Distributed optical fiber monitoring system and method for faults of carrier rollers of belt conveyor

Abstract

The invention relates to the technical field of belt conveyor carrier roller fault monitoring, and discloses a belt conveyor carrier roller fault distributed optical fiber monitoring system which comprises a double-light-source DAS system, optical fiber information acquisition, filtering preprocessing, time domain analysis, time-frequency domain analysis, phase analysis, multi-mode feature fusion, fault classification and fault position and grade output. By building a dual-light-source polarization diversity DAS system and combining an adaptive filtering preprocessing technology, high signal-to-noise ratio acquisition and noise reduction processing of vibration signals are realized, dual-light-source orthogonal polarization state transmission effectively suppresses polarization fading, wavelet packet denoising and variable-step LMS filtering collaboratively filter environmental noise and pulse interference, and the noise reduction performance of the vibration signals is improved. Meanwhile, the application of the dynamic weighted root-mean-square value and the continuous wavelet transform ensures the accurate capture of the time domain energy characteristic and the frequency domain time-varying characteristic, and lays a reliable foundation for the multi-modal characteristic fusion.

Description

Technical Field

[0001] The present invention relates to the technical field of belt conveyor roller fault monitoring, and more particularly discloses a distributed optical fiber monitoring system and method for belt conveyor roller fault monitoring. Background Art

[0002] A belt conveyor, also known as a rubber belt conveyor or belt conveyor, is a machine that uses friction drive to continuously transport materials. It is mainly composed of a frame, conveyor belt, rollers, drums, tensioning devices, transmission devices, etc., and uses an endless conveyor belt as a load-bearing and traction component to transport materials from the loading point to the unloading point. It is widely used in various industries such as home appliances, electronics, tobacco, injection molding, post and telecommunications, printing, and food. As the core component of the belt conveyor, the roller is easily damaged during long-term use. Therefore, real-time monitoring of the roller's operating status is particularly important.

[0003] The patent document with authorization announcement number CN114739503A in the prior art discloses "a distributed optical fiber monitoring system and method for belt conveyor roller faults", which includes: first building a distributed optical fiber acoustic wave sensing system; first performing data preprocessing on the original signal collected by the DAS system; performing time domain analysis on the DAS collected data, calculating the root mean square value of the original data, and obtaining the root mean square value-position corresponding curve graph; analyzing the frequency domain and time-frequency domain of the data to determine the fault point on the time-frequency graph, and determining the location of the belt conveyor belt fault through the curve graph of the signal root mean square value and the corresponding position; and analyzing the frequency domain of the DAS signal to determine the roller operation conditions under different working conditions.

[0004] In addition, the patent document with the authorization announcement number CN112033669B discloses a "DAS-based belt conveyor trough roller fault monitoring method", which includes the following steps: by collecting the initial vibration signal along the optical fiber corresponding to the trough roller, performing phase demodulation, obtaining the phase waterfall diagram of the belt conveyor, and determining the roller fault point. If the location of the roller fault point is consistent with the pre-determined fault location, extract the current vibration signal of the roller fault point in the initial vibration signal, and draw a curve of the phase difference changing with time based on the current vibration signal and the reference vibration signal, respectively, to obtain the current time domain signal and the reference time domain signal, perform short-time Fourier transform, and obtain the current frequency domain signal corresponding to the current time domain signal and the reference frequency domain signal corresponding to the reference time domain signal. The roller fault point is monitored based on the current frequency domain signal and the reference frequency domain signal.

[0005] Although the existing technology has achieved the monitoring of roller faults through distributed fiber optic sensing technology, and at the same time, by building a distributed fiber optic acoustic wave sensing system, combining time domain root mean square value analysis and frequency domain short-time Fourier transform, the preliminary positioning of the fault point is achieved, and the Hilbert transform phase demodulation and short-time Fourier transform are used to improve the fault monitoring accuracy by comparing the phase waterfall diagram and frequency domain characteristics. However, the reliance on the Hilbert transform makes it insufficient in noise resistance in complex vibration environments, and fault location requires the fault location to be determined in advance, which lacks adaptability. In addition, the root mean square value analysis is sensitive to pulse noise, and the frequency domain analysis relies only on FFT, which makes it difficult to capture the time-varying characteristics of early faults. Moreover, the existing technology has not achieved adaptive fusion of multimodal features and deep learning intelligent classification, resulting in a low recognition rate of complex faults. Summary of the Invention

[0006] The present invention mainly provides a distributed optical fiber monitoring system and method for belt conveyor roller faults, which can solve the problems raised by the above background technology.

