Database-based boiler heat-resistant alloy hardness monitoring method and system

Abstract

The invention provides a boiler heat-resistant alloy hardness monitoring method and system based on a database, and is applied to the technical field of industrial equipment nondestructive testing and industrial data processing. The monitoring method comprises the following steps: acquiring boiler multi-source data to construct a feature vector, and utilizing K-means clustering to separate working conditions; establishing a hardness model in combination with the decision tree and performing structured storage; calculating a hardness slope through a sliding window to identify an abnormal trend; a spatial distribution map is constructed to retrieve neighborhood data, similarity is calculated, a difference report is generated when the similarity is lower than a threshold value, and accurate monitoring of the boiler hardness is achieved; according to the invention, high-precision dynamic monitoring of the hardness of the heat-resistant alloy of the boiler can be realized, local abnormity and systematic change are effectively distinguished, the early fault early warning capability is improved, and the problems of data island, poor real-time performance and space correlation deficiency in traditional detection are solved.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing of industrial equipment and industrial data processing technology, and in particular to a database-based method and system for monitoring the hardness of boiler heat-resistant alloys. Background Technology

[0002] Boiler heating surface tubes operate under high temperature and high pressure for a long time, and their material properties will degrade due to creep and metallographic structure evolution. Hardness, as an indicator of material mechanical properties, is directly related to the structural integrity and operational safety of the equipment.

[0003] In existing technologies, Leeb hardness testers are typically used to conduct on-site hardness testing of areas such as boiler superheaters, reheaters, and water-cooled walls during shutdown and maintenance. However, this method has significant limitations in practical applications. First, for small-diameter, thin-walled pipe fittings, insufficient support stiffness makes the impact rebound signal susceptible to vibration interference, resulting in low measurement results and poor repeatability. Second, traditional testing relies solely on measured values ​​at a single time point, lacking fusion analysis with real-time operating data, thus failing to achieve continuous state assessment during operation. Furthermore, the test data is mostly recorded in discrete form, without a unified data management architecture, making it difficult to effectively distinguish data characteristics between different areas. More importantly, when an abnormal hardness is observed at a monitoring point, existing technologies cannot determine whether the abnormality is caused by local defects or regional degradation triggered by systemic overheating, lacking a neighborhood comparison mechanism based on spatial topological relationships, thus limiting the accuracy of fault diagnosis.

[0004] Therefore, there is an urgent need for a hardness monitoring method that can combine high-precision physical measurement with multi-source data modeling and introduce spatial correlation analysis to improve the comprehensiveness and reliability of the condition assessment of boiler heat-resistant alloy materials.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a database-based method and system for monitoring the hardness of boiler heat-resistant alloys. It aims to solve the technical problems in existing technologies, such as low measurement accuracy of boiler heat-resistant alloy hardness testing under small pipe diameter conditions, fragmented multi-source monitoring data forming information silos, lack of real-time soft measurement capabilities during operation, and inability to achieve spatial correlation diagnosis. By constructing a dual-closed-loop monitoring system combining physical calibration and digital twins, it integrates rigid support coupling technology, metallographically corrected hardness conversion models, K-means clustering-based operating condition separation, region-specific dynamic modeling, sliding window time-series trend analysis, and quadtree spatial topology correlation analysis. This enables high-precision perception, real-time simulation, abnormal trend warning, and intelligent identification of localized deterioration of the hardness state of the boiler heating surface alloy material throughout its entire life cycle.

[0007] This invention provides a database-based method for testing the hardness of boiler heat-resistant alloys, the method comprising: Acquire hardness test data of boiler superheater, reheater and water-cooled wall area, acquire operating condition data of area in corresponding operating cycle, operating condition data including temperature load data and working pressure data, associate and integrate hardness test data and operating condition data to generate multi-dimensional feature vector containing hardness value, temperature coefficient and pressure parameter. The K-means clustering algorithm is used to iteratively calculate the multidimensional feature vectors, and the data is separated into classification datasets based on the distribution characteristics of the hardness values, temperature coefficients and pressure parameters contained in the multidimensional feature vectors. Temperature load features and working condition features are extracted from the region labels in the classification dataset. A region-specific hardness model is trained using a decision tree algorithm. The region-specific hardness model is then used to perform feature mapping on historical test data and store it in a structured database. The system extracts hardness time series data arranged in chronological order from a structured database, calculates the slope of hardness change using a sliding window, and marks time periods in which the slope of hardness change exceeds a preset abnormal threshold as abnormal trend data. Based on the spatial location information in the abnormal trend data, and using the indexing mechanism to construct a spatial distribution map reflecting the data density, and to retrieve the adjacent area data of the area corresponding to the abnormal trend data in the spatial distribution map; The similarity of hardness values ​​between the region corresponding to the abnormal trend data and the region corresponding to the adjacent data is calculated. When the hardness value similarity is lower than the preset similarity threshold, an analysis report containing the difference features is generated to determine the safety status of the boiler heat-resistant alloy.

