Carrier roller fault monitoring method and system based on optical fiber auscultation

Abstract

The invention relates to a carrier roller fault monitoring method based on optical fiber auscultation, and the method comprises the steps: collecting and preprocessing a carrier roller operation sound signal in real time, and extracting three types of complementary acoustic features in the preprocessed sound signal in parallel to obtain three types of feature maps; performing normalization and size unification on the three types of feature maps, and then performing fusion to form a three-channel fusion image; inputting the three-channel fusion image into a pre-constructed deep learning model, performing feature learning and classification, and outputting probability distribution of the running state of the carrier roller; and according to the probability distribution of the running state of the carrier roller, abnormal alarm information is output, and the position of the carrier roller is marked. Compared with the prior art, the system has the advantages of being high in system integration degree and high in practicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring of coal conveying in thermal power plants, and in particular to a method and system for monitoring faults of a roller based on optical fiber auscultation. BACKGROUND

[0002] A belt conveyor is a key device for material transportation in coal power, metallurgy, mining and port, and its operation safety is closely related to the state of the roller. The roller has a large number of quantities, is widely distributed, and has a complex operating environment. Common abnormalities include bearing damage, jamming, and eccentric wear. These faults are often characterized by "abnormal sound" in the early stage.

[0003] The traditional manual inspection method has the problems of low efficiency, high missed detection rate and poor safety. The point sensor scheme can detect vibration or temperature, but has complex wiring and high maintenance cost, and cannot realize continuous monitoring. The existing optical fiber auscultation system is mainly used for pipeline leakage or perimeter intrusion detection, and lacks targeted optimization in roller monitoring, and has the problems of insufficient sensitivity, high false alarm rate and poor environmental adaptability.

[0004] Therefore, there is an urgent need for an integrated optical fiber auscultation system that is strong in anti-interference and can accurately locate roller abnormalities.

[0005] Defects of the prior art: 1. Insufficient signal sensitivity: the traditional Φ-OTDR system uses linear optical fiber laying, which cannot fully respond to local acoustic vibration; 2. Limited recognition accuracy: conventional algorithms focus on spectral analysis, and have poor anti-interference ability to field noise, wind noise and belt impact sound; 3. System dispersion: the auscultation host, sensing unit and upper computer platform are not in closed loop linkage, and have poor data consistency and real-time performance; 4. Insufficient adaptability: ordinary sensing schemes have insufficient stability in high temperature, high dust and high noise environments in mining areas and power plants; complex deployment: point sensors have a large number, complex power supply and high maintenance cost.

[0006] Patent CN120123974A discloses a lightweight intelligent auscultation system for building equipment that integrates model distillation. The system includes: a voiceprint data acquisition module for real-time acquisition of voiceprint data and equipment status data generated during the operation of building equipment; a voiceprint data processing and feature extraction module for preprocessing and extracting features from the raw voiceprint data; a lightweight model construction module for constructing a lightweight model suitable for edge devices based on the extracted features and equipment status data, serving as a lightweight student model; a model distillation module for transferring knowledge from the teacher model to the lightweight student model through distillation; a lightweight model optimization module for further optimizing the lightweight student model to obtain an optimized model; and an edge deployment module for deploying the optimized model to edge devices, enabling real-time inference of whether the building equipment is faulty. While ensuring real-time performance, this system improves the real-time performance, accuracy, and computational efficiency of the building equipment auscultation system. However, it does not consider the signal fidelity extraction problem during long-distance transmission. Summary of the Invention

[0007] The purpose of this invention is to overcome the defects of the prior art by providing a method and system for monitoring idler roller faults based on fiber optic auscultation.

[0008] The objective of this invention can be achieved through the following technical solutions: A method for monitoring idler roller faults based on fiber optic auscultation, the method comprising the following steps: Real-time acquisition and preprocessing of idler roller operation sound signals; parallel extraction of three types of complementary acoustic features from the preprocessed sound signals to obtain three types of feature maps. The three types of feature maps are normalized and their sizes are unified, and then fused to form a three-channel fused image. The three-channel fused image is input into a pre-built deep learning model for feature learning and classification, and the probability distribution of the idler roller's running state is output. Based on the probability distribution of the idler roller's operating status, an abnormal alarm message is output and the location of the idler roller is marked.

[0009] Furthermore, the three types of feature maps specifically include: log-Mel spectrogram, Mel frequency cepstral coefficient map, and spectral contrast map; the log-Mel spectrogram is used to simulate the auditory perception characteristics of the human ear and reflects the change of the energy distribution of the sound signal on the Mel frequency scale over time. The Mel frequency cepstral coefficient diagram is used to characterize the timbre features of a sound and remove the correlation between different features; The spectral contrast diagram is used to reflect the clarity, harmonic structure, and noise level of the sound spectrum.

