079] The autocorrelation function of the multi-scale component IMF(x) and the autocorrelation function of the original signal S(x) are calculated, and the autocorrelation function is calculated as follows:

[0080]

[0081] Wherein, x(t) is a multi-scale component or an original signal, x(t-τ) is obtained by reversing x(t) along the time dimension, R x (τ) is the autocorrelation function calculated.

[0082] Then, the autocorrelation function R IMFj of the jth component IMF(x) is calculated, and the autocorrelation function R x of the original signal is calculated.

[0083]

[0084] Wherein, N is the number of signal points, and then the correlation coefficient is compared with a preset threshold to obtain the first n effective components IMF'(x).

[0085] Further, the effective component IMF'(x) is subjected to Bayesian shrinkage wavelet threshold denoising processing to obtain a denoising component, and then the denoising signal D(x) is obtained by superposition, including the following steps:

[0086] The wavelet threshold T is calculated using the prior information of the generalized Gaussian distribution of the effective component IMF'(x):

[0087]

[0088] Wherein, σ 2 is the noise variance, σ x 2 is the signal variance.

[0089] The noise variance σ 2 can be calculated as follows:

[0090]

[0091] Wherein, IMF max (x) is the effective component under the maximum decomposition scale.

[0092] The signal variance σ x 2 can be obtained as follows:

[0093]

[0094] wherein IMF'(x) is the effective component, is its corresponding mean value.

[0095] According to the obtained wavelet threshold T corresponding to the effective component, the effective component IMF'(x) is contracted by using a threshold function, and the calculation process is as follows:

[0096]

[0097] wherein IMF'(x) k is the kth effective component, is the de-noised component after threshold function processing.

[0098] The de-noised components are superimposed to obtain a de-noised signal D(x):

[0099]

[0100] The time-frequency diagram of the signal before and after wavelet threshold de-noising is shown in Figure 4 It can be seen that the de-noised signal retains the complete high-frequency non-stationary periodic characteristics and removes most of the noise.

[0101] Further, after obtaining the de-noised signal D(x), 11 time-frequency domain feature parameters are extracted, including:

[0102] 6 time domain parameters are calculated: the mean value reflects the average level of the signal, representing the overall offset; the standard deviation measures the fluctuation intensity of the signal; the kurtosis (used to describe the degree of signal peak or abnormality; the peak-to-peak value represents the difference between the maximum and minimum values of the signal, reflecting the dynamic range; the root mean square measures the energy or power level of the signal; and the peak factor is the ratio of the maximum absolute value to the RMS;

[0103] 5 frequency domain parameters are calculated: the center of gravity frequency reflects the weighted average position of the signal energy distribution, which can be used to identify the dominant vibration frequency; the spectral bandwidth is calculated by the standard deviation of the spectral energy, which quantifies the dispersion degree of the energy distribution, and the smaller the value, the more concentrated the energy is around the center of gravity frequency; the mean square frequency reflects the energy weighted root mean square value of the frequency component, which is sensitive to high frequency components and can effectively capture the high frequency energy changes caused by impact failures; the 300Hz-700Hz frequency band energy ratio and the 1.8kHz-2.2kHz frequency band energy ratio are calculated by calculating the proportion of the energy of a specific frequency band to the total energy, which accurately locates the energy aggregation characteristics of the fault feature frequency band.

[0104] After obtaining the 11 time-frequency domain feature parameters of the denoised signal D(x), the Kullback-Leibler divergence (KL divergence) corresponding to each feature parameter among the roller fault samples is calculated:

[0105]

[0106] Where P1 and P2 are the probability distributions of two different samples. For a sample composed of discrete signals, its probability distribution is calculated as follows:

[0107]

[0108] Wherein, P(ν) i Let c(ν) be the probability distribution of the sample. i ) represents each ν in the sample i The number of times ν appears i These are the possible values ​​in the sample.

[0109] Specifically, a larger KL distance indicates a stronger discriminative power for that feature parameter. In this embodiment, the KL distance is used to filter the six most fault-sensitive feature parameters F(m): kurtosis, root mean square, peak factor, centroid frequency, spectral bandwidth, and energy ratio in the 1.8kHz-2.2kHz frequency band. Figure 5 As shown, these six characteristic parameters are used as the characteristic parameters F(m) of the idler roller fault sample set.

[0110] Furthermore, the most sensitive feature parameters for fault identification were identified as kurtosis, root mean square (RMS), crescendo factor, centroid frequency, spectral bandwidth, and the energy ratio of the 1.8kHz-2.2kHz frequency band. Kurtosis describes the degree of signal spikes or anomalies, reflecting the periodic impact characteristics of the fault signal; RMS measures the energy or power level of the signal, reflecting the energy changes caused by the fault; crescendo factor is the ratio of the maximum absolute value to the RMS, reflecting the relative fluctuations in signal amplitude; centroid frequency measures the weighted average position of the signal energy distribution, reflecting the dominant oscillation frequency of the fault signal; spectral bandwidth calculates the standard deviation of spectral energy, quantifying the dispersion of energy distribution and reflecting the energy concentration at fault characteristic frequencies; and the energy ratio of the frequency band accurately reflects the energy concentration characteristics of the fault characteristic frequency band.