[0008] In some optional embodiments, the hardness test data is obtained through the following calibration process: For different grades of heat-resistant alloys, a Leeb-Britt hardness conversion model is pre-established based on different heat treatment states and different metallographic structures. For pipe wall samples with a mass less than a preset mass threshold, the pipe wall samples are coupled to a rigid support platform with a mass greater than a preset reference mass through a coupling medium to obtain the Leeb hardness value. The Leeb hardness value is input into the Leeb-Britt hardness conversion model, and the Brinell hardness value is calculated and used as hardness test data.

[0009] In some optional embodiments, the step of coupling the tube wall sample to a rigid support platform with a mass greater than a preset reference mass via a coupling medium for Leeb hardness measurement further includes: Vaseline was used as the coupling medium and coated on the contact surface between the tube wall sample and the rigid support platform to form a coupling layer; Press the tube wall sample firmly onto the rigid support platform until the coupling layer is of uniform thickness and free of air bubbles. An impact force was applied in a direction perpendicular to the surface of the tube wall sample to collect the Leeb hardness signal.

[0010] In some optional embodiments, the step of pre-establishing a Leeb-Britt hardness conversion model based on different heat treatment states and different metallographic structures further includes: Metallographic images of heat-resistant alloys with different aging degrees were collected, and the characteristics of pearlite spheroidization level were extracted; The basic conversion formula is modified based on the characteristics of pearlite spheroidization level, and a Leeb-Britt hardness conversion relationship model including metallographic correction coefficients is generated.

[0011] In some optional embodiments, the step of iteratively calculating the multidimensional feature vector using the K-means clustering algorithm includes: The number of clusters is set to 3, corresponding to the three structural parts: superheater, reheater, and water-cooled wall. Calculate the Euclidean distance between each data point and the initial cluster center, and assign each data point to the cluster containing the initial cluster center that is closest to it; In each iteration, the centroid of each cluster is recalculated until the distance the cluster center moves is less than the preset convergence threshold.

[0012] In some optional embodiments, the step of separating the data into categorical datasets based on the distribution characteristics of the hardness values, temperature coefficients, and pressure parameters contained in the multidimensional feature vector includes: When the average hardness value of a cluster in the clustering results deviates from the theoretical hardness range of the corresponding boiler component by more than the preset standard deviation, the weight of the temperature coefficient in the multidimensional feature vector is adjusted and the K-means clustering algorithm is re-executed to iterate the multidimensional feature vector.

[0013] In some optional embodiments, the step of training a region-specific hardness model using a decision tree algorithm includes: A region-specific hardness model is constructed using a classification and regression tree algorithm; The ratio of temperature to load is selected as the splitting attribute of the tree node; The information gain ratio at different split points is calculated, and the point with the largest information gain ratio is selected as the split threshold. The classification dataset is then recursively divided into a high-temperature, high-load subset and a low-temperature, low-load subset.

[0014] In some optional embodiments, the step of training a region-specific hardness model using a decision tree algorithm further includes: At the leaf nodes of the decision tree, a multivariate nonlinear regression equation is used to fit the predicted hardness values. The multivariate nonlinear regression equation includes a temperature coefficient term, a load coefficient term, and a material composition coefficient term. The temperature coefficient term is directly proportional to the negative exponent of temperature, and the load coefficient term is directly proportional to the positive exponent of pressure.

[0015] In some optional embodiments, the step of calculating the slope of the hardness change using a sliding window includes: Set the duration of the sliding window to the preset window duration; Within each sliding window, the least squares method is used to linearly fit the change of hardness time series data over time; The slope of the fitted line obtained from the linear fit is used as the slope of the hardness change at that sliding window moment.

[0016] In some optional embodiments, the step of marking time periods in which the slope of hardness change exceeds a preset abnormal threshold as abnormal trend data includes: Compare the slope of the hardness change with the preset normal attenuation rate range; When the slope of the hardness change is negative and the absolute value of the slope of the hardness change is greater than the preset rapid decrease threshold, the time period is determined to be the stage of abnormal hardness decrease.

[0017] In some optional embodiments, the step of constructing a spatial distribution map reflecting data density using an indexing mechanism includes: The spatial region of the boiler's heating surface is divided into a multi-level grid structure; Assign a unique quadtree node index to each grid in a multi-level grid structure; The number of hardness test data points falling into each grid is counted, and the data density value is calculated by combining the grid area; A spatial distribution map is generated using a heatmap rendering algorithm based on the data density values.

