Fault monitoring method and system for slewing mechanism of stacker-reclaimer based on dynamic threshold value
By dynamically adjusting the wavelet threshold and combining deep learning and empirical mode decomposition, the problem of noise interference in the vibration signal of the stacker-reclaimer rotary mechanism was solved, enabling accurate identification and real-time diagnosis of early minor faults.
Patent Information
- Application Number
- CN202511547131.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-28
- Publication Date
- 2026-01-23
AI Technical Summary
Existing technologies suffer from severe noise interference in the vibration signals of stacker-reclaimer rotary mechanisms, leading to a decrease in the accuracy of fault diagnosis, especially in the early stages of minor faults.
A dynamic threshold-based method is adopted. By acquiring vibration signal feature data, using a deep learning model to predict noise intensity, and dynamically adjusting the wavelet threshold, effective denoising of vibration signals and preservation of fault features are achieved. Combined with empirical mode decomposition and wavelet transform, the accuracy and robustness of fault diagnosis are improved.
It significantly improves the accuracy and real-time performance of fault diagnosis for stacker-reclaimer rotary mechanisms, effectively overcomes noise interference, and ensures the identification of early, minor faults.
Smart Images
Figure CN121376646A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of fault monitoring, in particular to a dynamic threshold-based fault monitoring method and system for a slewing mechanism of a stacker-reclaimer. BACKGROUND
[0002] As a high-efficiency equipment for continuous loading and unloading of bulk materials in modern industry, a stacker-reclaimer has been widely applied to the stacking and reclaiming operations in bulk material storage yards such as ports, wharfs, coal storage yards and power plants. The stacker-reclaimer includes a portal frame, a slewing mechanism, a cantilever frame driven by the slewing mechanism and other upper structures such as a bucket wheel.
[0003] The slewing mechanism is a core component of the stacker-reclaimer. The transmission components in the slewing mechanism, such as meshing gears, bear heavy mechanical loads. Vibration analysis is a commonly used means for fault detection of the slewing mechanism of the stacker-reclaimer. If a fault occurs in the slewing bearing during operation, such as wear of the rolling elements or damage to the raceway, the characteristics of the vibration signal will change. By installing vibration sensors at key positions of the slewing mechanism, such as the inner ring, outer ring and bearing seat of the slewing bearing, vibration signals can be collected. Subsequently, the running state of the slewing mechanism can be obtained through the vibration signals.
[0004] Due to the complex port environment, there are a large amount of mechanical noise and electromagnetic noise, which can mix into the vibration signals of the slewing mechanism, causing the fault characteristic signals to be submerged, and reducing the accuracy of fault diagnosis based on vibration analysis. In particular, in the early weak fault stage, the fault characteristic signals are already very weak, and are more susceptible to noise, making it difficult to accurately identify. SUMMARY
[0005] The present application aims to provide a dynamic threshold-based fault monitoring method and system for a slewing mechanism of a stacker-reclaimer, which can effectively overcome the limitations of existing methods in dealing with noise interference in the vibration signals of the slewing mechanism of the stacker-reclaimer, and significantly improve the accuracy, robustness and real-time performance of fault diagnosis.
