A bolt axial force health state comprehensive monitoring system
Patent Information
- Application Number
- CN202512025471.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-05-26
Smart Images
Figure CN122084183A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of bolt monitoring technology, and in particular to a comprehensive monitoring system for the health status of bolt axial force. Background Technology
[0002] High-strength bolts are core connecting components of wind turbine units. Their connection reliability directly determines the safety, stability and service life of the unit. Bolts are affected by multiple factors such as alternating loads, temperature changes, vibration and shock, and environmental corrosion, and are prone to problems such as preload loss, fatigue damage or even fracture failure.
[0003] Traditional bolt axial force monitoring methods mostly rely on manual periodic inspections and online monitoring. They are often designed for single types of bolts or specific scenarios, and their monitoring accuracy and stability are insufficient under complex environmental interference. The accuracy of condition monitoring is also low, and problems are often only discovered after a failure occurs, leading to serious safety hazards and economic losses. They cannot meet the needs of large equipment for refined and intelligent monitoring of bolt health status. Summary of the Invention
[0004] To address the aforementioned technical issues, this application provides a comprehensive monitoring system for the health status of bolt axial forces. By setting multiple bolt categories and constructing separate state analysis models, it enables accurate classification of bolts of different types and under different scenarios, and the formulation of targeted monitoring strategies. By combining piezoelectric ultrasonic sensing units and electromagnetic ultrasonic enhanced sensing units, it improves the acquisition quality and accuracy of axial force-related data under complex working conditions, obtains real-time status results, and fine-tunes the real-time status results based on predicted status levels for future periods, further improving the timeliness and reliability of early warnings, and realizing intelligent comprehensive monitoring and management of bolt axial force health status.
[0005] In some embodiments of this application, a comprehensive monitoring system for the health status of bolt axial force is provided, including: The module is used to define multiple bolt categories and build a state analysis model for each bolt category. The monitoring module is used to set the monitoring strategy for the bolt to be monitored based on the condition analysis model, obtain axial force related data according to the monitoring strategy, analyze it, and obtain real-time condition results. The early warning module is used to generate predicted status levels for future periods and fine-tune the real-time status results. Based on the fine-tuned real-time status results, it determines whether to generate an alarm command.
[0006] In some embodiments of this application, the monitoring module further includes: A piezoelectric ultrasonic sensing unit is used to monitor the bolt to be monitored in real time according to the monitoring strategy. Electromagnetic ultrasound enhancement sensing unit, including a customized probe and signal enhancement algorithm; The customized probe is designed based on the structural characteristics of different bolts to generate a multi-channel sensor array for signal acquisition.
[0007] In some embodiments of this application, a state analysis model is constructed for each bolt category, including: Obtain the basic parameters and scenario parameters of the bolt, and set several basic and scenario indicators based on the corresponding monitoring impact. Several bolt categories are obtained by randomly combining basic and scenario indicators. Construct historical monitoring data packages for each bolt category; Generate historical status risk values for the corresponding bolt categories based on historical monitoring data packets; Based on historical risk values, several monitoring areas and corresponding monitoring time intervals are set for the corresponding bolt categories. Each monitoring area includes several monitoring points. Generate training data packets for each monitoring area of each bolt category, and construct a state model for each bolt category; A state analysis model for each bolt category is generated based on the state model for each bolt category and the corresponding monitoring time interval.
[0008] In some embodiments of this application, a state model is constructed for each bolt category, including: For each bolt category, historical axial force-related data in the historical monitoring data package is mapped to the monitoring area to obtain the initial training data package for each monitoring area. Preprocessing is performed on the initial training data packets of the same monitoring area to obtain training data packets, which include several historical data sequences. The preprocessing includes data cleaning, smoothing, and sequence construction. Generate an axial force correlation dataset for the corresponding monitoring area from historical axial force correlation data at the same historical time point in different historical data sequences; Generate several axial force-related datasets for each monitoring area; A dataset-axial force value mapping table is pre-built, and the historical actual axial force value of each axial force-related dataset is generated based on the dataset-axial force value mapping table; Historical state coefficients are generated based on actual historical axial force values; Using each axial force-related dataset of each monitoring area for the same bolt category as training input data and the corresponding historical state coefficients as training output data, a neural network is trained to obtain the state model for the corresponding bolt category.
[0009] In some embodiments of this application, axial force-related data are obtained and analyzed according to a monitoring strategy to obtain real-time status results, including: The parameters of the bolt to be monitored and the real-time scene parameters are obtained and compared with each bolt category to obtain several contrast parameters. The real-time bolt category is determined based on contrast, and a monitoring strategy is generated based on several monitoring areas of the real-time bolt category and the corresponding monitoring time interval. According to the monitoring strategy, real-time axial force related data of several monitoring points in each monitoring area are obtained. The real-time axial force related data includes ultrasonic time, relative change of time, rate of change, sound velocity, amplitude attenuation rate, and strain value. Input the real-time axial force related data into the real-time bolt category state model to obtain the real-time state coefficient of each monitoring point; Real-time status results are generated based on the real-time status coefficients of all monitoring points.
