Early fault diagnosis system for coal mine machinery based on vibration analysis

Abstract

The present application relates to the technical field of mechanical equipment fault diagnosis, and discloses a coal mine machinery early fault diagnosis system based on vibration analysis, comprising: a data acquisition module for acquiring vibration signals of key components of coal mine machinery; a data processing module connected with the data acquisition module, for carrying out frequency band separation on the vibration signals based on variational mode decomposition, and decomposing the vibration signals into a plurality of modal components; and a feature extraction module connected with the data processing module, for carrying out noise modeling and feature signal extraction on the modal components based on sparse Bayesian inference. The technical scheme of frequency separation of the vibration signals based on variational mode decomposition achieves the technical effect of effectively separating fault features and noise in non-stationary signals. Compared with the problem of mode aliasing in the commonly used empirical mode decomposition in the prior art, the deficiencies of fuzzy frequency range of modal components and difficulty in extracting key fault features are solved.

Description

Technical Field

[0001] This invention relates to the field of mechanical equipment fault diagnosis technology, specifically to an early fault diagnosis system for coal mine machinery based on vibration analysis. Background Technology

[0002] When coal mining machinery operates under complex conditions for extended periods, its core components, such as bearings and gearboxes, are prone to early failures due to factors such as load fluctuations and poor lubrication. These failures often manifest as vibration signals. However, vibration signals themselves have obvious non-stationary and multi-noise characteristics, making the extraction and diagnosis of fault features particularly difficult. To address this issue, many vibration signal analysis and diagnosis methods have been proposed in the existing technology, but these methods have significant shortcomings when dealing with complex coal mining conditions.

[0003] In existing technologies, empirical mode decomposition (EMD) is a common non-stationary signal decomposition method that can decompose vibration signals into a series of modal components. However, EMD is prone to mode aliasing, which leads to blurred boundaries between different frequency features. This problem is particularly prominent when the signal contains high noise. This makes it difficult to clearly separate key fault features from noise components in the spectrum. In addition, EMD decomposition is highly dependent on the signal. When the mechanical operating conditions change, the decomposition results are often inconsistent, resulting in unstable diagnostic results.

[0004] In terms of feature extraction, existing methods often use fixed threshold filtering or frequency band energy ratio to extract fault feature signals. Although these methods can achieve certain results in low-noise environments, in the complex environment of coal mines, random noise and environmental interference are strong. Fixed threshold filtering methods often lack adaptability and may lead to the filtering out of key fault feature signals or the misidentification of noise components as fault features. This method that relies on fixed rules has significant limitations in practical applications.

[0005] Furthermore, most existing fault diagnosis technologies are based on classification models that directly diagnose feature signals. While commonly used methods such as Support Vector Machines (SVM) or neural networks can achieve a certain degree of pattern recognition, these methods have high requirements for the consistency of the distribution of input features. When there is a deviation between the feature distribution of the signal and the training data, the accuracy of the classification model will decrease significantly. Especially under the complex and variable conditions of coal mines, the signal distribution of different fault modes varies greatly, and direct classification methods are unable to cope with the distribution offset problem, resulting in insufficient reliability of the diagnostic results. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an early fault diagnosis system for coal mine machinery based on vibration analysis. This system solves the problem that under complex and variable working conditions in coal mines, the signal distribution of different fault modes varies greatly, and direct classification methods are insufficient to address the distribution offset issue.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an early fault diagnosis system for coal mine machinery based on vibration analysis, comprising: The data acquisition module is used to collect vibration signals from key components of coal mining machinery; The data processing module, connected to the data acquisition module, is used to perform frequency band separation on the vibration signal based on variational mode decomposition, decomposing the vibration signal into multiple modal components; The feature extraction module, connected to the data processing module, is used to perform noise modeling and feature signal extraction of modal components based on sparse Bayesian inference. The fault diagnosis module, connected to the feature extraction module, is used to match and diagnose the feature signal distribution with the fault mode distribution based on the target distribution optimization theory. The output module, connected to the fault diagnosis module, is used to output the fault type and early fault diagnosis results of the coal mining machinery.

[0008] Preferably, the data processing module decomposes the vibration signal based on the variational mode decomposition algorithm, specifically including the following steps: Construct a modal decomposition objective function for the input vibration signal to decompose the vibration signal into multiple modal components; The frequency center and signal amplitude of the modal components are iteratively updated using optimization methods; The modal components are divided into high-frequency components and low-frequency components. The high-frequency components contain noise signals, while the low-frequency components contain mechanical fault characteristic signals.