[0006] A dynamic threshold-based fault monitoring method for a slewing mechanism of a stacker-reclaimer, comprising:
[0007] Obtaining the vibration signals of the slewing mechanism of the stacker-reclaimer to be tested in the current detection period;
[0008] Performing feature extraction on the vibration signals to obtain feature data of the current detection period; the feature data includes instantaneous signal-to-noise ratio, signal energy change rate, total signal energy, high-frequency energy proportion, peak factor, kurtosis coefficient, effective value, peak difference, main frequency amplitude, frequency barycenter and vibration displacement standard deviation;
[0009] input the feature data of the current detection period and the feature data of N detection periods before the current detection period into the feature prediction network to obtain a noise intensity prediction value of the current detection period;
[0010] obtain a basic threshold for wavelet threshold denoising based on the instantaneous signal-to-noise ratio of the current detection period, and obtain a correction factor for wavelet threshold denoising based on the noise intensity prediction value of the current detection period;
[0011] obtain a wavelet threshold of the current detection period based on the basic threshold and the correction factor;
[0012] perform empirical mode decomposition on the vibration signal to obtain a plurality of initial single-component signals, and perform wavelet transform on each of the initial single-component signals to obtain a plurality of initial wavelet coefficient sets;
[0013] perform denoising processing on each of the initial wavelet coefficient sets based on the wavelet threshold of the current detection period to obtain a plurality of denoised wavelet coefficient sets;
[0014] perform reconstruction on each of the denoised wavelet coefficient sets to obtain a plurality of denoised single-component signals;
[0015] perform reconstruction on each of the denoised single-component signals to obtain a denoised signal, and input the denoised signal into a fault diagnosis model to obtain an operating state of the slewing mechanism of the stacker-reclaimer.
[0016] Optionally, the basic threshold is greater than the size of the instantaneous signal-to-noise ratio multiplied by the inverse ratio, and the size of the correction factor is proportional to the size of the noise intensity prediction value.
[0017] Optionally, the ratio of the effective signal power to the noise power of the vibration signal is the instantaneous signal-to-noise ratio, the effective signal power is obtained by subtracting the noise power from the total power of the vibration signal, and the noise power is obtained by estimation using the minimum mean square error criterion;
[0018] obtain the difference between the energy of the middle-late segment signal and the energy of the middle-early segment signal of the vibration signal, divide the difference by the energy of the middle-early segment signal of the vibration signal, and obtain the signal energy change rate;
[0019] the total signal energy is the energy of the vibration signal;
[0020] the high-frequency energy proportion is the ratio of the energy of the vibration signal with a frequency greater than 5 kHz to the total signal energy;
[0021] the effective value is the root mean square value of the signal;
[0022] the peak factor is the ratio of the peak value to the effective value of the vibration signal;
[0023] The kurtosis is a ratio of a square of a fourth central moment of the vibration signal to a square of a second central moment of the vibration signal.
[0024] The peak difference is a difference between a maximum value and a minimum value of the vibration signal.
[0025] The main frequency amplitude is a frequency value corresponding to a maximum amplitude in a power spectrum of the vibration signal.
[0026] The frequency barycenter is a barycenter frequency of the power spectrum of the vibration signal.
[0027] The vibration displacement standard deviation is obtained by performing twice integration on the vibration signal to obtain a displacement signal, and obtaining a standard deviation of the displacement signal.
[0028] Optionally, the feature prediction network adopts a three-layer LSTM network structure.
[0029] Optionally, the feature prediction network is trained by using an early stopping method.
[0030] Optionally, the wavelet threshold value of the current detection period is obtained based on the basic threshold value and the correction factor, and the method comprises the following steps:
[0031] The high-frequency wavelet threshold value of the current detection period is obtained based on the basic threshold value and the correction factor, and the expression is as follows:
[0032] H a = Q x Y x D a ;
[0033] The low-frequency wavelet threshold value of the current detection period is obtained based on the basic threshold value and the correction factor, and the expression is as follows:
[0034] H b = Q x Y x D b ;
[0035] Wherein, H a is the high-frequency wavelet threshold value, H b is the low-frequency wavelet threshold value, Q is the basic threshold value, Y is the correction factor, D a is the high-frequency weight coefficient, and D b is the low-frequency weight coefficient.
[0036] The wavelet threshold value comprises the high-frequency wavelet threshold value and the low-frequency wavelet threshold value; an intermediate boundary value is determined based on a boundary layer number of wavelet transform, and a layer number greater than the intermediate boundary value is high frequency, and a layer number less than or equal to the intermediate boundary value is low frequency.
