A thermal power belt conveyor fault early warning method and system

Abstract

This invention discloses a fault early warning method and system for thermal power plant conveyor belts, belonging to the field of electromechanical equipment fault diagnosis technology. The method collects multi-dimensional operating data of the conveyor belt, including vibration, temperature, and current. After noise reduction, time alignment, and removal of transient start-stop data, a steady-state dataset is constructed. Fault-sensitive features are extracted, load-vibration correlation curves are fitted, and the normal operating range is defined. The residual deviation between measured and theoretical vibration is calculated, and the fault type and severity are identified by combining the sensitive features, ultimately generating graded early warning information. This invention effectively eliminates interference from load and operating condition fluctuations, reduces false alarms, achieves early identification of progressive faults in conveyor belts, improves the operational safety of thermal power plant coal conveying equipment, and reduces operation and maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for electromechanical equipment, specifically to a fault early warning method and system for thermal power plant belt conveyors. Background Technology

[0002] Coal transportation in thermal power plants primarily relies on belt conveyors, whose operational status directly impacts plant safety and power generation efficiency. Current belt conveyor maintenance largely employs manual inspections or single-parameter threshold alarms. Manual inspections are labor-intensive, have blind spots and are often delayed, making it difficult to detect early potential faults. Single-parameter threshold alarms do not consider the impact of belt conveyor load changes on operating parameters, easily generating false alarms during load increases or startup, leading to desensitization of alarm information by maintenance personnel. Existing technologies cannot effectively distinguish between normal operating fluctuations and abnormal equipment conditions, lacking effective early identification methods for progressive faults such as belt misalignment and idler wear. Faults are often only discovered after they have caused significant impact, increasing the risk of unplanned downtime and maintenance costs. Summary of the Invention

[0003] The technical problems to be solved by this invention include at least one of the following: manual inspection has blind spots and lag, making it impossible to identify potential progressive faults of belt conveyors in the early stage; single parameter threshold alarms are easily affected by load and start-up / shutdown conditions, resulting in a large number of false alarms; it is impossible to distinguish between normal parameter fluctuations caused by load and abnormal parameter changes caused by equipment failure; fault types cannot be accurately identified, fault severity cannot be quantified, and early warning lacks a graded basis.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for early warning of faults in thermal power plant conveyor belts includes the following steps:

[0006] S1: Deploy sensors to collect vibration data, temperature data, and current data. After noise reduction and time alignment, remove transient start-stop data to form a standard dataset for steady-state operation of the belt conveyor.

[0007] S2: Extract vibration and temperature features from the standard dataset, integrate and normalize them, and then filter out the features to construct a set of fault-sensitive feature parameters;

[0008] S3: Fit the load-vibration correlation curve based on historical health data, calculate the confidence interval, and divide the normal fluctuation area of ​​equipment vibration;

[0009] S4: Calculate the residual deviation between measured and theoretical vibration, determine abnormal conditions, identify fault types, and calculate fault severity index;

[0010] S5: Classify the warning level according to the fault type and severity index, create corresponding prompts, and complete the warning information release;

[0011] Step S3 further includes:

[0012] Step 3-1: Extract historical steady-state data of equipment health periods, preprocess it, and construct a historical health dataset covering the full load.

[0013] Step 3-2: Using the motor current as the load parameter and the effective vibration value as the response parameter, plot a scatter plot showing the correlation between the two.

[0014] Step 3-3: Perform regression analysis on the scatter data to fit the baseline correlation curve model between load and vibration;

[0015] Steps 3-4: Calculate the confidence interval of the baseline model and divide the equipment vibration into the normal fluctuation zone and the suspected fault zone.

[0016] In one embodiment of the present invention, the sensor includes a triaxial accelerometer, a contact thermocouple, an ambient temperature sensor, and a current transformer.

[0017] In one embodiment of the present invention, the historical health dataset includes operating data of the belt conveyor under different load conditions from no load to full load.

[0018] In one embodiment of the present invention, the horizontal axis of the scatter plot represents the motor current value, and the vertical axis represents the effective vibration value at the corresponding moment. Each data point represents the combination of current and vibration at a moment.

[0019] In one embodiment of the present invention, in steps 3-4, the confidence interval of the load-vibration reference function is calculated, and the area within the confidence interval is defined as the normal fluctuation area, and the area above the upper limit of the confidence interval is defined as the suspected fault area.

[0020] In one embodiment of the present invention, step S4 includes the following steps:

[0021] Step 4-1: Substitute the real-time current into the load-vibration reference function, calculate the theoretical vibration value, and obtain the residual deviation between the measured vibration and the actual vibration.

[0022] Step 4-2: Compare the residual deviation with the preset threshold to determine whether the equipment is in a healthy state or a suspected fault state;

[0023] Step 4-3: Analyze and determine the specific fault type of the belt conveyor by combining the changing patterns of sensitive characteristic parameters;

[0024] Step 4-4: Based on the residual deviation and the temperature change rate, calculate and correct the quantitative index of fault severity.

[0025] In one embodiment of the present invention, the sensitive feature parameters include kurtosis index, envelope spectrum energy value, transverse vibration to longitudinal vibration amplitude ratio, and temperature characteristics.

[0026] The present invention also provides a fault early warning system for thermal power plant belt conveyors, comprising:

[0027] The system includes: a multi-dimensional data acquisition and steady-state dataset construction unit, a fault-sensitive feature extraction and feature set construction unit, a load vibration correlation modeling and normal operation zone delineation unit, an anomaly identification, fault diagnosis and severity quantification unit, and a graded early warning information generation and dissemination unit.

[0028] The input of the multidimensional data acquisition and steady-state dataset construction unit is the raw data collected in real time by sensors deployed at various measuring points of the belt conveyor, and the output is a steady-state standard dataset to the fault-sensitive feature extraction and feature set construction unit and the load vibration correlation modeling and normal operation area delineation unit.

[0029] The fault-sensitive feature extraction and feature set construction unit extracts vibration time-domain features, vibration frequency-domain features, and temperature features based on the input standard dataset. After normalization and redundant feature removal, it outputs a set of fault-sensitive feature parameters to the anomaly identification, fault diagnosis, and severity quantification unit.

[0030] The load-vibration correlation modeling and normal operation zone delineation unit obtains the load-vibration benchmark correlation model through data fitting, calculates the confidence interval and delineates the normal fluctuation zone and the suspected fault zone, and outputs the load-vibration benchmark correlation model, vibration deviation threshold and normal operation zone boundary to the anomaly identification, fault diagnosis and severity quantification unit.