[0007] To solve the above technical problems, the present invention provides the following technical solutions, more specifically a distributed optical fiber monitoring method for belt conveyor roller faults, comprising: first, building a dual-light source polarization diversity DAS system, using the orthogonal polarization state transmission of a 1550nm main light source and a 1310nm reference light source to suppress polarization fading, and combining the wavelet packet denoising and variable step-size LMS filtering of an adaptive filtering preprocessing unit to perform noise reduction processing on the collected vibration signal; then, multimodal feature fusion is achieved by extracting time domain features through dynamic weighted root mean square value, analyzing time-frequency domain features through continuous wavelet transform combined with energy entropy, and extracting phase features through Hilbert transform combined with dynamic phase difference change rate; then, with the help of a CNN-LSTM fusion model and the introduction of an attention mechanism, fault classification is performed on the feature vector, and at the same time, accurate positioning and type identification of the fault point are achieved through the spatiotemporal joint criterion of dynamic phase difference analysis and the similarity comparison of adjacent features of the self-supervised fault location mechanism.

[0008] Furthermore, in the dual-light source polarization diversity DAS system, the 1550nm main light source and the 1310nm reference light source are transmitted in orthogonal polarization states through a polarization controller. The detection light is converted into a light pulse with a pulse width of 100ns by an acousto-optic modulator, amplified to 20dBm by an erbium-doped fiber amplifier, and then injected into the sensing fiber. The back-scattered Rayleigh light interferes with the reference light and is converted into an electrical signal by a photodetector.

[0009] Furthermore, in the adaptive filtering preprocessing, the wavelet packet denoising uses db4 wavelet to perform 5-layer decomposition, the high frequency sub-band sets a soft threshold of 3 times the noise standard deviation, and the step size factor of the variable step size LMS filter is satisfy:

[0010] in is the conveyor belt speed.

[0011] Furthermore, when the dynamic weighted root mean square value is used to calculate the time domain characteristics, the vibration signal is processed through a sliding window mechanism, and the weight factor within the window is:

[0012] in , is the window length, and the center data weight is strengthened to suppress edge noise. The calculation formula is:

[0013] Achieve accurate capture of roller vibration energy anomalies.

[0014] Furthermore, the continuous wavelet transform combined with energy entropy analysis of time-frequency domain features is to use Morlet wavelet to perform time-frequency decomposition on the signal, divide the time-frequency graph into 100-300Hz, 500-800Hz, and 1-2kHz frequency bands, and calculate the energy entropy of each frequency band by the following function:

[0015] in is the energy ratio of the frequency band.

[0016] Furthermore, the Hilbert transform combined with the dynamic phase difference change rate to extract phase features is used to perform fault detection through the following dynamic phase difference change rate:

[0017] Also set the sliding window length:

[0018] in is the belt speed, is the roller rotation period.

[0019] Furthermore, the multimodal feature fusion is to adaptively fuse the time domain features, time-frequency domain features, and phase features, and the feature vector is:

[0020] The time domain feature weight is:

[0021] The time-frequency feature weight is:

[0022] The phase feature weight is:

[0023] This enables multimodal feature fusion.

[0024] Furthermore, the CNN-LSTM fusion model includes 3 convolutional layers and 128-neuron LSTM layer, and the attention mechanism weight calculation formula is:

[0025] in is the LSTM hidden state, Output features for CNN.

[0026] Furthermore, in the self-supervised fault location mechanism, the similarity of feature vectors of adjacent rollers is calculated by cosine similarity, and the formula is as follows:

[0027] in The value range is [−1,1]. The closer the value is to 1, the more similar the eigenvector directions are. Otherwise, the difference is greater. and Position and The roller feature vector of .

[0028] According to another aspect of the present invention, a distributed optical fiber monitoring system for belt conveyor roller faults is provided. The system is implemented based on the above-mentioned distributed optical fiber monitoring method for belt conveyor roller faults, and specifically includes: First, optical fiber information is collected through a dual-light source DAS system. Polarization fading is suppressed by the orthogonal polarization state transmission of a 1550nm main light source and a 1310nm reference light source. The collected vibration signal is then filtered and preprocessed, and noise reduction is achieved through wavelet packet denoising and variable-step-size LMS filtering in an adaptive filter preprocessing unit. Time domain analysis, time-frequency domain analysis, and phase analysis are then performed. In time domain analysis, energy features are extracted through dynamic weighted root mean square value. In time-frequency domain analysis, frequency characteristics are mined using continuous wavelet transform combined with energy entropy. In phase analysis, phase mutations are captured using Hilbert transform combined with dynamic phase difference change rate. Multi-dimensional features are then fused to form a vector containing time domain, time-frequency domain, and phase features. Fault classification is then performed on the feature vector using a CNN-LSTM fusion model and the introduction of an attention mechanism. Finally, the fault point is accurately located and graded by outputting a level by comparing the spatiotemporal joint criterion of dynamic phase difference analysis with the similarity of adjacent features in the self-supervised fault location mechanism.

