Bucket wheel machine coal blockage acoustic monitoring method based on multi-sub-band envelope

Abstract

The invention discloses a bucket wheel machine coal blockage acoustic monitoring method based on multi-subband envelope. The method comprises the following steps: firstly, constructing a bucket wheel machine normal voiceprint template library; secondly, sound signals of operation of the bucket wheel machine are collected in real time and subjected to low-pass filtering, complete self-adaptive noise set empirical mode decomposition is carried out, and sub-band signals are obtained and subjected to directional wavelet threshold noise reduction; envelopes of the denoised sub-band signals are obtained, a first-order difference gradient sequence of each envelope is solved, multi-dimensional dynamic time warping (DTW) is carried out in combination with a normal voiceprint template library, a DTW distance set is calculated, dynamic weighting based on two-factor decision is carried out on the DTW distance set, results are accumulated to obtain a difference sum, and a self-adaptive threshold value is set. And finally, comparing the sum of the difference values with an adaptive threshold value to obtain a signal classification result, and completing monitoring. The method can adapt to severe working conditions of a coal yard, and can capture signal energy and waveform changes at the same time, thereby improving the sensitivity to weak faults.

Description

Technical Field

[0001] This invention belongs to the field of acoustic detection technology and industrial equipment fault monitoring, specifically involving an acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-subband envelopes, which is applicable to real-time monitoring and early warning of coal blockage problems in bucket wheel excavators in coal conveying systems of thermal power plants. Background Technology

[0002] Bucket wheel stacker-reclaimers (hereinafter referred to as bucket wheel stackers) play a crucial role in the coal conveying system of thermal power plants, transporting coal from transport equipment to storage areas or boilers. However, during the operation of bucket wheel stackers, especially at the connection between the conveyor belt and the coal hopper, coal blockage can easily occur due to the moisture and viscosity of the coal, or abnormal operation of the conveyor belt. Blockage can prevent coal from being transported smoothly, thus affecting the stable operation of the thermal power plant. Traditional methods for monitoring coal blockage mainly rely on mechanical sensors (such as rotary paddle level switches and capacitive level switches), which suffer from high false alarm rates, high false alarm rates, and high maintenance costs. While image recognition-based monitoring schemes can intuitively identify blockages, environmental factors such as dust in the coal yard and changes in lighting can significantly reduce the accuracy of image recognition. Furthermore, methods based on coal flow detection require the deployment of multiple sets of sensors to monitor the coal flow, resulting in severe equipment wear, high maintenance costs, and difficulty in guaranteeing data accuracy.

[0003] In recent years, voiceprint recognition technology has been introduced into the field of fault monitoring due to its advantages such as good environmental adaptability and strong anti-interference ability. When a fault occurs in electromechanical equipment, the vibration characteristics of the internal mechanical structure change, and the voiceprint signal changes significantly accordingly. Early warning of faults can be achieved by analyzing voiceprint characteristics. However, the application of existing voiceprint detection technology in monitoring coal blockage in bucket wheel excavators still faces the following challenges:

[0004] (1) Severe background noise interference: There are strong interferences such as train noise and equipment vibration noise in the working environment of the bucket wheel excavator, which affect the effective extraction of the acoustic signal; (2) Signal non-stationarity: When coal blockage occurs, the acoustic signal characteristics change in a complex manner, and traditional time-frequency analysis methods (such as short-time Fourier transform) are difficult to capture dynamic features; (3) Multi-target signal aliasing: The sound field inside the coal bucket of the bucket wheel excavator is complex, and multi-path reflection leads to signal aliasing, which reduces the monitoring accuracy.