[0031] The anomaly identification, fault diagnosis, and severity quantification unit receives steady-state standard data output in real time from the multi-dimensional data acquisition and steady-state dataset construction unit, a set of fault-sensitive feature parameters output from the fault-sensitive feature extraction and feature set construction unit, and a benchmark model and judgment threshold output from the load vibration correlation modeling and normal operation area delineation unit. After residual calculation, feature matching, and severity correction, it outputs the fault type judgment result and the fault severity index in the 0-100 range to the graded early warning information generation and release unit.

[0032] The graded early warning information generation and release unit classifies fault levels and generates corresponding prompts.

[0033] In one embodiment of the present invention, the load vibration correlation modeling and normal operation area delineation unit completes the construction of the load-vibration function relationship and the delineation of the normal operation area based on the health historical steady-state data; the fault-sensitive feature extraction and feature set construction unit first extracts the most recent steady-state data of the equipment from the historical operation database to construct a historical health dataset covering the entire load from no-load to full-load; then, using the drive motor current value as the load characterization parameter and the vibration effective value as the vibration response parameter, a scatter plot of the correlation between the two is drawn, and a second-order polynomial regression analysis is performed on the scatter data using the least squares method to fit the benchmark correlation curve model of load and vibration; after completing the model fitting, the fault-sensitive feature extraction and feature set construction unit calculates the confidence interval of the benchmark model and divides the equipment vibration normal fluctuation area and the fault-suspected area.

[0034] The fault severity index S=S base ×k, where k is the temperature correction coefficient, S base Based on the severity of the fault, S base =(|Residual Deviation|-Normal Critical Residual) / Effective Quantization Interval Width, Effective Quantization Interval Width = Maximum Fault Residual - Normal Critical Residual.

[0035] Compared to existing technologies, the beneficial effects of this invention are as follows: Compared to manual inspection, this invention eliminates blind spots and lags in inspections through multi-dimensional data acquisition from sensors and steady-state data screening, enabling early detection of equipment fault characteristics; Compared to single-parameter threshold alarms, this invention establishes a load-vibration correlation benchmark, eliminating interference from load and start-up / shutdown conditions, thus reducing false alarms at the source; This invention extracts fault-sensitive features and combines them with residual analysis to distinguish between normal operating condition fluctuations and equipment anomalies, effectively identifying progressive faults such as idler wear, belt misalignment, and roller bearing failure; This invention achieves accurate fault type diagnosis and severity quantification, coupled with a graded early warning mechanism, reducing the risk of unplanned downtime and lowering equipment maintenance costs. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the fault early warning method for thermal power plant belt conveyors provided by the present invention;

[0037] Figure 2 A schematic diagram of the steady-state standard dataset provided by this invention;

[0038] Figure 3 A schematic diagram of sample data for extracting fault-sensitive features provided by the present invention;

[0039] Figure 4 A schematic diagram of the load-vibration correlation curve and confidence interval provided by the present invention;

[0040] Figure 5This is a schematic diagram of the structure of the thermal power plant belt conveyor fault early warning system provided by the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. In the drawings or description, similar or identical parts are referred to by the same reference numerals, and in practical applications, the shape, thickness, or height of each component may be enlarged or reduced. The embodiments listed in this invention are merely illustrative and not intended to limit the scope of the invention. Any obvious modifications or changes made to this invention do not depart from the spirit and scope of the invention.

[0042] In one embodiment, such as Figure 1 As shown in the figure, this embodiment illustrates a fault early warning method for thermal power plant belt conveyors based on vibration load reference correlation. The method includes the following steps:

[0043] Step 1: Acquisition of multi-dimensional operational data and construction of steady-state dataset. Sensors are deployed to collect vibration, temperature, and current data. After noise reduction and time alignment, transient data during start-up and shutdown are removed to form a standard dataset for steady-state operation of the belt conveyor.

[0044] Step 2: Fault-Sensitive Feature Extraction and Set Construction. Vibration and temperature features are extracted from the steady-state standard dataset, integrated and normalized, and then filtered to construct a fault-sensitive feature parameter set.

[0045] Step 3: Establish the load-vibration function relationship and delineate the normal operating range. Fit the load-vibration correlation curve based on historical health data, calculate the confidence interval, and delineate the normal fluctuation range of equipment vibration.

[0046] Step 4: Anomaly Identification, Fault Diagnosis, and Severity Quantification. Calculate the residual deviation between measured and theoretical vibrations, determine the abnormal state, identify the fault type, and calculate the quantification index of fault severity.

[0047] Step 5: Generation and dissemination of tiered early warning information. Early warning levels are determined based on fault type and severity indicators. Corresponding prompts are created, and the early warning information is disseminated.

[0048] In step 1, various sensors are arranged at key locations on the belt conveyor to collect various data reflecting the operating status of the equipment, and the raw data is cleaned and organized to remove invalid interference information.

[0049] Furthermore, vibration sensors and temperature sensors are installed on the drive roller bearing housing, the redirecting roller bearing housing, and the idler roller brackets of key sections of the belt conveyor, respectively, and a current transformer is connected to the drive motor control circuit.

[0050] Preferably, the vibration sensor is a triaxial accelerometer, capable of simultaneously acquiring vibration signals in three mutually perpendicular directions. The temperature sensor uses a contact thermocouple, embedded inside the bearing to directly measure the core operating temperature. Simultaneously, an ambient temperature sensor is placed in the belt conveyor corridor for subsequent temperature compensation. A current transformer is used to collect the operating current of the drive motor, thereby characterizing the actual load on the belt conveyor.

[0051] More preferably, the sampling frequency of the vibration signal is not less than 1000Hz, and the sampling frequency of the temperature and current signals is not less than 1Hz. All collected data are time-stamped for subsequent data alignment processing. Preferably, at the outlet of the guide chute and the roller reversing position where belt misalignment is prone to occur, the number of vibration sensors is increased to improve the accuracy of misalignment fault identification.

[0052] The acquired raw vibration signal was denoised to remove high-frequency electromagnetic interference and environmental noise, retaining only the effective signal components reflecting the mechanical operating state of the equipment. The denoising process employed wavelet transform, decomposing the raw vibration signal into sub-band signals of different frequencies by selecting appropriate wavelet basis functions and decomposition levels. Thresholding was applied to the sub-band signals containing high-frequency noise to remove the noise components. The processed sub-band signals were then reconstructed to obtain the denoised vibration signal. Correspondingly, the temperature and current signals were also subjected to simple smoothing to eliminate instantaneous random fluctuations and make the signal variation trends clearer.