[0018] In some optional embodiments, the step of retrieving neighboring region data includes: The preset search radius range is determined by taking the grid node corresponding to the region of abnormal trend data as the center; Use the quadtree node index to traverse the neighboring grid nodes within the preset search radius; Historical detection data corresponding to neighboring grid nodes are extracted from the structured database as adjacent area data.

[0019] In some optional embodiments, the step of calculating the similarity of hardness values ​​includes: Construct time series vectors for regions corresponding to abnormal trend data and time series vectors for regions corresponding to adjacent region data, respectively. Calculate the Pearson correlation coefficient between the time series vector of the region corresponding to the abnormal trend data and the time series vector of the region corresponding to the data of the adjacent region; The root mean square error of the time series vector of the region corresponding to the abnormal trend data and the time series vector of the adjacent region at the corresponding time point is calculated in parallel. When the Pearson correlation coefficient is less than 0.75 and the root mean square error is greater than the preset error limit, the hardness value similarity is determined to be lower than the preset similarity threshold.

[0020] In some optional embodiments, the step of determining the safety status of the boiler heat-resistant alloy includes: Calculate the rate of change of hardness and the current average hardness in the superheater, reheater, water-cooled wall and economizer regions respectively; Weighting coefficients are assigned based on the importance of each region in the boiler system; The overall hardness index is obtained by weighted summation of the current average hardness values ​​of each region. The overall hardness index is compared with a safety threshold set based on the material fatigue limit to determine the safety status level.

[0021] In some optional embodiments, the above method further includes a prediction step: Based on the historical hardness evolution trend in the structured database, the Monte Carlo simulation algorithm is used for iterative calculation; Simulate the hardness decay trajectory within a preset time period, calculate the probability that the hardness value will drop to the failure threshold, and generate a suggested maintenance schedule based on the probability.

[0022] This invention provides a database-based boiler heat-resistant alloy hardness testing system to implement the aforementioned database-based boiler heat-resistant alloy hardness testing method. The system includes: The multi-source data aggregation module is configured to acquire hardness test data of each heating surface area of ​​the boiler, acquire operating condition data of the area in the corresponding operating cycle, including temperature load data and working pressure data, and associate and integrate the hardness test data and operating condition data to generate a multi-dimensional feature vector, and execute the K-means clustering algorithm to separate the multi-dimensional feature vector into independent datasets corresponding to the superheater, reheater and water-cooled wall respectively according to the distribution characteristics of physical parameters. The structured modeling engine is configured to extract working condition features from independent datasets, run decision tree algorithms to train region-specific hardness models, and use region-specific hardness models to complete feature mapping and structured storage of historical data. The time series trend analyzer is configured to perform sliding window operations on stored time series data, calculate the slope of hardness change, and mark periods in which the slope of hardness change exceeds a preset abnormal threshold as abnormal trend objects. The spatial topology association unit is configured to establish a spatial distribution map index that reflects data density, retrieve spatial adjacent area data of abnormal trend objects based on the spatial distribution map index, and trigger a difference analysis report when the hardness similarity between the abnormal trend object and the spatial adjacent area data is lower than a preset standard. The holographic visualization interface can be configured to parse the difference analysis report, calculate the overall safety index, and render a 3D boiler safety status view that identifies abnormal areas.

[0023] In some optional embodiments, the system further includes a precision calibration terminal communicatively connected to the multi-source data aggregation module, the precision calibration terminal comprising: The physical coupling device includes a rigid support platform with a mass greater than a preset reference mass and an acoustic coupling layer disposed on the surface of the rigid support platform. The physical coupling device is used to fix small-diameter samples to eliminate measurement rebound error. The metallographic correction unit stores a Leeb-Britt hardness conversion table based on different pearlite spheroidization levels. The metallographic correction unit is configured to convert the physical signals collected by the physical coupling device into calibrated Brinell hardness digital signals and transmit them to the multi-source data aggregation module.

[0024] In some optional embodiments, the spatial topology association unit has a built-in multi-level grid indexer, which is configured as follows: The physical space of the boiler heating surface is discretized into a quadtree node structure; In each node of the quadtree node structure, the hardness statistical feature value of the region and the physical neighborhood pointer are stored; When a query request for an abnormal trend object is received, the geometrically adjacent and thermally symmetrical grid nodes are directly locked through the physical neighborhood pointer.

[0025] In some optional embodiments, the holographic visualization interface integrates a dynamic risk rendering engine, which is configured as follows: The remaining life prediction value is calculated based on the hardness decay rate of each region; The remaining life prediction value is mapped onto the 3D model using color gradients. For areas with normal hardness values ​​but abnormal slope of change, the dynamic risk rendering engine uses high-frequency flashing textures for early warning marking.

[0026] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention.