[0005] To address the aforementioned issues, there is an urgent need for a robust coal blockage monitoring technology that can adapt to complex industrial environments. This technology would solve key technical challenges such as complex noise suppression, feature extraction optimization, and multi-objective observability improvement, providing reliable support for the intelligent upgrading of coal conveying systems in thermal power plants. Summary of the Invention

[0006] This invention proposes an acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-subband envelopes. This invention uses acoustic methods to distinguish between fault signals and normal signals related to coal blockage in bucket wheel excavators, enabling real-time monitoring and early warning of blockages. Acoustic fingerprint signals collected during bucket wheel excavator operation are acquired using an acoustic sensor array and processed for signal analysis to obtain subband signal envelopes. A dynamic weighting method is used to enhance the characteristics of subband signals with high signal-to-noise ratios in specific frequency bands. Furthermore, by establishing a threshold model based on sliding window statistics, threshold adaptation is achieved, better adapting to the complex background noise conditions of coal yards. In terms of broadband signal processing, existing methods mostly use EMD-like decomposition of the original broadband signal, followed by noise reduction through correlation processing of the subband signal IMFs (Intrinsic Mode Functions), and then reconstruct the complete signal from the processed IMFs for signal decision-making. This invention directly uses the envelopes of multiple decomposed IMFs for comparison and decision-making, avoiding the possibility of the reconstruction process masking local features of the IMFs and reintroducing incompletely suppressed noise. Furthermore, by reducing the number of reconstruction steps, it reduces additional computational load. Addressing the complex background noise environment of bucket wheel excavators, a dynamic weighted difference mechanism based on two-factor decision-making (DTW) is proposed to solve the two core problems of high-frequency noise interference and rigid fixed weights in existing methods. By fusing the physical characteristics of the equipment (frequency band sensitivity) with the real-time signal-to-noise ratio of the environment, the robustness of the decision under complex operating conditions is significantly improved. Moreover, by constructing a sliding window-based statistical model, the threshold is automatically adjusted according to environmental noise fluctuations, making the algorithm more adaptable to complex background noise environments.

[0007] The present invention adopts the following technical solution:

[0008] An acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-subband envelopes, comprising the following steps:

[0009] Step 1: Construct a normal sound signature template library for bucket wheel excavators using the sound signals during normal operation.

[0010] Step 2: Acquire the acoustic signals of the bucket wheel excavator in real time through an acoustic sensor array, and perform signal conditioning using a low-pass filter (below 2000Hz).

[0011] Step 3: Perform fully adaptive ensemble empirical mode decomposition (CEEMDAN) on the filtered acoustic signal to be detected obtained in Step 2 to obtain the subband signal IMF.

[0012] Step 4: Perform directional wavelet threshold denoising on the sub-band signal IMF obtained in Step 3.

[0013] Step 5: Take the envelope of the noise-reduced subband IMF signal obtained in Step 4. Obtain the envelope of the subband signal by performing Hilbert transform on the subband signal, and calculate the first-order difference gradient sequence of each envelope.

[0014] Step 6: Perform multi-dimensional DTW (Dynamic Time Warping) on ​​the sub-band IMF signal envelope and envelope first-order differential gradient sequence obtained in Step 5 with the normal voiceprint template library to calculate the DTW distance set.

[0015] Step 7: Perform dynamic weighting on the DTW distance set obtained in Step 6 based on two-factor decision-making. The difference is dynamically and adaptively weighted according to the two factors of frequency band fault sensitivity of sub-band envelope and real-time signal-to-noise ratio of sub-band envelope, and the results are accumulated to obtain the sum of the differences Sum.

[0016] Step 8: Construct a dynamic statistical model using the DTW difference of historical normal signals and set a reasonable threshold.

[0017] Step 9: Compare the sum of differences Sum obtained in Step 7 with the adaptive threshold Threshold calculated in Step 8 to obtain the signal classification result and complete the acoustic monitoring of coal blockage in the bucket wheel excavator.

[0018] The present invention has the following beneficial effects:

[0019] 1. This invention uses acoustic signal processing to monitor coal blockage in bucket wheel excavators, which can adapt to the harsh working conditions of high temperature, high humidity and high dust in coal yards.

[0020] 2. This invention employs a multi-subband envelope comparison signal technology. By independently processing and comparing multiple subband envelopes, it solves the problems of information loss and noise coupling caused by reconstruction in current subband decomposition-based reconstructed signal comparison methods. It is particularly suitable for complex signal comparison scenarios that require high precision and high robustness.