[0010] In some embodiments of this application, generating real-time status results based on the real-time status coefficients of all monitoring points includes: Generate a set of associated points for each monitoring point; The state probability of the real-time state coefficient of the corresponding monitoring point is generated based on the set of associated points; Determine whether to correct the real-time state coefficient of the corresponding monitoring point based on the state probability. If so, generate the corrected real-time state coefficient based on the set of associated points and update it. Update the real-time status coefficient of each monitoring point sequentially; The average value of the real-time state coefficients of all monitoring points in the same monitoring area is obtained by averaging the updated real-time state coefficients of the corresponding monitoring area. The average state coefficient of all monitoring areas of the bolt to be monitored is weighted to obtain the comprehensive state coefficient. The state level and the corresponding real-time state results are determined based on the comprehensive state coefficient.
[0011] In some embodiments of this application, determining the state level and the corresponding real-time state result based on the comprehensive state coefficient includes: Several preset comprehensive state coefficient ranges are pre-defined, and each preset comprehensive state coefficient range is mapped to a corresponding preset state level. Based on the correspondence between the comprehensive state coefficient and the preset comprehensive state coefficient range, the corresponding preset state level is selected as the current state level. The preset state level includes a first preset state level, a second preset state level, and a third preset state level. When the current status level is the first preset status level, the corresponding real-time status result is that danger exists; When the current status level is the second preset status level, the corresponding real-time status result is that an alert is required; When the current status level is the third preset status level, the corresponding real-time status result is healthy.
[0012] In some embodiments of this application, generating a predicted state level for a future time period includes: Generate several real-time data sequences for each monitoring point in the current monitoring period; Trend features are extracted from real-time data sequences to obtain their time-domain and frequency-domain features. Similarity analysis is then performed with the historical trend feature database of real-time bolt categories to determine the trend type of each real-time data sequence. The historical trend feature library includes several preset trend types, and each preset trend type is mapped to a corresponding preset time domain feature and a corresponding preset frequency domain feature. Based on the preset time series prediction model corresponding to the trend type, the model outputs the predicted data sequence for the corresponding real-time data sequence in the future period. Several predicted axial force related datasets are generated based on all predicted data sequences of the same monitoring point and input into the state model to obtain several predicted state coefficients of the corresponding monitoring point in the future period. Generate several comprehensive forecast state coefficients for future time periods and determine several corresponding forecast state levels.
[0013] In some embodiments of this application, fine-tuning of the real-time status results includes: The distribution of predicted state levels is determined based on several predicted state levels for future time periods. Collect the frequency of occurrence and duration of each predicted state level in the future period, and calculate the early warning index; Pre-set early warning index thresholds; When the warning index is less than the warning index threshold, no fine-tuning is performed on the real-time status results. When the warning index is not less than the warning index threshold, the real-time status result is downgraded by one level.
[0014] In some embodiments of this application, determining whether to generate an alarm command based on the fine-tuned real-time status result includes: When the real-time status result after fine-tuning indicates that there is a danger, an alarm command is generated; When the real-time status result after fine-tuning requires an alert, an alert command is generated. No warning command is generated when the real-time status result after fine-tuning is healthy.
[0015] The comprehensive monitoring system for bolt axial force health status according to an embodiment of this application has the following advantages compared with the prior art: By setting multiple bolt categories and constructing separate state analysis models, it is possible to accurately classify bolts of different types and in different scenarios and formulate targeted monitoring strategies. By combining piezoelectric ultrasonic sensing units and electromagnetic ultrasonic enhanced sensing units, the acquisition quality and state analysis accuracy of axial force-related data under complex working conditions are improved, real-time state results are obtained, and the real-time state results are fine-tuned by combining the predicted state level for future periods, further improving the timeliness and reliability of early warning, and realizing intelligent comprehensive monitoring and management of bolt axial force health status. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of a comprehensive monitoring system for the health status of bolt axial force in an embodiment of this application. Detailed Implementation
[0017] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.
[0018] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.
[0019] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.
[0020] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.
[0021] like Figure 1 As shown in the figure, a comprehensive monitoring system for the health status of bolt axial force according to an embodiment of this application includes: The module is used to define multiple bolt categories and build a state analysis model for each bolt category. The monitoring module is used to set the monitoring strategy for the bolt to be monitored based on the condition analysis model, obtain axial force related data according to the monitoring strategy, analyze it, and obtain real-time condition results. The early warning module is used to generate predicted status levels for future periods and fine-tune the real-time status results. Based on the fine-tuned real-time status results, it determines whether to generate an alarm command.
[0022] In some embodiments of this application, the monitoring module further includes: A piezoelectric ultrasonic sensing unit is used to monitor the bolt to be monitored in real time according to the monitoring strategy. Electromagnetic ultrasound enhancement sensing unit, including a customized probe and signal enhancement algorithm; The customized probe is designed according to the structural characteristics of different bolts and is used to generate a multi-channel sensor array for signal acquisition.