[0010] Furthermore, the process of generating the log-Mel spectrum includes: Perform a short-time Fourier transform on the preprocessed audio signal: Where STFT{x(t)}(τ, ω) is the short-time Fourier transform result, representing the complex amplitude at time τ and angular frequency ω; x(t) is the original sound signal, i.e., the acoustic signal generated when the roller is running; w(t - τ) is a window function that moves at time τ to achieve short-time analysis of the signal; τ is a time parameter, representing the center position of the window function; ω is an angular frequency parameter. It is a complex rotational basis, where j is the imaginary unit.

[0011] Calculate the power spectrum for each frame: Where P(τ, ω) is the power spectrum, representing the signal power at time τ and angular frequency ω. STFT(τ, ω) is the short-time Fourier transform result.

[0012] The obtained power spectrum is obtained by weighted summation of the power spectrum using a Mel filter bank: Where S(m, t) is the Mel spectrum, representing the energy value at Mel frequency m and per unit time; P(k, t) is the power spectrum value at frequency k and per unit time; Hm(k) is the coefficient of the m-th filter in the Mel filter bank; N is the total number of frequency points; and m is the Mel frequency index. Taking the logarithm of the Mel spectrum yields the logarithmic Mel spectrum. Wherein, LogMel(m, t) is the log-Mel spectrum, representing the logarithmic energy value at Mel frequency m and unit time; S(m, t) is the Mel spectrum value; ε is a small constant used to avoid zero or negative values ​​when taking the logarithm.

[0013] Furthermore, the process of generating the Mel frequency cepstral coefficient map includes: Based on the logarithmic Mel spectrum, a discrete cosine transform is performed to obtain the Mel frequency cepstral coefficients. Arranging these coefficients frame by frame forms the Mel frequency cepstral coefficient diagram. The calculation formula is as follows: Where MFCC(n, t) is the Mel frequency cepstral coefficient, representing the value at the nth coefficient and unit time; N is the index of the cepstral coefficient; M is the number of Mel filter banks; LogMel(m, t) is the value of the log-Mel spectrum at Mel frequency m and unit time; m is the Mel frequency index.

[0014] Furthermore, the spectral contrast diagram reflects the clarity, harmonic structure, and noise level of the sound spectrum by calculating the contrast between peak and valley energy in different frequency bands of the sound signal's spectrum; the calculation formula is as follows: in, ε is the spectral contrast, representing the contrast value at frequency band b and unit time; X(k, t) is the value of the spectrum of the sound signal at frequency k and time t; Bb is the b-th frequency band in the spectrum; k is the frequency index; ε is a small constant used to avoid division by zero or taking the logarithm of zero.

[0015] Furthermore, the process of collecting the sound signal of the idler roller operation specifically includes: When the belt conveyor is running, the vibration and abnormal noise signals generated by the idler rollers are converted into fiber optic phase disturbance signals by the fiber optic detector; The auscultation fiber optic host is based on the principle of phase-sensitive optical time-domain reflectometry. It sends laser pulses and receives Rayleigh scattering signals through pulse digital encoding. The Rayleigh scattering signals are demodulated to recover the fiber phase disturbance signals, and then digitally converted to obtain the idler roller running sound signals.

[0016] Furthermore, the Rayleigh scattering signal is transmitted to the auscultatory fiber optic host via the host fiber optic terminal box, and then the auscultatory fiber optic host performs phase demodulation and digitization conversion to obtain the idler roller running sound signal; the idler roller running sound signal is uploaded to the monitoring platform via Ethernet or fiber optic network; The fiber optic detector adopts a wound-type sensitivity enhancement structure and is installed on the guardrails on both sides of the idler roller to improve the response sensitivity and positioning accuracy to local abnormal noises.

[0017] Furthermore, the deep learning model is a ResNet18 convolutional neural network; the deep learning model is trained using labeled idler roller sound data through supervised learning methods and a cosine annealing strategy.

[0018] Furthermore, the process of outputting the abnormal alarm information specifically includes: judging based on the probability distribution of the idler roller's operating status: when the probability of the abnormal category exceeds a preset threshold, the system automatically triggers an alarm and generates a work order; The specific operating states of the idler rollers include: normal operation, abnormal noise from the idler rollers, bearing wear, damaged rubber coating, excessively tight cleaner, and dust entering the bearings.

[0019] A monitoring system for a fiber optic auscultation-based idler roller fault monitoring method as described above, the system comprising: An optical fiber enhanced detector array is arranged along both sides of the conveyor belt idler roller to convert the sound waves and vibration signals generated by the operation of the idler roller into optical fiber phase perturbation signals. The host fiber optic terminal box is used to aggregate the fiber phase disturbance signals of the fiber optic sensitizer array and to achieve splicing and protection. The auscultation fiber optic host, based on the principle of phase-sensitive optical time-domain reflectometry, is used to send laser pulses through pulse digital encoding, receive Rayleigh scattering signals and demodulate them to recover the fiber phase disturbance signal, and then obtain the idler roller running sound signal through digital conversion. The data upload module is used to transmit the idler running sound signal output by the auscultation fiber optic host to the intelligent monitoring and analysis platform via Ethernet or fiber optic network; The intelligent monitoring and analysis platform is used to receive and process the idler roller running sound signals from the auscultation fiber optic host, perform acoustic feature extraction, output the probability distribution of the idler roller running status and abnormal alarms, and visualize the idler roller running status, idler roller position and abnormal alarm information on the platform interface and user interface. The user-end display module allows users to view the idler roller's operating status, position, acoustic waveform, and alarm information in real time on a PC or mobile device, enabling remote intelligent operation and maintenance.