[0053] The denoised vibration, temperature, and current signals undergo time alignment to map data from different sampling frequencies onto a unified time axis. Because the vibration, temperature, and current signals have different sampling frequencies, data collected at the same time may have slight temporal deviations. Preferably, time alignment uses the vibration signal acquisition time point as a reference, and linear interpolation is performed on the temperature and current signals to obtain temperature and current values ​​that perfectly correspond to the vibration signal time point. After time alignment, each time point contains corresponding vibration, temperature, and current data. Furthermore, distorted data that significantly exceeds the physical measurement range is removed; for example, vibration signals showing values ​​beyond the sensor's measurement range, or temperature signals showing extreme values ​​that are not realistic.

[0054] The cleaned dataset is extracted based on the start and stop logic signals of the belt conveyor. Transient process data during motor start-up and shutdown are removed, and only the data during the stable operation of the belt conveyor is retained as the standard dataset.

[0055] During the start-up and shutdown of a belt conveyor, operating parameters may experience drastic, transient changes. These changes are normal transitional conditions and not caused by equipment malfunction. Including this transient data in the analysis can interfere with fault diagnosis and lead to false alarms. By capturing data from stable operating periods, subsequent analysis can be based on the steady-state operation of the equipment, improving the accuracy of fault identification. Preferably, the criteria for a stable operating period are that the motor current fluctuates by no more than 5% for five consecutive minutes, and the belt speed remains between 95% and 105% of the rated speed.

[0056] This step involves collecting a standard dataset, which specifically includes the following three types of data:

[0057] The first category consists of triaxial vibration data that has been denoised by wavelet transform, time-aligned, and with the start-stop transient process removed. This data represents the vibration velocity signals of the belt conveyor drive roller bearing housing, redirecting roller bearing housing, and key section idler support positions.

[0058] The second category is temperature data that has been smoothed, time-aligned, and distorted values ​​removed. This data includes internal bearing contact temperature measurement data and belt corridor ambient temperature data.

[0059] The third category is drive motor current data that has been time-aligned and has had transient fluctuations removed. This data represents the actual load of the belt conveyor with the current value during stable operation, without the severe fluctuations during startup and shutdown.

[0060] All three types of data are mapped to a unified time axis, have precise time stamps, and retain only valid data from the steady-state operation period of the belt conveyor.

[0061] In one embodiment, data is collected from a belt conveyor in the coal conveying system of a 600MW coal-fired power unit. Figure 2 This is a schematic diagram of the standard dataset obtained after multidimensional raw operating data collection and preprocessing in step 1. The data are all measured values ​​during the steady-state operation of the equipment, and noise reduction, time alignment, and removal of start-stop transient data have been completed.

[0062] Figure 2The data collection scenario is a 600MW unit coal conveyor belt (width 1.4m, belt speed 2.5m / s, drive motor rated current 100A). Measurement point 1 is the drive drum bearing housing, the core monitoring point; measurement point 2 is the redirecting drum bearing housing; and measurement point 3 is the central key idler support. Data dimensions include sampling time, triaxial vibration RMS values ​​(X / Y / Z axes, unit: mm / s), bearing temperature (unit: ℃), and drive motor current (unit: A, representing load). In the collected vibration data, the triaxial vibration RMS values ​​are all within the normal vibration range of the coal conveyor belt drums and idlers in thermal power plants, with no abnormal exceeding the standard. In the collected temperature data, the bearing temperature is between 45℃ and 60℃, which is within the healthy operating temperature range of the rolling bearings of the coal conveyor belt, with no overheating phenomenon. Under fault-free conditions, the kurtosis K = 1.5~2.5, and the peak value Cp = 1.2~1.8.

[0063] Data correlation: Under the same sampling time, the data from the three measurement points correspond completely with the motor current. After time alignment processing, there is no time deviation.

[0064] Step 2 is based on the standard dataset obtained in Step 1. Various feature parameters that can reflect the mechanical health status of the belt conveyor are extracted from it, and features that are sensitive to faults are selected to form a set of sensitive feature parameters.

[0065] In step 2, the time-domain statistical characteristics of the vibration data in the standard dataset are first calculated. The calculated time-domain statistical characteristics include the effective value v of the vibration velocity signal. rms Kurtosis index K and peak index C p .

[0066] In the formula, v i The vibration velocity signal is the instantaneous value at the i-th sampling point, ranging from 0 to 20 mm / s; N is the total number of sampling points within a single sampling period, and in this invention, N is preferably 1000 (corresponding to the sampling frequency f). s =1000Hz). Under normal conditions, v in a healthy state rms The value range is 1.5~6.5 mm / s. Values ​​exceeding 6.5 mm / s are considered to be severely excessive vibrations.

[0067] , The arithmetic mean of the vibration velocity signal. ; The summation of the fourth-order central moments of the vibration velocity signal; This is the sum of the second-order central moments of the vibration velocity signal. Generally, in a healthy state, K ranges from 1.5 to 2.5, close to the kurtosis value of a normal distribution of 3. When early wear occurs in the bearing, the K value rapidly increases to 3-6; in severe wear, K > 6.

[0068] , v is the maximum absolute value of the instantaneous vibration velocity signal; rms This represents the effective value of the aforementioned vibration velocity signal. Under normal circumstances, the healthy state C... p The value ranges from 1.2 to 1.8, which is close to the normal distribution kurtosis value of 3; when sudden faults such as jamming or breakage occur, C p The value surged to over 2.5.

[0069] The above three types of time-domain characteristics reflect the operating status of the equipment from three different perspectives: vibration energy, impact sharpness, and instantaneous impact intensity.

[0070] Next, frequency domain analysis was performed on the vibration data in the standard dataset, converting the time-domain signal into a frequency-domain signal to extract frequency-domain features that reflect the operating characteristics of the equipment components. The frequency domain analysis employed the Fast Fourier Transform (FFT) method, decomposing the time-domain vibration signal into a superposition of sinusoidal signals of different frequencies, resulting in a signal spectrum. From the spectrum, the amplitude at the rotational frequency, the amplitude at the harmonic frequencies, and the envelope spectrum energy values ​​in the high-frequency band were extracted. The rotational frequency refers to the rotational frequency of the conveyor belt drum or idler roller; when a component malfunctions, the amplitude at its corresponding characteristic frequency will significantly increase. Envelope spectrum analysis can extract low-frequency fault features modulated by a high-frequency carrier wave. Through frequency domain feature analysis, the specific component where the fault occurred can be located.

[0071] Then, temperature characteristics are calculated from the temperature data in the standard dataset. These characteristics include absolute temperature values ​​and the rate of temperature change. The absolute temperature value is the measured temperature at the current moment, compared to the equipment's maximum permissible operating temperature to determine if overheating exists. The rate of temperature change is calculated by measuring the temperature change per unit time, reflecting the speed of temperature rise. When equipment malfunctions, increased friction leads to a rapid temperature rise; therefore, the rate of temperature change can detect potential faults earlier. Correspondingly, the temperature difference between different idlers within the same idler group is calculated. By comparing the temperature differences between adjacent components, localized overheating areas can be identified more accurately.