[0027] The database-based method and system for monitoring the hardness of boiler heat-resistant alloys provided by this invention have the following beneficial effects: This invention achieves dynamic monitoring of the hardness of boiler heat-resistant alloys throughout their entire lifecycle by integrating high-precision physical testing with multi-source data modeling. Rigid support and metallographic correction techniques improve the accuracy of on-site measurements for small-diameter pipes; combining K-Means clustering and decision tree regression to construct region-specific models enhances the adaptability of soft hardness measurements; and the introduction of sliding window time-series analysis and quadtree spatial topology correlation mechanisms enables accurate identification and false alarm suppression of local anomaly areas. This effectively addresses the shortcomings of traditional methods in terms of real-time performance, data silos, and spatial correlation analysis, significantly improving the reliability of hardness degradation trend early warning and the scientific basis of maintenance decisions. Attached Figure Description

[0028] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0029] Figure 1 This is a flowchart of a database-based boiler heat-resistant alloy hardness monitoring method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a database-based boiler heat-resistant alloy hardness monitoring system according to an embodiment of the present invention. Detailed Implementation

[0030] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0031] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0032] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined. Therefore, the actual execution order may change depending on the specific circumstances.

[0033] Under high-temperature and high-pressure operating conditions, the mechanical property evolution of boiler heat-resistant alloy materials is influenced by multiple factors, including temperature, stress, and microstructure evolution. Hardness changes are not only related to the current operating conditions but also exhibit time-cumulative effects and spatial distribution characteristics. By collecting data from multiple sensors to construct a multi-dimensional feature vector containing hardness, temperature, pressure, and time-varying parameters, a comprehensive characterization of the material's state can be achieved. Unsupervised clustering algorithms are used to automatically partition high-dimensional data based on the inherent distribution patterns of physical parameters, effectively separating typical operating modes in different heated regions and providing a classification basis for subsequent modeling. Furthermore, a decision tree method is employed to segmentally model the hardness response behavior under nonlinear operating conditions, establishing predictive relationships adapted to local characteristics within different operating condition subspaces and improving the model's generalization ability. For the performance degradation trend over time, a sliding window combined with linear regression is used to extract slope features, quantifying the dynamic rate of material softening and identifying abnormal evolution processes deviating from the normal decay trajectory. Further, a spatial topology analysis mechanism is introduced, mapping monitoring points to a grid index with clearly defined geometric neighborhood relationships. Based on a quadtree structure, efficient spatial querying and neighborhood data retrieval are achieved. By calculating the similarity indicators (such as Pearson correlation coefficient) and numerical differences (such as root mean square error) between the target area and surrounding areas over time series, local anomalies and systematic changes can be effectively distinguished, eliminating misjudgments caused by environmental fluctuations or measurement interference. This method, by integrating temporal dynamic analysis and spatial correlation verification, achieves a leap from single-point detection to regional collaborative diagnosis, improving the accuracy and reliability of fault identification while ensuring data consistency.

[0034] like Figure 1 As shown in the figure, this invention provides a database-based method for testing the hardness of boiler heat-resistant alloys. The method includes the following steps: Step S100: Acquire multi-source sensor data and construct a multi-dimensional feature vector.

[0035] In one embodiment, the system performs source-specific acquisition and integration of execution data. Specifically, the system first retrieves hardness test data for a specific monitoring area (such as a section of a superheater tube) from the historical maintenance database via a data interface. This hardness test data was obtained through physical testing during the most recent shutdown maintenance. Simultaneously, the system extracts operating condition data for the monitoring area within the corresponding operating cycle (i.e., the service period before hardness testing) from the field sensor network (or DCS historical database) distributed in the boiler superheater, reheater, and water-cooled wall areas. This data includes tube wall temperature, operating pressure, and derived parameters related to changes in operating conditions (such as temperature coefficients and stress conversion parameters).

[0036] Next, the system performs data association and integration: discrete, static hardness test data are used as labels or state variables, and logically mapped with statistical operating condition data (such as average temperature and weighted pressure load within the cycle) to integrate a multi-dimensional feature vector for each monitoring point at a specific service stage. This feature vector serves as the basic input for subsequent data analysis and undergoes normalization during data preprocessing to eliminate dimensional differences. In other optional implementations, the feature vector may also include additional dimensions such as ambient humidity, flue gas velocity, or historical maintenance records to enhance the model's context awareness.

[0037] Step S200: Implement data classification and separation based on clustering algorithms.

[0038] In one embodiment, a clustering algorithm from unsupervised learning is used to iteratively process the constructed multidimensional feature vector set. Based on the distribution patterns of each data point in the hardness, temperature coefficient, and pressure parameter spaces, subsets of data with similar physical behaviors are automatically identified. Through multiple iterations of optimization, the original mixed data stream is divided into several categorized datasets, each corresponding to a typical working condition region or material response mode. The clustering process continues until the convergence condition is met, ensuring that the classification results are stable and reliable. In other alternative implementations, Gaussian mixture models (GMMs) or hierarchical clustering can be used instead of the K-means algorithm to adapt to non-spherical data structures.