[0020] Compared with the prior art, the present invention has the following advantages: (1) This invention works by converting the vibration and abnormal noise signals generated by the idler rollers into optical phase disturbances via a fiber optic detector when the belt conveyor is running. The signals are then transmitted to the auscultation host via the host fiber optic terminal box, where the auscultation fiber optic host performs phase demodulation and digital conversion. The converted acoustic signals are then uploaded to the intelligent monitoring and analysis platform via the network. The platform uses feature extraction and intelligent recognition algorithms for analysis, and finally outputs the idler roller running status and triggers an alarm on the platform and at the user end.

[0021] This system integrates a fiber optic enhanced detector array, a main fiber optic terminal box, a stethoscope fiber optic host, an intelligent monitoring and analysis platform, and a user-end display module. It combines fiber optic sound pickup, signal processing, intelligent recognition, and visual management, demonstrating excellent industrial applicability. By providing an integrated idler roller fault stethoscope monitoring system, it achieves long-distance, multi-point continuous acoustic monitoring; realizes idler roller-level acoustic positioning, real-time waveform and sound playback, video linkage, and work order closed-loop functions; and integrates detection, alarm, and maintenance on a single platform, significantly reducing the workload of manual inspection.

[0022] (2) This invention employs a wound-type optical fiber layout near the sound source of the belt idler, significantly improving the response sensitivity and positioning accuracy to local abnormal noises; compared with the traditional straight-line optical fiber laying method, the sensitivity is increased by about 3 times, solving the problem of weak signal of Φ-OTDR technology in a high-noise industrial environment. The fusion of multiple complementary features, especially the feature sensitive to signal details, helps to capture the weak acoustic signal changes generated by the idler in the early stage of the fault, thereby realizing early warning of the fault.

[0023] (3) This invention employs PDC signal modulation and phase demodulation algorithms on the host side to overcome the contradiction between spatial resolution and transmission distance; it achieves continuous monitoring over a long distance of 1.2 km while maintaining a positioning accuracy of ±3.5 m. Furthermore, on the platform side, a three-channel acoustic feature fusion method is adopted, which integrates logarithmic Mel spectrum, MFCC and spectral contrast map into a deep convolutional neural network; effectively improving the recognition accuracy of roller noise patterns and realizing fine-grained intelligent classification of multiple types of faults. Attached Figure Description

[0024] Figure 1 This is a flowchart of an integrated method for monitoring idler roller faults based on fiber optic auscultation provided in an embodiment of the present invention; Figure 2 This is a system structure diagram of an integrated method for monitoring idler roller faults based on fiber optic auscultation provided in an embodiment of the present invention; Figure 3 This is a detailed system flowchart of an integrated method for monitoring idler roller faults based on fiber optic auscultation, provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0026] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0027] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0028] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed during use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0029] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0030] Furthermore, terms such as "horizontal" and "vertical" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0031] Example 1 like Figure 1 As shown, this invention provides an integrated method for monitoring idler roller faults based on fiber optic auscultation, comprising the following steps: S1: Real-time acquisition and preprocessing of idler roller running sound signals, and parallel extraction of three types of complementary acoustic features from the preprocessed sound signals to obtain three types of feature maps; Specifically, the system uses distributed optical fiber acoustic sensing technology (DAS) to collect the sound of the idler rollers running in real time; the sampling rate is 16 kHz, and the continuous signal stream is processed in segments of fixed duration (such as 3 seconds).

[0032] Preferably, three types of complementary acoustic features are extracted in parallel for each sound segment: Log-Mel spectrogram—This feature simulates the auditory perception characteristics of the human ear, reflecting the change of the energy distribution of a sound signal on the Mel frequency scale over time, and is sensitive to changes in pitch and energy. Its extraction process includes: 1. Perform a Short-Time Fourier Transform (STFT) on the original audio signal: Where STFT{x(t)}(τ, ω) is the short-time Fourier transform result, representing the complex amplitude at time τ and angular frequency ω; x(t) is the original sound signal, i.e., the acoustic signal generated when the roller is running; w(t - τ) is a window function that moves at time τ to achieve short-time analysis of the signal; τ is a time parameter, representing the center position of the window function; ω is an angular frequency parameter. It is a complex rotational basis, and j is the imaginary unit.

