Grouping-based cooperative acoustic real-time monitoring method for coal belt conveyor roller fault

Abstract

The application discloses a belt coal conveyor roller fault acoustic real-time monitoring method based on grouping cooperation, which first arranges acoustic sensors based on a topological constraint grouping principle, establishes a standard roller operation signal voiceprint library based on sound signals collected by the acoustic sensors, and calculates a center frequency band through energy characteristics. Secondly, the collected sound signals are decomposed into sub-bands by using VMD, and the VMD parameters are optimized by a quantum constraint optimization engine driven by a quantum-chaos resonance factor in the process. Then, it is judged whether the center frequency band is abnormal by using mutual information entropy, and it is decided whether to enter voiceprint comparison and fault diagnosis. Finally, based on a hash matching algorithm, voiceprint comparison and fault diagnosis are carried out to complete real-time monitoring. The feature extraction precision of the application jumps, and the dynamic working condition is self-adaptive, solving the problems of high monitoring calculation complexity, low positioning accuracy and high hardware cost of large-scale roller groups.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of acoustic signal processing and mechanical fault diagnosis, specifically involving an acoustic monitoring method for belt conveyor idler roller faults based on a grouped collaborative architecture and a quantum-constrained variational mode decomposition (VMD) optimization mechanism (Quantum-Constrained PSO, QC-PSO). Background Technology

[0002] In actual operation of belt conveyors, idlers experience complex stresses. They bear not only the oblique lateral forces exerted by the belt and materials above, but also the constraint forces of the idler axis and the frictional forces on the idler surface. Furthermore, belt conveyors are often exposed to dust, humidity, and other environmental factors, and operate under high-speed, heavy-load conditions. Idler rollers are critical components most prone to failure. Common idler roller failures include bearing damage, roller jamming, and central shaft deformation. Idler jamming is often caused by dust, water vapor in the air, etc., entering the bearing, preventing the idler from operating normally. The continuous friction between a faulty idler and the belt causes the roller surface temperature to rise. Since coal dust is flammable, it can ignite the conveyor belt or coal dust, potentially causing a fire in the belt conveyor. Currently, idler roller failure detection mainly relies on manual foot inspections, which results in a large workload for on-site maintenance and high inspection difficulty. Therefore, real-time monitoring technology for idler roller operation is particularly important.

[0003] Currently, conveyor belt idler roller fault monitoring mainly focuses on two methods: vibration and sound. Compared to acoustic methods, vibration monitoring primarily analyzes narrowband signals, discarding broadband information and resulting in a missed detection rate of over 30% for early, subtle faults (such as micron-level wear). Contact-type vibration sensors require system shutdown for installation and are susceptible to mechanical impact damage, with deployment costs reaching as high as 100,000 yuan per kilometer. Acoustic monitoring, on the other hand, offers advantages such as retaining fault characteristic information from broadband signals, significantly simplifying the installation process through non-contact measurement, and improving the sensitivity of acoustic signatures to early wear faults. Traditional vibration monitoring methods require individual analysis of each idler roller. When the number of idlers is large, the system's computational complexity increases exponentially with the number of rollers (single-node computation time > 5 seconds), making real-time monitoring difficult and requiring an excessive number of sensors. Existing methods are limited by computing power bottlenecks and inaccurate operating conditions, while the group collaborative architecture of this invention solves the problem of large-scale monitoring, and the QC-PSO optimization algorithm overcomes the feature fidelity problem under dynamic disturbances. Together, they provide the only feasible high-precision real-time monitoring solution for 10,000-level idler roller clusters. Summary of the Invention

[0004] To address the challenges of acoustic signal feature extraction and fault location under complex operating conditions of belt conveyors, this invention proposes an acoustic-based idler roller fault monitoring method. This method achieves synergistic optimization of computational efficiency and positioning accuracy through a grouping and processing strategy of acoustic sensors, making it suitable for real-time detection and fault early warning of idler roller operating status in coal mine transportation systems. It solves the technical difficulties of high computational complexity, low positioning accuracy, and high hardware cost in large-scale idler roller group monitoring, and is applicable to real-time fault early warning and operation and maintenance optimization in coal mine transportation systems. The method includes the following steps:

[0005] 1. Based on the topological constraint grouping principle, acoustic sensors are deployed and installed in the middle of the idler roller support.

