Ripple compensator multi-project management and early warning analysis system based on cloud platform

Abstract

The invention relates to the technical field of intelligent early warning, and discloses a ripple compensator multi-project management and early warning analysis system based on a cloud platform, and the system comprises a sensor module which comprises a temperature sensor and a displacement sensor; the wireless communication module is connected with the sensor module and is used for transmitting temperature data and / or displacement data; the data processing module collects expansion and contraction data of the corrugated pipe, obtains the fatigue times of the corrugated pipe and determines the residual fatigue times of the corrugated compensator; collecting temperature data, comparing the temperature data with a temperature threshold value, judging whether a leakage risk exists or not, and when the risk exists, extracting a feature vector data set, processing based on a multi-head time sequence attention mechanism, and determining a comprehensive time sequence feature; and the early warning module performs risk early warning according to the comprehensive time sequence characteristics. According to the invention, the identification capability of early fatigue degradation and slow change leakage is improved, the false report and missing report rate is reduced, and multi-project and cross-regional unified management is realized through cloud centralized deployment.

Description

Technical Field

[0001] The present invention relates to the field of intelligent early warning technology, and in particular to a cloud platform-based multi-project management and early warning analysis system for ripple compensators. Background Art

[0002] As urban heating networks continue to expand, bellows compensators, as key flexible connectors in thermal pipelines, face a significant impact on network safety and pipeline life. Bellows compensators primarily compensate for axial displacement caused by thermal expansion and contraction. However, over long-term operation, due to frequent hot-cold cycles and environmental changes, they are prone to fatigue damage, structural deformation, or seal failure. These issues can lead to leakage, energy loss, equipment corrosion, and even major safety incidents.

[0003] To ensure the stable operation of the heating system, some pipeline networks have deployed online monitoring methods for compensators based on temperature or displacement sensors, attempting to determine abnormal conditions through thresholds. However, existing technologies have several major shortcomings. Traditional solutions often use static alarm thresholds set by rules, lack the ability to mine potential risk trends and multi-dimensional interactive information in sensor time series, and are unable to accurately determine early signs of leakage or fatigue degradation. Most systems are based on single-point signals or simple fluctuation judgments, lack the ability to model long-term evolution processes, and are prone to false alarms or missed alarms, especially in the case of high-frequency small deformations or slow temperature rises, resulting in a low alarm signal-to-noise ratio. In large-scale pipeline networks, existing systems struggle to achieve unified data management and risk normalization analysis across regions and projects, hindering systematic maintenance and resource scheduling.

[0004] Therefore, it is necessary to design a multi-project management and early warning analysis system for ripple compensators based on a cloud platform to solve the problems existing in current technology. Summary of the Invention

[0005] In view of this, the present invention proposes a multi-project management and early warning analysis system for ripple compensators based on a cloud platform, aiming to solve the current problems of low utilization efficiency of monitoring data, poor risk identification accuracy and lack of multi-project collaborative management mechanism.

[0006] The present invention proposes a multi-project management and early warning analysis system for ripple compensators based on a cloud platform, including:

[0007] Sensor module, including temperature sensor and displacement sensor;

[0008] a wireless communication module connected to the sensor module, the wireless communication module being used to transmit temperature data and / or displacement data;

[0009] a data processing module configured to collect bellows expansion and contraction data and obtain bellows fatigue times to determine the remaining fatigue times of the bellows compensator; collect temperature data and compare it with a temperature threshold to determine whether there is a leakage risk; and, when it is determined that there is a leakage risk or a fatigue damage risk, collect data from all sensors within a preset time period, extract a feature vector data group, and process the feature vector data group based on a multi-head temporal attention mechanism to determine a comprehensive temporal feature;

[0010] The early warning module is configured to issue a risk early warning based on the comprehensive time series characteristics.

[0011] Furthermore, the sensor module includes:

[0012] There are at least four temperature sensors, which are arranged on both sides of the external movable end of the bellows compensator, with at least two temperature sensors arranged on each side;

[0013] At least four displacement sensors are provided, and the displacement sensors are arranged on both sides of the protective bellows inside the bellows compensator.

[0014] Furthermore, the data processing module collects the expansion and contraction data of the bellows and obtains the fatigue times of the bellows, and determines the remaining fatigue times of the bellows compensator, including:

[0015] The expansion and contraction data of the bellows in the bellows compensator are collected based on a preset sampling period to obtain an axial expansion and contraction displacement sequence. The axial expansion and contraction displacement sequence is divided into sliding windows to construct a displacement change data set within the time window, identify peaks and valleys, and obtain tension and compression cycles.

[0016] The amplitude and average displacement of each pair of tension-compression half-cycles were extracted based on rainflow counting;

[0017] Collect the elastic modulus of the bellows and the stress-strain conversion coefficient of the compensator structure to obtain the equivalent stress amplitude;

[0018] Obtaining a standard number of fatigue cycles according to the SN fatigue life curve corresponding to the equivalent stress amplitude and the bellows material;

[0019] The fatigue consumption of the identified tension-compression cycles is accumulated based on the Miner linear damage accumulation method, and compared with the standard fatigue cycle number to obtain the remaining fatigue number.

[0020] Furthermore, after the data processing module determines the remaining fatigue times of the corrugated compensator, it further includes:

[0021] The data processing module compares the remaining fatigue times with the remaining fatigue threshold, and determines whether the corrugated compensator has a fatigue damage risk based on the comparison result;

[0022] When the remaining fatigue times are less than the remaining fatigue threshold, the data processing module determines that the corrugated compensator is at risk of fatigue damage; when the remaining fatigue times are greater than or equal to the remaining fatigue threshold, the data processing module determines that the corrugated compensator is not at risk of fatigue damage.

[0023] Furthermore, the data processing module collects temperature data and compares it with the temperature threshold to determine whether there is a leakage risk, including:

[0024] When the temperature data is less than or equal to the temperature threshold, the data processing module determines that there is no leakage risk; when the temperature data is greater than the temperature threshold, the data processing module determines that there is a leakage risk.

[0025] Furthermore, the data processing module obtains the temperature threshold by the following method:

[0026] The data processing module collects temperature data for N consecutive sampling periods, constructs a reference temperature curve for each sensor measuring point, and obtains a normal temperature change baseline of the compensator in a non-leakage state;

[0027] Collecting the soil temperature of the external environment of the compensator, and performing an offset correction on the normal temperature change baseline based on the soil temperature, wherein the correction factor is obtained by linear regression;

[0028] The standard temperature maximum and standard temperature minimum are obtained according to the corrected normal temperature change baseline, and the temperature threshold is obtained according to the standard temperature minimum and a safety margin, where the safety margin is 2-5°C.

[0029] Furthermore, the data processing module processes the feature vector data group based on the multi-head temporal attention mechanism to determine the comprehensive temporal features, including:

[0030] The feature vector data set includes displacement data, temperature data, acceleration change of displacement data and acceleration change of temperature data;

[0031] Establishing the feature vector data set as a time series feature matrix;

[0032] Configure the number of attention heads and perform multi-head temporal attention processing on the time series feature matrix in sequence. Each attention head includes an independent Query, Key, and Value linear mapper.

[0033] Calculate the attention weight of each time frame to other frames in the time series;

[0034] The representations of all heads are concatenated and linearly combined to generate comprehensive temporal features.

[0035] Furthermore, the data processing module processes the feature vector data group based on the multi-head temporal attention mechanism to determine the comprehensive temporal features, and further includes: For the h-th attention head, its attention output is: ; in, Represents the similarity scoring matrix; represents the scaling factor, represents the Value matrix, represents the Query matrix, represents the Key matrix, represents the normalization function, Represents the attention output result of the h-th attention head; Among them, the softmax result is the attention weight matrix: ; in, represents the degree to which the i-th time frame in the h-th attention head sequence pays attention to the j-th frame when generating its output representation, Represents the Query vector of the i-th time step; represents the Key vector of the jth time step; j represents the target time frame number for calculating attention, Represents the index variable used for normalization in the softmax denominator; T represents the length of the time series.

[0036] Furthermore, when the early warning module issues a risk early warning based on the comprehensive time series characteristics, it includes:

[0037] The early warning module performs risk warning based on a fully connected network and a softmax activation function and according to the comprehensive time series features;

[0038] p risk =softmax(W c y fused +b c );

[0039] Among them, p risk represents the predicted probability of each risk level, y fused represents the comprehensive time series characteristics, W c represents the weight of the fully connected neural network classifier, b c represents the bias vector, and p risk ∈R C , C represents the risk level number, C = 4;

[0040] The early warning module uses the risk level with the highest predicted probability as the output risk level and simultaneously outputs a determination result, wherein the determination result includes whether there is a leakage risk and / or whether there is a fatigue damage risk.