[0021] 3. This invention uses multi-dimensional DTW for similarity measurement, integrating both signal amplitude and gradient change characteristics. Compared to traditional single-channel DTW algorithms, which only compare envelope amplitude and ignore phase and shape differences, multi-dimensional DTW can simultaneously capture signal energy and waveform changes, improving sensitivity to subtle faults.

[0022] 4. This invention employs dynamic weighting and a two-factor decision-making approach for the distance difference obtained from DTW comparison. The weights are jointly controlled by two factors: the frequency band fault sensitivity of the sub-band envelope and the real-time signal-to-noise ratio of the sub-band envelope. Furthermore, the frequency band fault sensitivity is dynamically adjusted by the kurtosis of the sub-band envelope. While preserving the core characteristics of the low-frequency sub-band of the coal blockage signal, this invention reduces the susceptibility of the high-frequency sub-band signal to random noise interference. It can better adapt to the dynamic fluctuations in the sub-band signal-to-noise ratio caused by changes in coal ash and humidity in the bucket wheel excavator operating environment, and has better adaptability compared to fixed weights.

[0023] 5. This invention constructs an adaptive threshold based on sliding window statistics. It uses the DTW difference of historical normal signals to build a dynamic statistical model. The threshold is automatically adjusted according to the fluctuation of environmental noise, which can better adapt to the noise change of the environment where the bucket wheel excavator is located over time. At the same time, adding the noise peak ratio to the sensitivity coefficient can suppress the influence of abnormal operating condition noise on the decision threshold and reduce the false alarm rate of the system. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the principle of an acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-subband envelope.

[0025] Figure 2 This is a schematic diagram of the principle of fully adaptive noise set empirical mode decomposition;

[0026] Figure 3 This is the IMF diagram after subband decomposition;

[0027] Figure 4 This is the IMF plot of the subband signal after wavelet thresholding and denoising.

[0028] Figure 5 This is a comparison chart of the root mean square error before and after noise reduction;

[0029] Figure 6 This is a comparison chart of the accuracy rates of this invention with other methods. Detailed Implementation

[0030] The present invention will now be described in further detail with reference to the accompanying drawings.

[0031] This invention presents an acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-subband envelopes. Addressing the shortcomings of existing bucket wheel excavator coal blockage monitoring technologies—such as low efficiency and difficulty of manual inspections, the inability of image recognition methods to handle the complex environment of coal yards, and the high cost and severe equipment wear associated with monitoring coal flow—this invention employs an acoustic monitoring method that effectively copes with the harsh environment in which bucket wheel excavators operate. Recognizing the information loss and noise coupling issues caused by partial subband signal reconstruction in acoustic fault detection, this invention uses the envelopes of multi-subband signals directly for subsequent feature analysis, reducing information loss and noise coupling while preserving signal features as much as possible. Furthermore, addressing the issue that using fixed weights for fusion of distance difference features obtained after subband envelope DTW makes it difficult to highlight the characteristics of the low-frequency fault-sensitive frequency band during bucket wheel excavator coal blockage, and the susceptibility of high-frequency subbands to random noise, this invention employs dynamic weighting and a two-factor decision mechanism to preserve the low-frequency characteristics of the signal under test while suppressing noise in the high-frequency subbands. To address the difficulty of using fixed thresholds in signal decision-making to handle the time-varying effects of ambient noise in the bucket wheel excavator's environment, this invention constructs an adaptive threshold based on sliding window statistics, which can better adapt to complex background noise conditions. The overall algorithm flowchart is attached. Figure 1 .

[0032] I. Acoustic Signal Processing Method for Bucket Wheel Excavators Based on Multi-Sub-belt

[0033] 1.1 Establishment of a Normal Voiceprint Template Library

[0034] One hundred sets of normal acoustic signatures were collected by an acoustic sensor during the operation of the bucket wheel excavator. The average of the 100 sets of acoustic signature signals was then summed and averaged. The averaged signal was used as a normal acoustic signature template library and served as a benchmark for subsequent signal comparison.