[0006] 2. Based on the acoustic signals collected by the acoustic sensors, establish a standard idler roller running signal acoustic pattern library, and calculate the center frequency band through energy characteristics.

[0007] 3. The acquired acoustic signal is decomposed into subbands using VMD, and the quantum-chaotic resonance factor (VMD) is used in the process. The quantum constraint optimization engine QC-PSO, driven by quantum constraints, optimizes VMD parameters.

[0008] 4. After VMD parameter optimization, the similarity between the center frequency sub-band signal of the decomposed signal and the center frequency sub-band signal of the standard signal is calculated using mutual information entropy. This determines whether there are any anomalies in the center frequency band. If no anomalies are found, the process proceeds directly to acoustic signature comparison and fault diagnosis. If anomalies are found in the center frequency band, it is determined whether there are any suspended idlers, thus eliminating shaft frequency anomalies caused by suspended idlers. If suspended idlers are found, the idler group is continuously tracked, and the acoustic signal data when the suspended idler is in contact with the belt and rotating normally is used as the signal to be detected, and this step is repeated. If no suspended idlers are found, the process proceeds to acoustic signature comparison and fault diagnosis.

[0009] 5. Entering Voiceprint Comparison and Fault Diagnosis. The test signal is matched against standard / fault voiceprint templates using an AF-Hash algorithm based on the idler roller group's voiceprint. If an anomaly is found in the center frequency band, it is directly matched against the fault voiceprint template; if no anomaly is found, it is first matched against the standard voiceprint template, and if that fails, it is then matched against the fault voiceprint template. If a fault is found, the idler roller group fusion positioning algorithm is used to locate the fault, outputting the fault type, the location of the faulty idler roller group, and triggering a tiered alarm.

[0010] Furthermore, the topological constraint grouping principle: by dividing adjacent idlers in the belt conveyor into collaborative analysis units based on the sensor detection range, and by modeling the spatial correlation between adjacent idlers, the monitoring unit is upgraded from the single idler level to the group level, and a "sensor group-signal collaborative analysis" model is constructed, which reduces the data processing scale per kilometer from 2,000 independent signals to 400 groups, with a data compression rate of 80%, breaking through the computational bottleneck of traditional single-node analysis.

[0011] Furthermore, the Quantum Constrained Optimization Engine (QC-PSO) uses a quantum constrained particle swarm optimization algorithm to establish a cross-scale fault model of "quantum tunneling-chaotic evolution": it captures microcracks in materials through the quantum tunneling effect and provides early warning of system instability through chaotic dynamics, using the quantum-chaotic resonance factor (QC-PSO). Constructing dynamic hard boundary constraints ( ), in the case of sudden load changes (>10%), strong noise ( It achieves parameter self-optimization under complex working conditions such as ) and solves the problem of modal aliasing in traditional methods. The detection rate of micron-level damage is increased to 90%, and the adaptability to working conditions is improved by 3 times compared with traditional methods.

[0012] Furthermore, Voiceprint Hash Matching (AF-Hash): A fast voiceprint matching technique based on Locality Sensitive Hashing is proposed, overcoming the computational bottleneck of the traditional Dynamic Time Warping (DTW) algorithm. Feature compression encoding compresses the multidimensional features after VMD decomposition into 128-bit binary voiceprints. A fast matching mechanism calculates fingerprint similarity using Hamming distance, reducing the time complexity from O(n²) to O(1).