[0009] Preferably, the feature extraction module processes the modal components through sparse Bayesian inference, specifically including the following steps: Assuming that the modal components consist of sparse signals and Gaussian noise, we construct the prior distribution of the sparse signals and the noise distribution, respectively. The posterior distribution of sparse signals in modal components is solved by Bayesian inference. The weights and noise weights of sparse signals are dynamically updated to suppress high-frequency noise and optimize the extraction of low-frequency feature signals.

[0010] Preferably, the fault diagnosis module processes the extracted feature signals based on the target distribution optimization theory, specifically including the following steps: A probability density function model is constructed for the extracted feature signals to characterize the distribution of the feature signals; Establish a probability distribution model for the target failure modes; Calculate the transmission distance between the characteristic signal distribution and the target fault mode distribution; By optimizing the transmission distance, the distribution of characteristic signals is brought closer to the target distribution, thus completing fault mode matching and diagnosis.

[0011] Preferably, the data acquisition module includes multiple vibration sensors, which are installed in key parts of the coal mining machinery, including bearings, gearboxes, and drive motors, to collect acceleration signals and vibration frequency signals.

[0012] Preferably, the early fault diagnosis results output by the system include the fault type, severity, and location of key components of the coal mining machinery.

[0013] Preferably, the data processing module performs frequency segmentation processing on the vibration signal, wherein the low-frequency signal is used to extract the operating status characteristics of the equipment, and the high-frequency signal is used to identify random noise or environmental interference.

[0014] Preferably, the signal weights of the sparse Bayes inference are updated using the following formula: the weights of the sparse signal are dynamically adjusted based on the ratio of the signal amplitude to the covariance matrix, and the update rule is as follows:

[0015] in, The number of signal samples. The square of the amplitude of the characteristic signal, Let be the signal covariance matrix.

[0016] Preferably, when the fault diagnosis module optimizes the transmission distance between the characteristic signal distribution and the target distribution, it uses distance to calculate the deviation of the signal distribution, and the optimization target is:

[0017] in, Represents the probability distribution of the characteristic signal. Represents the probability distribution of the target failure mode. and These are the cumulative distribution functions of the characteristic signal and the target distribution, respectively.

[0018] Preferably, the system includes a health status prediction module, which predicts the health status of coal mining machinery and calculates its remaining service life based on extracted feature signals and diagnostic results.

[0019] This invention provides an early fault diagnosis system for coal mine machinery based on vibration analysis. It has the following beneficial effects: 1. This invention adopts a technical solution based on variational mode decomposition (VMD) to separate the frequency of vibration signals, which achieves the technical effect of effectively separating fault features and noise in non-stationary signals. Compared with the empirical mode decomposition (EMD) commonly used in the prior art, which is prone to mode aliasing, this invention solves the shortcomings of ambiguous frequency range of modal components and difficulty in extracting key fault features.

[0020] 2. This invention introduces Sparse Bayesian inference (SBI) for feature signal extraction, which enhances the sparse representation of mechanical fault feature signals. Compared with the existing technology that uses fixed threshold filtering or simple frequency band segmentation, it solves the problems of feature signals being easily contaminated by noise and lacking adaptive modeling capabilities, and provides support for robust fault diagnosis under multiple working conditions.

[0021] 3. This invention adopts a fault diagnosis method based on optimal transmission theory. It uses target distribution matching to achieve accurate alignment between fault feature signals and predefined fault modes, thus achieving high-precision fault identification. Existing methods that rely on direct feature classification are easily affected by inconsistent signal distribution under complex working conditions. This invention solves the problem of low accuracy in fault diagnosis of vibration signals in complex coal mine machinery.

[0022] 4. This invention achieves the technical effect of improving the usability and intuitiveness of the fault diagnosis system through multi-module linkage and data visualization design of the output module. Compared with the technical solution of traditional systems that can only provide simple alarm information, it solves the problem that users cannot intuitively understand the operating status of equipment and the cause of faults. At the same time, it enhances the system's scalability through report generation and historical data storage. Attached Figure Description

[0023] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0024] The technical solutions in 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see the appendix Figure 1 This invention provides an early fault diagnosis system for coal mine machinery based on vibration analysis, comprising: The data acquisition module is used to collect vibration signals from key components of coal mining machinery; The data processing module, connected to the data acquisition module, is used to perform frequency band separation on the vibration signal based on variational mode decomposition, decomposing the vibration signal into multiple modal components; The feature extraction module, connected to the data processing module, is used to perform noise modeling and feature signal extraction of modal components based on sparse Bayesian inference. The fault diagnosis module, connected to the feature extraction module, is used to match and diagnose the feature signal distribution with the fault mode distribution based on the target distribution optimization theory. The output module, connected to the fault diagnosis module, is used to output the fault type and early fault diagnosis results of the coal mining machinery.