[0029] The present invention is based on a distributed optical fiber monitoring system and method for belt conveyor roller faults, and has the following beneficial effects: by building a dual-light source polarization diversity DAS system and combining it with adaptive filtering preprocessing technology, high signal-to-noise ratio acquisition and noise reduction processing of vibration signals are achieved; the orthogonal polarization state transmission of the dual light sources effectively suppresses polarization fading; wavelet packet denoising and variable step-size LMS filtering collaboratively filter out environmental noise and pulse interference, thereby providing a pure data basis for subsequent feature extraction; at the same time, the application of dynamic weighted root mean square value and continuous wavelet transform ensures the accurate capture of time domain energy characteristics and frequency domain time-varying characteristics, laying a reliable foundation for multimodal feature fusion; Through the collaboration of adaptive fusion of multimodal features and the CNN-LSTM intelligent model, high-precision classification and positioning of roller faults are achieved. The dynamic phase difference change rate combined with the joint spatiotemporal criterion improves the false alarm resistance of fault point identification. The self-supervised fault location mechanism compares the features of adjacent rollers through cosine similarity, breaking through the reliance of existing technical methods on prior fault locations, and thus realizing graded identification of fault types (jamming / fracture / bearing wear) and severity. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0031] Figure 1 It is a schematic diagram of the principle process; Figure 2 Schematic diagram of the method flow chart. DETAILED DESCRIPTION

[0032] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0033] According to one aspect of the present invention, Figure 1-2 As shown in the figure, a distributed optical fiber monitoring system and method for belt conveyor roller faults are provided. First, a dual-light source polarization diversity DAS system is built. The orthogonal polarization state transmission of the 1550nm main light source and the 1310nm reference light source is used to suppress polarization fading. The collected vibration signal is denoised by combining the wavelet packet denoising and variable step-size LMS filtering of the adaptive filter preprocessing unit. In the dual-light source polarization diversity DAS system, a 1550nm main light source and a 1310nm reference light source are transmitted in orthogonal polarization states through a polarization controller. The probe light is converted into a 100ns pulse width optical pulse by an acousto-optic modulator, amplified to 20dBm by an erbium-doped fiber amplifier, and then injected into the sensing fiber. The backscattered Rayleigh light interferes with the reference light and is converted into an electrical signal by a photodetector. In the adaptive filter preprocessing, wavelet packet denoising uses db4 wavelet to perform 5-layer decomposition, and the high-frequency sub-band sets a soft threshold of 3 times the noise standard deviation, and the step size factor of the variable step size LMS filter is 1. satisfy:

[0034] in The filter parameters can be adjusted dynamically with the conveyor belt speed, and the excellent noise reduction performance can be maintained when the speed fluctuates. Among them, 0.01 is used as a coefficient to control the step value within a reasonable range. When the exponential function has no scaling coefficient, its output value has a large range of variation, and the coefficient of 0.01 can make the step value The values taken at different conveyor belt speeds meet the system requirements, avoiding system instability caused by too large a step length, or slow convergence due to too small a step length. In addition, the value of 2.5 is a representative belt speed value in actual operation of the belt conveyor and is often regarded as a standard belt speed. When the belt speed is around 2.5m / s, the system operation is relatively stable and the noise characteristics are also in a relatively typical state. When the belt speed deviates from 2.5m / s, the step length is calculated through the exponential function. It will be dynamically adjusted accordingly (for example, when the belt speed is higher than 2.5m / s, the step size will be reduced accordingly to adapt to the changes in noise characteristics during high-speed operation and improve the filtering effect; conversely, when the belt speed is lower than 2.5m / s, the step size will also be adjusted accordingly) to ensure that the system can effectively filter out noise under different working conditions.