[0039] Step S300: Train the region-specific hardness evolution model and complete the structured storage.

[0040] In one embodiment, for each classified dataset, its unique temperature load and operating condition features are extracted and used as modeling inputs. A nonlinear mapping model is constructed using a decision tree-based machine learning algorithm. This model reflects the dynamic evolution of material hardness with operating parameters under specific operating conditions. After model training, it is applied to the time series of historical detection data, performing feature mapping operations to transform the original observations into standardized predicted outputs. These outputs, along with time labels and spatial location information, are stored in a relational database or time series database that supports transaction management, forming a structured dataset with spatiotemporal indexing capabilities. In other optional embodiments, the leaf node regression part can use support vector regression (SVR) or a neural network local fitting module to replace the multivariate nonlinear regression equation, thereby improving the approximation accuracy of complex nonlinear relationships.

[0041] Step S400: Sliding window analysis identifies abnormal hardness trends.

[0042] In one embodiment, single-point or multi-point hardness time series data arranged chronologically are extracted from a structured database. A sliding window mechanism is used to segment the series, and the rate of change of hardness over time is calculated within each time window. This rate of change is estimated using a linear or nonlinear fitting method to characterize the dynamic hardness decay within the current time period. The calculated rate of change is compared with a preset anomaly threshold; when it exceeds a set range, the time period is determined to be "abnormal trend data" with significant performance degradation, and a corresponding identifier is added to the database. In other optional embodiments, Bayesian change point detection or a state-space model can be introduced to replace least-squares fitting to improve sensitivity to sudden softening events.

[0043] Step S500: Construct a distribution map based on the spatial index and retrieve neighborhood data.

[0044] In one embodiment, an indexing mechanism is used to logically divide the spatial regions of the boiler's heating surface, establishing a spatial distribution map reflecting data density. This index structure supports rapid location of geographically neighboring units for any given region. When a region is identified as an abnormal trend object, the system triggers a neighborhood query based on its spatial location information, retrieving historical detection data of adjacent regions within a preset geometric or thermal adjacency range. This process relies on an efficient index traversal strategy, avoiding the latency caused by full table scans. In other optional implementations, the spatial index can be implemented using an R-tree, octree, or hash grid structure, the specific choice depending on the dimensional complexity of the equipment layout.

[0045] Step S600: Calculate the hardness similarity between regions and generate a difference analysis report.

[0046] In one embodiment, time-series vectors are constructed for the target region marked as exhibiting an abnormal trend and its retrieved neighboring regions. A numerical similarity index is calculated between the two to measure the consistency of their hardness evolution trajectories. If the similarity is below a preset standard, the target region is considered to exhibit unique degradation behavior that significantly deviates from its surrounding environment, potentially indicating local defects or abnormal thermal loads. In this case, the system automatically generates an analysis report containing comparison curves, difference metrics, and spatial location information for review by maintenance personnel. In other optional implementations, similarity assessment can be combined with methods such as Dynamic Time Warping (DTW) distance or cosine similarity to enhance the ability to identify asynchronous change patterns.

[0047] The aforementioned technical features form a closed-loop collaborative mechanism: First, multi-source data fusion and clustering separation decouple data from different physical regions, laying the foundation for subsequent differentiated modeling; second, the establishment of region-specific models enables precise characterization of the hardness response characteristics of various components during actual operation, compensating for the insufficient generalization of traditional single models; third, sliding window trend analysis focuses on the rate of change rather than static threshold judgment, effectively capturing early performance degradation signals; finally, the spatial topology association mechanism eliminates false alarms caused by global operating condition fluctuations through neighborhood comparison, improving the reliability of diagnostic conclusions. The entire process achieves a leap from "point measurement" to "area-domain intelligent discrimination," solving the detection blind spots and misjudgments caused by data fragmentation, lack of dynamic modeling, and spatial association in existing technologies.

[0048] Through the above scheme, this embodiment can achieve high-precision, continuous and intelligent monitoring of the hardness state of boiler heat-resistant alloy materials. It not only overcomes the limitations of traditional shutdown detection cycles and incomplete coverage, but also significantly improves the early fault detection capability and the scientific nature of maintenance decisions by integrating physical measurement and data-driven models to deduce the material performance evolution trend under the condition of boiler shutdown.

[0049] In one specific implementation, the hardness test data is obtained through the following calibration process: For different grades of heat-resistant alloy materials, a corresponding Leeb-Britt hardness conversion model is first established. This model comprehensively considers the influence of the evolution of metallographic structure of the material under different heat treatment states and during service.