[0026] Data acquisition module The data acquisition module is mainly used to collect vibration signals during the operation of coal mining machinery. To ensure the quality of the acquired signals, the module uses high-sensitivity accelerometers. These sensors can capture vibration signals from key components of the coal mining machinery (such as bearings, gearboxes, and motors) in real time. The sensor installation locations are carefully arranged to ensure that the acquisition points cover all critical components that may malfunction.

[0027] The collected vibration signals are transmitted to the processing module in real time through the data acquisition terminal. The signal sampling frequency and resolution can cover the working frequency range of coal mine machinery and equipment, and ensure the ability to capture early weak fault characteristics. This step provides the foundation for subsequent signal processing. Compared with conventional acquisition equipment, this module has a higher dynamic range and anti-interference ability, and can effectively resist the interference of complex noise in coal mine working conditions.

[0028] Data processing module The data processing module decomposes the acquired vibration signal using variational mode decomposition (VMD), a non-recursive adaptive decomposition method that can decompose a non-stationary signal into several narrowband modal signals. Each modal signal is concentrated in a specific frequency band, thereby isolating different components in the vibration signal.

[0029] Specifically, VMD transforms the decomposition problem of vibration signals into an optimization problem by constructing a Lagrangian objective function. The objective function includes frequency constraints and signal reconstruction constraints of the modal signals, and is solved iteratively using the Alternating Direction Multiplier Method (ADMM). The resulting modal signals are divided into high-frequency noise components and low-frequency characteristic components according to their frequency range.

[0030] This processing module effectively eliminates environmental noise, random interference, and non-faulty vibrations inside the equipment, while retaining key fault signals. This method is more stable than traditional empirical mode decomposition (EMD), and the decomposition results are not affected by mode aliasing, thus having higher resolution.

[0031] Complex vibration signals are decomposed into analyzable modal components.

[0032] It improves the separation accuracy of fault characteristic signals and background noise.

[0033] This ensures the accuracy of subsequent algorithms and the reliability of diagnosis.

[0034] Feature extraction module The feature extraction module employs Sparse Bayesian Inference (SBI) to model and optimize the decomposed modal signals. Each modal signal is assumed to consist of fault feature signals and random noise. By introducing a sparse prior, the fault feature signals are modeled as a sparse distribution, while the noise signals are modeled as a Gaussian distribution. The module dynamically optimizes the weight parameters to gradually eliminate noise components. In this module, Bayesian inference is used to maximize the estimation of the posterior probability of the signal. The optimization objective is to extract the sparse part of the modal signal while weakening the influence of the noise distribution. The specific formulas for dynamically updating the sparse signal weights and noise weights are as follows: Sparse signal weights:

[0035] in The number of signal samples. The square of the sparse signal amplitude, Let be the signal covariance matrix.

[0036] Noise weighting:

[0037] in The total number of samples, The original modal signal, It is a sparse feature signal.

[0038] It reduces the interference of random noise and enhances the sparsity of fault characteristic signals.

[0039] Dynamic optimization improves applicability under different operating conditions.

[0040] It provides high-quality feature signal input for signal distribution optimization and matching.

[0041] Through this optimization process, high-frequency noise is further suppressed, and low-frequency fault feature signals are represented in a sparser manner. The extracted signals have higher characterization capabilities, laying the foundation for subsequent fault diagnosis.

[0042] Fault diagnosis module The fault diagnosis module uses an algorithm based on target distribution optimization theory to further optimize the distribution and pattern matching of the extracted feature signals. This module completes fault mode recognition by calculating the distance between the feature signal distribution and the target distribution. In order to measure the difference between the distributions, the first-order Wasserstein distance (i.e., the optimal transmission distance) is used, which can reflect the global deviation of the distribution on the probability density.

[0043] The calculation method is as follows: First, construct the characteristic signal distribution. and target distribution The target distribution is obtained in advance through modeling of equipment health signals and typical fault signals.

[0044] Then, based on the Wasserstein distance formula:

[0045] in, and The cumulative distribution functions of the characteristic signal and the target distribution are respectively.

[0046] Finally, by optimizing the transmission distance, the distribution of characteristic signals is brought closer to the target distribution, thereby achieving fault mode matching.