[0033] 2. Calculate the power spectrum for each frame: Where P(τ, ω) is the power spectrum, representing the signal power at time τ and angular frequency ω. STFT(τ, ω) is the short-time Fourier transform result.

[0034] 3. The power spectrum is obtained by weighted summation using a Mel filter bank: Where S(m, t) is the Mel spectrum, representing the energy value at Mel frequency m and unit time; P(k, t) is the power spectrum value at frequency k and time t; Hm(k) is the coefficient of the m-th filter in the Mel filter bank; N is the total number of frequency points; and m is the Mel frequency index. 4. Take the logarithm of the Mel spectrum to obtain the logarithmic Mel spectrum: Wherein, LogMel(m, t) is the log-Mel spectrum, representing the logarithmic energy value at Mel frequency m and unit time; S(m, t) is the Mel spectrum value; ε is a small constant used to avoid zero or negative values ​​when taking the logarithm.

[0035] Mel-frequency cepstral coefficients (MFCCs) are coefficients obtained by performing a discrete cosine transform (DCT) on the logarithmic Mel spectrum. They effectively characterize the timbre of a sound and remove correlations between different features. Arranging them frame by frame forms an MFCC diagram. The calculation formula is as follows: Where MFCC(n, t) is the Mel frequency cepstral coefficient, representing the value at the nth coefficient and unit time; N is the index of the cepstral coefficient; M is the number of Mel filter banks; LogMel(m, t) is the value of the log-Mel spectrum at Mel frequency m and unit time; m is the Mel frequency index.

[0036] Spectral Contrast – This feature reflects the clarity, harmonic structure, and noise level of a spectrum by calculating the contrast between peak and valley energies across different frequency bands. High contrast typically corresponds to a clear narrowband signal, while low contrast may indicate broadband noise. The calculation formula is as follows: Where Cb(t) is the spectral contrast, representing the contrast value at frequency band b and unit time; t is the time frame index; X(k, t) is the value of the spectrum of the sound signal at frequency k and time t; Bb is the b-th frequency band in the spectrum; k is the frequency index; ε is a small constant used to avoid division by zero or taking the logarithm of zero.

[0037] S2: Normalize and unify the size of the three types of feature maps, then fuse them to form a three-channel fused image; input the three-channel fused image into a pre-built deep learning model to perform feature learning and classification, and output the probability distribution of the roller running status; Specifically, Normalize and unify the size of the three types of feature maps (e.g., 640×640 pixels). The R / G / B channels are superimposed to form a three-channel fused image, which serves as the input for the deep learning model.

[0038] Preferred, Feature learning and classification were performed using a ResNet18 convolutional neural network. The network input is a three-channel fused feature map, and the output is the probability distribution of the corresponding idler roller operating state. The model supports 11 state classifications (normal operation, bearing wear, rubber coating damage, abnormal noise of the idler roller, dust ingress, and excessive tightness of the cleaner).

[0039] Preferred, Supervised training was performed using labeled idler roller sound data; During the model training phase, a labeled set of idler roller sound samples was used. The training parameters included a batch size of 24, an initial learning rate of 0.1, a total number of rounds of 300, and a cosine annealing learning rate decay strategy. During the real-time inference phase, the collected sound signals are input into the model through the same feature extraction process, and the model outputs the roller anomaly category and confidence level.

[0040] The deep convolutional neural network model is trained through supervised learning. The input is a fused three-channel acoustic feature image, and the output is a multi-class label of the idler roller's running status.

[0041] The model training phase uses a set of labeled idler roller sound samples. S3: Based on the probability distribution of the idler roller's operating status, output abnormal alarm information and mark the location of the idler roller.

[0042] Specifically, the platform triggers anomaly alarms and marks the position of the idler rollers based on the identification results; users can view alarm information, waveforms, video footage and historical data in real time to achieve closed-loop management.

[0043] Preferably, the idler roller operating status includes multiple categories such as normal operation, abnormal roller noise, bearing wear, damaged rubber coating, excessively tight cleaner, and dust ingress into the bearing. The output of the method is combined with the idler roller position mapping data to realize idler roller-level acoustic positioning and alarm identification on a visual interface.

[0044] Example 2 This embodiment provides a monitoring system for a fiber optic auscultation-based idler roller fault monitoring method as described in any of Embodiment 1. The system includes: An optical fiber optic enhanced detector array is arranged along both sides of the conveyor belt idler rollers to pick up the acoustic signals generated during the rollers' operation; it employs a wound optical fiber structure. It can convert the acoustic and vibration signals generated by the rollers into optical fiber phase perturbation signals, achieving high-sensitivity sound pickup.