[0041] Furthermore, it also includes:

[0042] a data compression module configured to, when the data processing module determines that there is no leakage risk and no fatigue damage risk, perform aggregation processing on the collected temperature data and displacement data, and use a sliding average to generate representative sampling points to replace the original high-frequency sampling sequence; and, when the data processing module determines that there is a leakage risk or a fatigue damage risk, keep the collected temperature data and displacement data uncompressed and recorded;

[0043] The sampled data sequence is structured-coded and then stored, wherein the structured coding includes difference coding, interval statistics coding and lightweight time series coding.

[0044] Compared with the prior art, the beneficial effects of the present invention are: by setting a sensor module composed of a temperature sensor and a displacement sensor, real-time collection of key physical parameters of the bellows compensator (such as axial expansion and contraction and soil temperature changes) is realized, and remote, high-frequency, low-power data transmission is realized through a wireless communication module, which effectively adapts to the widely distributed urban heating pipe network scenario; the data processing module constructs a bellows expansion and displacement sequence based on the sampled data, obtains the current fatigue state of the bellows through the rain flow counting method and the Miner fatigue damage model, and combines temperature data with the dynamic threshold model to realize the identification and judgment of potential leakage risks; when risk signs are triggered, multi-channel sensor data within a certain time window is automatically called, and the key frame change features are extracted through the multi-head temporal attention mechanism to generate a fused comprehensive temporal feature representation for accurately identifying small trend anomalies or multi-variable interactions; finally, the early warning module outputs the risk level based on the fused representation, realizing a closed-loop architecture from "physical acquisition-intelligent analysis-active early warning". Compared with existing solutions that rely on static thresholds and cannot model evolution trends, it improves the ability to identify early fatigue degradation and slow-changing leakage, reduces false alarm and missed alarm rates, and realizes unified management of multiple projects and cross-regions through centralized cloud deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0046] Figure 1A structural block diagram of a multi-project management and early warning analysis system for ripple compensators based on a cloud platform provided by an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of the arrangement of temperature sensors in a multi-project management and early warning analysis system for ripple compensators based on a cloud platform provided by an embodiment of the present invention;

[0048] Figure 3 Schematic diagram of the arrangement of displacement sensors in the cloud platform-based multi-project management and early warning analysis system for ripple compensators provided in an embodiment of the present invention.

[0049] Among them, 100 is the internal protective bellows; 200 is the external movable end; 300 is the displacement sensor; and 400 is the temperature sensor. DETAILED DESCRIPTION

[0050] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0051] Traditional bellows compensator monitoring systems, based on static threshold judgments, rely on fixed temperature and displacement thresholds for anomaly detection, making it difficult to capture the multidimensional temporal correlations implicit in sensor data and effectively identify slowly varying risk signals caused by material fatigue accumulation or minor leaks. The system lacks the ability to analyze the dynamic evolution of long time series, leading to high-frequency, small displacement fluctuations and slow temperature rises being easily misidentified as noise. This makes it impossible to accurately distinguish between normal thermal expansion and contraction and early signs of damage. In pipeline networks with multiple compensators operating in coordination, siloed data storage and analysis further exacerbates the probability of missed detection of cross-project risk signatures.

[0052] For example, in a cross-regional heating network, multiple corrugated compensators are subjected to periodic thermal stress and soil pressure for a long time. Traditional monitoring systems independently perform threshold comparisons on the displacement and temperature data of each compensator. When the amplitude of a single displacement fluctuation of a compensator does not exceed the preset threshold due to material fatigue, but the frequency of fluctuations in adjacent time windows increases, it is impossible to identify the abnormal pattern through isolated data points. At the same time, due to the diurnal fluctuations in ambient temperature, the surface temperature monitoring value of the compensator has a baseline drift, and the static threshold cannot be adaptively adjusted, resulting in the initial characteristics of temperature-slow leakage being masked by environmental noise. In a multi-project management scenario, the operating data of compensators in different geographical regions lacks a unified time series modeling, and it is impossible to discover abnormal evolution trends through historical data comparison.

[0053] If these issues are not addressed, slowly-varying fatigue damage and minor leaks cannot be effectively identified at an early stage, leading to the continued accumulation of compensator structural damage until sudden failure. In multi-project management scenarios, isolated data analysis will prevent horizontal comparison of the risk characteristics of compensators in different regions, delaying systematic maintenance decisions. The accumulated damage caused by long-term, high-frequency, minor displacements can trigger cascading structural failures, significantly increasing the risk of pipeline network leaks and outages. Furthermore, false alarms caused by static thresholds will reduce the efficiency of operation and maintenance responses and increase manual verification costs.

[0054] When faced with the above problems, this application first considers how to break through the limitations of traditional static thresholds and identify slow-changing risks by capturing multi-dimensional time series correlation features. Existing single-point monitoring cannot distinguish between normal fluctuations and early damage, so it is necessary to build a dynamic analysis model to perform joint time series modeling of displacement and temperature data. In response to the high-frequency micro-displacement and temperature slow-changing phenomena, this application attempts to combine the displacement change data set within the time window with the temperature change baseline to extract tension and compression cycle characteristics and environmental correction factors. A multi-head time series attention mechanism is introduced to simultaneously process multi-dimensional features such as displacement acceleration and temperature change rate, and the abnormal patterns of key time frames are dynamically captured through self-attention weights to form a comprehensive time series analysis framework that integrates fatigue damage assessment and leakage risk determination.

[0055] See Figure 1As shown, the present application proposes a multi-project management and early warning analysis system for bellows compensators based on a cloud platform, comprising: a sensor module, including a temperature sensor and a displacement sensor. A wireless communication module, connected to the sensor module, the wireless communication module is used to transmit temperature data and / or displacement data. A data processing module, configured to collect the expansion and contraction data of the bellows and obtain the fatigue times of the bellows, and determine the remaining fatigue times of the bellows compensator. The temperature data is collected and compared with the temperature threshold to determine whether there is a leakage risk, and when it is determined that there is a leakage risk or a fatigue damage risk, the data of all sensors within a preset time period are collected, and the feature vector data group is extracted. The feature vector data group is processed based on the multi-head time series attention mechanism to determine the comprehensive time series features. The early warning module is configured to issue a risk warning based on the comprehensive time series features.

[0056] Specifically, temperature and displacement sensors are devices used to monitor the operating status and environmental changes of the bellows compensator in real time. Temperature detection can be achieved using thermocouples, RTDs, or fiber Bragg grating temperature sensors, while displacement measurement can be achieved using laser or capacitive displacement sensors. A multi-point deployment improves the comprehensiveness of data collection and the sensitivity of anomaly detection. A wireless communication module transmits temperature and displacement data, utilizing LoRa, NB-IoT, or 5G communication protocols for low-power, long-distance data transmission, ensuring real-time upload of multi-item monitoring data to a cloud platform. The bellows fatigue calculation module, within the data processing module, assesses the extent of structural fatigue damage based on axial expansion and contraction displacement sequences and material properties. Specifically, the rainflow counting method is used to extract the number of tensile and compressive cycles. Remaining life prediction is achieved by combining the SN fatigue life curve and the Miner linear damage accumulation method, addressing the problem that traditional static thresholds fail to capture long-term fatigue degradation. Leakage risk assessment involves comparing temperature data with dynamically adjusted temperature thresholds. Historical temperature data is used to establish a baseline for normal temperature variation, which is then corrected using linear regression based on ambient soil temperature to generate safety margin thresholds tailored to different operating conditions, avoiding misjudgments caused by a single threshold. Comprehensive time series feature generation involves processing multidimensional sensor data based on a multi-head time series attention mechanism. Specifically, multiple independent attention heads are used to capture the temporal correlations between displacement, temperature, and acceleration changes. The attention weight matrix quantifies the dependencies between different time frames, improving the ability to identify early leaks and minor deformations. The early warning module's risk warning involves multi-level risk classification based on comprehensive time series features. Specifically, a fully connected neural network combined with a softmax activation function is used to output risk probability distributions, achieving an end-to-end mapping from data features to risk levels, enhancing the accuracy of early warnings under complex working conditions.