[0035] Then, the time domain features are extracted by dynamic weighted root mean square value, the time-frequency domain features are analyzed by continuous wavelet transform combined with energy entropy, and the phase features are extracted by Hilbert transform combined with dynamic phase difference change rate to achieve multimodal feature fusion. When the dynamic weighted root mean square value is used to calculate the time domain characteristics, the vibration signal is processed through a sliding window mechanism, and the weight factor in the window is:

[0036] in , is the window length, and the center data weight is strengthened to suppress edge noise. The calculation formula is:

[0037] Achieve accurate capture of roller vibration energy anomalies; The continuous wavelet transform combined with energy entropy analysis of time-frequency domain characteristics is to use Morlet wavelet to perform time-frequency decomposition of the signal, and divide the time-frequency graph into 100-300Hz (low frequency band, corresponding to the fundamental frequency characteristics of normal operation of the roller), 500-800Hz (medium frequency band, reflecting the impact characteristics of early bearing wear), and 1-2kHz frequency band (high frequency band, corresponding to the violent vibration characteristics of roller jamming or breakage). The energy entropy of each frequency band is calculated using the following function:

[0038] in is the energy ratio of the frequency band; In addition, the Hilbert transform is combined with the dynamic phase difference change rate to extract the phase feature for fault detection through the following dynamic phase difference change rate:

[0039] Also set the sliding window length:

[0040] in is the belt speed, is the rotation period of the roller. When the roller is stuck, unbalanced or other faults occur, the phase difference change rate Sudden changes may occur due to vibration phase disorder. Combined with continuous monitoring within the window, it can effectively distinguish between normal belt speed fluctuations and phase anomalies caused by real faults. Finally, multimodal feature fusion is to adaptively fuse time domain features, time-frequency domain features, and phase features. The feature vector is:

[0041] The time domain feature weight is:

[0042] Where 0.1 is the dynamic weighted RMS value ( ), which physically means the vibration energy dividing point between normal operation and abnormal fault of the roller. The exponential coefficient of 5 is set to make the weight function form a steep nonlinear transition near the critical value of 0.1, ensuring that the weight is sensitive to the change of fault energy and avoiding misjudgment of normal fluctuations. The time-frequency feature weight is:

[0043] The phase feature weight is:

[0044] 5 is used to change the dynamic phase difference rate Normalized to the reasonable range of [0,1], due to roller failure Typical values are in the range of 0-5. This coefficient makes the phase feature weight comparable with other modal feature weights (such as time domain and time-frequency domain); This enables multimodal feature fusion.

[0045] The CNN-LSTM fusion model is then used to introduce an attention mechanism to classify faults based on feature vectors. Furthermore, the dynamic phase difference analysis's spatiotemporal joint criterion and the self-supervised fault location mechanism's adjacent feature similarity comparison enable accurate fault location and type identification. The CNN-LSTM fusion model consists of three convolutional layers (with kernel sizes of 5, 7, and 9, respectively), an LSTM layer with 128 neurons, and the attention mechanism weight calculation formula is:

[0046] in is the LSTM hidden state, Output features of CNN. During the propagation process, CNN first extracts the local spatial features of the vibration signal and outputs LSTM learns the dynamic evolution of feature sequences based on temporal relationships. , the attention mechanism is based on and The correlation calculation weight Dynamically weight features at different times and spatial locations to enhance the influence of key fault-related features. The weighted fusion features are then input into the fully connected layer, and the Softmax function is used to implement probabilistic classification of fault types. This allows the model to focus more on the unique temporal-spatial coupling characteristics of roller faults, improving classification accuracy. In the self-supervised fault location mechanism, the similarity of feature vectors of adjacent rollers is calculated by cosine similarity. The formula is as follows:

[0047] in The value range is [−1,1]. The closer the value is to 1, the more similar the eigenvector directions are. Otherwise, the difference is greater. and Position and The roller feature vector of .

[0048] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by ordinary technicians in this technical field within the essential scope of the present invention also fall within the scope of protection of the present invention.

Claims

1. A distributed optical fiber monitoring method for belt conveyor roller faults, characterized in that: The method includes: first, building a dual-light source polarization diversity DAS system, utilizing the orthogonal polarization state transmission of a 1550nm main light source and a 1310nm reference light source to suppress polarization fading, and combining wavelet packet denoising and variable step-size LMS filtering of an adaptive filtering preprocessing unit to perform noise reduction processing on the collected vibration signal; then, extracting time domain features through dynamic weighted root mean square value, analyzing time-frequency domain features through continuous wavelet transform combined with energy entropy, and extracting phase features through Hilbert transform combined with dynamic phase difference change rate to achieve multimodal feature fusion; then, using a CNN-LSTM fusion model and introducing an attention mechanism to classify faults on feature vectors, and simultaneously achieving accurate positioning and type identification of fault points through a spatiotemporal joint criterion of dynamic phase difference analysis and a similarity comparison of adjacent features of a self-supervised fault location mechanism.