[0050] For pipe wall samples with a mass below a preset threshold, to avoid measurement distortion due to insufficient structural stiffness, the sample is mechanically coupled to a rigid support platform with a mass greater than a preset reference mass via a coupling medium to form a stable reaction force support system. Subsequently, an impact pulse is applied in a controlled direction, and the raw Leeb hardness signal is acquired. In this embodiment, the preset reference mass is set to 5 kg to ensure that the support platform has sufficient inertial mass to resist the impact force of the Leeb hardness tester.

[0051] The obtained Leeb hardness value is input into a pre-built conversion model for the corresponding material. After parameter correction, the equivalent Brinell hardness value is calculated and used as the valid hardness test data in subsequent data processing. Specifically, the conversion model is built based on experimental calibration data and can reflect the influence of material microstructure on hardness response characteristics. The rigid support platform is made of high-density metal material, and its geometric dimensions and mass distribution ensure that no significant displacement or elastic vibration occurs at the moment of impact.

[0052] In some other alternative implementations, the coupling medium may be a gel-like material with moderate viscosity and good acoustic impedance matching characteristics; the input variables of the conversion model may also include other microstructure characteristics parameters other than metallographic level, such as grain size, precipitate coverage, etc., to further improve the conversion accuracy.

[0053] Through the above scheme, this embodiment can effectively eliminate the interference of structural elastic deformation in the field hardness testing process of small-diameter thin-walled components. At the same time, by introducing a multi-maintenance positive factor related to the actual aging state of the material, it significantly improves the accuracy and consistency of converting Leeb hardness values ​​to Brinell hardness values, thereby providing a highly reliable basic data source for the database.

[0054] In one specific embodiment, the step of coupling the pipe wall sample to a rigid support platform with a mass greater than a preset reference mass via a coupling medium for Leeb hardness measurement further includes: Vaseline was used as the coupling medium and coated on the contact surface between the tube wall sample and the rigid support platform to form a continuous and thin coupling layer. Specifically, the operator used a scraper or coating tool to evenly apply an appropriate amount of Vaseline to the contact area, ensuring that the support surface corresponding to the entire area to be tested was covered.

[0055] Press the tube wall sample firmly onto a rigid support platform, apply axial pressure and keep it stable until visual inspection confirms that the coupling layer thickness is consistent and there are no visible air bubbles.

[0056] The Leeb hardness tester impact device is activated in a direction perpendicular to the sample surface of the tube wall, so that the impact rod strikes the sample surface at a standard speed, and the rebound signal is collected and converted into a Leeb hardness value.

[0057] In this embodiment, the preset reference mass is set to 5 kg to ensure that the support platform has sufficient inertial mass to resist the impact force of the Leeb hardness tester.

[0058] In some other alternative implementations, the coupling medium can be replaced with silicone grease, a specialized ultrasonic coupling agent, or a low-viscosity grease, as long as it has good acoustic conduction properties and interface filling capabilities.

[0059] Through the above solution, this embodiment can effectively eliminate vibration interference caused by insufficient self-weight of small-diameter samples and significantly reduce systematic errors caused by energy loss during measurement. At the same time, Vaseline, as a common industrial auxiliary material, has the advantages of high chemical stability, non-volatile properties, and non-corrosiveness to metal surfaces. It can maintain the integrity of the coupling layer in complex field environments, thereby ensuring the consistency and repeatability of multiple measurements.

[0060] In one specific implementation, the step of pre-establishing a Leeb-Britt hardness conversion model based on different heat treatment states and different metallographic structures further includes: Samples of 12CrMoV and P91 heat-resistant alloy pipe sections with different service durations were collected, and metallographic samples were prepared and images were captured under an optical microscope at 500x magnification. Specifically, an image segmentation algorithm was used to binarize the pearlite region, and the proportion of spheroidal cementite to the original lamellar structure area was calculated. This proportion was mapped to the pearlite spheroidization level G (divided into 5 levels, G=1 indicates no spheroidization, and G=5 indicates complete spheroidization).

[0061] For each spheroidization stage, a comparative test of Leeb hardness (HL) and standard Brinell hardness (HBW) was conducted simultaneously, with each stage repeated at least 30 times to eliminate random errors. Then, a basic linear relationship was fitted using the least squares method. And calculate the measured HBW and Systematic deviation between This deviation is used as a metallographic correction factor. Establish a lookup table relationship: When G=1, When G=2, When HBW;G=3, When HBW;G=4, When HBW;G=5, HBW. The final conversion model includes metallographic correction terms. The model is then embedded into the calibration module of the on-site testing terminal. In other alternative implementations, the extraction of pearlite spheroidization levels can be achieved automatically through deep learning semantic segmentation networks (such as the U-Net architecture); or higher resolution carbide morphology data can be obtained through scanning electron microscopy (SEM) for correction coefficient modeling; or a continuous variable of spheroidization percentage can be introduced into the model to replace the hierarchical discrete variables in order to improve interpolation accuracy.