[0013] Furthermore, a roller group positioning algorithm is proposed: a roller group fusion positioning algorithm is suggested, which locates the faulty group by taking the middle position of spatially continuous roller groups with the same fault type. This algorithm solves the inter-group interference problem caused by the grouping strategy and improves positioning accuracy.

[0014] Beneficial effects of this invention:

[0015] Significantly reduced computational load: Traditional methods require individual analysis of each idler roller, with computational load increasing exponentially with the number of idler rollers (single-node computation time > 5 seconds). This invention uses a grouping processing strategy to divide adjacent idler rollers into groups, theoretically reducing the data processing scale per kilometer of idler rollers by 80%. Combined with VMD to extract core frequency band features, it reduces useless information and achieves millisecond-level response.

[0016] Improved accuracy in fault feature extraction: Traditional VMD methods suffer from mode aliasing due to fixed parameters. The quantum constraint optimization mechanism of this invention transforms the physical mechanism into mathematical constraints, resulting in a significant improvement in feature extraction accuracy and enabling adaptation to dynamic operating conditions.

[0017] Significantly optimized sensor deployment costs: Traditional contact vibration sensors cost up to 100,000 yuan per kilometer to deploy and require downtime for installation. This invention uses non-contact acoustic sensors (1 per 5 meters) to avoid contact with the roller, improving the signal-to-noise ratio and reducing overall deployment costs by more than 60%. Attached Figure Description

[0018] Figure 1 This is the overall system flowchart;

[0019] Figure 2 This is a flowchart of the feature extraction and fault diagnosis algorithm. Detailed Implementation

[0020] A real-time acoustic monitoring method for belt conveyor idler roller faults based on group collaboration is described below. Figure 1 and Figure 2 A detailed explanation is provided, including the following procedures:

[0021] I. Sensor Deployment

[0022] Sensor placement should follow the principle of proximity, placing them as close as possible to the fault source to improve the signal-to-noise ratio. For vibration or acoustic emission sensors, they can be directly mounted on the bearing housing or end cap, ensuring the sensor's measurement direction aligns with the primary propagation path of the vibration or acoustic emission. Temperature sensors can be embedded between the bearing housing and the outer ring of the bearing, or attached to the outer surface of the bearing housing, but care should be taken to avoid contact with the bearing cage. Strain sensors should be installed in stress concentration areas, such as near the connecting bolt holes of the bearing housing, and reliably sealed against water and oil. Furthermore, the economic efficiency and operability of sensor placement should be comprehensively considered, minimizing the number of sensors and optimizing the placement scheme.

[0023] Since this invention employs a fault monitoring scheme based on acoustic sensors, the acoustic sensors are positioned in the middle of the idler roller support (node ​​density 1 / 5m) according to the principle of proximity. This aims to ensure that the sensors are as close as possible to the fault source in physical space while avoiding contact with the roller, thereby improving the signal-to-noise ratio and ensuring equipment safety.

[0024] II. Feature Extraction and Fault Diagnosis Algorithms

[0025] 1. Feature extraction and fault diagnosis algorithm flow, such as Figure 2 As shown:

[0026] (1) While establishing the voiceprint database, the center frequency band and the neighborhood range are determined by energy characteristic calculation.

[0027] (2) The VMD is used to decompose the signal to be detected in the frequency domain to obtain subband information. The VMD decomposition process is then adaptively optimized using the quantum constraint optimization engine QC-PSO.

[0028] (3) Extract the feature information of the center frequency band subband of the signal to be detected by mutual information entropy, and initially filter out faultless idlers to reduce the amount of subsequent calculation.

[0029] (4) Perform AF-Hash matching between the signal to be detected and the normal / fault signals in the acoustic fingerprint library to diagnose whether the idler roller is faulty and the type of fault.

[0030] (5) The idler roller group fusion positioning algorithm uses the number of consecutive abnormal groups ( Calculate the positioning interval It suppresses interference from neighboring signals, improving the accuracy of fault location. It locates the fault and initiates a system alarm.