[0045] The main fiber optic terminal box is responsible for the aggregation, splitting, and protection of signals from multiple detectors. It features a built-in fusion splice tray and connection adapter interface to ensure low-loss optical signal transmission. It is used to aggregate optical signals from multiple detectors and to achieve splicing and protection. The auscultation fiber optic host, based on the Φ-OTDR optical time-domain reflectometry principle, demodulates the fiber optic echo signal. Pulse digital coding (PDC) is used to enhance echo energy, improving the signal-to-noise ratio and spatial resolution. The host has multi-channel acquisition capabilities, with a single channel covering 1.2 km and a positioning accuracy of ±3.5 m. It transmits laser pulses and receives Rayleigh scattering signals via PDC, performing phase demodulation to recover the idler roller's operating sound signal. The data upload module uploads the digital signals output by the host to the platform server in real time via Ethernet or fiber optic network. It supports TCP / IP protocol and secure encrypted transmission. It is used to transmit the digital acoustic data output by the auscultation fiber optic host to the platform server via Ethernet or fiber optic communication. The intelligent monitoring and analysis platform, deployed on the server side, integrates modules such as data access, feature analysis, model recognition, and alarm management. It provides functions such as "one-click auscultation," "real-time waveform display," "video linkage," and "closed-loop processing of abnormal work orders." It receives and processes acoustic data from the auscultation fiber optic host, performs acoustic feature extraction, fault identification, and alarm logic, and visually displays the idler roller's operating status on the interface.

[0046] The user-end display module allows users to view the status of idlers, acoustic waveforms, and alarm results in real time on a PC or mobile device, enabling remote intelligent operation and maintenance.

[0047] The system forms a closed-loop monitoring system through fiber optic links and network communication to realize real-time detection, anomaly identification, and alarm output of the idler roller's operating status.

[0048] The system processing flow includes: Sound signal acquisition: The sound signals of the belt conveyor rollers during operation are acquired through an array of fiber optic hypersensitive detectors; Optical signal demodulation: The auscultation fiber optic host performs phase demodulation on the Rayleigh scattering signal based on the Φ-OTDR principle to obtain the acoustic signal of the idler roller running; Sound segmentation processing: The continuous sound stream is divided into segments of fixed duration, and each segment is an analysis unit; Acoustic feature extraction: Parallel extraction of log-Mel spectra, Mel frequency cepstral coefficients, and spectral contrast plots; Feature fusion and recognition: The three types of feature maps are normalized and uniformly sized, then superimposed to form a three-channel fused image, which is then input into a trained deep convolutional neural network model for classification and recognition; Alarm and visualization display: Based on the identification results, the system generates roller status categories. When the probability of an abnormal category exceeds the threshold, the system automatically triggers an alarm and displays it on the monitoring platform.

[0049] Specifically, the fiber optic enhanced detector adopts a wound fiber optic structure and is installed on the guardrails on both sides of the conveyor belt roller to improve the local acoustic signal pickup sensitivity and positioning accuracy.

[0050] Specifically, the auscultation fiber optic host adopts a multi-channel design, with a single-channel monitoring distance of 1.2 km and a positioning accuracy of ±3.5 m, and supports parallel access of multiple detector arrays.

[0051] Specifically, the intelligent monitoring and analysis platform includes: a feature extraction module, a feature fusion module, a classification and recognition module, and an alarm output module. The feature extraction module is used to generate log-Mel spectrum, Mel frequency cepstral coefficients (MFCC), and spectral contrast feature map. The feature fusion module is used to normalize, unify the size of, and fuse the three feature maps into a three-channel image to form the input image. The classification and recognition module uses a deep convolutional neural network to learn and classify the input feature map, and outputs a probability distribution of the idler roller's operating status. The alarm output module is used to trigger an alarm and generate a work order when the probability of an abnormal category exceeds a preset threshold.

[0052] The intelligent monitoring and analysis platform supports roller-level anomaly location, sound playback, video linkage, and work order closed-loop processing, enabling remote intelligent operation and maintenance.

[0053] A specific application of a fiber optic auscultation-based idler roller fault monitoring system: Application scenario: Power plant coal conveyor belts The system deploys fiber optic enhanced detector arrays along the belt conveyor in the transfer station area, with a spacing of about 7 m on each side, for a total of 12 sets of detectors. The auscultation fiber optic host adopts a dual-channel configuration, with a single channel covering a distance of 1.2 km; The host output signal is uploaded to the platform server via Ethernet; The platform performs real-time analysis and visualization of the idler roller noise, and automatically generates alarm events when abnormal categories such as idler roller noise and bearing wear are identified. Users can play back the sound and confirm the alarm through the client, and the system will generate maintenance work orders in a synchronized manner to achieve remote closed-loop management.