[0047] The feature signals, after distribution optimization, can be mapped to the fault mode space more accurately. The system analyzes the degree of matching between different modes and outputs the fault type and its corresponding severity.

[0048] By optimizing the distribution, the accuracy of fault mode identification was improved.

[0049] The optimal transmission method enhances the robustness of the diagnostic system and is suitable for a variety of complex operating conditions.

[0050] Output module The output module connects to the fault diagnosis module, presenting the diagnostic results in a visual format. The output includes: fault type, faulty component, fault severity, and equipment health status assessment results. The visualization interface intuitively displays the spectral characteristics, modal distribution, and fault evolution trend of the vibration signal. Diagnostic results support export in multiple formats for easy subsequent analysis and maintenance.

[0051] Module 1: Detailed Implementation of the Data Acquisition Module The data acquisition module is one of the core modules of this invention. Its function is to efficiently acquire vibration signals from key components of coal mining machinery, providing raw data input for subsequent signal processing and fault diagnosis. Generally, the operating environment of coal mining machinery is complex, and vibration signals may be interfered with by various factors such as environmental noise and fluctuations in equipment operating conditions. Therefore, this module needs to optimize signal integrity, anti-interference capability, and sensor arrangement. As a crucial part connecting with the subsequent data processing module, the data acquisition module must ensure that the quality and spectral characteristics of the acquired signals meet the input requirements of subsequent algorithms.

[0052] In this embodiment, the data acquisition module mainly includes multiple high-sensitivity vibration sensors, a signal acquisition unit, and a data transmission interface. The function and implementation of each component are as follows: Specifically, vibration sensors are used to capture vibration signals from key components of coal mining machinery, including but not limited to bearings, gearboxes, and drive motors. In one possible implementation, the sensor employs a high-precision accelerometer, which can measure the acceleration signal of mechanical vibration in real time and capture the impact characteristics and minute vibration changes during mechanical operation. Generally, the sensitivity range of these sensors is between 10mV / g and 1000mV / g, which can meet the measurement requirements of different vibration intensities.

[0053] As an option, the sensor installation location is determined based on the structural characteristics of the mechanical equipment. For example, for bearings, sensors are preferentially installed near the contact surface of the inner and outer rings of the bearing to more accurately capture the vibration frequency characteristics of the bearing; for gearboxes, sensors are generally installed on the surface of the housing, as close as possible to the gear meshing point, in order to record the vibration signal during gear meshing; for motors, sensors are installed on the fixed end of the motor housing to directly reflect the rotational imbalance or impact characteristics of the motor. The specific number of sensors and installation method can be adjusted according to the specific type of coal mining machinery.

[0054] During data acquisition, the signal acquisition unit is used to sample and digitize the analog signal output by the sensor. Generally, the sampling frequency of the signal acquisition unit needs to be higher than twice the frequency of the target vibration signal to meet the sampling theorem requirements. In some embodiments, the sampling frequency can be set between 10kHz and 100kHz, with the specific value determined according to the frequency range of the vibration signal. The sampling accuracy is generally 16 bits or higher to ensure the amplitude resolution of the signal. In possible implementations, the signal acquisition unit also includes preprocessing functions, such as filtering and anti-aliasing processing. For example, a low-pass filter can be used to remove high-frequency noise or unwanted frequency components.

[0055] After signal acquisition is completed, the sampled signal is transmitted to the subsequent data processing module through the data transmission interface. Specifically, the data transmission interface supports multiple transmission methods. For example, the wired method can use industrial Ethernet or RS485 communication interface to ensure the stability of data transmission; the wireless method can use Wi-Fi or ZigBee protocol, which is suitable for scenarios that require remote monitoring.

[0056] In one possible implementation, to enhance the anti-interference capability of the data acquisition module, a shielding layer can be added to the sensor signal line to reduce external electromagnetic noise interference. Alternatively, a dynamic range extension function can be integrated into the signal acquisition unit to enable it to handle strong and weak vibration signals under different working conditions.

[0057] In addition, to adapt to the multi-noise interference problem in the complex coal mine environment, this module can add an environmental parameter acquisition unit to record environmental information such as temperature, humidity and dust concentration. This environmental data can be recorded together with the vibration signal to provide a reference for subsequent noise modeling and signal separation.

[0058] Generally, the collected vibration signal data is stored in time series format, and its mathematical expression is as follows:

[0059] in, This indicates the collected vibration signal. This represents the actual vibration signal of the mechanical component. This could be due to environmental noise or random interference.