[0037] Optionally, the wavelet threshold value of the current detection period is used to perform denoising processing on each initial wavelet coefficient set to obtain a plurality of denoising wavelet coefficient sets, and the method comprises the following steps:
[0038] Based on the intermediate demarcation value, each of the initial wavelet coefficient sets is divided to obtain a plurality of high-frequency coefficient sets and a plurality of low-frequency coefficient sets;
[0039] For each of the high-frequency coefficient sets, coefficients greater than or equal to the high-frequency wavelet threshold value are retained, and coefficients less than or equal to the high-frequency wavelet threshold value are set to zero; for each of the low-frequency coefficient sets, coefficients greater than the low-frequency wavelet threshold value are shrunk, and coefficients less than the low-frequency wavelet threshold value are set to zero, to obtain a plurality of the denoising wavelet coefficient sets.
[0040] The application also provides a dynamic threshold-based stacker-reclaimer slewing mechanism fault monitoring system, which comprises:
[0041] A data acquisition module is configured to acquire a vibration signal of a to-be-tested stacker-reclaimer slewing mechanism in a current detection period;
[0042] A feature extraction module is configured to perform feature extraction on the vibration signal to obtain feature data of the current detection period; the feature data comprises an instantaneous signal-to-noise ratio, a signal energy change rate, a total signal energy, a high-frequency energy proportion, a peak factor, a kurtosis coefficient, an effective value, a peak difference, a main frequency amplitude, a frequency barycenter, and a vibration displacement standard deviation.
[0043] A strength prediction module is configured to input the feature data of the current detection period and the feature data of N previous detection periods of the current detection period into a feature prediction network to obtain a noise strength prediction value of the current detection period.
[0044] A threshold factor module is configured to acquire a basic threshold value for wavelet threshold denoising based on the instantaneous signal-to-noise ratio of the current detection period; and acquire a correction factor for wavelet threshold denoising based on the noise strength prediction value of the current detection period.
[0045] A wavelet threshold module is configured to acquire a wavelet threshold value of the current detection period based on the basic threshold value and the correction factor.
[0046] A decomposition transformation module is configured to perform empirical mode decomposition on the vibration signal to obtain a plurality of initial single-component signals; and perform wavelet transformation on each of the initial single-component signals to obtain a plurality of initial wavelet coefficient sets.
[0047] A denoising module is configured to perform denoising processing on each of the initial wavelet coefficient sets based on the wavelet threshold value of the current detection period to obtain a plurality of denoising wavelet coefficient sets.
[0048] A wavelet reconstruction module is configured to reconstruct each of the denoising wavelet coefficient sets to obtain a plurality of denoising single-component signals.
[0049] A fault diagnosis module is configured to reconstruct each of the denoised single-component signals to obtain a denoised signal, and input the denoised signal into a fault diagnosis model to obtain an operating state of the slewing mechanism of the stacker-reclaimer under test.
[0050] Effects of the present application are as follows:
[0051] The stacker-reclaimer slewing mechanism fault monitoring method based on a dynamic threshold value of the present application constructs an adaptive threshold function based on real-time statistical characteristics of a vibration signal. The noise intensity trend is predicted by a deep learning model, and the wavelet threshold value is adjusted in advance to achieve a dynamic balance of "increasing the threshold value (strengthening noise reduction) when the noise is strong, and reducing the threshold value (preserving features) when the fault features are obvious". BRIEF DESCRIPTION OF DRAWINGS
[0052] Figure 1 is a flow chart of the stacker-reclaimer slewing mechanism fault monitoring method based on a dynamic threshold value of the present application. DETAILED DESCRIPTION
[0053] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings.
[0054] Figure 1 is a flow chart of the stacker-reclaimer slewing mechanism fault monitoring method based on a dynamic threshold value of the present application, as Figure 1 indicated, the present application provides a stacker-reclaimer slewing mechanism fault monitoring method based on a dynamic threshold value, which comprises:
[0055] S1, obtaining a vibration signal of the slewing mechanism of the stacker-reclaimer under test in a current detection period. A high-frequency vibration sensor is installed at key positions such as the slewing bearing bearing seat and the gear meshing area to synchronously collect three-dimensional vibration acceleration signals. The sampling frequency of the high-frequency vibration sensor is 20 kHz.