[0072] Next, the calculated vibration time-domain features, vibration frequency-domain features, and temperature features are integrated to construct a vibration feature vector matrix, and all feature values ​​in the feature vector matrix are normalized. Then, based on the statistical analysis results of historical fault data, the feature parameters most sensitive to common belt conveyor faults are selected to form a set of sensitive feature parameters. The selection criteria for sensitive feature parameters are that the value of the feature differs significantly between the normal and fault states of the equipment, and can stably reflect changes in the health status of the equipment.

[0073] The process of step 2 will be illustrated using the light-load, medium-load, and heavy-load steady-state data of measuring point 1 (drive drum bearing housing) from the data collected from a belt conveyor in the coal conveying system of a 600MW coal-fired power unit.

[0074] Select 3 sets of typical steady-state data (e.g.) Figure 3 (As shown) serves as the basis for feature extraction.

[0075] The temporal feature extraction process is as follows:

[0076] (Triaxial composite total effective value, characterizing vibrational energy), v X v Y v Z The instantaneous vibration velocities in the X, Y, and Z directions are respectively (X represents the axial / longitudinal direction of the belt, Y represents the lateral / deviation direction of the belt, and Z represents the vertical radial / up-down direction). The calculated light-load v... rms1 ≈3.55mm / s, medium load v rms2 ≈4.83mm / s; heavy load v rms3 ≈6.25mm / s.

[0077] Light load C p1 ≈1.31, Medium Load C p2 ≈1.42, Heavy load C p3 ≈1.53.

[0078] The frequency domain feature extraction process is as follows:

[0079] The time-domain signal is converted to the frequency domain using Fast Fourier Transform (FFT), and the frequency conversion amplitude A is extracted. f High-frequency envelope spectrum energy E e :

[0080] Light load frequency amplitude A f1 =120μm / s, E e1 =0.05J;

[0081] Medium-load frequency conversion amplitude A f2 =150μm / s, E e2 =0.06J;

[0082] Heavy-load frequency amplitude A f3 =180μm / s, E e3 =0.07J;

[0083] The extraction of temperature features mainly includes the following three aspects:

[0084] Absolute temperature value T: taken directly from the temperature measurement value in the dataset.

[0085] Temperature change rate ΔT / Δt: Calculates the temperature rise per unit time (minutes). For health-related devices, the temperature change rate is low during steady-state operation. For example, in the light, medium, and heavy-load steady-state data of this embodiment, this value is 0.3℃ / min.

[0086] Temperature difference ΔT among idler rollers in the same group roll Calculate the temperature difference between different idlers within the same idler group (taking measuring point 3 as an example). In the data of this embodiment, the difference is 0.2℃ under light, medium, and heavy load conditions.

[0087] Construct an 8-dimensional eigenvector matrix X=[v rms ,K,C p A f E e ,T,ΔT / v,ΔT roll The 3×8 feature matrix M calculated for measuring point 1 (drive roller bearing housing) is as follows, corresponding to the 3 sets of steady-state samples of measuring point 1: row 1: light load steady state (08:00), row 2: medium load steady state (08:05), row 3: heavy load steady state (08:10).

[0088] .

[0089] After normalization, the normalized characteristic matrix M is obtained. norm as follows:

[0090] .

[0091] Schematic diagram: The characteristic parameters most sensitive to common belt conveyor faults are selected according to the following rules:

[0092] Numerical discrimination rule: The relative value difference of the characteristic parameter between the healthy steady state and the fault state of the equipment is ≥50%, and the increase of the characteristic value of the fault state is not less than 50% compared with the healthy baseline value.

[0093] Fault spectrum rule: The characteristic parameters all show significant responses to three typical faults of belt conveyors (idler bearing wear, belt misalignment, and drum bearing failure).

[0094] Operating condition robustness rule: When the characteristic parameters are running under normal steady-state conditions across the entire load range of light, medium, and heavy loads, the numerical fluctuation range shall be ≤10%.

[0095] Based on the above rules, the temperature difference ΔT among the characteristic idler rollers in the same group is eliminated. roll The final set of sensitive feature parameters S={v rms ,K,C p A f E e ,T,ΔT / Δt}.

[0096] Step 3 further includes the following steps:

[0097] Step 3-1: Extract historical steady-state data of the equipment during the health period, and construct a historical health dataset covering the full load after preprocessing.

[0098] Step 3-2: Using the motor current as the load parameter and the effective vibration value as the response parameter, draw a scatter plot showing the correlation between the two.

[0099] Step 3-3: Perform regression analysis on the scatter data to fit the baseline correlation curve model between load and vibration.

[0100] Steps 3-4: Calculate the confidence interval of the baseline model and divide the equipment vibration into the normal fluctuation zone and the suspected fault zone.

[0101] In step 3-1, a historical standard dataset representing periods when the equipment was in a healthy state and had no fault records is extracted from the historical operation database of the belt conveyor. This historical health dataset covers the belt conveyor's operation data under different load conditions, from no-load to full-load. The data extraction time range is three months of stable operation data following the equipment's most recent comprehensive overhaul, during which time the equipment was in optimal operating condition. The extracted historical data undergoes the same preprocessing operations as in step 1, including noise reduction, time alignment, and effective data extraction. Preferably, the historical health dataset contains no fewer than ten thousand samples.

[0102] In step 3-2, the drive motor current value is used as the load characterization parameter, and the effective vibration value is used as the vibration response parameter. A scatter plot of the current value and the effective vibration value is then plotted. By plotting the scatter plot, the overall trend of the effective vibration value changing with the current value can be observed intuitively. The horizontal axis of the scatter plot represents the motor current value, and the vertical axis represents the effective vibration value at the corresponding time. Each data point represents the combination of current and vibration at a given time.

[0103] In step 3-3, regression analysis is performed on the data points in the scatter plot to fit a baseline curve describing the relationship between the current value and the effective vibration value. The baseline curve characterizes the natural increase in vibration energy with increasing load under normal mechanical clearance and good lubrication conditions of the belt conveyor. The regression analysis uses the least squares method, and the fitted function is a second-order polynomial. During the fitting process, outlier data points that significantly deviate from the overall trend are removed. These outliers may be caused by instantaneous disturbances or brief fluctuations in operating conditions, which can affect the fitting accuracy of the baseline curve. The mathematical relationship obtained after fitting is the baseline correlation model between vibration and load.

[0104] Figure 4The diagram schematically shows the scatter distribution of current values ​​and effective vibration values ​​plotted for historical standard datasets of measuring points 1-3, as well as the fitted baseline curve describing the relationship between current values ​​and effective vibration values.