[0062] Through the above scheme, this embodiment can effectively compensate for the impact of microstructure evolution caused by long-term material service on the hardness conversion results, making the Brinell hardness value after model conversion of the Leeb hardness in the field closer to the actual mechanical property level. Compared with the prior art, this embodiment significantly improves the accuracy of non-destructive hardness assessment by introducing quantifiable metallographic feature correction terms without adding complicated laboratory testing procedures, and is especially suitable for the refined monitoring of alloy performance degradation under high-temperature aging environment.

[0063] In one specific implementation, the step of iteratively calculating the multidimensional feature vector using the K-means clustering algorithm includes: First, the clustering processing of the multidimensional feature vectors is based on the typical distribution characteristics of the boiler's heating surface physical structure to set the initial number of clusters. Specifically, the number of clusters is initialized to 3, corresponding to three regions in the boiler system with significantly different operating conditions and material response characteristics: the superheater, the reheater, and the water-cooled wall. The initial cluster center for each cluster is set by selecting representative sample points from the historical dataset to ensure coverage of the typical hardness, temperature, and pressure combinations of each region.

[0064] Then, for each data point acquired at any given time, the Euclidean distance between it and each initial cluster center is calculated. This distance metric is based on the standardized hardness measurement value, pipe wall temperature, working pressure, temperature change rate, and equivalent stress parameters contained in the feature vector. According to the minimum distance principle, the data point is assigned to the cluster containing the nearest cluster center. After one round of assignment, the mean value of all member points within each cluster is recalculated and used as the new cluster center location.

[0065] Next, the cluster centers are continuously updated during each iteration, and data point redistribution is performed again until the maximum movement distance of all cluster centers between two adjacent iterations is less than a preset convergence threshold, such as 0.01 (in normalized feature space units). At this point, the clustering process is considered to have reached a stable state, and the final classification result is output. This classification result achieves automatic separation of data streams from different heated surface regions in a high-dimensional feature space, forming independent datasets with clear physical meaning.

[0066] In some other alternative implementations, the similarity measure of the clustering algorithm can be replaced by Manhattan distance or cosine similarity; the convergence criterion can also be that the change in the objective function (such as the sum of squares within the cluster) is below a certain tolerance value; the number of clusters can also be set automatically within a certain range by the Elbow Method or the silhouette coefficient method, rather than being fixed at 3.

[0067] Through the above scheme, this embodiment enables automated grouping and processing of multi-source heterogeneous detection data without prior labels, effectively identifying data subsets reflecting the operational patterns of different components. This provides a data foundation with a clear structure and explicit physical meaning for the subsequent construction of region-specific hardness evolution models. Compared with existing technologies, this embodiment avoids subjective bias caused by manual data region division, improves the consistency and repeatability of data preprocessing, and enhances the model's ability to characterize the evolution of material properties under complex working conditions.

[0068] In one specific implementation, the step of separating data into categorized datasets based on the distribution characteristics of hardness values, temperature coefficients, and pressure parameters contained in the multidimensional feature vector includes: after the K-means clustering algorithm completes one iteration, the system first verifies the average hardness value of each generated cluster. Specifically, the average hardness value of the current cluster is compared with the theoretical hardness range of the boiler component corresponding to that cluster; for example, the theoretical Brinell hardness range of the heat-resistant alloy in the superheater region under normal service conditions is 200–230 HBW, in the water-cooled wall region it is 180–210 HBW, and in the reheater region it is 190–220 HBW. If the average hardness value of a cluster deviates from the theoretical range of its corresponding component by more than a preset standard deviation (e.g., 2σ, where σ is the standard deviation obtained based on historical data statistics), then the clustering result is determined to have a physical and logical deviation.

[0069] Next, the system automatically adjusts the weighting factor of the temperature coefficient in the multidimensional feature vector to enhance the influence of temperature-related parameters in distance calculation. For example, each dimension in the original feature vector is normalized with equal weights. At this time, the weight of the temperature coefficient (such as the temperature rise rate ΔT or cumulative high-temperature exposure time) is increased from 1.0 to 1.5-2.0, and the iterative process of K-means clustering is re-executed. By improving the discriminative power of the temperature dimension, data points with similar thermal histories but abnormal mechanical performance are more reasonably classified into categories consistent with their thermal conditions, thereby correcting misclassification caused by uneven material aging or sensor drift.

[0070] In other alternative implementations, the weighting strategy can employ a dynamic weighting mechanism, adjusting the temperature coefficient weight linearly or piecewise linearly based on the deviation magnitude; alternatively, in addition to the temperature coefficient, pressure change rate or stress cycle count can be introduced as auxiliary weighting terms under specific conditions to further improve the physical consistency of the clustering results. Furthermore, the theoretical hardness range can be set based on material grades (e.g., setting different benchmarks for P91 and 12CrMoV), and the recommended hardness range for the corresponding material can be obtained in real time through a database query interface.