[0111] Furthermore, a sample set is constructed based on the selected feature parameter F(m) to train a machine learning classification model.

[0112] It should be noted that the machine learning model architecture adopted in the embodiment is gradient boosting decision tree (GBDT) + random forest ensemble, the input is 600 1-second signal samples, of which 300 are normal and 300 are faults, five-fold cross-validation is used to optimize the parameters during training, and the output layer uses Softmax classification (0: normal, 1: fault).

[0113] Further, the feature parameters of the vibration signals collected in real time and processed through steps S2-S6 are input into the trained model, and the roller fault type and position are output. Specifically, it includes:

[0114] Real-time acquisition of vibration signals V n (t) at each position on the line; n represents the sensing position; the V n (t) is processed through steps S2-S3 to obtain a denoising signal D n (t);

[0115] According to step S44, it is confirmed that the six feature parameters to be calculated are kurtosis, root mean square, peak factor, center of gravity frequency, spectral bandwidth, and 1.8kHz-2.2kHz frequency band energy ratio; the denoising signal D n (t) obtained in S32 is calculated for feature parameters, and then a feature vector F n (m) is formed.

[0116] The feature vector F n (m) corresponding to V n (t) is input into the trained machine learning model in step S5 to obtain a fault classification result P fault (n), at this time, P fault (n) can reflect the fault position and fault type of the rollers on the entire line. As shown in Figure 6 , the roller fault prediction and positioning information on the entire line are displayed.

[0117] In summary, the embodiment discloses an intelligent belt conveyor roller fault monitoring method based on DAS, which realizes full-distributed, on-site networking-free sensing of the rollers on the entire belt conveyor by using DAS, denoising and reconstruction of the roller operating signals by using WT-CC-BWT, calculation of 11 time-frequency feature parameters of the denoising and reconstructed signals, screening of six feature parameters with the best resolution capability as samples by using KL distance, and realization of roller fault classification and prediction by using XGBoost (machine learning), which has important potential value for the maintenance and detection of the belt conveyor.

[0118] The above are only preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical scheme falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled in the art, some improvements and refinements without departing from the principles of the present application shall be considered as the protection scope of the present application.

Claims

1. A DAS-based intelligentized belt conveyor idler fault monitoring method, characterized in that, The method comprises the following steps: S1, the sensing optical cable is arranged on the side support of the belt conveyor roller group, a distributed optical fiber sensing system DAS is used to inject an optical pulse into the sensing optical cable, a backscattering Rayleigh signal carrying the vibration information of the roller is collected, and a full-line roller vibration signal S(x) is demodulated by a phase demodulation algorithm; S2, in combination with the fault characteristic frequency distribution and the system sampling rate, a suitable wavelet basis function and a decomposition layer number are selected, the vibration signal S(x) is decomposed into a series of components IMF(x) by using a wavelet decomposition algorithm, the correlation coefficients of the autocorrelation functions of each component IMF(x) and the original vibration signal S(x) are calculated, and effective components IMF'(x) are screened according to a preset coefficient threshold value; S3, the effective components IMF'(x) are subjected to Bayesian shrinkage wavelet threshold denoising processing to obtain denoised components, and the denoised signal D(x) is obtained by superposition; S4, 11 time-frequency domain characteristic parameters of the denoised signal D(x) are extracted, the KL distance between the denoised fault signal sample and the denoised normal signal sample is calculated, and the first six characteristic parameters with the largest distance are taken as six characteristic parameters F(m) most sensitive to the fault; S5, the six characteristic parameters F(m) screened out are taken as a single sample to construct a sample set and train a machine learning classification model; S6, the roller vibration signals collected in real time on the full line of the belt conveyor are input, the characteristic parameters processed by steps S2 to S4 are input into the trained machine learning classification model, and the roller fault prediction results at each position are output.

2. The DAS-based intelligentized belt conveyor idler failure monitoring method according to claim 1, characterized in that, Step S2 comprises the following sub-steps: S21, the time-frequency characteristic of the typical fault roller in the sample library is confirmed to determine the fault characteristic frequency distribution interval; S22, the Daubechies 4 wavelet is selected for six-layer decomposition according to the fault characteristic frequency distribution and in combination with the system sampling rate, and a series of scale components IMF(x) containing high-frequency transients, low-frequency trends and mechanical vibrations are obtained; S23, the autocorrelation functions of each scale component IMF(x) and the autocorrelation function of the original signal S(x) are calculated, and the autocorrelation function is calculated as follows: where x(t) is the multiscale component or the original signal, x(t - τ) is the reversal of x(t) along the time dimension, R x (τ) is the autocorrelation function calculated S24, calculating the autocorrelation function of the jth component IMF(x) correlation coefficient ξ(j) of the autocorrelation function R of the original signal x correlation coefficient ξ(j) of the autocorrelation function R of the original signal Wherein, N is the number of signal points, the correlation coefficient ξ(j) is compared with the preset threshold value to obtain the first n effective components IMF'(x).