[0056] S2, performing feature extraction on the vibration signal to obtain feature data of the current detection period. The feature data includes instantaneous signal-to-noise ratio, signal energy change rate, total signal energy, high-frequency energy proportion, peak factor, kurtosis coefficient, effective value, peak difference, main frequency amplitude, frequency gravity center and vibration displacement standard deviation.
[0057] Specifically, the ratio of the effective signal power to the noise power of the vibration signal is the instantaneous signal-to-noise ratio; the effective signal power is obtained by subtracting the noise power from the total power of the vibration signal, and the noise power is obtained by using the minimum mean square error criterion estimation. The instantaneous signal-to-noise ratio directly reflects the proportion of useful information and noise in the vibration signal. The lower the instantaneous signal-to-noise ratio, the stronger the noise interference, and the threshold value needs to be increased to strengthen noise reduction.
[0058] The difference between the energy of the middle-back section signal and the energy of the middle-front section signal of the vibration signal is obtained, and the difference is divided by the energy of the middle-front section signal of the vibration signal to obtain a signal energy change rate. The vibration signal is converted to a time-frequency domain through a short-time Fourier transform, and the energy changes of each frequency component are weighted and summed to highlight the energy fluctuations of the fault high-frequency characteristics. The signal energy change rate is used to capture the sudden change of the energy of the vibration signal, and when the change rate exceeds 20%, it may indicate the burst of noise or fault characteristics.
[0059] The total signal energy is the energy of the vibration signal. When the stacker reclaimer slewing mechanism is in normal operation, the energy is relatively stable, and a sudden increase in noise or an increase in fault will cause the total energy to rise significantly, and the total signal energy provides a basis for judging the overall strength of the signal.
[0060] The high-frequency energy proportion is the ratio of the energy of the frequency > 5 kHz in the vibration signal to the total signal energy. Early fault characteristics (such as cracks and wear) also often exhibit high-frequency vibration, and the high-frequency energy proportion is used to distinguish whether the high-frequency component is noise or a fault signal, and when the proportion is too high, other characteristics need to be combined to determine whether the high-frequency threshold needs to be increased.
[0061] The effective value is the root mean square value of the signal. The effective value reflects the average energy level of the signal. The effective value increases with the increase of the noise intensity or the fault degree, and is a stable indicator for measuring the overall amplitude of the signal, which provides a reference for setting the basic threshold.
[0062] The peak factor is the ratio of the peak value of the vibration signal to the effective value. Noise impact or component impact (such as gear tooth breakage) will significantly increase the peak value. The peak factor is sensitive to sudden noise and impact faults, and is the core input of the LSTM model to identify noise mutation scenarios.
[0063] The kurtosis coefficient is the ratio of the square of the fourth-order central moment to the square of the second-order central moment of the vibration signal. Noise impact or fault impact will increase the kurtosis, while smooth noise may decrease the kurtosis. The kurtosis coefficient is used to distinguish the impact of the signal, and when the kurtosis rises sharply, the threshold needs to be increased to suppress impact noise.
[0064] The peak difference value is the difference between the maximum value and the minimum value of the vibration signal. Noise interference or component jamming will cause the peak difference value to suddenly expand, and the peak difference value combined with the peak factor can more comprehensively describe the extreme value of the amplitude of the signal.
[0065] The main frequency amplitude is the frequency value corresponding to the maximum amplitude in the power spectrum of the vibration signal. Noise interference may introduce a new main frequency, while fault characteristics will appear at a specific frequency. The main frequency amplitude is used to identify the dominant frequency component of the signal.