[0105] In steps 3-4, the confidence interval of the load-vibration reference function is calculated. The area within the confidence interval is defined as the normal fluctuation area, and the area above the upper limit of the confidence interval is defined as the suspected fault area.

[0106] In a simple and practical implementation, the confidence interval is calculated using statistical methods, with the upper and lower boundaries determined based on the standard deviation of the residuals. The confidence level is set to 95%, meaning that under normal operating conditions, 95% of the vibration RMS data points will fall within the confidence interval. If the vibration RMS data point at a certain moment falls above the upper boundary of the confidence interval, it indicates that the vibration energy at that moment exceeds the expected range under normal load, potentially indicating a equipment malfunction.

[0107] Schematic, the formulas for calculating the upper and lower bounds of the confidence interval are as follows: y high To establish the upper bound, y low To establish the lower bound, y fit σ is the theoretical vibration value, σ is the standard deviation of the health data residuals, and 1.96 = 95% confidence critical coefficient.

[0108] In step 4, the residual deviation between the measured vibration value and the theoretical effective vibration value is calculated to determine whether there is an abnormality in the equipment. The specific fault type is then identified by combining this with a set of sensitive characteristic parameters, and the severity of the fault is assessed. Step 4 further includes the following steps:

[0109] Step 4-1: Substitute the real-time current into the load-vibration reference function, calculate the theoretical vibration value, and obtain the residual deviation between the theoretical vibration value and the measured vibration value.

[0110] Step 4-2: Compare the residual deviation with the preset threshold to determine whether the equipment is in a healthy state or a suspected fault state.

[0111] Step 4-3: Analyze and determine the specific fault type of the belt conveyor by combining the changing patterns of sensitive characteristic parameters.

[0112] Step 4-4: Based on the residual deviation and the temperature change rate, calculate and correct the quantitative index of fault severity.

[0113] In step 4-1, the current motor current value collected in real time is input into the load-vibration reference function constructed in step 3 to calculate the theoretical effective vibration value under the current value; then the effective vibration value collected in real time at the same time is retrieved for subsequent residual calculation.

[0114] Under normal operating conditions, the residual deviation should fluctuate within a small range. When equipment malfunctions, the fit of mechanical components changes, leading to an abnormal increase in vibration energy and a significant increase in residual deviation. Residual deviation analysis can effectively distinguish between normal vibration increases caused by load and abnormal vibration increases caused by malfunctions.

[0115] To illustrate, a set of real-time collected data is as follows:

[0116] Light load condition: Real-time acquisition of motor current 23.1A, real-time acquisition of measured effective vibration value 2.15mm / s;

[0117] Under medium load conditions: real-time acquisition of motor current of 37.2A and real-time acquisition of measured effective vibration value of 2.91mm / s;

[0118] Heavy load conditions: Real-time acquisition of motor current 43.8A, real-time acquisition of measured effective vibration value 3.72mm / s.

[0119] Load-vibration reference function: v 理 =f(I)=aI 2 +bI+c, where I is the real-time acquired drive motor current. a is the quadratic coefficient, representing the acceleration of vibration as the load increases, with a value ranging from 0.0001 to 0.0002; b is the linear coefficient, representing the linear rate of vibration as the load increases, with a value ranging from 0.02 to 0.03; c is the constant term, representing the foundation vibration level under no-load conditions, with a value ranging from 1.5 to 2.0; coefficients are calculated by fitting historical health data using the least squares method.

[0120] For the above real-time acquired data, the load-vibration reference function is:

[0121] v 理 =f(I)=aI 2 +bI+c=0.000118I 2 +0.0219I+1.7625.

[0122] Residual deviation: Δv = v 实 -v 理 v 实 This represents the measured effective value of the vibration.

[0123] v 理 =0.000118×23.1 2 +0.0219×23.1+1.7625=2.3314mm / s;

[0124] Light load condition: Δv=v 实 -v 理=2.15-2.3314=-0.1814mm / s;

[0125] Medium load condition: Δv = 0.1695 mm / s;

[0126] Heavy load condition: Δv = 0.7719 mm / s.

[0127] In step 4-2, the calculated residual deviation is compared with a preset deviation threshold. The deviation threshold is determined based on the upper bound of the 95% confidence interval of the load-vibration second-order benchmark fitting model and is used to define the maximum allowable deviation range of normal vibration.

[0128] If the residual deviation does not exceed the deviation threshold, it indicates that the current vibration deviation of the equipment is within a reasonable fluctuation range, and the vibration energy conforms to the normal operating characteristics under the same load. The equipment is judged to be in a healthy operating state, and the system continues to perform real-time data acquisition and monitoring.

[0129] If the residual deviation exceeds the deviation threshold, it indicates that the measured vibration energy is significantly higher than the theoretical expected value under the same load conditions, the vibration deviates abnormally from the healthy baseline, and the equipment is judged to have entered a suspected fault state.

[0130] For periods when a fault is suspected, the system further retrieves other monitoring indicators from the set of sensitive feature parameters for secondary analysis. Combining the changing patterns of multi-dimensional feature parameters, it identifies invalid anomalies such as instantaneous operating condition disturbances and data acquisition interference, determines whether the equipment has a real fault, and identifies the specific fault type.

[0131] To illustrate, we will use the data from the three different load conditions in step 4-1 above as an example.

[0132] The overall judgment rule is as follows: when the residual deviation is less than or equal to 0.15 mm / s, the vibration deviation of the equipment is within the normal fluctuation range, and the equipment is judged to be in a healthy operating state; when the residual deviation is greater than 0.15 mm / s, the measured vibration energy exceeds the normal expected range under the same load, and the equipment is judged to be in a suspected fault state.

[0133] The first group is under light load conditions, with a residual deviation of -0.1814 mm / s. Its absolute value slightly exceeds the deviation threshold of 0.15 mm / s, indicating that the equipment has entered a suspected fault state.

[0134] The second group represents the medium-load operating condition, with a corresponding residual deviation of 0.1695 mm / s. This value slightly exceeds the deviation threshold and deviates slightly from the normal fluctuation range corresponding to the 95% confidence interval, indicating that the equipment has entered a suspected fault state.

[0135] The third group represents the heavy-load operating condition. The calculated residual deviation is 0.7719 mm / s. This value significantly exceeds the preset deviation threshold, indicating that the current vibration energy of the equipment is significantly higher than the theoretical expected value under the same load and healthy operating condition. The vibration deviates severely from the normal range of the benchmark model, and the equipment is judged to have entered a suspected fault state.

[0136] In step 4-3, when the equipment is determined to be in a suspected fault state, the fault type is further identified by combining the kurtosis index, envelope spectrum energy value, transverse vibration to longitudinal vibration amplitude ratio, and temperature characteristics in the sensitive feature parameter set.