[0060] In another possible implementation, the acquired signal can be directly subjected to spectrum analysis to determine the frequency range of the main vibration characteristics in the signal. For example, the spectrum distribution of the vibration signal can be obtained through Fast Fourier Transform (FFT), and the results of spectrum analysis can assist the mode decomposition algorithm in subsequent modules to determine the optimal frequency division range.

[0061] Through the above design, this module can provide high-quality raw vibration signal input for subsequent signal processing modules, ensuring that fault characteristic signals are not lost due to distortion or noise interference during the acquisition process. The technical solution of this module is fully disclosed and can provide complete guidance for those skilled in the art to reproduce it.

[0062] Module 2: Detailed Implementation of the Data Processing Module After the data acquisition module collects the vibration signal, the core task of the data processing module is to decompose the signal to extract the effective signal with fault characteristics, while suppressing environmental noise and random interference. This module uses the variational mode decomposition (VMD) method to decompose the complex vibration signal into several narrowband modal components. Generally, VMD can handle non-stationary and noisy vibration signals in coal mining machinery with high stability, making the decomposed modal signals more suitable as input for the subsequent feature extraction module.

[0063] As an alternative, the data processing module mainly optimizes and decomposes the frequency and time characteristics of the input signal. After decomposition, the center frequency and spectral range of each modal component are clear, and noise components and mechanical characteristic signals can be effectively isolated. Through the processing of this module, the feature extraction module can be provided with frequency-domain accurate signal component input.

[0064] In this embodiment, the data processing module uses variational mode decomposition to analyze the vibration signal. Specifically, VMD decomposes the original signal into several modal signals by optimizing the objective function. The objective function is defined as follows:

[0065] in, For input vibration signal; For the first One modal signal; The center frequency of the modal signal; This represents the total number of modal signals; For time differential operators; express Norm.

[0066] The objective function described above is solved using the Alternating Direction Multiplier Method (ADMM). Generally, the initial frequency... The settings can be adjusted based on the signal's spectral characteristics, thereby improving the convergence speed of the decomposition algorithm. During the optimization process, modal components... and center frequency Alternate updates are performed to gradually approach the optimal solution.

[0067] Specifically, the optimization process includes the following key steps: First, the vibration signal... Initialized to several modal components .

[0068] Alternatively, the initial frequency range can be estimated from the signal's spectral distribution. This ensures that each modal component is focused on a different frequency band.

[0069] Next, at a fixed center frequency In the case of updating the modal signal The updated modal components can more accurately reflect the vibration characteristic signals of the corresponding frequency band.

[0070] Then, based on the currently updated modal signal Recalculate the center frequencies of the modal components. .

[0071] The center frequency can be updated using the following formula:

[0072] in, The center frequency of the modal signal; This represents the energy distribution of the modal signal.

[0073] Finally, the convergence of the decomposition results is evaluated by calculating the signal residuals. If the residuals are below a preset threshold, the decomposition process terminates; otherwise, iterative updates continue. In some embodiments, the total number of modal signals can be dynamically adjusted to further optimize the decomposition results. For example, in high-noise environments, the number of modes can be increased to more finely separate noise and characteristic signals; in low-noise environments, the number of modes can be reduced to improve decomposition efficiency. Through these steps, the vibration signal is decomposed into several modal components. Generally, high-frequency modal signals mainly contain environmental noise and random interference, while low-frequency modal signals reflect the operating status and fault characteristics of mechanical equipment. In one possible implementation, the signals can be filtered according to their frequency range, eliminating high-frequency noise and retaining only the low-frequency modal signals related to fault characteristics.

[0074] Module 3: Detailed Implementation of the Feature Extraction Module After the data processing module completes the modal decomposition of the vibration signal, the main task of the feature extraction module is to extract key feature signals related to coal mine machinery faults from the decomposed modal signals, while suppressing noise interference as much as possible. Since the vibration signals of coal mine machinery are usually affected by complex working conditions and contain strong noise and multiple interference components, the extraction of effective feature signals needs to maintain a balance between robustness and accuracy. Generally, this module uses the Sparse Bayesian Inference (SBI) method to dynamically separate noise signals and fault feature signals and generate feature signals that can be used by the subsequent fault diagnosis module.

[0075] As an alternative, this module performs feature modeling and optimization on multiple modal signals output by the data processing module. These modal signals have been separated into narrowband components in the frequency domain. The feature extraction module further utilizes the sparsity of the signal to separate the effective signal from the noise through modeling, ensuring that the extracted feature signals can reflect the mechanical operating state.