[0066] The frequency center of gravity is the center of gravity frequency of the power spectrum of the vibration signal. When noise dominates, the frequency center of gravity shifts to high frequency, and when a fault develops, it may gather around a specific frequency. The frequency center of gravity reflects the distribution center of signal energy in the frequency domain, and assists in judging the frequency characteristics of noise or faults.
[0067] The vibration displacement standard deviation is obtained by performing a second integration on the vibration signal to obtain a displacement signal, and obtaining the standard deviation of the displacement signal. The vibration displacement standard deviation reflects the stability of the vibration amplitude of the rotary mechanism. The vibration intensification caused by noise or faults will increase the value, and it is especially suitable for evaluating the stability of low-frequency vibration.
[0068] S3, input the feature data of the current detection period and the feature data of the N detection periods before the current detection period into the feature prediction network to obtain the noise intensity prediction value of the current detection period.
[0069] Preferably, the feature prediction network selects a three-layer LSTM network structure. The input layer of the feature prediction network includes 32 neurons, the hidden layer of the feature prediction network includes 64 neurons, and the output layer of the feature prediction network includes 1 neuron.
[0070] The model training process needs to incorporate the noise characteristics of the port scene: the training data set contains vibration signals under normal working conditions, different fault types, and different noise intensities; the early stopping method is used to prevent overfitting; the model volume is reduced by 60% through model quantization compression technology to ensure real-time operation on the edge computing unit.
[0071] To cope with noise mutation scenarios, the model introduces an attention mechanism, giving higher weights to features such as "peak factor" and "high-frequency energy proportion" that are highly correlated with noise mutation, so that the model can quickly capture noise mutation signals.
[0072] S4, based on the instantaneous signal-to-noise ratio of the current detection period, obtain the basic threshold value of wavelet threshold denoising; based on the noise intensity prediction value of the current detection period, obtain the correction factor of wavelet threshold denoising. Specifically, the basic threshold value is greater than the size of the instantaneous signal-to-noise ratio multiplied by the inverse ratio; the size of the correction factor is directly proportional to the size of the noise intensity prediction value.
[0073] S5, based on the basic threshold value and the correction factor, obtain the wavelet threshold value of the current detection period.
[0074] S5 includes:
[0075] S51, based on the basic threshold value and the correction factor, obtain the high-frequency wavelet threshold value of the current detection period; the expression is:
[0076] H a =Q×Y×D a .
[0077] S52, obtaining a low-frequency wavelet threshold of the current detection period based on the basic threshold and the correction factor; the expression is:
[0078] H b = Q x Y x D b ;
[0079] wherein, H a is a high-frequency wavelet threshold, H b is a low-frequency wavelet threshold, Q is a basic threshold, Y is a correction factor, D a is a high-frequency weight coefficient, and D b is a low-frequency weight coefficient.
[0080] The wavelet threshold includes a high-frequency wavelet threshold and a low-frequency wavelet threshold; an intermediate boundary value is determined based on the number of boundary layers of the wavelet transform; the number of layers greater than the intermediate boundary value is high frequency, and the number of layers less than or equal to the intermediate boundary value is low frequency.
[0081] S6, performing empirical mode decomposition on the vibration signal to obtain a plurality of initial single-component signals; performing wavelet transform on each initial single-component signal to obtain a plurality of initial wavelet coefficient sets.
[0082] S7, performing denoising processing on each initial wavelet coefficient set based on the wavelet threshold of the current detection period to obtain a plurality of denoised wavelet coefficient sets.
[0083] S7 is specifically: dividing each initial wavelet coefficient set based on the intermediate boundary value to obtain a plurality of high-frequency coefficient sets and a plurality of low-frequency coefficient sets.
[0084] For each high-frequency coefficient set, coefficients greater than or equal to the high-frequency wavelet threshold are retained, and coefficients less than or equal to the high-frequency wavelet threshold are set to zero; for each low-frequency coefficient set, coefficients greater than the low-frequency wavelet threshold are shrunk, and coefficients less than the low-frequency wavelet threshold are set to zero, to obtain a plurality of denoised wavelet coefficient sets.