[0137] Different types of faults exhibit different combinations of characteristics. By analyzing these combinations of characteristics, we can accurately distinguish common fault types such as idler bearing wear, belt misalignment, and roller bearing failure.

[0138] Indicatively:

[0139] When the residual deviation exceeds the threshold and the kurtosis index increases significantly, and the high-frequency envelope spectrum energy value shows modulation characteristics, it is determined to be a roller bearing wear fault.

[0140] When idler roller bearings wear out, they will generate periodic impact vibrations, leading to an increase in kurtosis and the appearance of corresponding fault characteristic frequencies in the envelope spectrum.

[0141] When the residual deviation exceeds the threshold and low-order harmonic components of the rotational frequency appear in the vibration spectrum, and the ratio of the lateral vibration amplitude to the longitudinal vibration amplitude exceeds the preset deviation coefficient, it is determined to be a belt misalignment fault. When the belt misaligns, the lateral pressure of the belt on the roller or idler increases, resulting in a significant increase in lateral vibration energy. At the same time, it will cause uneven rotation of the roller, generating low-order harmonics of the rotational frequency.

[0142] When the residual deviation exceeds a threshold and the amplitude at the corresponding characteristic frequency of the roller bearing increases significantly, while the bearing temperature continues to rise, it is determined to be a roller bearing failure. The characteristic frequency of roller bearing failure is related to the structural parameters of the bearing and can be accurately identified through spectrum analysis.

[0143] In step 4-4, a fault severity quantification index is calculated based on the magnitude of the residual deviation. The larger the residual deviation, the more severe the equipment degradation and the higher the fault severity quantification index.

[0144] Preferably, the residual deviation is uniformly mapped to a standard scoring range of 0 to 100 to achieve quantitative standardization. A value of 0 represents that the equipment is completely healthy, with no abnormal vibrations or component deterioration; a value of 100 represents that the equipment malfunction is extremely serious, and component failure or shutdown is imminent.

[0145] Based on the 0.15mm / s abnormal threshold preset in step 4-2, the mapping benchmark is defined: when the residual deviation is ≤0.15mm / s, the equipment is in a healthy state and the basic fault severity is 0; using the maximum allowable residual during normal operation and the critical residual of the fault as the interval boundary, the residual deviation exceeding the threshold is linearly converted to the 0~100 interval.

[0146] To address the issue of one-sided assessment using a single vibration residual, a more preferred embodiment of the present invention introduces a temperature change rate to dynamically correct the severity of the basic fault.

[0147] During equipment failure and deterioration, component wear, jamming, and lubrication failure can exacerbate frictional heat generation. The faster the temperature rises, the faster the failure progresses and the more rapidly the potential problem worsens. A fixed temperature rise threshold is preset, and the short-term temperature change rate of the bearing is calculated in real time. If the temperature change rate does not exceed the threshold, the failure is progressing smoothly and no correction is needed. If the temperature change rate exceeds the threshold, the larger the temperature change rate, the larger the corresponding correction coefficient, and the greater the improvement in failure severity, thus reflecting the dynamic development trend of the failure.

[0148] In an illustrative implementation, the normal critical threshold is 0.15 mm / s, the fault saturation threshold is 0.80 mm / s, and the effective quantization interval width of the true anomaly = maximum fault residual - normal critical residual = 0.80 - 0.15 = 0.65 mm / s. An illustrative linear mapping calculation formula is as follows: In the formula S base Based on the severity of the basic fault.

[0149] It should be noted that the normal critical threshold of 0.15 mm / s and the fault saturation threshold of 0.80 mm / s are conservative and stringent requirements. In other application scenarios, the normal critical threshold and fault saturation threshold can be adjusted according to the circumstances.

[0150] For example, in step 4-1, the first group is under light load conditions, with an absolute value of 0.1814 mm / s for the residual deviation, slightly exceeding the normal critical threshold of 0.15 mm / s, indicating that the equipment is in a state of suspected minor fault. Substituting this into the linear mapping formula, the basic fault severity is calculated as: S base ≈4.8, under this operating condition, the bearing temperature is stable with no significant temperature rise, the temperature change rate is 0, the correction factor is 1.0, and the final failure severity score is approximately 4.8 points. base The first group is within the normal range of 0-20. The second group is under medium load conditions, with a residual deviation of 0.1695 mm / s, slightly exceeding the abnormal threshold, indicating a slight vibration anomaly. Based on linear mapping calculations, S... base≈3, the bearing temperature rises steadily, the temperature change rate does not exceed the threshold, the correction coefficient is 1.0, and the final fault severity score is approximately 3.0. The equipment is in normal condition, with vibration slightly deviating from the healthy baseline, and no substantial potential fault. The third group is a heavy-load condition, with a residual deviation of 0.7719 mm / s, significantly exceeding the abnormal threshold, and the vibration deviates greatly from the baseline. Through linear mapping calculation, S... base The bearing temperature is approximately 95.7, and the temperature change rate exceeds the preset temperature rise threshold. The correction factor is 1.15. After correction, the final fault severity is approximately 110.0 points (recorded as the upper limit of 100 points), which is judged as a high-risk fault state.

[0151] In a more preferred embodiment, a preset temperature rise threshold of 1°C / min is set; if this threshold is exceeded, a correction is initiated. For every 0.5°C / min increase in the temperature change rate, the correction factor increases by 0.05. That is, the final fault severity is: S = S base ×k, where: k is the temperature correction coefficient (k≥1, preferably 1.0~1.5); k=1 when the temperature rise is ≤1℃ / min; k increases by 0.05 for every 0.5℃ / min increase in temperature rise.

[0152] Step 5 generates corresponding graded early warning information based on the fault type determination result and fault severity quantification index obtained in step 4.

[0153] Based on the fault type determination results and the fault severity quantification indicators, the early warning level is divided into three levels: attention, early warning, and alarm.

[0154] Unrestricted:

[0155] S base <20: The equipment is in normal condition;

[0156] 20≤S base <40 indicates a relatively minor fault; the equipment can still operate normally, but it needs to be closely monitored during subsequent inspections.

[0157] 40≤S base <70, at this point the fault characteristics are obvious and there is a tendency for it to expand further, requiring on-site inspection and handling within the specified time;

[0158] S base ≥70 indicates a basic fault severity of 70 or higher, posing a significant safety hazard to the equipment and requiring immediate shutdown for inspection.

[0159] like Figure 5As shown, another embodiment of the present invention provides a fault early warning system for thermal power belt conveyors based on vibration load reference correlation. The system consists of a multi-dimensional data acquisition and steady-state dataset construction unit, a fault-sensitive feature extraction and feature set construction unit, a load vibration correlation modeling and normal operation area delineation unit, an anomaly identification, fault diagnosis and severity quantification unit, and a graded early warning information generation and release unit.