[0057] It's understandable that by integrating multiple project management capabilities into the cloud platform, combined with dynamically modified temperature thresholds, fatigue damage accumulation models, and a multi-head time-series attention mechanism, refined monitoring of the ripple compensator's operating status and early risk warnings are possible. Through time series analysis of multi-dimensional sensor data and machine learning algorithms, the poor adaptability of static thresholds and weak correlation between multi-dimensional data in traditional solutions are effectively addressed. This improves the accuracy of leak and fatigue damage detection and the timeliness of warnings, while also supporting the unified management and analysis of large-scale pipeline network data.

[0058] The working process and principle of this application are as follows: the multi-project management and early warning analysis system for bellows compensators based on the cloud platform includes a sensor module, a wireless communication module, a data processing module and an early warning module. The sensor module includes a temperature sensor and a displacement sensor, which are used to collect the temperature and displacement data of the bellows compensator. The wireless communication module is connected to the sensor module and is responsible for transmitting the collected temperature data and displacement data. The data processing module first collects the expansion and contraction data of the bellows and obtains the fatigue times of the bellows, and determines the remaining fatigue times of the bellows compensator by calculation. Then the temperature data is collected and compared with the preset temperature threshold to determine whether there is a leakage risk. When it is determined that there is a leakage risk or a fatigue damage risk, the data processing module will collect data from all sensors within a preset time period and extract a feature vector data group. Then, the feature vector data group is processed based on the multi-head temporal attention mechanism to determine the comprehensive temporal features. Finally, the early warning module issues a risk warning based on the comprehensive temporal features.

[0059] By collecting bellows expansion and contraction data and temperature data, combined with a multi-head temporal attention mechanism, we can effectively capture temporal correlations within the data, enabling early identification of potential risks in bellows compensators. This multi-head temporal attention mechanism simultaneously processes multi-dimensional features such as displacement and temperature. By calculating attention weights across different time frames, it dynamically captures abnormal patterns at key time points. Compared to traditional static thresholding, this mechanism can more accurately identify slowly changing risk signals caused by material fatigue accumulation or minor leaks.

[0060] It's understandable that using a multi-head time-series attention mechanism to process feature vector data sets can effectively extract long-term dependencies in time-series data, overcoming the difficulty traditional methods have in capturing the dynamic evolution patterns of long time series. By comprehensively analyzing multidimensional features, it can more accurately distinguish between normal thermal expansion and contraction and early signs of damage, improving the accuracy of risk warnings.

[0061] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0062] The sensor module includes multiple temperature sensors and displacement sensors installed at key locations on the bellows compensator. The temperature sensors monitor the surface temperature of the bellows, while the displacement sensors measure the axial displacement of the bellows.

[0063] The wireless communication module uses low-power Bluetooth technology to regularly transmit the data collected by the sensor to the cloud platform.

[0064] The data processing module first processes the bellows' expansion and contraction data. The sliding window method analyzes the axial expansion and contraction displacement sequence to identify tension-compression cycles. The rainflow counting method is then used to extract the amplitude and average displacement for each pair of tension-compression half-cycles. Based on the bellows' material parameters, the equivalent stress amplitude is calculated, and the standard fatigue cycle number is determined based on the SN fatigue life curve. Fatigue consumption is calculated using the Miner linear damage accumulation method, ultimately resulting in the remaining fatigue cycle.

[0065] For temperature data, establish a baseline for normal temperature variations. Collect temperature data for multiple consecutive sampling periods, combine it with the external soil temperature, and use linear regression to derive a correction factor to offset the baseline. Based on this corrected baseline, determine the temperature threshold for leak risk assessment.

[0066] When a risk is identified, a multi-head temporal attention mechanism is triggered. This mechanism combines displacement data, temperature data, and acceleration changes into a feature vector data set, creating a temporal feature matrix. Multiple attention heads are configured, each containing independent query, key, and value linear mappers. The attention weight of each time frame in the time series is calculated for each other. The representations of all heads are concatenated and linearly combined to generate a comprehensive temporal feature.

[0067] The early warning module uses a fully connected network and softmax activation function to predict risk levels using comprehensive time series features. It outputs the risk level with the highest predicted probability and determines whether there is a risk of leakage or fatigue damage.

[0068] The above solution effectively addresses the challenges inherent in traditional bellows compensator monitoring systems. By analyzing multidimensional temporal correlation features, the system overcomes the limitations of static thresholds and accurately identifies slow deformation risk signals caused by material fatigue accumulation or minor leaks. The introduction of a multi-head temporal attention mechanism enables the system to simultaneously process multi-dimensional features such as displacement acceleration and temperature change rate. By dynamically capturing abnormal patterns in key time frames through self-attention weighting, the system improves its ability to identify high-frequency, small displacement fluctuations and slow temperature rises.

[0069] The cloud-based platform design enables unified data management and risk analysis across regions and projects, overcoming the siloed data storage and analysis issues of traditional systems. It can identify abnormal evolution trends through historical data comparisons, effectively supporting systematic maintenance decisions.

[0070] This application's solution integrates fatigue damage assessment and leakage risk determination to form a comprehensive time-series analysis framework, improving the accuracy and timeliness of early risk identification for bellows compensators. This helps prevent the accumulation of structural damage leading to sudden failure, reduces false alarm rates, improves operational and maintenance response efficiency, and reduces manual verification costs.

[0071] In some of the above-mentioned schemes of the present application, there is an insufficient number of temperature and displacement sensors or incomplete position coverage when the sensor modules are arranged, resulting in the inability to effectively capture the dynamic changes of the external movable end 200 of the bellows compensator and the internal protective bellows 100. In particular, it is difficult to distinguish between normal fluctuations and abnormal deviations under complex working conditions, resulting in local deviations in the monitoring data and affecting the accuracy of leakage and fatigue damage judgment.

[0072] See Figure 2-3 As shown, the present application further proposes that at least four temperature sensors 400 are provided. The temperature sensors 400 are arranged on both sides of the external movable end 200 of the bellows compensator, with at least two temperature sensors 400 arranged on each side. At least four displacement sensors 300 are provided. The displacement sensors 300 are arranged on both sides of the protective bellows 100 inside the bellows compensator.

[0073] The temperature sensors 400 are located on both sides of the bellows compensator's external movable end 200, with at least two temperature sensors 400 positioned on each side, forming a symmetrical monitoring layout. The displacement sensors 300 are located on both sides of the bellows compensator's internal protective bellows 100, with at least two displacement sensors 300 positioned on each side, forming a structure that simultaneously monitors axial and radial displacements. The displacement sensors 300 are located on both sides of the bellows compensator. Through this bilaterally symmetrical layout, the temperature sensors 400 can capture temperature gradient changes on both sides of the movable end, while the displacement sensors 300 can simultaneously capture deformation differences on both sides of the protective bellows, eliminating errors caused by unilateral data collection. For example, two temperature sensors 400 are located on each side of the external movable end 200, forming a four-sensor redundant monitoring system. If one sensor on one side experiences an anomaly, the sensor on the other side can still provide valid data. Two displacement sensors 300 are located on each side of the internal protective bellows 100, enabling simultaneous monitoring of axial expansion and contraction and radial displacement, avoiding the biased nature of single-direction displacement data.

[0074] Specifically, the temperature sensors 400 on both sides of the external movable end 200 can identify local temperature anomalies caused by sealing failure through multi-point coverage, such as the temperature rise caused by the accumulation of leaking media on one side. The displacement sensors 300 on both sides of the internal protective bellows 100 are symmetrically arranged and can synchronously record the displacement differences on both sides of the bellows, such as axial offset and radial distortion caused by uneven soil settlement. When the data processing module collects data, the data of the sensors on both sides are compared by weighted averaging or difference to eliminate environmental interference and extract the true displacement and temperature changes. For example, when the data of the displacement sensor 300 on one side deviates significantly from the other side, it can be determined that there is a local deformation anomaly. The redundant arrangement of the temperature sensors 400 on both sides further reduces the risk of misjudgment due to single-point failure and improves the robustness of the monitoring system. Therefore, by optimizing the number and position of sensors, the complementarity and reliability of multi-dimensional data are enhanced, providing high-precision input for subsequent fatigue damage and leakage risk analysis.

[0075] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0076] The sensor module includes a temperature sensor 400 and a displacement sensor 300. Four temperature sensors 400 are provided, distributed on both sides of the external movable end 200 of the bellows compensator, with two temperature sensors 400 on each side. Four displacement sensors 300 are provided, distributed on both sides of the internal protective bellows 100 of the bellows compensator.