[0054] This invention provides an integrated idler roller fault auscultation and monitoring system, comprising a fiber optic auscultation host, a fiber optic enhanced detector array, fiber optic connection components, and an intelligent monitoring platform. Based on the Φ-OTDR fiber optic acoustic detection principle, it achieves high-fidelity acquisition and intelligent identification of abnormal noise signals from idler roller operation. The system employs a wound-type enhanced sensitivity structure and PDC pulse digital coding technology to improve sound pickup sensitivity, signal-to-noise ratio, and anti-interference capability. Combined with the TSA time-domain signal enhancement algorithm and acoustic pattern recognition model, it can achieve real-time detection, precise location, and remote alarm of idler roller anomalies. This system is particularly suitable for the complex operating environment of power plant belt conveyors, characterized by long distances, high noise, and high dust. It features high integration, high identification accuracy, and simple deployment, significantly reducing the frequency of manual inspections and improving equipment operation safety and maintenance efficiency. It achieves continuous acoustic monitoring over long distances and at multiple points; the fiber optic wound-type enhanced sensitivity structure improves sound pickup sensitivity and enhances sound wave response. It utilizes pulse digital coding and signal enhancement algorithms to improve the signal-to-noise ratio and positioning accuracy; it achieves abnormal noise detection, visual positioning, and remote alarm through an intelligent platform; it adapts to complex industrial environments, reduces wiring and maintenance costs, and enables online, unmanned operation and maintenance.

[0055] The present invention has the following significant advantages: High integration: The system integrates fiber optic audio pickup, signal processing, intelligent recognition, and visual management; High sensitivity and high signal-to-noise ratio: Employs a wound-type sensitivity enhancement structure and PDC coding enhancement technology; High accuracy and robustness: By fusing log-Mel spectra, MFCC plots, and spectral contrast Figure Three By combining complementary features with the powerful nonlinear feature learning capabilities of ResNet18, different fault modes can be effectively distinguished, significantly improving the accuracy and robustness of fault identification in complex industrial noise environments.

[0056] Fine-grained fault classification: It can effectively distinguish up to 11 specific idler roller-related fault types and normal operating conditions (on demand), providing a precise basis for predictive maintenance and precision repair.

[0057] Intelligent operation and maintenance: realizes roller-level positioning, sound playback and closed-loop management of abnormal work orders, replacing manual inspection.

[0058] Early fault warning: The fusion of multiple complementary features, especially those sensitive to signal details, helps to capture subtle acoustic signal changes generated by the idler roller in the early stages of a fault, thereby enabling early warning of faults.

[0059] The innovations of this invention in the field of idler roller noise monitoring are mainly reflected in the following five aspects: 1. System integration design For the first time, the fiber optic pickup unit, the enhanced detector array, the fiber optic connection components, and the intelligent analysis platform are deeply integrated at both the structural and functional levels.

[0060] Based on the separation of traditional fiber optic auscultation and algorithms, a complete link of "front-end perception - host demodulation - platform identification - user operation and maintenance" has been formed to realize intelligent monitoring of the entire life cycle of coal conveyor belts in power plants.

[0061] 2. Fiber-optic wound type enhanced detection structure A wound fiber optic layout is used near the sound source of the belt idler roller, which significantly improves the response sensitivity and positioning accuracy to local abnormal noises. Compared to traditional linear fiber optic cable laying methods, the sensitivity is improved by about 3 times, solving the problem of weak signal in Φ-OTDR technology in noisy industrial environments.

[0062] 3. PDC pulse digital coding and demodulation enhancement technology The host side employs PDC (Pulse Digital Coding) signal modulation and phase demodulation algorithms to overcome the contradiction between spatial resolution and transmission distance. It achieves continuous monitoring over a distance of 1.2 km while maintaining a positioning accuracy of ±3.5 m.

[0063] 4. Multidimensional acoustic feature fusion algorithm The platform adopts a three-channel acoustic feature fusion method, which integrates the log-Mel spectrum, MFCC and spectral contrast map into a deep convolutional neural network (ResNet18). It effectively improves the accuracy of identifying abnormal noise patterns on idlers and enables fine-grained intelligent classification of multiple types of faults.

[0064] 5. Intelligent positioning and visualized closed-loop operation and maintenance at the idler roller level It achieves acoustic positioning at the idler roller level, real-time waveform and sound playback, video linkage, and work order closed-loop functions; it integrates detection, alarm, and operation and maintenance on the same platform, significantly reducing the workload of manual inspection.

[0065] The core technical points claimed in this invention include, but are not limited to, the following: A fault monitoring system for idler rollers based on fiber optic auscultation includes a fiber optic enhanced detector, a fiber optic auscultation host, a data upload module, and an intelligent monitoring platform; the modules are connected to each other through fiber optic links and network communication to form a closed-loop monitoring system.

[0066] The fiber optic detector uses a wound-type sensitivity enhancement structure and is installed on the guardrails on both sides of the belt idler. It picks up the acoustic signal of the idler running through fiber optic phase perturbation.

[0067] The auscultation fiber optic host is based on the Φ-OTDR principle. It enhances the optical signal energy through pulse digital coding (PDC) and performs phase demodulation to obtain a high-fidelity sound signal.