[0076] In this embodiment, the feature extraction module uses sparse Bayesian inference to perform feature separation on the modal signal. Specifically, the input modal signal... Assumed to be composed of characteristic signals and noise signals Composition, its mathematical expression is:

[0077] in, Indicates the first One modal signal; It is a sparse feature signal in modal signals. For noise signals in modal signals, generally, characteristic signals The signal is modeled as a sparsely distributed signal, following a zero-mean Gaussian distribution, and its prior probability is:

[0078] in, These are sparse weights, used to characterize the sparsity of the feature signals; It is an identity matrix.

[0079] Meanwhile, noise signal It is modeled as a Gaussian distribution with zero mean and is used to describe random disturbances or background noise.

[0080] As one possible implementation, this module optimizes the feature signal using Bayesian inference methods. Specifically, by constructing joint posterior probabilities This allows for the estimation of characteristic components in modal signals. During the optimization process, the sparse weights are adjusted. and noise weights To gradually improve the accuracy of feature signals while reducing noise interference, another possible implementation involves using sparse weights for the feature signals to improve algorithm efficiency. It can be updated using the following formula

[0081] in, This represents the number of samples in the modal signal. The square of the amplitude of the characteristic signal; Let be the covariance matrix of the characteristic signal.

[0082] Weights of noise signals The data is then updated based on the relationship between noise energy and the total number of samples, thereby achieving dynamic noise suppression. Through the above sparse Bayesian inference process, the final extracted feature signal is obtained. This method can accurately reflect the key vibration characteristics of mechanical operation and avoid interference from noise components. In practical applications, the extracted feature signals can include key fault information such as bearing fault frequency components, gear meshing frequency components and their harmonic characteristics, providing reliable input data for subsequent fault mode identification. The above technical solution is fully disclosed, including complete modeling and optimization details, ensuring that those skilled in the art can reproduce and apply it based on the disclosed content.

[0083] Module 4: Detailed Implementation of the Fault Diagnosis Module After the feature extraction module extracts the key feature signals, the main task of the fault diagnosis module is to identify and diagnose the fault type based on the feature signals. This module adopts the target distribution optimization theory to identify the operating status of coal mining machinery and potential early faults by analyzing the matching relationship between the distribution characteristics of the feature signals and the fault modes. In general, there may be some differences between the extracted feature signals and the distribution of the target fault modes. This module uses the optimal transmission theory to align the distributions to ensure the accuracy and robustness of the diagnostic results.

[0084] Alternatively, this module processes sparse feature signals output by the feature extraction module as input. These signals already contain key vibration features during equipment operation. By establishing a mapping relationship between the feature signal distribution and the target fault distribution, pattern matching and classification of different fault types can be achieved. The output of this module includes fault type, fault severity, and fault-related feature descriptions.

[0085] In this embodiment, the fault diagnosis module optimizes the matching between the distribution of characteristic signals and the target fault distribution using optimal transmission theory. Specifically, the distribution of characteristic signals... Distribution of target failure modes The first-order Wasserstein distance is used as a metric. The Wasserstein distance is an indicator of the difference between two distributions, particularly suitable for analyzing signal characteristics with differences in probability density. Generally, the characteristic signal distribution... Modeling can be performed by extracting the probability density function (PDF) of the signal, as an alternative, for the target fault mode distribution. It can be pre-trained based on historical equipment data or fault models. For example, common bearing fault modes (such as rolling element defects and outer ring defects) can be established using historical vibration data to create a target distribution. Finally, the Wasserstein distance can be used to generate the distribution. The mathematical expression for the deviation between two distributions is as follows:

[0086] in, The probability density distribution of the characteristic signal. The probability density distribution of the target failure mode; and The cumulative distribution function (CDF) is the characteristic signal and the target distribution, respectively. In one possible implementation, to optimize diagnostic efficiency, the module sorts the fault modes according to the size of the Wasserstein distance. Generally, the smaller the distance, the more likely the mode is to match the current device state.

[0087] First, probability density modeling is performed on the sparse feature signal output by the feature extraction module. Alternatively, the probability density function of the feature signal can be constructed using kernel density estimation (KDE). Kernel density estimation is suitable for nonparametric distribution modeling and can efficiently estimate the distribution characteristics of the signal without prior distribution assumptions. Next, the module calls the pre-stored target fault mode distribution. And compare the distribution of characteristic signals .