[0035] 1.2 Subband Decomposition Based on CEEMAN

[0036] Subband decomposition breaks down the signal to be detected into several subband signals for subsequent processing. The subband decomposition method used in this invention is Fully Adaptive Empirical Mode Decomposition with Accumulated Noise (CEEMDAN). CEEMDAN adds pairs of white noise with opposite positive and negative values ​​to the original signal based on EMD, which can suppress the mode mixing problem present in EMD. The algorithm principle is shown in the appendix. Figure 2 The EMD algorithm steps are as follows:

[0037] (1) Use cubic spline functions to find the maximum and minimum values ​​of the original signal X(t) and fit them to obtain the upper and lower envelopes of the signal.

[0038] (2) Calculate the average value M1(t) for the upper and lower envelopes;

[0039] (3) Subtract the original signal X(t) from the envelope mean M1(t) to obtain the remaining signal d1(t):

[0040] (4) Repeat the above process for the remaining signal d1(t) until the condition SD of the modal component is less than the threshold value, and then stop to obtain the remaining signal c1(t) of the first IMF component. The SD condition is:

[0041]

[0042] (5) Calculate the difference between the signal X(t) and c1(t) to obtain the first-order residual r1(t). Replace X(t) with r1(t) and repeat the above steps n times to obtain the nth-order modal component c1(t) and the nth-order residual r. n (t)

[0043]

[0044] The expression for the original signal EMD decomposition is:

[0045]

[0046] In the formula, X(t) represents the original data; c n (t) represents the intrinsic mode function; r n (t) represents the difference; N represents the order.

[0047] After subband decomposition, several subband signals IMF are obtained. Each IMF component represents the variation state of the original data at different frequency stages.

[0048] 1.3 Wavelet Threshold-Oriented Noise Reduction Algorithm

[0049] The subband-decomposed IMF still contains considerable noise. To ensure the accuracy of subsequent algorithms, denoising of the subband signals is necessary. This invention uses a wavelet threshold denoising algorithm. Since the characteristics of coal blockage signals are mostly concentrated in the low-frequency range, to preserve the low-frequency components of the signal as much as possible, wavelet threshold denoising is performed only on subbands with a center frequency exceeding 30Hz, and a hard threshold is selected for denoising.

[0050] The following is the process of wavelet threshold denoising. After the noisy signal is decomposed by Discrete Wavelet Transform (DWT), its corresponding wavelet coefficients exhibit significant differences: the real signal energy is mainly distributed in the coefficient region with larger amplitude, while the noise component is concentrated in the coefficient range with smaller amplitude. The mathematical model of this method can be expressed by the following formula system:

[0051] The model for a noisy signal is shown in the following equation:

[0052] s(t)=f(t)+n(t)

[0053] In the formula, f(t) is the original noiseless signal, n(t) is the noise component, and s(t) is the observed signal.

[0054] The noisy signal s(t) is subjected to Discrete Wavelet Transform (DWT) and converted into coefficient form, as shown in the following equation.

[0055] W (j,k) =q (j,k) +n (j,k)

[0056] In the formula, W (j,k) To observe the wavelet coefficients of the signal, q (j,k) For the real signal components, n (j,k) Noise component

[0057] The steps for wavelet thresholding denoising are as follows:

[0058] (1) Signal decomposition. Select appropriate wavelet basis functions and decomposition levels, perform discrete wavelet transform on the observed signal, and obtain wavelet coefficient sequences at each resolution scale.

[0059] (2) Coefficient thresholding. Based on the statistical characteristics of noise, a threshold function is constructed to filter high-frequency coefficients:

[0060] Hard threshold: Forces the coefficient below the threshold to be zero.

[0061] Soft thresholding: Shrinking the portion exceeding the threshold. Key parameters include the selection of global / hierarchical adaptive thresholding strategy and threshold function.