[0023] In this embodiment, the piezoelectric ultrasonic sensing unit includes a piezoelectric crystal, which converts electrical signals into ultrasonic waves through the inverse piezoelectric effect and converts the ultrasonic echo reflected by the bolt into electrical signals through the direct piezoelectric effect, thereby realizing the direct sensing of changes in the axial stress of the bolt.
[0024] In this embodiment, the customized probe of the electromagnetic ultrasonic enhancement sensing unit is based on the principle of electromagnetic induction. It is designed with an arc-shaped array of coils and permanent magnets to address the different structural features of common wind turbine bolts such as hexagonal head bolts and flange bolts.
[0025] In this embodiment, the signal enhancement algorithm specifically adopts an adaptive filtering algorithm based on wavelet threshold denoising and Kalman filtering. First, the original sensing signal is decomposed into different frequency bands through wavelet decomposition, and the noise coefficient is thresholded to suppress high-frequency environmental vibration noise. Then, Kalman filtering is used to dynamically track and predict the filtered signal, further eliminating temperature drift and electromagnetic interference, improving the signal-to-noise ratio by at least 15dB, and ensuring the stability of axial force related data acquisition under complex working conditions.
[0026] In some embodiments of this application, a state analysis model is constructed for each bolt category, including: Obtain the basic parameters and scenario parameters of the bolt, and set several basic and scenario indicators based on the corresponding monitoring impact. Several bolt categories are obtained by randomly combining basic and scenario indicators. Construct historical monitoring data packages for each bolt category; Generate historical status risk values for the corresponding bolt categories based on historical monitoring data packets; Based on historical risk values, several monitoring areas and corresponding monitoring time intervals are set for the corresponding bolt categories. Each monitoring area includes several monitoring points. Generate training data packets for each monitoring area of each bolt category, and construct a state model for each bolt category; A state analysis model for each bolt category is generated based on the state model for each bolt category and the corresponding monitoring time interval.
[0027] In this embodiment, the basic parameters of a bolt refer to the set of parameters inherent to the bolt itself, used to characterize its physical properties and structural features. These parameters include, but are not limited to, the bolt's nominal diameter, pitch, length, yield strength, tensile strength, thread type, and head type (such as hexagonal head, cylindrical head, countersunk head, etc.). The degree of monitoring influence refers to the sensitivity analysis of each parameter to the axial force monitoring results. The greater the sensitivity to the axial force monitoring results, the greater the degree of monitoring influence, and vice versa.
[0028] In this embodiment, a quantified weight value is set for each parameter according to the degree of monitoring impact, and basic parameters with a weight coefficient greater than a preset threshold (0.35 in this application) are set as basic indicators, and the basic indicators include several preset basic parameter ranges.
[0029] In this embodiment, the scenario parameters refer to external parameters related to the working environment of the bolt, such as ambient temperature, humidity, vibration frequency, concentration of corrosive medium, load type and amplitude, and standard value of installation torque and preload. The monitoring influence of scenario parameters refers to the correlation analysis between scenario parameters and axial force changes in historical data. The greater the correlation, the greater the monitoring influence, and vice versa.
[0030] In this embodiment, scenario parameters that have a significant impact on the axial force monitoring results (also using a weighting coefficient greater than the preset threshold of 0.35) are selected as scenario indicators, and several preset scenario parameter ranges are divided for each scenario indicator. For example, the ambient temperature can be divided into ranges such as -40℃~-20℃, -20℃~50℃, 50℃~100℃, and 100℃~150℃.
[0031] In this embodiment, the classification standard refers to the combination of a preset basic parameter range for each basic indicator and a preset scenario parameter range for each scenario indicator. The combination of these parameter ranges constitutes a specific bolt category. Through this multi-dimensional index combination classification, several bolt categories with different combinations of bolt characteristics and working environments can be covered, ensuring that the state analysis model can be accurately adapted to the characteristics of the corresponding bolt category.
[0032] In this embodiment, several historical state coefficients of multiple preset positions for each bolt category in a historical period are obtained, and the difference is calculated with the preset state coefficient threshold to obtain several historical state coefficient differences for the same preset position. Based on the number of historical state coefficient differences of the same preset position in a historical period that are less than 0, the absolute value of the values less than 0, the frequency of occurrence, and the duration, a sub-state risk value for the corresponding preset position is generated. Based on the sub-state risk values at all preset positions and the weight coefficient of the corresponding preset position, a historical state risk value for the bolt category is generated. Specifically, the more historical state coefficient differences that are less than 0, the larger the absolute value of the values less than 0, the higher the frequency of occurrence, and the longer the duration, the higher the corresponding sub-state risk value, i.e., the higher the historical state risk value; conversely, the lower the historical state risk value. The preset positions are key monitoring points pre-set based on the structural characteristics of the bolt and the stress concentration area, such as the middle of the bolt shank, the thread engagement section, and the transition fillet between the head and the shank. The weight coefficients of different preset positions are determined based on their contribution to the force and the failure risk level during the bolt's load-bearing process. Positions with a greater contribution to the force and a higher failure risk level have a larger weight coefficient.