[0077] Specifically, the temperature sensor 400 uses a PT100 platinum resistance temperature sensor with a measurement range of -50°C to 200°C and an accuracy of ±0.1°C. Four temperature sensors 400 are installed symmetrically on either side of the bellows compensator's external movable end 200, two on each side. The displacement sensor 300 uses a linear displacement sensor with a measurement range of 0-100mm and a resolution of 0.01mm. Four displacement sensors 300 are installed symmetrically on either side of the bellows compensator's internal protective bellows 100, two on each side.

[0078] Through the above-mentioned technical solution, this application achieves comprehensive monitoring of the temperature and displacement of key parts of the bellows compensator. The symmetrical arrangement of temperature sensors 400 accurately captures the temperature distribution of the bellows compensator's external movable end 200, facilitating the timely detection of abnormal local temperature rise. The symmetrical arrangement of displacement sensors 300 accurately measures the axial and radial deformation of the bellows, facilitating the assessment of the bellows' fatigue state. This multi-point measurement arrangement improves the reliability and representativeness of the data, providing comprehensive basic data support for subsequent data processing and risk analysis.

[0079] In some of the above-mentioned schemes of this application, the data processing module performs fatigue analysis by collecting the expansion and contraction data of the bellows, but the existing methods are difficult to accurately quantify the fatigue consumption process of the bellows compensator, resulting in insufficient assessment accuracy of the remaining fatigue times and inability to effectively judge the risk of early fatigue damage.

[0080] The present application further proposes that a data processing module collects the expansion and contraction data of the bellows and obtains the fatigue times of the bellows, and determines the remaining fatigue times of the bellows compensator, including: collecting the expansion and contraction data of the bellows in the bellows compensator based on a preset sampling period, obtaining an axial expansion and contraction displacement sequence, dividing the axial expansion and contraction displacement sequence into sliding windows, constructing a displacement change data set within the time window, identifying peaks and valleys, and obtaining tension and compression cycles. The amplitude and average displacement of each pair of tension-compression half-cycles are extracted based on rain flow counting. The elastic modulus of the bellows and the stress-strain conversion coefficient of the compensator structure are collected to obtain the equivalent stress amplitude. According to the equivalent stress amplitude and the SN fatigue life curve corresponding to the bellows material, the standard fatigue cycle number is obtained. The fatigue consumption of the identified tension and compression cycles is accumulated based on the Miner linear damage accumulation method, and compared with the standard fatigue cycle number to obtain the remaining fatigue number.

[0081] Among them, the sliding window division adopts a dynamic window mechanism of fixed time length or event triggering, and the window length is dynamically adjusted according to the design life and operating environment of the compensator. The rain flow counting algorithm combines peak and valley detection with loop closure conditions to convert asymmetric fluctuations into equivalent symmetric cycles. The equivalent stress amplitude is converted into the stress amplitude inside the material by multiplying the elastic modulus and the stress-strain conversion coefficient. The SN curve is fitted based on the material fatigue test data, and the standard fatigue cycle number is determined by interpolation or extrapolation. The Miner linear damage accumulation method superimposes the damage degrees corresponding to different stress amplitudes according to the weights, and determines that the fatigue life is exhausted when the cumulative damage degree reaches 1.

[0082] Specifically, by segmenting the displacement sequence through a sliding window, the bellows' expansion and contraction fluctuation patterns under different operating conditions can be dynamically captured, eliminating noise interference. The rainflow counting algorithm decomposes complex displacement fluctuations into independent tension and compression cycles, accurately quantifying the number of cycles in actual operation. The introduction of elastic modulus and stress-strain conversion coefficients converts displacement data into material stress parameters, matching the input requirements of the SN curve. Damage accumulation calculation based on the SN curve and the Miner criterion comprehensively considers the impact of different stress amplitudes on fatigue life, avoiding the limitations of traditional single threshold judgment. For example, when a fluctuation with a peak of +8mm and a valley of -5mm is detected in the axial expansion and contraction displacement sequence, the rainflow counting algorithm decomposes it into symmetrical cycles with an amplitude of 6.5mm. Combining the elastic modulus of 2.06×10^5MPa and the conversion factor of 0.15, the equivalent stress amplitude is calculated to be 195MPa, corresponding to a standard number of cycles in the SN curve of 1.2×10^5. If the accumulated damage degree in actual operation is 0.85, the remaining fatigue life is 1.8×10^4. This method achieves a refined assessment of fatigue life by combining dynamic data acquisition with material mechanical properties.

[0083] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0084] The data processing module collects expansion and contraction data from the bellows in the bellows compensator based on a preset sampling period, generating an axial expansion and contraction displacement sequence. This axial expansion and contraction displacement sequence is then partitioned into sliding windows to construct a dataset of displacement changes within the time window. Tension and compression cycles are then captured by identifying peaks and valleys.

[0085] Furthermore, the amplitude and average displacement of each pair of tension-compression half-cycles were extracted using the rainflow counting method. Specifically, a four-point rainflow counting method was used, with displacement data points read sequentially to determine whether a closed cycle had formed. For each closed cycle, the amplitude and average displacement values ​​were recorded.

[0086] From this, the elastic modulus of the bellows and the stress-strain conversion coefficient of the compensator structure are collected to obtain the equivalent stress amplitude. For example, the stress-strain conversion coefficient is determined through finite element analysis and the displacement amplitude is converted into a stress amplitude.

[0087] The standard fatigue cycle number is obtained based on the SN fatigue life curve corresponding to the equivalent stress amplitude and the bellows material. The SN curve is obtained through material fatigue testing and indicates the number of cycles the material can withstand under a specific stress level.

[0088] Finally, the fatigue consumption of the identified tension-compression cycles is accumulated using the Miner linear damage accumulation method and compared with the standard number of fatigue cycles to obtain the remaining fatigue life. Specifically, the fatigue damage of each cycle is calculated and added up to obtain the total damage. The total damage is subtracted from 1 to obtain the remaining life ratio. This ratio is then multiplied by the standard number of fatigue cycles to obtain the remaining fatigue life ratio.

[0089] Through the above technical solution, this application realizes the accurate assessment of fatigue damage of corrugated compensators. By collecting real-time expansion and contraction data and performing rain flow counting analysis, the stress cycle under actual working conditions is accurately identified. Combining the material SN curve and Miner cumulative damage theory, the remaining fatigue life is reliably predicted. This method avoids the limitations of traditional static threshold judgment, can dynamically track the evolution of fatigue damage, and promptly identify potential risks. At the same time, the solution takes into account the actual stress amplitude distribution, which is more accurate than simple cycle count statistics. This provides a reliable basis for preventive maintenance and replacement decisions of corrugated compensators, effectively reducing the risk of sudden failures.

[0090] In some of the above-mentioned schemes of this application, a method of evaluating the compensator status by collecting bellows expansion and contraction data and calculating the remaining fatigue times is proposed. However, due to the nonlinear cumulative characteristics of the compensator's fatigue consumption in actual operation, simply calculating the remaining fatigue times cannot directly determine whether the current state is close to the critical damage state, which can easily lead to maintenance delays or misjudgment risks.

[0091] This application further proposes that a data processing module compares the remaining fatigue count with a remaining fatigue threshold and, based on the comparison result, determines whether the bellows compensator is at risk of fatigue damage. If the remaining fatigue count is less than the remaining fatigue threshold, the data processing module determines that the bellows compensator is at risk of fatigue damage. If the remaining fatigue count is greater than or equal to the remaining fatigue threshold, the data processing module determines that the bellows compensator is not at risk of fatigue damage.

[0092] The residual fatigue threshold is set based on the fatigue life safety factor of the bellows material, determined through a combination of material test data and engineering experience. The comparison process utilizes a real-time dynamic threshold adjustment mechanism, dynamically revising the threshold range based on historical data on the compensator's ambient temperature and pressure fluctuations. The risk determination logic incorporates multiple validation mechanisms, including triggering a determination when the residual fatigue count for three consecutive sampling periods is below the threshold.

[0093] Specifically, after the data processing module completes the calculation of the remaining fatigue times based on the rain flow counting method and the Miner linear damage accumulation model, it compares the calculation results with the preset dynamic threshold in real time. When the remaining fatigue times are lower than the dynamically adjusted remaining fatigue threshold, the system automatically triggers the fatigue damage risk mark and passes the risk level data to the early warning module. By matching the threshold with the real-time value of the remaining fatigue times, the problem of misjudgment caused by material performance degradation or environmental mutation is effectively avoided. For example, under working conditions with frequent pressure fluctuations, the remaining fatigue threshold can be adjusted downward based on the average stress amplitude of the previous sampling period, so that the risk judgment is more in line with the actual operating status. After the judgment result is generated, the data storage module is triggered to fully record the displacement and temperature data before and after the judgment to provide data support for subsequent maintenance.