[0062] (3) Signal reconstruction. The processed coefficient sequence is reconstructed by inverse discrete wavelet transform (IDWT) to finally obtain the denoised time-domain signal.

[0063] 1.4. Low-frequency fault signal feature enhancement based on envelope demodulation

[0064] The subband signal IMF, after wavelet threshold denoising, already has a better signal-to-noise ratio compared to the original signal to be detected. Further extraction of its instantaneous amplitude and frequency using Hilbert transform, where these instantaneous amplitudes can be considered as a form of envelope, achieves results similar to traditional envelopes, further enhancing the low-frequency signal characteristics of the subband signal. The principle of Hilbert transform is as follows.

[0065] Given a time-domain signal x(t), its Hilbert transform is: The analytic signal obtained from the time-domain signal is:

[0066]

[0067] Similar to amplitude and phase in the Fourier transform, we can obtain the Hilbert transform as a function of amplitude-time and phase-time:

[0068]

[0069] In the Hilbert transform, there is also the concept of instantaneous frequency, which is the derivative of the phase-time function with respect to time. Through the Hilbert transform, a time-domain signal can be introduced into a three-dimensional space of time-frequency-amplitude or time-frequency-phase for analysis, enabling good time-frequency analysis of the signal.

[0070] After Hilbert transform, the envelopes of each subband signal are obtained, and the first-order difference gradient sequence of these envelopes is calculated.

[0071] II. Multi-subband envelope DTW matching and decision mechanism

[0072] 2.1 Multi-dimensional DTW Similarity Measurement

[0073] Following the previous step, we obtain the subband signal envelope and the first-order difference gradient sequence. This step performs a multi-dimensional DTW similarity measurement on the subband envelope and the first-order difference gradient sequence of the subband envelope.

[0074] The principle is as follows: Two sub-band envelopes, X and Y, where X is the first sub-band envelope of the voiceprint template library signal, and Y is the first sub-band envelope of the signal to be detected, with lengths n and m respectively.

[0075] X = {x1, x2, ..., x} i ,…,x n}

[0076] Y = {y1, y2, ..., y} j ,…,y m}

[0077] In the formula, x and y represent the values ​​at points i and j in the sub-band envelope sequences X and Y, respectively.

[0078] The DTW algorithm constructs a model by calculating the Euclidean distance between corresponding points in two sequences. Figure 4 The distance matrix shown is of size n×m, and the Euclidean distance between corresponding points is denoted by d(i,j).

[0079] After the distance matrix is ​​established, k paths can be found between the pre-defined starting point and ending point. The path that achieves the minimum value of the cumulative distance function is defined as the optimal normalized path between time series X and Y.

[0080] Find the shortest distance DTW using the following recursive method. min

[0081] DTW min=d(i,j)+min{d(i-1,j),d(i,j-1),d(i-1,j-1)}

[0082] d(i,j) represents the Euclidean distance between points i and j. Using the above method, the DTW distance between the first sub-band envelope X of the voiceprint template library signal and the first sub-band envelope Y of the signal to be detected is calculated. min Repeat the above steps to calculate the DTW distance set (DTW) between the envelopes of all subbands to be detected and the envelopes of the signal subbands in the speaker template library. amp Similarly, DTW matching is performed on the first-order gradient sequence of the sub-band envelope to obtain the distance set DTW. grad The DTW distance set is obtained by fusing the two channels according to the following formula. i .

[0083] DTW i =λ*DTW amp +(1-λ)*DTW grad

[0084] Where λ is dynamically adjusted by the sub-band center frequency (low frequencies emphasize amplitude, high frequencies emphasize gradient), DTW amp The subband envelope and the DTW distance of the voiceprint database, DTW grad The distance between the first-order gradient sequence of the subband envelope and the DTW distance of the speaker database is given.

[0085] Compared to traditional single-channel DTW algorithms, which only compare envelope amplitude and ignore phase and shape differences, multi-dimensional DTW can simultaneously capture signal energy and waveform changes, improving sensitivity to subtle faults.