[0085] S8, reconstructing each denoised wavelet coefficient set to obtain a plurality of denoised single-component signals.
[0086] S9, reconstructing each denoised single-component signal to obtain a denoised signal; inputting the denoised signal into a fault diagnosis model to obtain the running state of the slewing mechanism of the stacker-reclaimer to be tested.
[0087] Specifically, taking a slewing mechanism of a certain stacker-reclaimer as an example, a vibration sensor detects a sudden noise, a sliding window intercepts a signal segment containing the noise, and the instantaneous signal-to-noise ratio is calculated to be 15 dB, the normal value is 30 dB, the signal energy change rate is 80%, the normal value is less than 20%, the peak factor is 2, and the normal value is less than 5. These characteristics are integrated into a time sequence input together with the characteristics of the previous 5 periods.
[0088] After the LSTM model receives the feature input, it predicts that the noise intensity maintains a high level.
[0089] The basic threshold factor is set to 0.6 according to the instantaneous signal-to-noise ratio, 0.3 when the normal signal-to-noise ratio is 30 dB, and the correction factor is set to 1.4 when the noise intensity is high. The final high-frequency wavelet threshold is 0.6*1.4*1.5=1.26, and the low-frequency wavelet threshold is 0.6*1.4*0.9=0.756.
[0090] In the wavelet transform, the part greater than 1.26 in the high-frequency layer coefficient is retained, mainly the fault feature, and the part less than 0.756 is set to zero; the part greater than 0.756 in the low-frequency layer coefficient is shrinkage processing, which not only suppresses noise but also avoids feature distortion.
[0091] When the noise intensity decreases, the instantaneous signal-to-noise ratio rises to 35 dB, the LSTM model predicts that the noise will stabilize at a low level, and the dynamic threshold generator will automatically reduce the threshold, the high-frequency wavelet threshold will be reduced to 0.8, and the weak fault feature will be ensured not to be filtered too much.
[0092] The application also provides a dynamic threshold-based stacker-reclaimer slewing mechanism fault monitoring system, which comprises:
[0093] A data acquisition module is configured to acquire a vibration signal of a stacker-reclaimer slewing mechanism to be measured in a current detection period.
[0094] A feature extraction module is configured to extract features from the vibration signal to obtain feature data of the current detection period. The feature data includes an instantaneous signal-to-noise ratio, a signal energy change rate, a total signal energy, a high-frequency energy proportion, a peak factor, a kurtosis coefficient, an effective value, a peak difference, a main frequency amplitude, a frequency barycenter, and a vibration displacement standard deviation.
[0095] An intensity prediction module is configured to input the feature data of the current detection period and the feature data of N previous detection periods into a feature prediction network to obtain a noise intensity prediction value of the current detection period.
[0096] A threshold factor module is configured to obtain a basic threshold for wavelet threshold denoising based on the instantaneous signal-to-noise ratio of the current detection period, and obtain a correction factor for wavelet threshold denoising based on the noise intensity prediction value of the current detection period.
[0097] A wavelet threshold module is configured to obtain a wavelet threshold of the current detection period based on the basic threshold and the correction factor.
[0098] A decomposition and transformation module is configured to perform empirical mode decomposition on the vibration signal to obtain a plurality of initial single-component signals, and perform wavelet transform on each initial single-component signal to obtain a plurality of initial wavelet coefficient sets.
[0099] A denoising module is configured to perform denoising processing on each initial wavelet coefficient set based on a wavelet threshold of a current detection period to obtain a plurality of denoised wavelet coefficient sets.
[0100] A wavelet reconstruction module is configured to reconstruct each denoised wavelet coefficient set to obtain a plurality of denoised single-component signals.