[0031] 2. VMD Feature Extraction Algorithm for Frequency Bands

[0032] In signal processing, Variational Mode Decomposition (VMD) is a signal decomposition and estimation method. This method determines the frequency center and bandwidth of each component by iteratively searching for the optimal solution of the variational model during the acquisition of decomposed components, thereby adaptively achieving frequency domain partitioning of the signal and effective separation of each component. This is achieved by constructing a variational model and setting appropriate parameters (such as the number of modes). Bandwidth constraint parameters VMD can adaptively decompose complex non-stationary acoustic signals collected during roller operation into multiple eigenmode functions with definite center frequencies. ), each Signal components corresponding to specific frequency bands. For example, for characteristic frequency bands related to roller shaft frequency, harmonic frequency, and faults such as eccentricity and wear, adjustments can be made... Values ​​separated from corresponding values The component model enables accurate extraction of signals in the target frequency band. Compared with traditional decomposition methods, VMD's adaptive decomposition characteristics make it more advantageous in extracting specific frequency bands from nonlinear and non-stationary signals. It can effectively suppress mode aliasing, improve the purity and discriminability of characteristic frequency band signals, and provide a high-quality signal foundation for subsequent fault feature quantification and diagnosis based on mutual information entropy, thereby improving the accuracy and reliability of idler roller fault diagnosis.

[0033] 3. QC-PSO Parameter Optimization Algorithm

[0034] To address the complex and variable operating conditions of coal mine belt conveyors, traditional VMD decomposition methods with fixed parameters are ill-suited for extracting dynamically changing fault features. Therefore, this invention proposes an intelligent VMD parameter optimization algorithm based on a quantum-chaotic coupling model. To address the mode aliasing problem caused by fixed parameters under complex operating conditions, this invention pioneers a quantum-chaotic coupling constraint mechanism. This mechanism utilizes the quantum tunneling energy ratio (… Quantifying the microscopic damage of materials, combined with chaotic dynamics indicators ( Characterizing the macroscopic instability of the system, a single constraint factor with a clear physical meaning is constructed. Through quantum-chaotic resonance factor ( Constructing quantum constraints and combining them with particle swarm optimization (QC-PSO) algorithm to adaptively adjust the number of modes in VMD. and bandwidth constraint parameters The specific steps are as follows:

[0035] Step 3.1: After VMD decomposition, if the subband signal-to-noise ratio or load fluctuation exceeds a set threshold (set here, signal-to-noise ratio...), then... When load fluctuations exceed 10%, activate VMD parameter optimization;

[0036] Step 3.2: Parameters and mode number in VMD decomposition. and broadband constraints Perform initialization. Decision variable: number of modes. Broadband constraints .

[0037] 20 sets of parameter combinations are generated randomly. Each set of parameters is called a particle. Initialize particle velocity: Directional velocity range: [-3.5, +3.5] Directional velocity range: [-2000, +2000].

[0038] Step 3.3: Quantum constraint verification and fitness evaluation. Perform the following operations for each particle:

[0039] Perform VMD decomposition. Use current parameters. Perform VMD decomposition on the input parameters to obtain eigenmode functions ( ).

[0040] Calculate the quantum-chaotic resonance factor ( ):

[0041] ; This represents the quantum tunneling effect in microcracks of quantized materials. Characterizes the degree of chaos in the system.

[0042] ,in This is the bandwidth limit. The characteristic frequency band is the bandwidth of ±5% of the bearing damage characteristic frequency. This is the Fourier transform of the demodulated frequency signal of the subband after VMD decomposition. The quantum tunneling probability is calculated based on the roller material and vibration energy.

[0043] ,in This represents the instantaneous phase difference between the subband signal and the fundamental frequency. Standard deviation Mean.

[0044] Constraint verification. When A solution is considered valid if it is less than the set factor threshold (0.06 in this case), i.e., if it meets the following condition: .

[0045] Calculate the fitness of solutions that satisfy the constraints: .