[0094] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0095] The data processing module compares the remaining fatigue count with the remaining fatigue threshold and, based on the comparison result, determines whether the bellows compensator is at risk of fatigue damage. Specifically, if the remaining fatigue count is less than the remaining fatigue threshold, the data processing module determines that the bellows compensator is at risk of fatigue damage. If the remaining fatigue count is greater than or equal to the remaining fatigue threshold, the data processing module determines that the bellows compensator is not at risk of fatigue damage.

[0096] For example, the remaining fatigue threshold can be set to 1000 times. The data processing module calculates that the remaining fatigue count for the bellows compensator is 800 times. Therefore, the data processing module compares 800 times with 1000 times. Since 800 times is less than 1000 times, the data processing module determines that the bellows compensator is at risk of fatigue damage. Furthermore, if the calculated remaining fatigue count is 1200 times, since 1200 times is greater than 1000 times, the data processing module will determine that the bellows compensator is not at risk of fatigue damage.

[0097] Through the above technical solution, the present application can timely detect the fatigue damage risk of the bellows compensator. By setting a reasonable residual fatigue threshold, a warning can be issued in advance before the bellows compensator reaches the fatigue limit, reserving sufficient maintenance time for maintenance personnel. This judgment method based on the number of remaining fatigue times can more accurately reflect the actual use and life of the bellows compensator compared to simple displacement or temperature threshold judgments, effectively avoiding misjudgments and missed judgments. At the same time, this method also provides a reliable basis for the preventive maintenance of the bellows compensator, which helps to extend the service life of the equipment and reduce the risk of sudden failures.

[0098] In some of the above-mentioned solutions of the present application, the data processing module judges the leakage risk through the temperature threshold, but the existing static threshold setting method cannot adapt to changes in ambient temperature, resulting in an increased probability of misjudgment risk in scenarios where the soil temperature fluctuates or the seasonal temperature difference is large. In particular, when the compensator is in an unstable heat exchange state, the fixed threshold is difficult to accurately reflect the abnormal temperature rise caused by actual leakage.

[0099] The present application further proposes that a data processing module collects temperature data and compares it with a temperature threshold to determine whether there is a leakage risk, including: when the temperature data is less than or equal to the temperature threshold, the data processing module determines that there is no leakage risk; when the temperature data is greater than the temperature threshold, the data processing module determines that there is a leakage risk.

[0100] The temperature threshold is obtained through dynamic correction. Specifically, this involves constructing a baseline for normal temperature variation based on temperature data from continuous sampling cycles, performing offset correction based on the external soil temperature, and finally determining the standard temperature maximum and minimum values ​​based on the corrected baseline. The temperature threshold is then generated by adding a safety margin to the standard temperature minimum. The data comparison process uses logical judgment between real-time temperature data and dynamic thresholds. Once the judgment result is triggered, the subsequent feature vector data collection and processing process is linked.

[0101] Specifically, the data processing module first obtains the real-time temperature data of the current temperature sensor 400 measuring point, and then calls the preset dynamic temperature threshold for numerical comparison. If the real-time temperature does not exceed the threshold, it is determined that there is no leakage risk in the current state and the normal monitoring mode is maintained. If the real-time temperature exceeds the threshold, the leakage risk judgment is triggered, and the full collection and feature extraction of all sensor data within the preset time period are started simultaneously. This judgment logic reduces misjudgments caused by environmental interference through a dynamic threshold correction mechanism, and at the same time ensures early warning in the event of a slight temperature rise anomaly based on a safety margin setting. For example, when the soil temperature fluctuates due to the temperature difference between day and night, the dynamic threshold is automatically adjusted according to the correction factor to avoid misjudgment as a leak due to an increase in ambient temperature. When the actual leakage causes the local temperature to rise above the corrected threshold, the risk judgment is immediately triggered. The judgment result further drives the data processing module to perform multi-dimensional feature analysis to ensure the accuracy and timeliness of risk identification.

[0102] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0103] The data processing module collects temperature data and compares it with a temperature threshold to determine if there is a risk of leakage. Specifically, if the temperature data is less than or equal to the temperature threshold, the data processing module determines that there is no leakage risk. If the temperature data is greater than the temperature threshold, the data processing module determines that there is a leakage risk.

[0104] Furthermore, temperature data is collected by at least four temperature sensors 400, located on either side of the bellows compensator's outer movable end 200. For example, two temperature sensors 400 may be located on each side of the bellows compensator's outer movable end 200, for a total of four temperature sensors 400. This allows for comprehensive monitoring of the bellows compensator's temperature changes.

[0105] The data processing module first receives temperature data from temperature sensor 400. This temperature data can be collected in real time or preprocessed. The data processing module then compares the received temperature data with a preset temperature threshold. This threshold can be pre-set based on factors such as the bellows compensator's design parameters and the operating environment.

[0106] Specifically, the data processing module executes the following judgment logic: if the temperature data collected by any temperature sensor 400 is greater than the temperature threshold, it is determined that there is a leakage risk. If the temperature data collected by all temperature sensors 400 are less than or equal to the temperature threshold, it is determined that there is no leakage risk.

[0107] As a preferred embodiment, multiple temperature thresholds can be set, corresponding to different risk levels. For example, a low-risk threshold, a medium-risk threshold, and a high-risk threshold can be set. When the temperature data exceeds the low-risk threshold but does not exceed the medium-risk threshold, it is determined to be a low-risk leak. When the temperature data exceeds the medium-risk threshold but does not exceed the high-risk threshold, it is determined to be a medium-risk leak. When the temperature data exceeds the high-risk threshold, it is determined to be a high-risk leak.

[0108] Through the above technical solution, the present application can promptly detect the leakage risk of the bellows compensator. By comparing the temperature data with the preset temperature threshold, it is possible to quickly determine whether there is a leakage risk, thereby avoiding energy loss and equipment corrosion caused by leakage. In addition, the use of multiple temperature sensors 400 for monitoring improves the comprehensiveness and accuracy of monitoring. By setting multi-level temperature thresholds, the degree of leakage risk can be judged more precisely, which is conducive to taking corresponding preventive and treatment measures. This method is simple and intuitive, easy to implement, and can effectively improve the operational safety and reliability of the bellows compensator.

[0109] In some of the above-mentioned solutions of the present application, the traditional static temperature threshold does not take into account the impact of the external ambient temperature fluctuation of the compensator on leakage detection, which may result in misjudgment or missed detection when the soil temperature changes.

[0110] This application further proposes that the data processing module obtain the temperature threshold through the following method: collecting temperature data for N consecutive sampling cycles, constructing a reference temperature curve for each sensor measurement point, and obtaining the normal temperature change baseline of the compensator in a leak-free state. The soil temperature outside the compensator is collected, and an offset correction is performed on the normal temperature change baseline based on the soil temperature. The correction factor is obtained through linear regression. The standard temperature maximum and standard temperature minimum values ​​are obtained based on the corrected normal temperature change baseline. The temperature threshold is obtained based on the standard temperature minimum and a safety margin. The safety margin is 2-5°C.

[0111] The baseline temperature curve is constructed by statistically generating temperature data from a leak-free state over N consecutive cycles, ensuring that the baseline reflects the temperature fluctuation range of the compensator during normal operation. A linear regression correction for soil temperature is performed by mapping the compensator surface temperature to the ambient temperature, eliminating interference from external ambient temperature fluctuations on the baseline curve. The safety margin is determined through a combination of engineering experience and the leakage temperature rise rate. For example, the threshold sensitivity is adjusted within a range of 2-5°C based on the pipeline medium type.

[0112] Specifically, in the absence of leaks, the data processing module continuously collects compensator surface temperature data and generates a baseline temperature curve for each measurement point using a sliding average algorithm. When the ambient soil temperature fluctuates seasonally, a linear regression model is used, using soil temperature as the independent variable and compensator surface temperature as the dependent variable, to calculate the temperature offset and update the baseline curve. This modified baseline curve separates the combined effects of ambient temperature and leakage temperature rise. The extracted standard temperature minimum, combined with a 2-5°C safety margin, forms a dynamically adjusted temperature threshold. For example, when soil temperature drops sharply in winter, the correction factor lowers the baseline temperature minimum to avoid false leak detection in low-temperature environments. Conversely, during high summer temperatures, the baseline temperature minimum is raised to prevent ambient temperature rise from masking leak anomalies. The combination of dynamic thresholds and safety margins prevents minor leaks from being missed and suppresses false alarms caused by environmental interference.