[0160] The multidimensional data acquisition and steady-state dataset construction unit is used to collect multidimensional operating data of the belt conveyor and construct a steady-state dataset. This unit acquires data using sensors deployed at key locations such as the drive roller bearing housing, the redirecting roller bearing housing, and the idler roller supports in critical sections of the belt conveyor. After collecting the raw data, noise reduction is performed. Then, based on the vibration signal acquisition time, linear interpolation is performed on data from different sampling frequencies to achieve time alignment. Finally, distorted data exceeding the physical range is removed. This results in a standard dataset containing only steady-state operating data.

[0161] The fault-sensitive feature extraction and feature set construction unit, based on the standard dataset of the multi-dimensional data acquisition and steady-state dataset construction unit, completes the extraction of fault-sensitive features and the construction of a feature set. First, the unit calculates time-domain statistical features of the vibration data, including the effective value of the vibration velocity signal, kurtosis index, and peak value index, reflecting the equipment's operating status from the dimensions of vibration energy, impact sharpness, and instantaneous impact intensity. Then, it performs frequency-domain analysis using Fast Fourier Transform, converting the time-domain signal into a frequency-domain signal, extracting frequency-domain features such as the frequency conversion amplitude and high-frequency envelope spectrum energy value to locate the specific component where the fault occurs. Simultaneously, the unit calculates temperature features of the temperature data, including absolute temperature value, temperature change rate, and temperature difference between different idlers within the same idler group, reflecting potential equipment failure risks through temperature changes.

[0162] After completing various feature calculations, the fault-sensitive feature extraction and feature set construction unit integrates vibration time-domain, frequency-domain, and temperature features to construct a feature vector matrix. All feature values ​​are normalized, and feature parameters are then selected based on the statistical results of historical fault data. The selection criteria are that the features have significant differences in value between normal and fault states, can stably reflect changes in health status, have small fluctuations in value during normal operation across the full load range, and have obvious responses to common faults. Finally, redundant features are eliminated to form a set of fault-sensitive feature parameters.

[0163] The load vibration correlation modeling and normal operation area delineation unit completes the construction of the load-vibration function relationship and the delineation of the normal operation area based on the healthy historical steady-state data.

[0164] The fault-sensitive feature extraction and feature set construction unit first extracts steady-state data from the historical operation database showing no fault records for three consecutive months following the equipment's most recent comprehensive overhaul. This data forms a historical health dataset covering all load levels from no-load to full-load, with at least 10,000 samples. The data undergoes the same preprocessing as the steady-state data construction unit. Then, using the drive motor current as the load characterization parameter and the vibration effective value as the vibration response parameter, a scatter plot showing their correlation is plotted. Second-order polynomial regression analysis is performed on the scatter data using the least squares method to fit a baseline correlation curve model between load and vibration. During the fitting process, outlier data points deviating from the overall trend are removed to ensure model accuracy. After model fitting, the fault-sensitive feature extraction and feature set construction unit uses statistical methods to calculate the 95% confidence interval of the baseline model. The upper and lower boundaries are determined based on the residual standard deviation. The area within the confidence interval is defined as the normal vibration fluctuation zone, and the area above the upper boundary of the interval is defined as the suspected fault zone, thus clarifying the reasonable vibration fluctuation range of the equipment.

[0165] The anomaly identification, fault diagnosis, and severity quantification unit completes anomaly determination, fault identification, and severity quantification based on the benchmark correlation curve model of load and vibration and sensitive feature parameters.

[0166] The anomaly identification, fault diagnosis, and severity quantification unit first substitutes the real-time motor current value into the benchmark correlation curve model based on load and vibration, calculates the theoretical effective value of vibration for the corresponding load, retrieves the measured effective value of vibration at the same time and calculates the residual deviation, compares the residual deviation with the preset deviation threshold to determine the equipment status. If the residual does not exceed the threshold, it is in a healthy state; if it exceeds the threshold, it enters a suspected fault state.

[0167] Once the suspected state is entered, the anomaly identification, fault diagnosis, and severity quantification unit performs a secondary analysis based on sensitive feature parameters. The fault type is identified according to the feature combination rules: if the residual exceeds the standard and the kurtosis increases, and the high-frequency envelope spectrum energy shows modulation characteristics, it is determined to be a roller bearing wear fault; if the residual exceeds the standard and the spectrum shows low-order harmonics, and the ratio of lateral to longitudinal vibration amplitude meets the standard value, it is determined to be a belt misalignment fault; if the residual exceeds the standard and the characteristic frequency amplitude of the roller bearing increases, and the bearing temperature continues to rise, it is determined to be a roller bearing fault.

[0168] After the fault type is identified, the anomaly identification, fault diagnosis and severity quantification unit first maps the residual deviation to the standard range of 0 to 100 to calculate the basic fault severity, and then introduces dynamic correction of temperature change rate to finally obtain the corrected fault severity quantification index, thus completing the assessment from anomaly judgment to severity.

[0169] The graded early warning information generation and release unit generates and releases graded early warning information based on the fault type and fault severity quantification index output by the anomaly identification, fault diagnosis and severity quantification unit.

[0170] The multidimensional data acquisition and steady-state dataset construction unit takes as input the raw data collected in real time by sensors deployed at various measuring points on the belt conveyor, and outputs a steady-state standard dataset. This steady-state standard dataset is transmitted to two units simultaneously: one is the fault-sensitive feature extraction and feature set construction unit, and the other is the load vibration correlation modeling and normal operation area delineation unit.

[0171] The input to the fault-sensitive feature extraction and feature set construction unit is the steady-state standard dataset output by the multi-dimensional data acquisition and steady-state dataset construction unit. Based on the input data, vibration time-domain features, vibration frequency-domain features, and temperature features are extracted. After normalization and redundant feature removal, the fault-sensitive feature parameter set is output to the anomaly identification, fault diagnosis, and severity quantification unit.

[0172] The load-vibration correlation modeling and normal operation area delineation unit obtains the load-vibration benchmark correlation model through data fitting, calculates the 95% confidence interval and delineates the normal fluctuation area and the suspected fault area, and outputs the load-vibration benchmark correlation model, vibration deviation threshold, and normal operation area boundary, which is then unidirectionally transmitted to the anomaly identification, fault diagnosis and severity quantification unit.