2. The distributed optical fiber monitoring method for belt conveyor roller faults according to claim 1, characterized in that: In the dual-light source polarization diversity DAS system, a 1550nm main light source and a 1310nm reference light source are transmitted in orthogonal polarization states through a polarization controller. The detection light is converted into a 100ns pulse width optical pulse by an acousto-optic modulator, amplified to 20dBm by an erbium-doped fiber amplifier, and then injected into the sensing fiber. The backscattered Rayleigh scattered light interferes with the reference light and is converted into an electrical signal by a photodetector.

3. The distributed optical fiber monitoring method for belt conveyor roller faults according to claim 1 is characterized in that: In the adaptive filtering preprocessing, wavelet packet denoising uses db4 wavelet to perform 5-layer decomposition, and a soft threshold of 3 times the noise standard deviation is set for the high-frequency sub-band. The step size factor of the variable step size LMS filter is satisfy: . in is the conveyor belt speed.

4. The distributed optical fiber monitoring method for belt conveyor roller faults according to claim 1, characterized in that: When the dynamic weighted root mean square value is used to calculate the time domain characteristics, the vibration signal is processed through a sliding window mechanism, and the weight factor within the window is: . in , is the window length, and the center data weight is strengthened to suppress edge noise. The calculation formula is: . Achieve accurate capture of roller vibration energy anomalies.

5. The distributed optical fiber monitoring method for belt conveyor roller faults according to claim 1, characterized in that: The continuous wavelet transform combined with energy entropy analysis of time-frequency domain features is to use Morlet wavelet to perform time-frequency decomposition on the signal, divide the time-frequency graph into 100-300Hz, 500-800Hz, and 1-2kHz frequency bands, and calculate the energy entropy of each frequency band using the following function: . in is the energy ratio of the frequency band.

6. The distributed optical fiber monitoring method for belt conveyor roller faults according to claim 1, characterized in that: The Hilbert transform combined with the dynamic phase difference change rate to extract phase features is used to perform fault detection through the following dynamic phase difference change rate: . Also set the sliding window length: . in is the belt speed, is the roller rotation period.

7. The distributed optical fiber monitoring method for belt conveyor roller faults according to claim 1, characterized in that: The multimodal feature fusion is to adaptively fuse the time domain features, time-frequency domain features, and phase features. The feature vector is: . The time domain feature weight is: . The time-frequency feature weight is: . The phase feature weight is: . This enables multimodal feature fusion.

8. The distributed optical fiber monitoring method for belt conveyor roller faults according to claim 1, characterized in that: The CNN-LSTM fusion model contains 3 convolutional layers and 128-neuron LSTM layer. The attention mechanism weight calculation formula is: . in is the LSTM hidden state, Output features for CNN.

9. The distributed optical fiber monitoring method for belt conveyor roller faults according to claim 1, characterized in that: In the self-supervised fault location mechanism, the similarity of feature vectors of adjacent rollers is calculated by cosine similarity, and the formula is as follows: . in The value range is [−1,1]. The closer the value is to 1, the more similar the eigenvector directions are. Otherwise, the difference is greater. and Position and The roller feature vector of .

10. A distributed optical fiber monitoring system for belt conveyor roller faults, the system being implemented based on the distributed optical fiber monitoring method for belt conveyor roller faults according to any one of claims 1 to 8, and specifically comprising: First, optical fiber information is collected through a dual-light source DAS system. Polarization fading is suppressed by the orthogonal polarization state transmission of a 1550nm main light source and a 1310nm reference light source. The collected vibration signal is then filtered and preprocessed, and noise reduction is achieved through wavelet packet denoising and variable-step-size LMS filtering in an adaptive filter preprocessing unit. Time domain analysis, time-frequency domain analysis, and phase analysis are then performed. In time domain analysis, energy features are extracted through dynamic weighted root mean square value. In time-frequency domain analysis, frequency characteristics are mined using continuous wavelet transform combined with energy entropy. In phase analysis, phase mutations are captured using Hilbert transform combined with dynamic phase difference change rate. Multi-dimensional features are then fused to form a vector containing time domain, time-frequency domain, and phase features. Fault classification is then performed on the feature vector using a CNN-LSTM fusion model and the introduction of an attention mechanism. Finally, the fault point is accurately located and graded by outputting a level by comparing the spatiotemporal joint criterion of dynamic phase difference analysis with the similarity of adjacent features in the self-supervised fault location mechanism.

Images