[0113] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0114] The data processing module collects temperature data for 30 consecutive sampling cycles, constructs a baseline temperature curve for each sensor measurement point, and obtains the normal temperature variation baseline of the compensator in a leak-free state. Furthermore, the soil temperature of the environment outside the compensator is collected, and an offset correction is performed on the normal temperature variation baseline based on the soil temperature. The correction factor is obtained through linear regression. Specifically, the correlation coefficient between the soil temperature and the temperature of each measurement point is first calculated. Then, the linear regression equation is fitted using the least squares method to obtain the correction factor. Thus, the standard temperature maximum and standard temperature minimum values ​​are obtained based on the corrected normal temperature variation baseline. For example, the 95th percentile of the corrected baseline data is taken as the standard temperature maximum, and the 5th percentile is taken as the standard temperature minimum. Finally, the temperature threshold is determined based on the standard temperature minimum and a safety margin of 3°C.

[0115] Through the above-mentioned technical solution, this application achieves dynamic adjustment of the temperature threshold of the bellows compensator. By taking into account the influence of the external soil temperature and applying corresponding offset corrections, the obtained temperature threshold is more accurate and reliable. This method can adapt to temperature changes under different environmental conditions, reduce misjudgments caused by environmental factors, and improve the accuracy of leakage risk assessment. At the same time, by setting a safety margin, the reliability and stability of the system are enhanced. This dynamic adjustment mechanism enables the system to better adapt to the actual operating environment and improves the overall performance of the multi-project management and early warning analysis system for bellows compensators.

[0116] In some of the above-mentioned solutions of this application, when the data processing module uses feature vector data groups for risk analysis, due to the single feature dimension and lack of dynamic change information, it is unable to effectively capture the complex dynamic correlations in multi-dimensional time series data, affecting the accuracy of early risk identification.

[0117] This application further proposes that the feature vector data set includes displacement data, temperature data, acceleration change of displacement data, and acceleration change of temperature data. The feature vector data set is established as a time series feature matrix. The number of attention heads is configured, and the time series feature matrix is ​​processed with multi-head time series attention in sequence. Each attention head includes an independent query, key, and value linear mapper to calculate the attention weight of each time frame in the time series to other frames. The representations of all heads are spliced ​​and linearly combined to generate comprehensive time series features.

[0118] Among them, the feature vector data group introduces the acceleration change of displacement data and the acceleration change of temperature data, which can reflect the dynamic change trend of physical quantities and supplement the transient characteristics that static data cannot reflect. The construction of the time series feature matrix aligns multi-source heterogeneous data in the time dimension to form a unified data structure. In multi-head time series attention processing, each attention head generates a different Query, Key, and Value matrix through an independent linear mapper, allowing different heads to focus on association patterns at different time scales. The attention weight calculation quantifies the degree of influence between different time frames through the scaled dot product mechanism, such as the strength of the association between the current moment and the historical moment.

[0119] Specifically, the acceleration change in the feature vector data set is obtained through a differential operation. For example, the acceleration change in displacement data is the second-order difference of the displacement values ​​of adjacent sampling points. The row dimension of the time series feature matrix corresponds to the time step sequence, and the column dimension contains four feature channels: displacement, temperature, and its acceleration change. During multi-head processing, each attention head performs a linear transformation on the time series feature matrix, generating three subspace representations: query, key, and value. The attention weight matrix is ​​obtained by calculating the dot product similarity between the query and key and performing softmax normalization. This matrix reflects the strength of association between different time frames. The weight matrix is ​​multiplied by the value matrix to obtain the attention output of each head. Finally, the outputs of multiple heads are concatenated and linearly projected to form a comprehensive time series feature. For example, when processing high-frequency, small deformations, one attention head focuses on capturing the short-term sudden change pattern of displacement acceleration, while another head focuses on the long-term cumulative trend of temperature change. Through the collaborative analysis of multi-dimensional time series features, the accuracy of identifying complex risks is improved.

[0120] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0121] The data processing module processes the feature vector data group based on the multi-head temporal attention mechanism to determine the comprehensive temporal features, including the following steps:

[0122] First, the feature vector data set includes displacement data, temperature data, acceleration change of displacement data, and acceleration change of temperature data.

[0123] Secondly, the feature vector data set is built into a time series feature matrix. Specifically, the collected data are arranged in chronological order to form a matrix, where each row represents a time point and each column represents a feature.

[0124] Furthermore, the number of attention heads can be configured. For example, eight attention heads can be set. The time series feature matrix is ​​sequentially processed by multi-head time series attention, with each attention head including independent query, key, and value linear mappers. In this way, each attention head can focus on different feature combinations.

[0125] Then, the attention weight of each time frame in the time series to other frames is calculated. Specifically, the similarity is calculated by the dot product of the query and the key, and then the attention weight is normalized by the softmax function.

[0126] Finally, the representations of all heads are concatenated and linearly combined to generate a comprehensive temporal feature. This step combines the outputs of multiple attention heads to obtain a comprehensive feature representation containing multi-dimensional information.

[0127] Through the above technical solution, the present application can effectively capture the long-term temporal dependencies and feature interaction information in the multi-dimensional sensor data of the corrugated compensator. This improves the accuracy of identifying early signs of leakage and fatigue degradation of the corrugated compensator and reduces false positives and missed reports. Furthermore, the solution can adapt to complex scenarios such as high-frequency micro-deformations or slow temperature rise, improving the robustness of the early warning system under different operating conditions. At the same time, the cloud platform-based architecture realizes unified data management and risk normalization analysis across regions and projects, providing strong support for systematic maintenance and resource scheduling.

[0128] In some of the above-mentioned schemes of this application, when the data processing module processes the feature vector data group based on the multi-head temporal attention mechanism, the dependency relationship between different time frames is difficult to be fully captured, resulting in limited accuracy of temporal feature extraction and affecting the accuracy of subsequent risk warnings.

[0129] This application further proposes that for the hth attention head, its attention output is.

[0130]

[0131] Among them, Q (h) K (h)T Represents the similarity scoring matrix. represents the scaling factor, V (h) represents the Value matrix, Q (h) represents the Query matrix, K (h) represents the Key matrix, softmax represents the normalization function, Attention (h) (Q (h) ,K (h) ,V (h) ) represents the attention output result of the h-th attention head.

[0132] Among them, the softmax result is the attention weight matrix: ; in, represents the degree to which the i-th time frame in the h-th attention head sequence pays attention to the j-th frame when generating its output representation, Represents the Query vector of the i-th time step; represents the Key vector of the jth time step; j represents the target time frame number for calculating attention, Represents the index variable used for normalization in the softmax denominator; T represents the length of the time series.

[0133] Specifically, the product of the Query matrix and the Key matrix generates the original attention score matrix, and the scaling factor controls the gradient stability. The normalization function converts the original score into a probability distribution, so that the attention weight falls in the interval [0,1]. After the Value matrix is ​​multiplied by the attention weight matrix, the weighted feature representation is output. The similarity between the Query vector of each time frame and the Key vector of all time frames is calculated to generate a dynamic attention distribution. Through parallel calculation of multiple attention heads, the temporal dependency patterns of different subspaces are captured, and the outputs of each head are finally spliced ​​to form a comprehensive temporal feature. For example, the scaling factor is the inverse square root of the input vector dimension. When the input dimension is 64, the scaling factor is set to 1 / 8, which effectively prevents the dot product result from being too large and causing the gradient to vanish. The normalization function performs an exponential operation on each row and then sums and normalizes it row by row to ensure that the weight distribution in the time dimension meets the probabilistic characteristics.

[0134] As a preferred embodiment, the solution of this application is implemented as follows: During the multi-head temporal attention processing, a temporal feature matrix containing displacement data, temperature data, and their acceleration changes is input into the multi-head attention layer. Each attention head is configured with an independent linear mapping matrix, where the query matrix dimensions are set to 64×128, the key matrix dimensions are set to 128×128, and the value matrix dimensions are set to 128×64. A similarity score is calculated between the query vector of each time frame and the key vectors of all time frames through matrix multiplication. The result is divided by a scaling factor of √128 and then a row-level softmax function is applied to generate an attention weight matrix. For the third attention head, when calculating the output for the fifth time frame, its attention weights for time frames 2 to 7 are 0.18, 0.22, 0.15, 0.12, 0.20, and 0.13, respectively, indicating that this head focuses on capturing feature interactions in the medium-term time period. The outputs of all attention heads are concatenated and linearly combined through a 256-dimensional fully connected layer, ultimately generating a 256-dimensional comprehensive temporal feature vector for subsequent risk classification.