[0088] As an option, the target distribution can be constructed offline using historical vibration data, covering common fault types in coal mine machinery, such as bearing outer ring faults and gear wear. Then, the difference between the feature signal distribution and each target distribution is calculated based on the Wasserstein distance. To improve computational efficiency, the module can quickly solve for the optimal transmission path using the Kantorovich bilinear optimization method in optimal transmission theory. After calculation, the feature signal distribution is matched with the closest target distribution, and the corresponding fault diagnosis result is generated. Alternatively, this module can also combine the energy and frequency information of the feature signal to further assist in the verification of the diagnosis result. For example, when the spectrum of the feature signal is concentrated near the specific fault frequency of the bearing, the weight of fault mode matching can be further increased. In another possible implementation, this module can generate a confidence score for the diagnosis result. For example, a matching score is generated for each fault type based on the normalized value of the Wasserstein distance, and the fault type with the highest score is output as the diagnosis result. Through the above analysis and calculation, the final output of this module includes: The equipment's operating status, such as "normal" or "abnormal"; the fault type, such as "bearing inner ring defect" or "gear meshing failure"; and the fault severity, such as the degree of deviation of the fault characteristics from the target distribution represented by the cumulative Wasserstein distance.

[0089] The diagnostic results can also be displayed through a visualization interface, such as a comparison chart of the distribution of characteristic signals and the target distribution, thus providing an intuitive reference for equipment maintenance personnel.

[0090] The technical solutions of this module have been fully disclosed, including the specific implementation methods of feature signal modeling, distribution optimization matching and fault mode recognition. The above content can provide a complete basis for reproduction for those skilled in the art and can be widely applied to early fault diagnosis scenarios of coal mine machinery and equipment.

[0091] Module 5: Detailed Implementation of the Output Module After the fault diagnosis module identifies the operating status and fault type of the coal mine machinery, the output module is responsible for integrating and presenting the diagnostic results. Generally, the diagnostic results need to be presented to users in an intuitive and easy-to-understand form so that equipment maintenance personnel can take targeted repair measures in a timely manner. At the same time, the output module also needs to support the historical record of diagnostic results, visualization display, and multi-format output to meet the needs of different application scenarios.

[0092] Alternatively, this module processes diagnostic results data output by the fault diagnosis module as input, including fault type, fault severity, fault characteristic description, and corresponding confidence score. The output module needs to present this information graphically or textually according to user requirements and support connectivity with equipment monitoring platforms or industrial networks for easy system integration.

[0093] In this embodiment, the core task of the output module is to organize, analyze and display the output results of the fault diagnosis module. Specifically, this module includes the following sub-functions: display of diagnosis results, visualization of the comparison between feature signals and target distribution, storage and backtracking of historical data, and export support for multiple formats.

[0094] As one possible implementation, this module offers diverse output methods for diagnostic results. For example, for describing the operating status of coal mine machinery, a graphical interface can be used to intuitively present the fault location and type. The output results may include the following: The operating status of the equipment, such as "normal" or "abnormal"; Fault type, such as "bearing outer ring defect" or "gear meshing failure"; The severity of the fault can be represented, for example, by the accumulated deviation value, indicating the mismatch between the characteristic signal and the target distribution.

[0095] Specifically, the visualization function in this module can display the comparison results between the characteristic signal and the target distribution in the form of spectrum graphs, waveform graphs, or scatter plots. For example, by drawing a spectrum graph of the fault characteristic signal, the abnormal peaks of the characteristic signal in terms of frequency and their corresponding frequency components can be intuitively reflected. In another possible implementation, a heat map can also be used to display the energy distribution of the fault characteristic signal to help determine the specific location and severity of the fault.

[0096] In general, to provide detailed fault description information, this module also supports automatic report generation based on feature signals. For example, by combining the confidence score from the fault diagnosis module, this module can automatically generate a text-based diagnostic report, including the following information: A detailed description of the fault, such as "the gears are worn excessively, and it is recommended to replace the gears as soon as possible"; Preliminary analysis of the cause of the failure, such as "gear surface fatigue may be caused by insufficient lubrication"; Preventative maintenance recommendations include, for example, "It is recommended to check the viscosity and amount of gear lubricant every 300 hours."

[0097] Alternatively, this module can also export the generated diagnostic reports in formats such as PDF, CSV, or XML for user archiving or integration with other industrial monitoring systems. In some embodiments, the reports can also be transmitted to a remote monitoring center via the Industrial Internet to achieve centralized management and fault monitoring of multiple devices.

[0098] To further enhance the user experience, this module also supports the backtracking and querying of historical fault diagnosis data. By building a time series database, users can track the changes in the operating status of the equipment. For example, users can query the distribution of vibration signals and the corresponding diagnostic results within a specific time period through the interface, thereby analyzing the long-term performance degradation trend of the equipment.