[0101] A fault diagnosis module is configured to reconstruct each denoised single-component signal to obtain a denoised signal, and input the denoised signal into a fault diagnosis model to obtain an operation state of the slewing mechanism of the stacker-reclaimer under test.
[0102] The above-described embodiments are merely preferred embodiments of the present application, and are not intended to limit the scope of the present application. Various modifications and improvements to the technical solutions of the present application made by those of ordinary skill in the art without departing from the design spirit of the present application shall fall within the scope of protection of the present application.
Claims
1. A method for fault monitoring of the rotary mechanism of a stacker-reclaimer based on dynamic thresholds, characterized in that, It includes: Acquire the vibration signal of the rotary mechanism of the stacker-reclaimer under test within the current testing cycle; Feature extraction is performed on the vibration signal to obtain feature data for the current detection cycle. The feature data includes instantaneous signal-to-noise ratio, signal energy change rate, total signal energy, high-frequency energy ratio, peak factor, kurtosis coefficient, RMS value, peak difference, main frequency amplitude, frequency centroid, and vibration displacement standard deviation. The feature data of the current detection period and the feature data of the previous N detection periods are input into the feature prediction network to obtain the noise intensity prediction value of the current detection period. The basic threshold for wavelet threshold denoising is obtained based on the instantaneous signal-to-noise ratio of the current detection period; the correction factor for wavelet threshold denoising is obtained based on the predicted noise intensity value of the current detection period. The wavelet threshold for the current detection period is obtained based on the base threshold and the correction factor; The vibration signal is subjected to empirical mode decomposition to obtain several initial single-component signals; each of the initial single-component signals is subjected to wavelet transform to obtain several initial wavelet coefficient sets. Based on the wavelet threshold of the current detection period, each of the initial wavelet coefficient sets is denoised to obtain several denoised wavelet coefficient sets. Each of the denoised wavelet coefficient sets is reconstructed to obtain several denoised single-component signals; The denoised single-component signals are reconstructed to obtain denoised signals; the denoised signals are input into the fault diagnosis model to obtain the operating status of the stacker-reclaimer rotary mechanism under test.
2. The fault monitoring method for the rotary mechanism of a stacker-reclaimer based on dynamic thresholds according to claim 1, characterized in that, The magnitude of the base threshold is inversely proportional to the magnitude of the instantaneous signal-to-noise ratio; the magnitude of the correction factor is directly proportional to the magnitude of the predicted noise intensity value.
3. The method for fault monitoring of the rotary mechanism of a stacker-reclaimer based on dynamic thresholds according to claim 1, characterized in that, The ratio of the effective signal power to the noise power of the vibration signal is the instantaneous signal-to-noise ratio; the effective signal power is obtained by subtracting the noise power from the total power of the vibration signal, and the noise power is estimated using the minimum mean square error criterion. The energy difference between the middle and later segments of the vibration signal and the energy of the middle and earlier segments is obtained, and the difference is divided by the energy of the middle and earlier segments of the vibration signal to obtain the rate of change of the signal energy. The total signal energy is the energy of the vibration signal; The high-frequency energy ratio is the ratio of the energy with a frequency > 5kHz in the vibration signal to the total signal energy; The effective value is the root mean square value of the signal; The peak factor is the ratio of the peak value to the effective value of the vibration signal; The kurtosis coefficient is the ratio of the square of the fourth central moment to the square of the second central moment of the vibration signal. The peak difference is the difference between the maximum and minimum values of the vibration signal; The main frequency amplitude is the frequency value corresponding to the largest amplitude value in the power spectrum of the vibration signal; The centroid of the frequency is the centroid frequency of the power spectrum of the vibration signal; The vibration displacement standard deviation is obtained by performing a second integral on the vibration signal to obtain the displacement signal, and then obtaining the standard deviation of the displacement signal.
4. The fault monitoring method for the rotary mechanism of a stacker-reclaimer based on dynamic thresholds according to claim 1, characterized in that, The feature prediction network uses a three-layer LSTM network structure.