[0135] Through the above technical solution, this application effectively improves the time series correlation modeling capabilities of risk warning. By using a multi-head attention mechanism to capture the dynamic correlation patterns of displacement and temperature characteristics over different time spans, it addresses the problem of traditional static threshold methods' insufficient sensitivity in identifying slowly varying abnormal signals. This method automatically learns the weight distribution of key frames in a time series, eliminating redundant noise interference. It can accurately identify abnormal characteristics even in the early stages of pipeline media leakage or material fatigue, reducing the rate of misjudgment under complex working conditions.

[0136] In some of the above-mentioned schemes in this application, the data processing module processes the feature vector data group through a multi-head temporal attention mechanism to generate comprehensive temporal features. However, when calculating the degree of correlation between different time frames, traditional methods are difficult to effectively capture the dynamic interaction relationship of multidimensional features in the time series, resulting in uneven distribution of attention weights, affecting the accuracy of the comprehensive temporal features, and thus reducing the reliability of risk warnings.

[0137] This application further proposes that for the h-th attention head, its attention output is calculated through a similarity scoring matrix, and is normalized after adjusting the scaling factor, and the output results of each attention head are spliced ​​together to generate a comprehensive time series feature.

[0138] The similarity score matrix is ​​calculated by multiplying the query matrix by the key matrix. The scaling factor is set to the square root of the key vector dimension to control gradient stability. The normalization function uses row-wise softmax to ensure that the sum of the attention weights for each time frame is 1. The attention weight matrix is ​​generated by exponentially normalizing the dot product of the vectors between time frames and is used to represent the strength of associations at different time steps.

[0139] Specifically, by mapping the time series feature matrix into independent Query, Key, and Value vectors, each attention head can learn time series association patterns of different dimensions. For example, when the acceleration change of the displacement data suddenly changes in a certain time frame, the dot product of the corresponding Key vector and the Query vector of the subsequent time frame will generate a higher similarity score, which will enhance the correlation between the time frame and the subsequent frames in the attention weight matrix, thereby highlighting the impact of the sudden change event when the Value matrix is ​​weighted and summed. At the same time, the scaling factor avoids the gradient saturation problem of the softmax function by reducing the magnitude of the dot product value in the high-dimensional space. The independent mapping of each attention head enables the model to capture different feature patterns such as temperature change trends and periodic fluctuations in displacement in parallel. The spliced ​​comprehensive time series features are further fused with multi-dimensional information through the fully connected layer, ultimately improving the accuracy of risk level classification.

[0140] In some embodiments of the present application, the early warning module is based on a fully connected network and a softmax activation function and performs risk warning according to comprehensive time series features.

[0141] p risk =softmax(W c y fused +b c ).

[0142] Among them, p risk represents the predicted probability of each risk level, y fused represents the comprehensive time series characteristics, W c represents the weight of the fully connected neural network classifier, b c represents the bias vector, and p risk ∈R C , C represents the number of risk levels, C = 4. The early warning module uses the risk level with the highest predicted probability as the output risk level and simultaneously outputs the judgment result, which includes whether there is a leakage risk and / or whether there is a fatigue damage risk.

[0143] As a preferred embodiment, the solution of the present application is specifically implemented as follows: the early warning module adopts a fully connected neural network architecture with three hidden layers, and the input layer receives a 256-dimensional comprehensive time series feature vector generated by a multi-head time series attention mechanism. Each hidden layer is configured with 128, 64, and 32 neuron units in turn, and the activation function adopts ReLU. The output layer is provided with four neuron nodes, corresponding to the four levels of no risk, low risk, medium risk, and high risk, and the probability distribution of each risk category is calculated by the softmax function. In the model training stage, the cross entropy loss function with L2 regularization is adopted, the optimizer uses the Adam algorithm, and the initial learning rate is set to 0.001. When the system is running, after the comprehensive time series feature vector is calculated by forward propagation, the risk level corresponding to the maximum probability of the output layer will trigger a warning signal. At the same time, the risk level is combined with the fatigue damage risk flag and the leakage risk flag to form a status code containing a four-tuple, which is pushed to the operation and maintenance terminal in real time through the cloud platform interface.

[0144] Through the above technical solution, this application effectively solves the technical defects of the traditional static threshold method in the rough classification of risk levels and high error rate. By mapping the multi-dimensional time series characteristics to the nonlinear classification space, it can accurately capture the coupling relationship between temperature, displacement and their dynamic change patterns, and realize the coordinated discrimination of leakage risk and fatigue damage risk. The risk level classification mechanism based on probability output improves the recognition sensitivity of early minor abnormal conditions, avoids the problem of missed reports caused by single threshold triggering, and provides a quantitative basis for operation and maintenance decision-making through the gradient output of risk levels.

[0145] In some of the above-mentioned solutions of this application, after the cloud platform-based corrugated compensator multi-project management and early warning analysis system has completed risk assessment, if it is determined that there is no leakage risk and no fatigue damage risk, it still needs to continue to collect temperature and displacement data. The amount of high-frequency sampling data is huge, and direct transmission and storage will occupy a large amount of cloud platform resources, resulting in reduced data transmission efficiency and increased storage costs. However, existing solutions do not effectively compress redundant data in a risk-free state, and cannot balance the contradiction between data integrity and resource consumption.

[0146] The present application further proposes a data compression module, which is configured to perform aggregation processing on the collected temperature data and displacement data when the data processing module determines that there is no leakage risk and no fatigue damage risk, and use sliding average to generate representative sampling points to replace the original high-frequency sampling sequence. When the data processing module determines that there is a leakage risk or a fatigue damage risk, the collected temperature data and displacement data are recorded uncompressed. The sampled data sequence is structured and stored, and the structured coding includes difference coding, interval statistical coding, and lightweight time series coding.

[0147] The sliding average generates representative sampling points by calculating the mean of data within a preset time window. The window length is dynamically adjusted based on the sampling frequency. For example, raw data is aggregated into one representative point every 5 seconds. Difference coding reduces data redundancy by recording the change between adjacent sampling points instead of the original absolute value. Interval statistical coding extracts the maximum, minimum, and standard deviation within a specified time period to form compressed statistical features. Lightweight time series coding uses a binary bit allocation strategy to map continuous timestamps to offsets relative to the start time, reducing the storage space of time series labels.

[0148] Specifically, when the system determines there is no risk, the data compression module initiates a sliding average process, replacing the original high-frequency sequence with low-density representative points, achieving a compression ratio of over 80%. These representative points preserve overall trend characteristics and avoid loss of detail. If a risk is determined, the original data is recorded in its entirety, ensuring the integrity of the high-frequency information required for subsequent analysis. During structured encoding, difference encoding converts continuous changes in temperature or displacement into a difference sequence, which is then further compressed using Huffman coding. Interval statistical encoding divides data within each 10-minute period into independent intervals, extracting statistical indicators to replace the original sequence. Lightweight time series encoding uses a fixed bit width to store time offsets, such as 32-bit integers representing second-level offsets. This combined compression method reduces data volume to 15%-20% of its original size in a risk-free state, reducing storage and transmission load on the cloud platform while maintaining data parseability. When the system detects a risk event, the compression module automatically switches to non-compression mode to ensure that fault characteristics are not smoothed or omitted, providing a complete data foundation for subsequent multi-head time series attention analysis.

[0149] As a preferred embodiment, the solution of the present application is specifically implemented as follows: the data compression module executes a differentiated data processing process based on the risk determination result output by the data processing module. When it is determined that there is no leakage risk and no fatigue damage risk, the currently collected temperature data and displacement data are aggregated, and a sliding window mechanism is used to perform a sliding average operation with a window length of 30 seconds to generate a representative sampling point sequence containing mean and extreme values, replacing the original high-frequency sampling data stream of 10 times per second. For the sampled data sequence, a structured coding method is used for storage, in which the difference coding realizes data compression by recording the numerical difference between adjacent sampling points, the interval statistical coding divides the data within every 5 minutes into independent intervals and records its mean, maximum and minimum values, and the lightweight time series coding uses Delta coding to record the timestamp difference. When it is determined that there is a risk, the original temperature and displacement data are fully recorded in an uncompressed form, while the difference coding and interval statistical coding in the structured coding are still performed, but the timestamps and numerical sequences of all original data points are retained.