[0033] In this embodiment, when the historical state risk value is higher, more monitoring areas are set and the monitoring time interval within the monitoring areas is shorter; conversely, fewer monitoring areas are set and the monitoring time interval is longer. The monitoring areas cover the key stress-bearing parts of the bolts and areas prone to stress concentration.
[0034] In this embodiment, by determining several bolt categories and constructing corresponding state models, and combining the monitoring time interval, a state analysis model for each bolt category is obtained. This lays a data foundation for the monitoring accuracy and state assessment of the bolts to be monitored, and enables differentiated and precise monitoring strategy configuration and state assessment for different types of bolts.
[0035] In some embodiments of this application, a state model is constructed for each bolt category, including: For each bolt category, historical axial force-related data in the historical monitoring data package is mapped to the monitoring area to obtain the initial training data package for each monitoring area. Preprocessing is performed on the initial training data packets of the same monitoring area to obtain training data packets, which include several historical data sequences. The preprocessing includes data cleaning, smoothing, and sequence construction. Generate an axial force correlation dataset for the corresponding monitoring area from historical axial force correlation data at the same historical time point in different historical data sequences; Generate several axial force-related datasets for each monitoring area; A dataset-axial force value mapping table is pre-built, and the historical actual axial force value of each axial force-related dataset is generated based on the dataset-axial force value mapping table; Historical state coefficients are generated based on actual historical axial force values; Using each axial force-related dataset of each monitoring area for the same bolt category as training input data and the corresponding historical state coefficients as training output data, a neural network is trained to obtain the state model for the corresponding bolt category.
[0036] In this embodiment, the dataset-axial force value mapping table is constructed based on bolt axial force calibration experimental data. For different bolt types, their basic parameters and scenario parameters are simulated in a laboratory environment. Known axial force values are applied to the bolts using axial force calibration equipment, and axial force-related data are collected simultaneously, including ultrasonic time, relative change in time, rate of change, sound velocity, amplitude attenuation rate, and strain value. These known axial force values and corresponding axial force-related data are used to construct sample pairs. Through statistical analysis, feature engineering, and training and optimization of machine learning algorithms (such as neural networks and support vector machines) on massive sample pairs, a mapping relationship between input features (i.e., the axial force-related dataset) and output labels (i.e., the actual axial force value) is established—the dataset-axial force value mapping table. This mapping table can accurately deduce the corresponding actual axial force value of the bolt based on the input axial force-related dataset, providing a crucial conversion basis for the subsequent rapid and accurate calculation of historical actual axial force values based on monitoring data.
[0037] In this embodiment, the historical actual axial force value is compared with the preset baseline value (referring to the axial force value in the initial tightening state). Based on the comparison result and the axial force safety threshold range [minimum safety threshold, maximum safety threshold] determined according to historical data for the corresponding bolt type, the historical state coefficient is obtained. Specifically, when the historical actual axial force value is less than the preset baseline value, the historical state coefficient = 1 - (preset baseline value - historical actual axial force value) / (preset baseline value - minimum safety threshold); when the historical actual axial force value is greater than the preset baseline value, the historical state coefficient = 1 - (historical actual axial force value - preset baseline value) / (maximum safety threshold - preset baseline value); and when the historical actual axial force value is at the preset baseline value, the historical state coefficient is 1.
[0038] In this embodiment, the state model is a mathematical model constructed using a neural network algorithm. The neural network adopts a multi-layer perceptron architecture, with the number of neurons in the input layer matching the feature dimensions of the axial force-related dataset. Three to five hidden layers are set, and the number of neurons in each layer is optimized using a grid search method. The output layer consists of a single neuron used to output state coefficients. During training, mean squared error is used as the loss function, and the Adam optimizer is used for iterative parameter updates. Early stopping is employed to prevent overfitting. Historical monitoring data packets are divided into training and validation sets in a 7:3 ratio. The training set is used for model parameter learning, while the validation set is used for real-time monitoring of the model's generalization ability. Training stops when the validation set loss no longer decreases for a preset number of rounds (e.g., 10 rounds). Finally, the model with the smallest validation set loss is saved as the state model for this bolt category.
[0039] In this embodiment, the state model constructed in the above manner can fully learn the nonlinear mapping relationship between axial force-related data and state coefficients in different monitoring areas, providing accurate model support for the subsequent real-time state result analysis of the bolts to be monitored.