3. The DAS-based intelligentized belt conveyor idler failure monitoring method according to claim 1, characterized in that, Step S3 comprises the following sub-steps: S31, the wavelet threshold T is calculated by using the prior information of the generalized Gaussian distribution of the effective component IMF'(x): where σ 2 is the noise variance, σ x 2 is the signal variance; S32, the wavelet threshold T corresponding to the effective component is obtained, the effective component IMF'(x) is contracted by using a threshold function, and the calculation process is as follows: wherein IMF'(x) k is the kth effective component, is the de-noised component after threshold function processing; S33, the denoised signal D(x) is obtained by superimposing the denoised components:

4. The DAS-based intelligentized belt conveyor idler failure monitoring method according to claim 3, characterized in that, In step S31, the noise variance σ 2 is calculated as follows: wherein IMF max (x) is the effective component at the maximum decomposition scale; Signal variance σ x 2 The calculation is as follows: where IMF'(x) is the effective component, is its corresponding mean.

5. The DAS-based intelligentized belt conveyor idler failure monitoring method according to claim 1, characterized in that, Step S4 comprises the following sub-steps: S41, the samples in the roller fault sample set are processed according to steps S2-S3 to obtain the denoised signal D(x), and six time domain parameters are calculated, including mean, standard deviation, kurtosis, peak-to-peak value, root mean square, and peak factor; S42, five frequency domain parameters of the denoised signal D(x) are extracted, including the center of gravity frequency, the spectral bandwidth, the mean square frequency, the energy ratio of the 300Hz-700Hz frequency band, and the energy ratio of the 1.8kHz-2.2kHz frequency band. S43, 11 time-frequency domain feature parameters of the de-noised signal D(x) are obtained, and the KL divergence corresponding to each feature parameter between the roller fault samples is calculated, and the formula is as follows: Wherein, P1 and P2 are the probability distributions of two different samples, and for the sample composed of discrete signals, the probability distribution is calculated as follows: P(ν i ) = c(ν i ) / n i = 1,... k where P(v i ) is the probability distribution of the sample, c(v i ) represents the number of occurrences of each v i in the sample, and v i is the possible value in the sample; the greater the KL distance, the stronger the discrimination of the corresponding feature parameter; S44, select the 6 feature parameters with the largest KL distance as the feature parameters F(m) of the roller fault sample set.

6. The DAS-based intelligentized belt conveyor idler failure monitoring method according to claim 5, characterized in that, Among the 6 time domain parameters, the mean value is used to reflect the average level of the vibration signal, and the overall deviation is embodied; the standard deviation is used to measure the fluctuation intensity of the signal; the kurtosis is used to describe the peak degree or abnormality of the signal; the peak-peak value represents the difference between the maximum value and the minimum value of the signal, and is used to reflect the dynamic range; the root mean square is used to measure the energy or power level of the signal; the peak factor is the ratio of the maximum absolute value to the RMS.

7. The DAS-based intelligentized belt conveyor idler failure monitoring method according to claim 5, characterized in that, Among the 5 frequency domain parameters, the center of gravity frequency reflects the weighted average position of the signal energy distribution, and is used to identify the dominant vibration frequency; the spectral bandwidth is calculated by calculating the standard deviation of the spectral energy, and is used to quantify the dispersion degree of the energy distribution, and the smaller the value is, the more concentrated the energy is near the center of gravity frequency; the mean square frequency reflects the energy weighted root mean square value of the frequency component, and is sensitive to high frequency components, and is used to capture the high frequency energy change caused by impact failure; the two frequency band energy ratios are calculated by calculating the proportion of the energy of the specific frequency band to the total energy, and the energy aggregation characteristics of the fault characteristic frequency band are located.

8. The DAS-based intelligentized belt conveyor idler failure monitoring method according to claim 5, characterized in that, Among the 11 time-frequency domain feature parameters, the six feature parameters with the largest KL distance are: kurtosis, root mean square, peak factor, center of gravity frequency, spectral bandwidth and 1.8kHz-2.2kHz frequency band energy ratio.

9. The DAS-based intelligentized belt conveyor roller fault monitoring method according to claim 1, characterized in that, In step S5, the machine learning classification model is a lightweight integrated model, including: gradient boosting decision tree combined with random forest, XGBoost combined with support vector machine; five-fold cross-validation is used to optimize the hyperparameters during training, and the output layer uses softmax function processing.

10. The DAS-based intelligentized belt conveyor idler failure monitoring method according to claim 1, characterized in that, Step S6 includes the following sub-steps: S61, collecting vibration signals V of each position of the belt conveyor in real time n (t), n represents the sensing position; processing V n (t) by steps S2-S3 to obtain a denoised signal D n (t); S62, based on the six feature parameters with the largest KL distance obtained in step S4, the de-noised signal D obtained in S61 is processed to obtain a feature vector F n (t) performing feature parameter calculation to form a feature vector F n (m); S63, V n (t) the corresponding feature vector F n (m) into the trained machine learning classification model in step S5, to obtain the fault classification result P fault (n), reflecting the fault occurrence position and fault type of the whole line of rollers.

Images