[0099] In another possible implementation, this module supports real-time monitoring, updating and displaying the current operating status and diagnostic results of the equipment in the form of dynamic graphics. For example, using a dashboard-style interface to display the vibration amplitude, fault confidence, and operating risk level of the equipment can help users keep track of the equipment's operating status in real time.

[0100] During the output of signals and diagnostic results, this module also supports data encryption and access control functions to ensure the security of data during transmission and storage. As an option, the AES encryption algorithm can be used to encrypt the diagnostic results, and access permissions for the data can be controlled through a user authentication mechanism.

[0101] In this embodiment, the output module also supports integration with industrial control systems or equipment management systems. For example, diagnostic results can be transmitted to a PLC or SCADA system via Modbus or OPCUA protocols to achieve automated fault alarms and responses.

[0102] Through the above design, the output module can effectively improve the operability and practicality of the diagnostic system, ensuring that the diagnostic results can be quickly and accurately transmitted to the user or device management platform. The technical solution of this module has been fully disclosed, covering the complete implementation path from data visualization to report generation, historical query and data security, and has good reproducibility and scalability.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A vibration analysis-based early fault diagnosis system for coal mine machinery, characterized in that, include: The data acquisition module is used to collect vibration signals from key components of coal mining machinery; The data processing module, connected to the data acquisition module, is used to perform frequency band separation on the vibration signal based on variational mode decomposition, decomposing the vibration signal into multiple modal components; The feature extraction module, connected to the data processing module, is used to perform noise modeling and feature signal extraction of modal components based on sparse Bayesian inference. The fault diagnosis module, connected to the feature extraction module, is used to match and diagnose the feature signal distribution with the fault mode distribution based on the target distribution optimization theory. The output module, connected to the fault diagnosis module, is used to output the fault type and early fault diagnosis results of the coal mining machinery.

2. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, The data processing module decomposes the vibration signal based on the variational mode decomposition algorithm, specifically including the following steps: Construct a modal decomposition objective function for the input vibration signal to decompose the vibration signal into multiple modal components; The frequency center and signal amplitude of the modal components are iteratively updated using optimization methods; The modal components are divided into high-frequency components and low-frequency components. The high-frequency components contain noise signals, while the low-frequency components contain mechanical fault characteristic signals.

3. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, The feature extraction module processes the modal components through sparse Bayesian inference, specifically including the following steps: Assuming that the modal components consist of sparse signals and Gaussian noise, we construct the prior distribution of the sparse signals and the noise distribution, respectively. The posterior distribution of sparse signals in modal components is solved by Bayesian inference. The weights and noise weights of sparse signals are dynamically updated to suppress high-frequency noise and optimize the extraction of low-frequency feature signals.

4. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, The fault diagnosis module processes the extracted feature signals based on the target distribution optimization theory, specifically including the following steps: A probability density function model is constructed for the extracted feature signals to characterize the distribution of the feature signals; Establish a probability distribution model for the target failure modes; Calculate the transmission distance between the characteristic signal distribution and the target fault mode distribution; By optimizing the transmission distance, the distribution of characteristic signals is brought closer to the target distribution, thus completing fault mode matching and diagnosis.

5. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, The data acquisition module includes multiple vibration sensors, which are installed in key parts of the coal mining machinery, including bearings, gearboxes, and drive motors, to collect acceleration signals and vibration frequency signals.

6. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, The early fault diagnosis results output by the system include the fault type, severity, and location of key components of coal mining machinery.

7. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, The data processing module performs frequency segmentation processing on the vibration signal, where the low-frequency signal is used to extract the operating status characteristics of the equipment, and the high-frequency signal is used to identify random noise or environmental interference.

8. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, The signal weights in the sparse Bayes inference are updated using the following formula: the weights of the sparse signal are dynamically adjusted based on the ratio of the signal amplitude to the covariance matrix, and the update rule is as follows: in, The number of signal samples. The square of the amplitude of the characteristic signal, Let be the signal covariance matrix.

9. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, When optimizing the transmission distance between the characteristic signal distribution and the target distribution, the fault diagnosis module uses distance to calculate the deviation of the signal distribution, and the optimization target is: in, Represents the probability distribution of the characteristic signal. Represents the probability distribution of the target failure mode. and These are the cumulative distribution functions of the characteristic signal and the target distribution, respectively.

10. The early fault diagnosis system for coal mine machinery based on vibration analysis according to claim 1, characterized in that, The system includes a health status prediction module, which predicts the health status of coal mining machinery and calculates its remaining service life based on extracted feature signals and diagnostic results.

Images