[0040] In some embodiments of this application, axial force-related data are obtained and analyzed according to a monitoring strategy to obtain real-time status results, including: The parameters of the bolt to be monitored and the real-time scene parameters are obtained and compared with each bolt category to obtain several contrast parameters. The real-time bolt category is determined based on contrast, and a monitoring strategy is generated based on several monitoring areas of the real-time bolt category and the corresponding monitoring time interval. According to the monitoring strategy, real-time axial force related data of several monitoring points in each monitoring area are obtained. The real-time axial force related data includes ultrasonic time, relative change of time, rate of change, sound velocity, amplitude attenuation rate, and strain value. Input the real-time axial force related data into the real-time bolt category state model to obtain the real-time state coefficient of each monitoring point; Real-time status results are generated based on the real-time status coefficients of all monitoring points.
[0041] In this embodiment, contrast refers to the quantified value of the consistency between the bolt's own parameters and real-time scene parameters and the preset basic indicators and scene indicators in each bolt category. The consistency analysis is performed between the bolt's own parameters and the corresponding preset basic parameter range in the basic indicators of each bolt category to obtain a quantified consistency value. If it falls within the corresponding preset basic parameter range, the quantified value is 1; if it does not fall within the corresponding preset basic parameter range and the smaller the deviation, the closer the quantified value is to 1; conversely, the quantified value is closer to 0. The consistency quantification between the real-time scene parameters and the corresponding preset scene parameter range in the scene indicators of each bolt category is performed in the same way as above, and will not be repeated here. Based on the monitoring influence weights set when constructing the state analysis model for the basic indicators and scene indicators, the quantified consistency values are weighted and summed to obtain the contrast between the bolt to be monitored and the corresponding bolt category. The contrast value ranges from [0,1]. The closer the value is to 1, the higher the consistency between the bolt's own parameters and real-time scene parameters and the bolt category, meaning that the bolt category better represents the actual situation of the bolt to be monitored. The bolt category with the highest contrast is set as the real-time bolt category.
[0042] In this embodiment, ultrasonic time refers to the round-trip propagation time of ultrasonic waves. The relative change in time is equal to the real-time time minus the preset reference time. The rate of change is equal to (real-time time minus the preset reference time) / preset reference time * 100%. The speed of sound refers to the propagation speed of ultrasonic waves. The amplitude attenuation rate is a measure of energy loss when ultrasonic waves propagate in a bolt. The strain value is directly measured by a strain sensor to determine the degree of deformation on or inside the bolt surface. It is an intuitive physical quantity that reflects the magnitude of the axial force.
[0043] In this embodiment, by determining the real-time bolt category and selecting the corresponding state model, monitoring area, and monitoring time interval, it can be ensured that the monitoring strategy is highly matched with the actual characteristics and working conditions of the bolt to be monitored, laying the foundation for accurate assessment and risk warning of the bolt axial force health status in the future.
[0044] In some embodiments of this application, generating real-time status results based on the real-time status coefficients of all monitoring points includes: Generate a set of associated points for each monitoring point; The state probability of the real-time state coefficient of the corresponding monitoring point is generated based on the set of associated points; Determine whether to correct the real-time state coefficient of the corresponding monitoring point based on the state probability. If so, generate the corrected real-time state coefficient based on the set of associated points and update it. Update the real-time status coefficient of each monitoring point sequentially; The average value of the real-time state coefficients of all monitoring points in the same monitoring area is obtained by averaging the updated real-time state coefficients of the corresponding monitoring area. The average state coefficient of all monitoring areas of the bolt to be monitored is weighted to obtain the comprehensive state coefficient. The state level and the corresponding real-time state results are determined based on the comprehensive state coefficient.
[0045] In this embodiment, the associated point set refers to the set of monitoring points where the monitoring positions and axial forces of the same monitoring point and different monitoring points have a transmission relationship. It is determined by analyzing statistical indicators such as Pearson correlation coefficient, mutual information value, or Granger causality of the real-time state coefficients of each monitoring point in historical monitoring data. When the absolute value of the correlation coefficient between two monitoring points is greater than a preset correlation threshold (e.g., 0.6) or the mutual information value exceeds a preset mutual information threshold, it is determined that there is a significant correlation between the two, and they are included in each other's associated point set.
[0046] In this embodiment, the state probability is a quantifiable value of the reliability of the real-time state coefficient of the current monitoring point, calculated using a Bayesian probability model based on the distribution characteristics of the real-time state coefficients of each point in the associated point set. Specifically, firstly, the mean μ and standard deviation σ of the real-time state coefficients of all points in the associated point set are statistically analyzed to construct a normal distribution probability density function N(μ,σ²). Then, the real-time state coefficient of the current monitoring point is substituted into this probability density function to obtain its probability density value under this distribution, i.e., the state probability. A higher state probability indicates a stronger consistency between the current real-time state coefficient and the overall state trend of the associated points, and a higher reliability; conversely, a lower state probability may indicate data anomalies or localized sudden interference.