[0046] Step 3.4: Particle swarm iteration update.

[0047] Update the individual's historical best: Compare the current fitness with the particle's historical best fitness. Retain the better solution as the particle's individual best. .

[0048] Update the global historical optimum: Compare the optimal fitness of all individual particles. Retain the optimal solution as the global optimum. .

[0049] Speed-up calculation. , ∈(0,1) uniformly distributed random numbers.

[0050] Inertia weight , .

[0051] Speed ​​Update .

[0052] Speed ​​Update .

[0053] Location update calculation. .

[0054] Learning factor update .

[0055] Boundary constraint handling:

[0056] when Reset And reverse the direction of velocity;

[0057] when Reset And reverse the direction of velocity;

[0058] when At that time, reset according to the mirror principle. ;

[0059] when At that time, reset according to the mirror principle. ;

[0060] Step 3.5: Output and Storage of Optimal Parameters. The iteration terminates when the rate of change of the globally optimal fitness is less than 0.1% for 10 consecutive generations or when the maximum number of iterations (50 generations) is reached. The parameters are then stored in the database.

[0061] 4. Frequency band anomaly detection based on mutual information entropy

[0062] Mutual information entropy is a dimensionless statistical measure (usually in bits) that one random variable can provide about the changes in another random variable. In the patent for fault diagnosis of idler rollers in coal mine belt conveyors, mutual information entropy is used because it can effectively quantify the nonlinear correlation between signals, overcoming the shortcomings of traditional linear analysis methods (such as cross-correlation coefficients) in capturing the nonlinear modulation characteristics caused by faults such as roller eccentricity and bearing wear. By calculating the logarithmic difference between the joint probability distribution and the marginal probability distribution, it can significantly improve the sensitivity to weak features of early faults. Furthermore, dynamic binning technology (such as the Freedman-Diaconis rule) enhances noise resistance and robustness. In practical applications, the amplitude of continuous acoustic signals is discretized and approximated using the binning method. In the patent, it is combined with VMD to form a diagnostic framework of "group processing + adaptive decomposition + nonlinear feature quantization," achieving efficient feature extraction and accurate diagnosis of idler roller fault signals, improving fault identification rate and real-time performance, and reducing deployment costs.

[0063] 1. Definition of mutual information entropy for discrete variables: For discrete random variables... and The joint probability distribution is The marginal probability distribution is and Calculate mutual information entropy ;

[0064] 2. Feature extraction formula combined with VMD. The acoustic signal of the idler roller is decomposed into K eigenmode functions through variational mode decomposition (VMD). ) Extract each Mutual information entropy with reference mode: Under fault conditions, the target (e.g., corresponding to 2 times the frequency) The mutual information entropy of ) is significantly increased, which can be used as a fault characteristic quantity.

[0065] 5. Hash matching based on acoustic fingerprint of idler roller group (AF-Hash)

[0066] The storage structure of the standard idler roller operation signal acoustic fingerprint library is simplified to a hash code-fault type mapping table; the core frequency band after VMD decomposition... Feature extraction is performed on the components, and the feature quantity is: ;The characteristic quantity is Perform element-wise addition and generate voiceprint codes using Locality Sensitive Hashing (LSH). The standard signal voiceprint code corresponding to the standard idler roller running signal voiceprint library is: Based on voiceprint coding and Hamming distance is used Measuring voiceprint similarity When the voiceprint similarity is greater than a set voiceprint threshold, the signals are determined to be of the same type. Specifically:

[0067] Focusing on the feature encoding and fast matching of acoustic signals of idler rollers, the model is simplified to a "pure acoustic feature hash comparison" model by removing the topological position calculation module, and directly performing similarity calculation with standard / fault templates in the acoustic signature library.