[0086] 2.2 Dynamic Weighted Difference of DTW Based on Two-Factor Decision Making

[0087] The previous step has already obtained the combined DTW distance of the two channels. i This step is based on the frequency band fault sensitivity α of the sub-band envelope. i Real-time signal-to-noise ratio (SNR) of sub-band envelope i These two two-factor pairs DTW i Perform dynamic weighting, where α i The kurtosis value K of the subband signal i Dynamically set, according to the formula:

[0088]

[0089] α i (t)=α i (t-1)+η*(K i -K ref )

[0090] Among them, Weight i (t) represents the weight, α i (t) represents the frequency band fault sensitivity, α i (t-1) represents the sensitivity to the previous frequency band fault, SNR i Represents the subband envelope signal-to-noise ratio, SNR ref The signal-to-noise ratio represents the reference standard, η is the sensitivity coefficient, and K i K represents the kurtosis value of the subband envelope. ref This represents the historical normal kurtosis mean. For a discrete signal sequence x... i ={x1,x2,…,x n The formula for calculating kurtosis K is as follows:

[0091]

[0092] Where μ represents the mean of the signal, and N represents the signal length (number of sampling points).

[0093] In this way, higher weights are given to low-frequency and high signal-to-noise ratio subbands, while the weights of high-frequency and low signal-to-noise ratio subbands are reduced, and the frequency band fault sensitivity α is increased. i Automatic adjustment based on signal kurtosis enables weighting of frequency bands with sudden kurtosis increases, forming a closed-loop feedback of the physical characteristics of coal blockage in bucket wheel excavators.

[0094] Then, the difference between the DTW distance of each subband and the distance of the normal template library is multiplied by the corresponding weight, and the weighted distance differences are summed to obtain Sum, as shown in the following formula:

[0095]

[0096] Among them, DTW i The DTW distance between the subband envelope of the signal to be detected and the signal in the speaker template library is given. Sum represents the DTW distance from the normal signal to the speaker template library, and Sum is the weighted sum of the DTW distance differences. In this way, the features of each subband envelope are fused.

[0097] 2.3 Adaptive Threshold Based on Sliding Window Statistics

[0098] This step enables the construction of an adaptive threshold based on sliding window statistics, utilizing the DTW difference of historical normal signals. A dynamic statistical model is constructed to automatically adjust the threshold according to fluctuations in ambient noise. The sum of the sub-band DTW distances between all normal signals and the speaker template library over a set time period is stored. nomal The adaptive threshold is calculated using the following formula:

[0099] Threshold(t)=(μ(t)+k*σ(t))

[0100]

[0101] Where Threshold(t) represents the final generated adaptive threshold, and μ(t) is the DTW difference of the historical normal signal within the window. The summation means, σ(t) is the standard deviation, h is the benchmark coefficient (usually 3), k is the sensitivity coefficient, and P is the mean of the summation. noise P represents the peak ambient noise level at the current moment. hist This represents the peak noise level under normal operating conditions, and β is the attenuation coefficient. The sensitivity coefficient k is dynamically adjusted based on the noise peak ratio to suppress the influence of abnormal operating condition noise on the decision threshold (such as the instantaneous noise when a train passes).

[0102] Finally, the detection signal is judged. The weighted sum of DTW distance differences (Sum) obtained in the previous step is compared with the adaptive threshold (Threshold(t)) obtained in this step. If the sum of differences is greater than the adaptive threshold, the detection signal is considered a coal blockage fault signal. Otherwise, it is considered a normal signal, and the detection signal is added to the historical normal signals for subsequent adaptive threshold generation. This achieves real-time monitoring and early warning of coal blockage in bucket wheel excavators.

[0103] III. Performance Analysis

[0104] To verify the performance of the acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-subband envelope comparison and decision-making described in this invention, corresponding simulation experiments were conducted using MATLAB. The mathematical model of the sound signal during bucket wheel excavator operation can be composed of multiple sine waves with different frequencies and amplitudes, and superimposed with broadband Gaussian white noise as interference, as shown in the following equation, where S1(t) represents the sound signal model of the bucket wheel excavator during normal operation, and S2(t) represents the sound signal model when coal blockage occurs in the bucket wheel excavator.