[0150] Through the above technical solution, this application effectively resolves the contradiction between data storage resource consumption and the integrity of key information in traditional monitoring systems. When the equipment is operating safely, a dynamic data compression algorithm is used to reduce storage pressure while retaining trend characteristics. When a potential risk is detected, it automatically switches to full data recording mode to ensure the complete time series information required for subsequent fault analysis. This solution further achieves the unification of data storage formats through structured coding, so that compressed data and uncompressed data have the same parsing interface, improving data call and playback efficiency.

[0151] In the above embodiment, a sensor module consisting of temperature and displacement sensors enables real-time acquisition of key physical parameters of the bellows compensator (such as axial expansion and soil temperature changes). Wireless communication modules enable remote, high-frequency, and low-power data transmission, effectively adapting to widely distributed urban heating pipe networks. The data processing module constructs a bellows expansion and displacement sequence based on the sampled data. It uses the rainflow counting method and the Miner fatigue damage model to determine the current fatigue state of the bellows. It then combines temperature data with a dynamic threshold model to identify potential leakage risks. When a risk indicator is triggered, multi-channel sensor data within a specific time window is automatically retrieved. Keyframe change features are extracted using a multi-head temporal attention mechanism to generate a fused, comprehensive temporal feature representation, enabling accurate identification of subtle trend fluctuations or multivariate interactions. Finally, the early warning module outputs a risk level based on the fused representation, achieving a closed-loop architecture from "physical acquisition - intelligent analysis - proactive early warning." Compared to existing solutions that rely on static thresholds and fail to model evolutionary trends, this system improves the ability to identify early fatigue degradation and slowly varying leakage, reduces false alarms, and enables unified management across multiple projects and regions through centralized cloud deployment.

[0152] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or a combination of software and hardware embodiments. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0153] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0154] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A multi-project management and early warning analysis system for ripple compensators based on a cloud platform, characterized by: include: Sensor module, including temperature sensor and displacement sensor; a wireless communication module connected to the sensor module, the wireless communication module being used to transmit temperature data and / or displacement data; a data processing module configured to collect bellows expansion and contraction data and obtain bellows fatigue times to determine the remaining fatigue times of the bellows compensator; collect temperature data and compare it with a temperature threshold to determine whether there is a leakage risk; and, when it is determined that there is a leakage risk or a fatigue damage risk, collect data from all sensors within a preset time period, extract a feature vector data group, and process the feature vector data group based on a multi-head temporal attention mechanism to determine a comprehensive temporal feature; The early warning module is configured to issue a risk early warning based on the comprehensive time series characteristics.

2. The multi-project management and early warning analysis system for ripple compensators based on a cloud platform according to claim 1 is characterized in that: The sensor module includes: There are at least four temperature sensors, which are arranged on both sides of the external movable end of the bellows compensator, with at least two temperature sensors arranged on each side; At least four displacement sensors are provided, and the displacement sensors are arranged on both sides of the protective bellows inside the bellows compensator.

3. The multi-project management and early warning analysis system for ripple compensators based on a cloud platform according to claim 1 is characterized in that: The data processing module collects the expansion and contraction data of the bellows and obtains the fatigue times of the bellows to determine the remaining fatigue times of the bellows compensator, including: The expansion and contraction data of the bellows in the bellows compensator are collected based on a preset sampling period to obtain an axial expansion and contraction displacement sequence. The axial expansion and contraction displacement sequence is divided into sliding windows to construct a displacement change data set within the time window, identify peaks and valleys, and obtain tension and compression cycles. The amplitude and average displacement of each pair of tension-compression half-cycles were extracted based on rainflow counting; Collect the elastic modulus of the bellows and the stress-strain conversion coefficient of the compensator structure to obtain the equivalent stress amplitude; Obtaining a standard number of fatigue cycles according to the SN fatigue life curve corresponding to the equivalent stress amplitude and the bellows material; The fatigue consumption of the identified tension-compression cycles is accumulated based on the Miner linear damage accumulation method, and compared with the standard fatigue cycle number to obtain the remaining fatigue number.

4. The cloud platform-based multi-project management and early warning analysis system for ripple compensators according to claim 3 is characterized in that: After the data processing module determines the remaining fatigue times of the corrugated compensator, it also includes: The data processing module compares the remaining fatigue times with the remaining fatigue threshold, and determines whether the corrugated compensator has a fatigue damage risk based on the comparison result; When the remaining fatigue times are less than the remaining fatigue threshold, the data processing module determines that the corrugated compensator is at risk of fatigue damage; when the remaining fatigue times are greater than or equal to the remaining fatigue threshold, the data processing module determines that the corrugated compensator is not at risk of fatigue damage.

5. The cloud platform-based multi-project management and early warning analysis system for ripple compensators according to claim 4 is characterized in that: The data processing module collects temperature data and compares it with the temperature threshold to determine whether there is a leakage risk, including: When the temperature data is less than or equal to the temperature threshold, the data processing module determines that there is no leakage risk; when the temperature data is greater than the temperature threshold, the data processing module determines that there is a leakage risk.

6. The cloud platform-based multi-project management and early warning analysis system for ripple compensators according to claim 5 is characterized in that: The data processing module obtains the temperature threshold by the following method: The data processing module collects temperature data for N consecutive sampling periods, constructs a reference temperature curve for each sensor measuring point, and obtains a normal temperature change baseline of the compensator in a non-leakage state; Collecting the soil temperature of the external environment of the compensator, and performing an offset correction on the normal temperature change baseline based on the soil temperature, wherein the correction factor is obtained by linear regression; The standard temperature maximum and standard temperature minimum are obtained according to the corrected normal temperature change baseline, and the temperature threshold is obtained according to the standard temperature minimum and a safety margin, where the safety margin is 2-5°C.

7. The cloud platform-based multi-project management and early warning analysis system for ripple compensators according to claim 6 is characterized in that: The data processing module processes the feature vector data group based on the multi-head temporal attention mechanism to determine the comprehensive temporal features, including: The feature vector data set includes displacement data, temperature data, acceleration change of displacement data and acceleration change of temperature data; Establishing the feature vector data set as a time series feature matrix; Configure the number of attention heads and perform multi-head temporal attention processing on the time series feature matrix in sequence. Each attention head includes an independent Query, Key, and Value linear mapper. Calculate the attention weight of each time frame to other frames in the time series; The representations of all heads are concatenated and linearly combined to generate comprehensive temporal features.

8. The cloud platform-based multi-project management and early warning analysis system for ripple compensators according to claim 7 is characterized in that: The data processing module processes the feature vector data group based on the multi-head temporal attention mechanism to determine the comprehensive temporal features, and further includes: For the h-th attention head, its attention output is: ； in, Represents the similarity scoring matrix; represents the scaling factor, represents the Value matrix, represents the Query matrix, represents the Key matrix, represents the normalization function, Represents the attention output result of the h-th attention head; Among them, the softmax result is the attention weight matrix: ； in, represents the degree to which the i-th time frame in the h-th attention head sequence pays attention to the j-th frame when generating its output representation, Represents the Query vector of the i-th time step; represents the Key vector of the jth time step; j represents the target time frame number for calculating attention, Represents the index variable used for normalization in the softmax denominator; T represents the length of the time series.

9. The cloud platform-based multi-project management and early warning analysis system for ripple compensators according to claim 8 is characterized in that: When the early warning module issues a risk early warning based on the comprehensive time series characteristics, it includes: The early warning module performs risk warning based on a fully connected network and a softmax activation function and according to the comprehensive time series features; ； in, represents the predicted probability of each risk level, represents the comprehensive time series characteristics, represents the weight of the fully connected neural network classifier, represents the bias vector, and , C represents the risk level number, C=4; The early warning module uses the risk level with the highest predicted probability as the output risk level and simultaneously outputs a determination result, wherein the determination result includes whether there is a leakage risk and / or whether there is a fatigue damage risk.

10. The multi-project management and early warning analysis system for ripple compensators based on a cloud platform according to claim 1 is characterized in that: Also includes: a data compression module configured to, when the data processing module determines that there is no leakage risk and no fatigue damage risk, perform aggregation processing on the collected temperature data and displacement data, and use a sliding average to generate representative sampling points to replace the original high-frequency sampling sequence; and, when the data processing module determines that there is a leakage risk or a fatigue damage risk, keep the collected temperature data and displacement data uncompressed and recorded; The sampled data sequence is structured-coded and then stored, wherein the structured coding includes difference coding, interval statistics coding and lightweight time series coding.

Images