[0068] (1) The voiceprint database storage structure is simplified to a "hash code - fault type" mapping table. For example: , ;

[0069] (2) Acoustic feature extraction: for the core frequency band after VMD decomposition Feature extraction is performed on the components, and the feature quantity is: ;

[0070] (3) Acoustic fingerprint generation: Only acoustic features are fused, and a 128-bit fingerprint code is generated through Local Sensitive Hash (LSH): ;

[0071] (4) Hash similarity calculation: Hamming distance is used to measure fingerprint similarity, and the formula is: When the similarity is greater than the threshold, the signals are considered to be of the same type.

[0072] 6. Idler Roller Group Fusion Positioning Algorithm

[0073] Because sensors may receive signal interference from neighboring groups, a roller group fusion positioning algorithm is proposed to improve positioning accuracy and reduce the false alarm rate. The algorithm input includes:

[0074] The sequence of consecutive abnormal idler roller groups is as follows: .

[0075] The similarity between the acoustic signatures of consecutive abnormal idler roller groups and the fault acoustic signature templates is as follows: .

[0076] The root mean square (RMS) value of the VMD core band subband of a series of continuously abnormal idler roller groups Energy constituting fault characteristics .

[0077] Calculate the confidence level for each outlier group. If it exists satisfy Greater than the fault threshold Then directly locate For the fault group, if all Less than The center of a continuous abnormal group is the fault group.

[0078] III. Performance Analysis

[0079] 1. Experimental setup

[0080] To verify the effectiveness of this invention, multiple simulation experiments were designed, covering environments with different signal-to-noise ratios and idler roller sizes. The experiments were conducted on the MATLAB R2023a platform, with a hardware configuration of an AMD Ryzen 5 7500F processor and 16GB of memory. Specific parameters are as follows:

[0081] (1) Signal generation model:

[0082] Normal signal: fundamental frequency 200 Hz, containing 1×, 2×, and 3× harmonics, superimposed with pink noise ( ).

[0083] Fault signal: Bearing wear: High-frequency resonance (5×fundamental frequency) + periodic impact (interval = 1 / fundamental frequency period).

[0084] Eccentricity fault: speed fluctuation (±5%) + sideband modulation (carrier frequency ±0.15 Hz).

[0085] Stuck fault: Time-varying Poisson impact (λ=2~3.5) + asymmetric damped response.

[0086] Noise model: Gaussian white noise ( Range: 10 dB to -10 dB.

[0087] (2) Test scenario:

[0088] Signal-to-noise ratio test: Each group tested 1000 idler rollers.

[0089] Time calculation test: Number of idlers = [100, 200, 500, 1000]. Idler group = [20, 40, 100, 200].

[0090] Data distribution: Normal signals account for 55%, and each of the three types of faults accounts for 15%. This simulates the uneven distribution of fault types in actual working conditions.

[0091] 2. Fault Detection Performance Analysis

[0092] (1) Fault detection performance analysis

[0093] Table 1. Comparison of fault detection performance under different signal-to-noise ratios

[0094]

[0095] Table 1 compares the detection performance of the present invention and the conventional method under different signal-to-noise ratios. Parameter description: It is a true positive (a genuine case); It is a true negative (a true counterexample); This is a false positive (false positive / false alarm). False negatives (false negatives / missed reports); detection rate False alarm rate Classification accuracy

[0096] Key findings:

[0097] High signal-to-noise ratio environment ( Both methods achieved 100% detection rate and classification accuracy, indicating that the performance of the present invention is comparable to that of traditional methods under ideal working conditions.

[0098] Low signal-to-noise ratio environment ( Traditional methods have a false alarm rate that soars to 3% and a classification accuracy that plummets to 97%; while the false alarm rate of this invention is controlled at 0.89% and the classification accuracy remains at 99.5%.

[0099] Extreme noise environment ( The detection rate of this invention decreased by only 1.1%, while the classification accuracy remained at 72.5%, significantly better than the 61.3% of the traditional method.

[0100] Table 2 shows a comparison of execution times for different idler roller sizes.