5. The fault monitoring method for the rotary mechanism of a stacker-reclaimer based on dynamic thresholds according to claim 4, characterized in that, The feature prediction network is trained using the early stopping method.
6. The fault monitoring method for the rotary mechanism of a stacker-reclaimer based on dynamic thresholds according to claim 1, characterized in that, The process of obtaining the wavelet threshold for the current detection period based on the base threshold and the correction factor includes: The high-frequency wavelet threshold for the current detection period is obtained based on the aforementioned base threshold and correction factor; the expression is: H a =Q×Y×D a ; The low-frequency wavelet threshold for the current detection period is obtained based on the aforementioned base threshold and correction factor; the expression is: H b =Q×Y×D b ; Among them, H a H is the high-frequency wavelet threshold. b For low-frequency wavelet thresholding, Q is the base threshold, Y is the correction factor, and D is the threshold value. a D represents the high-frequency weighting coefficient. b Low-frequency weighting coefficients; The wavelet threshold includes the high-frequency wavelet threshold and the low-frequency wavelet threshold; the intermediate boundary value is determined based on the number of boundary layers of the wavelet transform, and the number of layers greater than the intermediate boundary value is high frequency, and the number of layers less than or equal to the intermediate boundary value is low frequency.
7. The fault monitoring method for the rotary mechanism of a stacker-reclaimer based on dynamic thresholds according to claim 6, characterized in that, The initial wavelet coefficient sets are denoised based on the wavelet threshold of the current detection period to obtain several denoised wavelet coefficient sets, specifically: Based on the intermediate boundary value, each of the initial wavelet coefficient sets is divided to obtain several high-frequency coefficient sets and several low-frequency coefficient sets; For each set of high-frequency coefficients, coefficients greater than or equal to the high-frequency wavelet threshold are retained, and coefficients less than or equal to the high-frequency wavelet threshold are set to zero; for each set of low-frequency coefficients, coefficients greater than the low-frequency wavelet threshold are shrunk, and coefficients less than the low-frequency wavelet threshold are set to zero, thus obtaining several sets of denoised wavelet coefficients.
8. A fault monitoring system for the rotary mechanism of a stacker-reclaimer based on dynamic thresholds, characterized in that, It includes: The data acquisition module is used to acquire the vibration signal of the rotary mechanism of the stacker-reclaimer under test within the current testing cycle; The feature extraction module is used to extract features from the vibration signal to obtain feature data for the current detection cycle. The feature data includes instantaneous signal-to-noise ratio, signal energy change rate, total signal energy, high-frequency energy ratio, peak factor, kurtosis coefficient, RMS value, peak difference, main frequency amplitude, frequency centroid, and vibration displacement standard deviation. The intensity prediction module is used to input the feature data of the current detection period and the feature data of the previous N detection periods into the feature prediction network to obtain the noise intensity prediction value of the current detection period. The threshold factor module is used to obtain the basic threshold for wavelet threshold denoising based on the instantaneous signal-to-noise ratio of the current detection period; and to obtain the correction factor for wavelet threshold denoising based on the predicted noise intensity value of the current detection period. The wavelet thresholding module is used to obtain the wavelet threshold for the current detection period based on the base threshold and the correction factor. The decomposition and transformation module is used to perform empirical mode decomposition on the vibration signal to obtain several initial single-component signals; and to perform wavelet transform on each of the initial single-component signals to obtain several initial wavelet coefficient sets. The denoising module is used to denoise each of the initial wavelet coefficient sets based on the wavelet threshold of the current detection period to obtain several denoised wavelet coefficient sets. The wavelet reconstruction module is used to reconstruct each of the denoised wavelet coefficient sets to obtain several denoised single-component signals. The fault diagnosis module is used to reconstruct each of the denoised single-component signals to obtain a denoised signal; The denoised signal is input into the fault diagnosis model to obtain the operating status of the rotary mechanism of the stacker-reclaimer under test.