[0173] The anomaly identification, fault diagnosis, and severity quantification unit inputs three types of data: steady-state standard data output in real time by the multi-dimensional data acquisition and steady-state dataset construction unit; a set of fault-sensitive feature parameters output by the fault-sensitive feature extraction and feature set construction unit; and a benchmark model and judgment threshold output by the load vibration correlation modeling and normal operation area delineation unit. After residual calculation, feature matching, and severity correction, this unit outputs the fault type judgment result and a fault severity quantification index in the 0-100 range. These two results are transmitted to the graded early warning information generation and release unit.

[0174] The graded early warning information generation and release unit receives the fault type and fault severity quantification indicators output by the anomaly identification, fault diagnosis and severity quantification unit; this unit divides the fault into three levels: attention, early warning and alarm according to the severity and generates corresponding prompts, and finally outputs graded early warning information.

[0175] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A thermal power belt conveyor fault early warning method, characterized in that, Includes the following steps: S1: Deploy sensors to collect vibration data, temperature data, and current data. After noise reduction and time alignment, remove transient start-stop data to form a standard dataset for steady-state operation of the belt conveyor. S2: Extract vibration and temperature features from the standard dataset, integrate and normalize them, and then filter out the features to construct a set of fault-sensitive feature parameters; S3: Fit the load-vibration correlation curve based on historical health data, calculate the confidence interval, and divide the normal fluctuation area of ​​equipment vibration; S4: Calculate the residual deviation between measured and theoretical vibration, determine abnormal conditions, identify fault types, and calculate fault severity index; S5: Classify the warning level according to the fault type and severity index, create corresponding prompts, and complete the warning information release; Step S3 further includes: Step 3-1: Extract historical steady-state data of equipment health periods, preprocess it, and construct a historical health dataset covering the full load. Step 3-2: Using the motor current as the load parameter and the effective value of vibration as the response parameter, plot a scatter plot showing the correlation between the two. Step 3-3: Perform regression analysis on the scatter data to fit the baseline correlation curve model between load and vibration; Steps 3-4: Calculate the confidence interval of the baseline model and divide the equipment vibration normal fluctuation zone and the suspected fault zone.

2. The method according to claim 1, characterized in that, The sensors include a triaxial accelerometer, a contact thermocouple, an ambient temperature sensor, and a current transformer.

3. The method according to claim 2, characterized in that, The historical health dataset includes operating data of the belt conveyor under different load conditions, from no load to full load.

4. The method for early warning of faults in thermal power plant conveyor belts according to claim 3, characterized in that, The horizontal axis of the scatter plot represents the motor current value, and the vertical axis represents the effective vibration value at the corresponding moment. Each data point represents the combination of current and vibration at a given moment.

5. A fault early warning method for thermal power plant conveyor belts according to claim 4, characterized in that, In steps 3-4, the confidence interval of the load-vibration reference function is calculated. The area within the confidence interval is defined as the normal fluctuation area, and the area above the upper limit of the confidence interval is defined as the suspected fault area.

6. A fault early warning method for thermal power plant conveyor belts according to claim 5, characterized in that, Step S4 includes the following steps: Step 4-1: Substitute the real-time current into the load-vibration reference function, calculate the theoretical vibration value, and obtain the residual deviation between the measured vibration and the actual vibration. Step 4-2: Compare the residual deviation with the preset threshold to determine whether the equipment is in a healthy state or a suspected fault state; Step 4-3: Analyze and determine the specific fault type of the belt conveyor by combining the changing patterns of sensitive characteristic parameters; Step 4-4: Based on the residual deviation and the temperature change rate, calculate and correct the quantitative index of fault severity.

7. A method for early warning of faults in thermal power plant conveyor belts according to claim 6, characterized in that, The sensitive characteristic parameters include kurtosis index, envelope spectrum energy value, ratio of transverse vibration to longitudinal vibration amplitude, and temperature characteristics.

8. A fault early warning system for thermal power plant belt conveyors, characterized in that, include: The system includes: a multi-dimensional data acquisition and steady-state dataset construction unit, a fault-sensitive feature extraction and feature set construction unit, a load vibration correlation modeling and normal operation zone delineation unit, an anomaly identification, fault diagnosis and severity quantification unit, and a graded early warning information generation and dissemination unit. The input of the multidimensional data acquisition and steady-state dataset construction unit is the raw data collected in real time by sensors deployed at various measuring points of the belt conveyor, and the output is a steady-state standard dataset to the fault-sensitive feature extraction and feature set construction unit and the load vibration correlation modeling and normal operation area delineation unit. The fault-sensitive feature extraction and feature set construction unit extracts vibration time-domain features, vibration frequency-domain features, and temperature features based on the input standard dataset. After normalization and redundant feature removal, it outputs a set of fault-sensitive feature parameters to the anomaly identification, fault diagnosis, and severity quantification unit. The load-vibration correlation modeling and normal operation zone delineation unit obtains the load-vibration benchmark correlation model through data fitting, calculates the confidence interval and delineates the normal fluctuation zone and the suspected fault zone, and outputs the load-vibration benchmark correlation model, vibration deviation threshold and normal operation zone boundary to the anomaly identification, fault diagnosis and severity quantification unit. The anomaly identification, fault diagnosis, and severity quantification unit receives steady-state standard data output in real time from the multi-dimensional data acquisition and steady-state dataset construction unit, a set of fault-sensitive feature parameters output from the fault-sensitive feature extraction and feature set construction unit, and a benchmark model and judgment threshold output from the load vibration correlation modeling and normal operation area delineation unit. After residual calculation, feature matching, and severity correction, it outputs the fault type judgment result and the fault severity index in the 0-100 range to the graded early warning information generation and release unit. The graded early warning information generation and release unit classifies fault levels and generates corresponding prompts.

9. A fault early warning system for thermal power plant conveyor belts according to claim 8, characterized in that: The load vibration correlation modeling and normal operation area delineation unit, based on historical steady-state health data, completes the construction of the load-vibration function relationship and the delineation of the normal operation area. The fault-sensitive feature extraction and feature set construction unit first extracts the most recent steady-state data of the equipment from the historical operation database to construct a historical health dataset covering the entire load from no-load to full-load. Then, using the drive motor current value as the load characterization parameter and the vibration effective value as the vibration response parameter, a scatter plot of the correlation between the two is drawn. The least squares method is used to perform second-order polynomial regression analysis on the scatter data to fit the benchmark correlation curve model of load and vibration. After completing the model fitting, the fault-sensitive feature extraction and feature set construction unit calculates the confidence interval of the benchmark model and divides the equipment vibration normal fluctuation area and the fault-suspected area.

10. A fault early warning system for thermal power plant conveyor belts according to claim 9, characterized in that, The failure severity index S = S base × k, where k is a temperature correction coefficient, S base is a basic failure severity, S base = (|residual deviation amount| - normal critical residual) / effective quantization interval width, effective quantization interval width = maximum failure residual - normal critical residual.

Images