[0105] S1(t)=(1+1.5*sin(2πf1t)+1.0*sin(2πf2t))*sin(2πf3t)+n(t)

[0106] S2(t)=(1+1.0*sin(2πf4t)+0.7*sin(2πf5t))*sin(2πf6t)+n(t)

[0107] In the formula, f1 = 20Hz, f2 = 130Hz, f3 = 30Hz, f4 = 15Hz, f5 = 100Hz, f6 = 20Hz, and n(t) is random broadband Gaussian white noise.

[0108] The sampling frequency was set to 2048Hz, the number of sampling points was 6000, and the signal-to-noise ratio condition range for the simulation was -15dB to 5dB (5dB step, 5 groups in total). 1000 simulations were performed respectively. The ratio of normal signals to coal blockage fault signals in the dataset was 1:1.

[0109] pass Figure 3 and Figure 4 Comparison of subband signal plots shows that the noise component of the IMF component is significantly suppressed after CEEMDAN subband decomposition and wavelet threshold denoising. Quantitative tests show that under different initial signal-to-noise ratio (SNR) conditions, the SNR of the denoised signal is improved to [-9.50, -4.58, 0.49, 5.66, 10.80] dB. This result demonstrates that the algorithm achieves a stable SNR improvement of 5.4–5.8 dB under various noise environments, verifying the effective suppression of high-frequency random noise by subband decomposition combined with wavelet threshold denoising. Furthermore, RMSE (Root Mean Square Error) is introduced to evaluate the denoising effect, such as... Figure 5 As shown, the root mean square error (RMSE) of the signals before and after denoising decreases with increasing signal-to-noise ratio (SNR), and the RMSE value is even lower after denoising, with an average reduction of about 48%. This indicates that the fitting error between the denoised signal and the original clean signal is significantly reduced, and the extraction rate of the effective signal is higher after denoising. These results demonstrate that the method of CEEMDAN subband decomposition combined with wavelet thresholding has a good effect on noise suppression.

[0110] Figure 6 This paper demonstrates a performance comparison of the multi-subband envelope alignment decision algorithm of this invention with other existing algorithms. Simulation results show that the dynamic weighted difference method based on two-factor decision-making (DTW) and the adaptive threshold method based on sliding window statistics described in this invention improves the signal detection accuracy to a certain extent compared with current fixed weight and fixed threshold methods. The combined use of the two methods results in a more significant improvement. In low to medium signal-to-noise ratio (SNR) environments with significant noise (SNR > -10dB), the method described in this invention achieves an accuracy of over 86%, and in environments with extremely low SNR (SNR = -15dB), it still maintains a recognition rate of over 70%.

[0111] In summary, the acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-subband envelope comparison and decision proposed in this invention has good anti-interference ability in complex noise environments. It can suppress complex background noise and improve the signal-to-noise ratio in low signal-to-noise ratio environments. The dynamic weighting method and adaptive threshold method proposed in this invention have good effects on bucket wheel excavator operating signals with low signal-to-noise ratios. They can effectively distinguish between normal signals and coal blockage signals, thereby realizing real-time monitoring and early warning of coal blockage in bucket wheel excavators.

[0112] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Those skilled in the art, within the technical scope disclosed in this application, can easily conceive of variations or substitutions, all of which should be included within the scope of protection of this application. Therefore, the scope of this application should be determined by the scope of the claims.