[0047] In this embodiment, if the state probability is less than a preset probability threshold (0.55 in this application), the real-time state coefficient of the current monitoring point is determined to be abnormal and needs to be corrected. The real-time state coefficients of each point in the associated point set are weighted and calculated with the state probability of the current point. The formula is: Corrected real-time state coefficient = Current real-time state coefficient × State probability + Σ(Real-time state coefficient of associated points × Association weight) × (1 - State probability). The association weight is obtained by normalizing the absolute value of the correlation coefficient between the associated point and the current point; the larger the absolute value of the correlation coefficient, the higher the association weight.
[0048] In this embodiment, when updating the real-time status coefficient of each monitoring point in sequence, an iterative correction mechanism is used. When the real-time status coefficient of a certain monitoring point is corrected, other points in its associated point set will synchronously reference the updated value in subsequent correction processes, thereby reducing the error propagation caused by correction.
[0049] In this embodiment, the weighting process assigns different weight coefficients to each monitoring area based on its importance in the overall stress safety of the bolt. These weight coefficients are positively correlated with the historical state risk values used when dividing the monitoring areas. The higher the historical state risk value, the larger the weight coefficient of the monitoring area, and vice versa.
[0050] In this embodiment, by performing a reliable assessment and correction on the real-time status coefficients of all monitoring points, the accuracy of the real-time status results is improved, the interference of abnormal data from a single monitoring point on the overall assessment results is avoided, and the health status of the bolt axial force is accurately reflected.
[0051] In some embodiments of this application, determining the state level and the corresponding real-time state result based on the comprehensive state coefficient includes: Several preset comprehensive state coefficient ranges are pre-defined, and each preset comprehensive state coefficient range is mapped to a corresponding preset state level. Based on the correspondence between the comprehensive state coefficient and the preset comprehensive state coefficient range, the corresponding preset state level is selected as the current state level. The preset state level includes a first preset state level, a second preset state level, and a third preset state level. When the current status level is the first preset status level, the corresponding real-time status result is that danger exists; When the current status level is the second preset status level, the corresponding real-time status result is that an alert is required; When the current status level is the third preset status level, the corresponding real-time status result is healthy.
[0052] In this embodiment, multiple preset comprehensive state coefficient intervals are set based on historical data. These intervals are set according to different historical state results. In this application, the comprehensive state coefficient intervals are divided into [0,0.3), [0.3,0.7), and [0.7,1], which correspond to the first preset state level, the second preset state level, and the third preset state level, respectively, and correspond to the state results of danger, need for warning, and health, respectively.
[0053] In this embodiment, the current monitoring time interval is also adjusted according to different status results. When there is danger, the subsequent monitoring time interval is shortened to 1 / 3 of the original monitoring time interval; when an early warning is required, the subsequent monitoring time interval is shortened to 1 / 2 of the original monitoring time interval; when the status is healthy, the original monitoring time interval is kept unchanged, so as to dynamically adjust the monitoring frequency to adapt to the actual health status of the bolt, improve monitoring efficiency and reduce the cost of invalid data collection.
[0054] In some embodiments of this application, generating a predicted state level for a future time period includes: Generate several real-time data sequences for each monitoring point in the current monitoring period; Trend features are extracted from real-time data sequences to obtain their time-domain and frequency-domain features. Similarity analysis is then performed with the historical trend feature database of real-time bolt categories to determine the trend type of each real-time data sequence. The historical trend feature library includes several preset trend types, and each preset trend type is mapped to a corresponding preset time domain feature and a corresponding preset frequency domain feature. Based on the preset time series prediction model corresponding to the trend type, the model outputs the predicted data sequence for the corresponding real-time data sequence in the future period. Several predicted axial force related datasets are generated based on all predicted data sequences of the same monitoring point and input into the state model to obtain several predicted state coefficients of the corresponding monitoring point in the future period. Generate several comprehensive forecast state coefficients for future time periods and determine several corresponding forecast state levels.
[0055] In this embodiment, the current monitoring cycle is set based on the bolt's own parameters, its operating conditions (such as vibration frequency, temperature fluctuation amplitude, load change rate, etc.), and preset monitoring accuracy requirements. For example, for critical load-bearing bolts in high-frequency vibration environments, the current monitoring cycle is preferably 2 hours.
[0056] In this embodiment, the time-domain features include mean, variance, maximum value, minimum value, and peak factor; the frequency-domain features include dominant frequency, energy spectral density, and spectral entropy; and the historical trend feature library contains typical trend feature templates formed by all monitoring points under this bolt category during the historical state change process. Each template is labeled with the corresponding trend type, time-domain features, and frequency-domain features.
[0057] In this embodiment, similarity analysis refers to calculating the cosine similarity between the time-domain features and frequency-domain features and the preset time-domain feature vectors and preset frequency-domain feature vectors corresponding to each preset trend type in the historical trend feature library, respectively, to obtain the time-domain similarity and frequency-domain similarity. Then, the weighted sum is obtained to obtain the comprehensive similarity, and the preset trend type with the highest comprehensive similarity is determined as the trend type of the real-time data sequence.