[0068] The intelligent monitoring platform performs multi-dimensional acoustic feature extraction and fusion algorithms (including log-Mel spectrum, MFCC, spectral contrast, etc.) on the collected sound signals, and classifies and identifies the operating status of the idler rollers based on a deep neural network model.

[0069] The system automatically locates and alarms abnormal noise sources on the idler rollers, and provides visualization, sound playback, alarm statistics, and work order closed-loop functions on the platform.

[0070] The system is suitable for continuous operation monitoring of coal conveying and transfer belt conveyor equipment in power plants, and has high sensitivity, high signal-to-noise ratio and anti-interference capabilities.

[0071] Definitions: Phase-sensitive optical time-domain reflectometry (OTDR) is a distributed fiber optic sensing technology based on backscattering Rayleigh scattering of coherent laser pulses. Its core principle is to locate, identify, and quantify the disturbance event by detecting the phase change of the scattered light caused by external disturbances (such as vibration, sound waves, and strain) acting on the sensing fiber. Unlike traditional OTDR, which relies solely on light intensity analysis, Φ-OTDR utilizes interference effects to extract phase information, exhibiting extremely high sensitivity and signal-to-noise ratio. It can detect fiber micro-strain ranging from nanometer to micrometer scales and is widely used in pipeline safety monitoring, perimeter security, seismic wave detection, and infrastructure health diagnostics.

[0072] Pulse Digital Coding (PDC) is an advanced modulation technique used in distributed fiber optic sensing (especially Φ-OTDR systems) to improve signal-to-noise ratio (SNR) and spatial resolution. Its core principle is to replace the traditional single optical pulse with a digital pulse sequence with a specific coding rule (such as Gray code, complementary Gray code, or linear frequency modulation code). At the receiving end, the backscattered Rayleigh signal is processed using corresponding decoding algorithms (such as matched filtering and cross-correlation). This technique, through time-domain coding gain, effectively increases the detection range, dynamic range, and anti-interference capability of the sensing system without significantly increasing peak optical power, overcoming the contradiction between SNR and spatial resolution caused by the power limitation of traditional single pulses.

[0073] Short-time Fourier transform (SFT): This is a time-frequency analysis method used to analyze the changes in the local frequency characteristics of non-stationary signals over time. Its core principle is to divide the signal into multiple approximately stationary short time intervals by sliding a window function (such as a Hanning window or a Gaussian window) across the signal, and then performing a Fourier transform on each of these intervals to obtain a two-dimensional time-frequency spectrum (spectral graph) showing the evolution of the signal's frequency components over time.

[0074] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for monitoring idler roller faults based on fiber optic auscultation, characterized in that, include: Real-time acquisition and preprocessing of idler roller operation sound signals; parallel extraction of three types of complementary acoustic features from the preprocessed sound signals to obtain three types of feature maps. The three types of feature maps are normalized and their sizes are unified, and then fused to form a three-channel fused image. The three-channel fused image is input into a pre-built deep learning model for feature learning and classification, and the probability distribution of the roller's operating state is output. Based on the probability distribution of the idler roller's operating status, an abnormal alarm message is output and the location of the idler roller is marked.

2. The method for monitoring idler roller faults based on fiber optic auscultation according to claim 1, characterized in that, The three types of feature maps specifically include: logarithmic Mel spectrogram, Mel frequency cepstral coefficient map, and spectral contrast map; the logarithmic Mel spectrogram is used to simulate the auditory perception characteristics of the human ear and reflects the change of the energy distribution of the sound signal on the Mel frequency scale over time. The Mel frequency cepstral coefficient diagram is used to characterize the timbre features of a sound and remove the correlation between different features; The spectral contrast diagram is used to reflect the clarity, harmonic structure, and noise level of the sound spectrum.

3. The method for monitoring idler roller faults based on fiber optic auscultation according to claim 2, characterized in that, The process of generating the logarithmic Mel spectrum includes: Perform a short-time Fourier transform on the preprocessed audio signal: Where STFT{x(t)}(τ, ω) is the short-time Fourier transform result, representing the complex amplitude at time τ and angular frequency ω; x(t) is the original sound signal, i.e., the acoustic signal generated when the idler roller is running; w(t - τ) is a window function that shifts at time τ to achieve short-time analysis of the signal; τ is a time parameter representing the center position of the window function; ω is an angular frequency parameter; and is a complex rotational basis, where j is the imaginary unit; Calculate the power spectrum for each frame: Where P(τ, ω) is the power spectrum, representing the signal power at time τ and angular frequency ω; STFT(τ, ω) is the short-time Fourier transform result; The obtained power spectrum is obtained by weighted summation of the power spectrum using a Mel filter bank: Where S(m, t) is the Mel spectrum, representing the energy value at Mel frequency m and per unit time; P(k, t) is the power spectrum value at frequency k and per unit time; Hm(k) is the coefficient of the m-th filter in the Mel filter bank; N is the total number of frequency points; and m is the Mel frequency index. Taking the logarithm of the Mel spectrum yields the logarithmic Mel spectrum. Where LogMel(m, t) is the log-Mel spectrum, representing the logarithmic energy value at Mel frequency m and unit time; S(m, t) is the Mel spectrum value; ε is a constant used to avoid zero or negative values ​​when taking the logarithm.