Claims

1. An acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-subband envelope, characterized in that, Includes the following steps: Step 1: Construct a normal sound signature template library for bucket wheel excavators using the sound signals during normal operation; Step 2: Acquire the acoustic signals of the bucket wheel excavator in real time using an acoustic sensor array and perform low-pass filtering; Step 3: Perform fully adaptive noise set empirical mode decomposition on the filtered acoustic signal to obtain sub-band signals, and perform directional wavelet threshold noise reduction on the sub-band signals; Step 4: Take the envelope of the noise-reduced sub-band signal, calculate the first-order difference gradient sequence of each envelope, combine it with the normal voiceprint template library to perform multi-dimensional dynamic time warping (DTW), and calculate the DTW distance set. Step 5: Perform dynamic weighting on the DTW distance set based on two-factor decision-making, accumulate the results to obtain the sum of differences, and set an adaptive threshold; Step 6: Compare the sum of the differences with the adaptive threshold to obtain the signal classification result and complete the acoustic monitoring of coal blockage in the bucket wheel excavator.

2. The acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-sub-band envelopes according to claim 1, characterized in that, The specific process for calculating the DTW distance set is as follows: Two sub-band envelopes, X and Y, are given, where X is the first sub-band envelope of the speaker template library signal and Y is the first sub-band envelope of the signal to be detected, with lengths n and m, respectively. The DTW algorithm constructs an n×m distance matrix by calculating the Euclidean distance between corresponding points in the two sequences. After establishing the distance matrix, find k paths between a pre-defined starting point and ending point. Define the path where the cumulative distance function reaches its minimum value as the optimal normalized path between time series X and Y. Then, find the shortest distance (DTW) recursively. min =d(i,j)+min{d(i-1,j),d(i,j-1),d(i-1,j-1)}, where d(i,j) represents the Euclidean distance between points i and j; Calculate the DTW distance between the first subband envelope X of the voiceprint template library signal and the first subband envelope Y of the signal to be detected. min Repeat the above calculations to obtain the DTW distance set (DTW) between the envelopes of all subbands to be detected and the envelopes of the signal subbands in the speaker template library. amp Similarly, DTW matching is performed on the first-order gradient sequence of the sub-band envelope to obtain the distance set DTW. grad Then, a weighted fusion is performed to obtain the dual-channel joint DTW distance set DTW. i .

3. The acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-sub-band envelopes according to claim 2, characterized in that, The specific process for obtaining the sum of differences, Sum, is as follows: Based on two factors—the frequency band fault sensitivity of the subband envelope and the real-time signal-to-noise ratio of the subband envelope—DTW i Perform dynamic weighting, weight Among them, SNR i Represents the subband envelope signal-to-noise ratio, SNR ref Represents the reference signal-to-noise ratio; α i (t) is the difference between the kurtosis value of the subband envelope and the historical normal kurtosis mean, multiplied by a weighting coefficient, plus the frequency band fault sensitivity α from the previous time step. i (t-1) is obtained; Then, the difference between the DTW distance of each subband and the distance of the normal template library is multiplied by the corresponding weight, and the weighted distance differences are summed to obtain Sum.

4. The acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-sub-band envelopes according to claim 3, characterized in that, The setting of the adaptive threshold specifically involves: utilizing the DTW distance set of historical normal signals. A dynamic statistical model is constructed to automatically adjust the threshold according to fluctuations in ambient noise; the sum of the sub-band DTW distance sets of all normal signals and the voiceprint template library within a set time period is stored. nomal Sum nomal Standard deviation multiplied by sensitivity coefficient k plus Sum nomal The average value of the sensitivity coefficient is used to obtain the adaptive threshold; Where P nosie P represents the peak ambient noise level at the current moment. hist The noise peak value represents the historical normal operating condition, h is the reference coefficient, β is the attenuation coefficient, and the sensitivity coefficient is dynamically adjusted according to the noise peak ratio.

5. The acoustic monitoring method for coal blockage in bucket wheel excavators based on multi-sub-band envelopes according to claim 4, characterized in that, Step 6 is specifically implemented as follows: the weighted sum of DTW distance differences is compared with the adaptive threshold. If the weighted sum of DTW distance differences is greater than the adaptive threshold, the signal to be detected is considered to be a coal blockage fault signal; otherwise, it is considered to be a normal signal. The signal to be detected is then placed in the historical normal signals for subsequent adaptive threshold generation, thereby realizing real-time monitoring and early warning of coal blockage in the bucket wheel excavator.

Images