[0101] (2) Verification of computational efficiency and scalability

[0102] Table 2: Comparison of execution time under different numbers of idlers

[0103]

[0104] Table 2 shows a comparison of execution time for different idler roller scales. It can be seen that the computational efficiency advantage is significant: with a data scale of 1000 idler rollers, the execution time of this invention is greatly reduced compared to traditional methods. The efficiency improvement mainly stems from:

[0105] Grouping processing strategy: Dividing idlers into groups reduces data processing scale by 80%;

[0106] Subband Focusing Optimization: VMD only decomposes the target frequency band, reducing computational complexity from O(N) 2 ) decreased to O( It also combines mutual information entropy to perform preliminary screening of normal signals, reducing the number of subsequent AF-Hash executions.

[0107] Hash matching based on acoustic fingerprints of idler rollers (AF-Hash). Fault features are hashed, converted to binary storage, and fingerprint similarity is calculated using Hamming distance. The computational complexity is reduced from O(N) 2 ) decreased to O( ).

[0108] The acoustic monitoring algorithm for idler roller faults in a belt conveyor based on group processing proposed in this invention has demonstrated the following advantages through systematic simulation experiments:

[0109] High-precision diagnostics: The subclassification accuracy was 72.5%, and the false alarm rate was <8%.

[0110] High-efficiency computing: The algorithm for scaling thousands of idler rollers is 9.69 times more efficient.

[0111] Strong robustness: It has good adaptability to noise interference.

[0112] The monitoring framework constructed in this invention, consisting of "group processing + adaptive decomposition + nonlinear feature quantization," provides a highly robust, low-computational-cost, and easily deployable solution for the intelligent operation and maintenance of coal mine belt conveyors. Experimental data demonstrate its effectiveness in... The system maintains a classification accuracy of 72.5% in this scenario, an improvement of 11.2% compared to traditional methods, and nearly 10 times the efficiency at the scale of thousands of idler rollers. This technology breaks through the efficiency and cost bottlenecks of large-scale mechanical cluster monitoring, and has significant industrial application value and academic innovation.

Claims

1. A method for real-time acoustic monitoring of idler roller faults in a belt conveyor based on group collaboration, characterized in that, Includes the following steps: Step 1: Deploy acoustic sensors based on topological constraint grouping principles; Step 2: Based on the acoustic signals collected by the acoustic sensors, establish a standard idler roller running signal acoustic pattern library, and calculate the center frequency band through energy characteristics; Step 3: Use VMD to decompose the acquired acoustic signal into subbands, and optimize the VMD parameters through a quantum-chaotic resonance factor driven quantum constraint optimization engine during the process; The quantum constraint optimization engine QC-PSO is specifically implemented as follows: A quantum tunneling-chaotic evolution cross-scale fault model is established: Based on the quantum tunneling effect, dynamic hard boundary constraints are constructed using the quantum-chaotic resonance factor (QCR) to achieve parameter self-optimization; after VMD decomposition, when the subband signal-to-noise ratio or load fluctuation exceeds a set threshold, VMD parameter optimization is activated, and an improved particle swarm optimization algorithm is used to optimize the parameters in VMD decomposition. The improved particle swarm optimization algorithm described herein is based on the conventional particle swarm optimization algorithm, but the calculation of particle fitness is improved as follows: Use the current parameter (K) i , i Perform VMD decomposition on the input parameters to obtain K. i There are 1 eigenmode functions (IMFs), where K i and i Given i modes, the modal number and bandwidth constraints; The quantum-chaotic resonance factor is calculated as the product of the quantum tunneling effect QTER of microcracks in quantized materials and the degree of chaos CDI. Set a factor threshold; when the QCRF is less than the set factor threshold, (K) i , i ) is considered a valid solution, and its fitness is calculated, i.e., the corresponding parameter (K) i , i The sum of squares of ) is weighted and summed with QCRF; Step 4: After VMD parameter optimization, determine whether there is an anomaly in the center frequency band by mutual information entropy, and decide whether to proceed to voiceprint comparison and fault diagnosis. Step 5: Based on the hash matching algorithm, perform voiceprint comparison and fault diagnosis to complete real-time monitoring.