[0058] In this embodiment, the trend types include stationary trends, linear growth trends, linear decay trends, and volatile trends. The preset time series prediction model corresponding to each trend type is trained and optimized based on historical data. For example, a stationary trend corresponds to the ARIMA model, a linear growth / decay trend corresponds to the linear regression model, and a volatile trend corresponds to the LSTM neural network model.
[0059] In this embodiment, following the same processing flow as generating real-time status results (including status probability calculation of associated point set, anomaly correction, regional mean processing and weighting), the comprehensive predicted status coefficient for future time periods is obtained. Based on the preset comprehensive status coefficient interval in which the comprehensive predicted status coefficient is located, the corresponding predicted status level and predicted status result are determined, thereby realizing a forward-looking early warning of the health status of bolt axial force.
[0060] In some embodiments of this application, fine-tuning of the real-time status results includes: The distribution of predicted state levels is determined based on several predicted state levels for future time periods. Collect the frequency of occurrence and duration of each predicted state level in the future period, and calculate the early warning index; Pre-set early warning index thresholds; When the warning index is less than the warning index threshold, no fine-tuning is performed on the real-time status results. When the warning index is not less than the warning index threshold, the real-time status result is downgraded by one level.
[0061] In this embodiment, the warning index threshold is determined based on the worst state level, frequency of occurrence, and duration in the future period. In this application, the warning index threshold is set to 0.45.
[0062] In this embodiment, the early warning index = Σ(early warning weight corresponding to the predicted state level × frequency of occurrence × duration) / total duration of the future period, and the early warning weights are set to 0 (third preset state level), 0.5 (second preset state level), and 1.0 (first preset state level).
[0063] In this embodiment, downgrading the real-time status result by one level means that if the original real-time status result was healthy, it is adjusted to require a warning; if the original real-time status result was required a warning, it is adjusted to indicate that there is danger.
[0064] In this embodiment, by introducing the predicted state level for future time periods to dynamically fine-tune the current real-time state results, it is possible to intervene in risk management in advance by combining the development trend of bolt health status, thereby improving the early warning sensitivity of the monitoring system.
[0065] In some embodiments of this application, determining whether to generate an alarm command based on the fine-tuned real-time status result includes: When the real-time status result after fine-tuning indicates that there is a danger, an alarm command is generated; When the real-time status result after fine-tuning requires an alert, an alert command is generated. No warning command is generated when the real-time status result after fine-tuning is healthy.
[0066] In this embodiment, the alarm command has a higher alarm intensity than the warning command, and the alarm command will be pushed to the management terminal simultaneously and the emergency response mechanism will be activated. The warning command is mainly used to remind maintenance personnel to carry out planned inspections and preventive maintenance to avoid further deterioration of the bolt axial force.
[0067] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application.
Claims
1. A comprehensive monitoring system for the health status of bolt axial force, characterized in that, include: The module is used to define multiple bolt categories and build a state analysis model for each bolt category. The monitoring module is used to set the monitoring strategy for the bolt to be monitored based on the condition analysis model, obtain axial force related data according to the monitoring strategy, analyze it, and obtain real-time condition results. The early warning module is used to generate predicted status levels for future periods and fine-tune the real-time status results. Based on the fine-tuned real-time status results, it determines whether to generate an alarm command.
2. The bolt axial force health status comprehensive monitoring system as described in claim 1, characterized in that, The monitoring module also includes: A piezoelectric ultrasonic sensing unit is used to monitor the bolt to be monitored in real time according to the monitoring strategy. Electromagnetic ultrasound enhancement sensing unit, including a customized probe and signal enhancement algorithm; The customized probe is designed according to the structural characteristics of different bolts and is used to generate a multi-channel sensor array for signal acquisition.
3. The bolt axial force health status comprehensive monitoring system as described in claim 2, characterized in that, A state analysis model is constructed for each bolt category, including: Obtain the basic parameters and scenario parameters of the bolt, and set several basic and scenario indicators based on the corresponding monitoring impact. Several bolt categories are obtained by randomly combining basic and scenario indicators. Construct historical monitoring data packages for each bolt category; Generate historical status risk values for the corresponding bolt categories based on historical monitoring data packets; Based on historical risk values, several monitoring areas and corresponding monitoring time intervals are set for the corresponding bolt categories. Each monitoring area includes several monitoring points. Generate training data packets for each monitoring area of each bolt category, and construct a state model for each bolt category; A state analysis model for each bolt category is generated based on the state model for each bolt category and the corresponding monitoring time interval.