4. The method for monitoring idler roller faults based on fiber optic auscultation according to claim 3, characterized in that, The process of generating the Mel frequency cepstral coefficient map includes: Based on the logarithmic Mel spectrum, a discrete cosine transform is performed to obtain the Mel frequency cepstral coefficients. Arranging these coefficients frame by frame forms the Mel frequency cepstral coefficient diagram. The calculation formula is as follows: Where MFCC(n, t) is the Mel frequency cepstral coefficient, representing the value at the nth coefficient and unit time; N is the index of the cepstral coefficient; M is the number of Mel filter banks; LogMel(m, t) is the value of the log-Mel spectrum at Mel frequency m and unit time; m is the Mel frequency index.

5. The method for monitoring idler roller faults based on fiber optic auscultation according to claim 2, characterized in that, The spectral contrast diagram reflects the clarity, harmonic structure, and noise level of the sound spectrum by calculating the contrast between peak and valley energy in different frequency bands of the sound signal's spectrum; the calculation formula is as follows: Where is the spectral contrast, representing the contrast value at frequency band b and unit time; X(k, t) is the value of the spectrum of the sound signal at frequency k and unit time; Bb is the b-th frequency band in the spectrum; k is the frequency index; ε is a small constant used to avoid division by zero or taking the logarithm of zero.

6. The method for monitoring idler roller faults based on fiber optic auscultation according to claim 1, characterized in that, The process of collecting the sound signal of the idler roller operation specifically includes: When the belt conveyor is running, the vibration and abnormal noise signals generated by the idler rollers are converted into fiber optic phase disturbance signals by the fiber optic detector; The auscultation fiber optic host is based on the principle of phase-sensitive optical time-domain reflectometry. It sends laser pulses and receives Rayleigh scattering signals through pulse digital encoding. The Rayleigh scattering signals are demodulated to recover the fiber phase disturbance signals, and then digitally converted to obtain the idler roller running sound signals.

7. The method for monitoring idler roller faults based on fiber optic auscultation according to claim 6, characterized in that, The Rayleigh scattering signal is transmitted to the auscultatory fiber optic host via the host fiber optic terminal box, and then the auscultatory fiber optic host performs phase demodulation and digital conversion to obtain the idler roller running sound signal. The sound signal of the idler roller operation is uploaded to the monitoring platform via Ethernet or fiber optic network; The fiber optic detector adopts a wound-type sensitivity enhancement structure and is installed on the guardrails on both sides of the idler roller to improve the response sensitivity and positioning accuracy to local abnormal noises.

8. The method for monitoring idler roller faults based on fiber optic auscultation according to claim 1, characterized in that, The deep learning model is a ResNet18 convolutional neural network; the deep learning model is trained using labeled idler roller sound data through supervised learning methods and a cosine annealing strategy.

9. The method for monitoring idler roller faults based on fiber optic auscultation according to claim 1, characterized in that, The process of outputting abnormal alarm information specifically includes: judging based on the probability distribution of the idler roller's operating status; when the probability of the abnormal category exceeds a preset threshold, the system automatically triggers an alarm and generates a work order; The specific operating states of the idler rollers include: normal operation, abnormal noise from the idler rollers, bearing wear, damaged rubber coating, excessively tight cleaner, and dust entering the bearings.

10. A monitoring system for a method of monitoring idler roller faults based on fiber optic auscultation as described in any one of claims 1-9, characterized in that, The system includes: An optical fiber enhanced detector array is arranged along both sides of the conveyor belt idler roller to convert the sound waves and vibration signals generated by the operation of the idler roller into optical fiber phase perturbation signals. The host fiber optic terminal box is used to aggregate the fiber phase disturbance signals of the fiber optic sensitizer array and to achieve splicing and protection. The auscultation fiber optic host, based on the principle of phase-sensitive optical time-domain reflectometry, is used to send laser pulses through pulse digital encoding, receive Rayleigh scattering signals and demodulate them to recover the fiber phase disturbance signal, and then obtain the idler roller running sound signal through digital conversion. The data upload module is used to transmit the idler running sound signal output by the auscultation fiber optic host to the intelligent monitoring and analysis platform via Ethernet or fiber optic network; The intelligent monitoring and analysis platform is used to receive and process the idler roller running sound signals from the auscultation fiber optic host, perform acoustic feature extraction, output the probability distribution of the idler roller running status and abnormal alarms, and visualize the idler roller running status, idler roller position and abnormal alarm information on the platform interface and user interface. The user-end display module allows users to view the idler roller's operating status, position, acoustic waveform, and abnormal alarm information in real time on a PC or mobile device, enabling remote intelligent operation and maintenance.

Images