2. The method for real-time acoustic monitoring of belt conveyor idler roller faults based on group collaboration as described in claim 1, characterized in that, The topological constraint grouping principle is as follows: by dividing adjacent idlers in the belt conveyor into collaborative analysis units based on the sensor detection range, the monitoring unit is upgraded from the single idler level to the group level.

3. The method for real-time acoustic monitoring of belt conveyor idler roller faults based on group collaboration as described in claim 1, characterized in that, Step 4 is specifically implemented as follows: After VMD parameter optimization, the similarity between the center frequency sub-band signal of the collected acoustic signal decomposed and the center frequency sub-band signal of the standard signal is calculated by mutual information entropy. The similarity of the center frequency sub-band signal is compared with the similarity threshold. If it is greater than the similarity threshold, it means that there is no abnormality and directly enters the voiceprint comparison and fault diagnosis. Otherwise, there is an abnormality in the center frequency band. If the center frequency band is abnormal, it is determined whether there is a suspended idler and the shaft frequency abnormality caused by the suspended idler is ruled out. If there is a suspended idler, the idler group is continuously tracked, and the sound signal data when the suspended idler is in contact with the belt and rotates normally is used as the signal to be detected and step 4 is repeated. If there is no suspended idler, the soundprint comparison and fault diagnosis are performed.

4. The method for real-time acoustic monitoring of belt conveyor idler roller faults based on group collaboration according to claim 1, characterized in that, Step 5 is specifically implemented as follows: the similarity matching between the signal under test and the standard / fault soundprint template is performed by the AF-Hash algorithm based on the soundprint of the idler group; if there is an anomaly in the center frequency band, it is directly matched with the fault soundprint template; if there is no anomaly, it is first matched with the standard soundprint template, and if the matching fails, it is then matched with the fault soundprint template. After matching the fault soundprint template, if a fault exists, the idler group fusion positioning algorithm is used to locate the fault, output the fault type, the location of the faulty idler group, and trigger a graded alarm.

5. The method for real-time acoustic monitoring of belt conveyor idler roller faults based on group collaboration according to claim 4, characterized in that, The specific process of the AF-Hash algorithm based on idler roller group acoustic signatures to perform similarity matching between the test signal and the standard / fault acoustic signature template is as follows: The storage structure of the standard idler roller running signal acoustic fingerprint library is simplified to a hash code-fault type mapping table. Feature extraction is performed on the IMF components of the core frequency band after VMD decomposition. The feature quantity is: 1,f2……f k ; The characteristic is 1,f2……f k Perform element-wise addition and generate voiceprint codes using Locality Sensitive Hashing (LSH). ; The standard signal corresponding to the voiceprint code in the standard idler roller operation signal voiceprint library is: ; Based on voiceprint coding and Hamming distance is used Measure voiceprint similarity s m When the voiceprint similarity is greater than the set voiceprint threshold, it is determined to be a signal of the same type.

6. The method for real-time acoustic monitoring of belt conveyor idler roller faults based on group collaboration according to claim 5, characterized in that, The specific implementation process for locating the fault is as follows: The sequence of m consecutive abnormal idler roller groups is G=[g1,g2,g3……g… m ]; The similarity between the acoustic signatures of m consecutive abnormal idler roller groups and the fault acoustic signature template is S=[s1,s2,s3……s m ]; The root mean square (RMS) value of the VMD core frequency band subband of m consecutive abnormal idler roller groups. m The fault characteristic energy E=[e1,e2,e3……e m ]; Based on matching similarity and fault characteristic energy Calculate the confidence level for each outlier group. If g exists i Satisfy c i If the fault value is greater than the fault threshold θ, then g is directly located. i For the fault group, if all c i If the value is less than θ, the center position of the continuous abnormal group is the fault group.

Images