4. The bolt axial force health status comprehensive monitoring system as described in claim 3, characterized in that, Construct a state model for each bolt category, including: For each bolt category, historical axial force-related data in the historical monitoring data package is mapped to the monitoring area to obtain the initial training data package for each monitoring area. Preprocessing is performed on the initial training data packets of the same monitoring area to obtain training data packets, which include several historical data sequences. The preprocessing includes data cleaning, smoothing, and sequence construction. Generate an axial force correlation dataset for the corresponding monitoring area from historical axial force correlation data at the same historical time point in different historical data sequences; Generate several axial force-related datasets for each monitoring area; A dataset-axial force value mapping table is pre-built, and the historical actual axial force value of each axial force-related dataset is generated based on the dataset-axial force value mapping table; Historical state coefficients are generated based on actual historical axial force values; Using each axial force-related dataset of each monitoring area for the same bolt category as training input data and the corresponding historical state coefficients as training output data, a neural network is trained to obtain the state model for the corresponding bolt category.
5. The bolt axial force health status comprehensive monitoring system as described in claim 3, characterized in that, The axial force-related data are obtained and analyzed according to the monitoring strategy to obtain real-time status results, including: The parameters of the bolt to be monitored and the real-time scene parameters are obtained and compared with each bolt category to obtain several contrast parameters. The real-time bolt category is determined based on contrast, and a monitoring strategy is generated based on several monitoring areas and corresponding monitoring time intervals of the real-time bolt category. According to the monitoring strategy, real-time axial force related data of several monitoring points in each monitoring area are obtained. The real-time axial force related data includes ultrasonic time, relative change of time, rate of change, sound velocity, amplitude attenuation rate, and strain value. Input the real-time axial force related data into the real-time bolt category state model to obtain the real-time state coefficient of each monitoring point; Real-time status results are generated based on the real-time status coefficients of all monitoring points.
6. The bolt axial force health status comprehensive monitoring system as described in claim 5, characterized in that, Real-time status results are generated based on the real-time status coefficients of all monitoring points, including: Generate a set of associated points for each monitoring point; The state probability of the real-time state coefficient of the corresponding monitoring point is generated based on the set of associated points; Determine whether to correct the real-time state coefficient of the corresponding monitoring point based on the state probability. If so, generate the corrected real-time state coefficient based on the set of associated points and update it. Update the real-time status coefficient of each monitoring point sequentially; The average value of the real-time state coefficients of all monitoring points in the same monitoring area is obtained by averaging the updated real-time state coefficients of the corresponding monitoring area. The average state coefficient of all monitoring areas of the bolt to be monitored is weighted to obtain the comprehensive state coefficient. The state level and the corresponding real-time state results are determined based on the comprehensive state coefficient.
7. The bolt axial force health status comprehensive monitoring system as described in claim 6, characterized in that, The state level and corresponding real-time state results are determined based on the comprehensive state coefficients, including: Several preset comprehensive state coefficient ranges are pre-defined, and each preset comprehensive state coefficient range is mapped to a corresponding preset state level. Based on the correspondence between the comprehensive state coefficient and the preset comprehensive state coefficient range, the corresponding preset state level is selected as the current state level. The preset state level includes a first preset state level, a second preset state level, and a third preset state level. When the current status level is the first preset status level, the corresponding real-time status result is that danger exists; When the current status level is the second preset status level, the corresponding real-time status result is that an alert is required; When the current status level is the third preset status level, the corresponding real-time status result is healthy.
8. The bolt axial force health status comprehensive monitoring system as described in claim 7, characterized in that, Generate predicted state levels for future time periods, including: Generate several real-time data sequences for each monitoring point in the current monitoring period; Trend features are extracted from real-time data sequences to obtain their time-domain and frequency-domain features. Similarity analysis is then performed with the historical trend feature database of real-time bolt categories to determine the trend type of each real-time data sequence. The historical trend feature library includes several preset trend types, and each preset trend type is mapped to a corresponding preset time domain feature and a corresponding preset frequency domain feature. Based on the preset time series prediction model corresponding to the trend type, the model outputs the predicted data sequence for the corresponding real-time data sequence in the future period. Several predicted axial force related datasets are generated based on all predicted data sequences of the same monitoring point and input into the state model to obtain several predicted state coefficients of the corresponding monitoring point in the future period. Generate several comprehensive forecast state coefficients for future time periods and determine several corresponding forecast state levels.
9. The bolt axial force health status comprehensive monitoring system as described in claim 8, characterized in that, Fine-tuning of real-time status results, including: The distribution of predicted state levels is determined based on several predicted state levels for future time periods. Collect the frequency of occurrence and duration of each predicted state level in the future period, and calculate the early warning index; Pre-set early warning index thresholds; When the warning index is less than the warning index threshold, no fine-tuning is performed on the real-time status results. When the warning index is not less than the warning index threshold, the real-time status result is downgraded by one level.
10. The bolt axial force health status comprehensive monitoring system as described in claim 9, characterized in that, Whether to generate an alarm command is determined based on the real-time status results after fine-tuning, including: When the real-time status result after fine-tuning indicates that there is a danger, an alarm command is generated; When the real-time status result after fine-tuning requires an alert, an alert command is generated. No warning command is generated when the real-time status result after fine-tuning is healthy.