[0071] Through the above scheme, this embodiment can effectively identify and correct the physical distortion problems that may occur in the pure numerical clustering process, ensuring that the separated classification dataset not only meets the mathematical compactness requirements, but also conforms to the actual operating rules of each heating surface area of ​​the boiler. Compared with the prior art, this embodiment improves the physical interpretability of multi-source data clustering, avoids model training bias caused by data distribution offset, and provides a reliable data foundation for the subsequent construction of a high-precision region-specific hardness evolution model.

[0072] In one specific implementation, based on the above embodiments, the structured modeling engine uses the Classification and Regression Tree (CART) algorithm to construct the model structure of a region-specific hardness model. Specifically, firstly, the clustered and separated classification dataset is input into the decision tree training process, using operating condition features as input variables and the calibrated Brinell hardness value at the corresponding time as the target output variable. Then, during each node split, composite operating condition parameters are selected as candidate split attributes, with priority given to calculating the information gain ratio (λ = T / Load) at different values ​​to evaluate its explanatory power for hardness variability. Next, all possible split points of this attribute are traversed, and the threshold that maximizes the information gain ratio is selected as the optimal split point, thereby recursively dividing the current dataset into significantly different subsets, such as two stable operating ranges: "high temperature and high load" and "low temperature and low load". Each non-leaf node undergoes binary splitting according to this rule until preset stopping conditions are met, including minimum sample size, upper limit of tree depth, or gain ratio below the convergence threshold. In other alternative implementations, the selection of splitting attributes can be extended to a weighted index based on multiple combined features; the calculation of information gain ratio can also be replaced by Gini impurity or other splitting quality evaluation criteria suitable for regression tasks to accommodate different types of data distribution characteristics.

[0073] Through the above scheme, this embodiment can achieve hierarchical modeling of the material hardness response behavior under complex working conditions, effectively capturing the stage-by-stage changes in the nonlinear degradation process. Using the temperature-to-load ratio as a splitting variable, the operating boundaries affecting the evolution of alloy properties can be accurately identified, improving the model's adaptability and predictive stability to different service environments. Compared with existing technologies, this embodiment avoids the fitting bias of a single global model in multi-condition scenarios. By constructing a physically meaningful branching structure, the interpretability of the hardness evolution path is enhanced, and a structural foundation is provided for subsequently introducing local fine-tuning fitting methods at leaf nodes.

[0074] In one specific implementation, the step of training a region-specific hardness model using a decision tree algorithm further includes: A multivariate nonlinear regression equation is constructed at the leaf nodes of the decision tree to fit the predicted hardness values. Specifically, historical data is first extracted from the stable operating condition subset represented by each leaf node, where temperature, pressure, and material composition conditions are relatively consistent. Then, based on the sample data in this subset, the parameters of the multivariate nonlinear function of the following form are fitted using the least squares method: in Real-time pipe wall temperature (unit: °C). Working pressure (unit: MPa). A constant vector characterizing the chemical composition of a material (e.g., a weighted combination of Cr, Mo, and V contents). , , , , These are the undetermined regression coefficients. The temperature term is included in the calculation in a negative exponential form, reflecting the physical law that the softening rate of materials accelerates with increasing temperature under high-temperature conditions; the pressure term is introduced in a positive exponential form, reflecting the work hardening or creep inhibition effects that may be induced under high stress. During the fitting process, an iterative optimization algorithm (such as the Levenberg-Marquardt algorithm) is used to solve the nonlinear least squares problem to ensure that the sum of squared residuals converges to a local minimum.

[0075] The regression coefficients obtained from training are stored in the model configuration file of the corresponding leaf node, enabling the system to calculate the expected hardness value under the current working condition in real time based on the current input parameters during the inference phase. In some other optional implementations, the temperature term can be replaced with an Arrhenius-type expression. To more accurately describe the thermal activation process; the pressure term can also adopt a piecewise power function or polynomial expansion form to adapt to the nonlinear response characteristics in different pressure ranges; the material composition coefficient term can be dimensionality reduced through principal component analysis to reduce the input dimension and eliminate collinearity effects.

[0076] Through the above scheme, this embodiment can establish a refined hardness prediction model that conforms to the laws of material physical evolution. While retaining the decision tree's ability to divide working conditions, it improves the local fitting accuracy within the leaf nodes and effectively reduces the systematic deviation caused by simple linear assumptions. It is especially suitable for long-term performance degradation modeling of complex alloy systems under high temperature and variable load conditions.

[0077] In one specific implementation, the step of calculating the slope of the hardness change using a sliding window includes: The time length of the sliding window is set to a preset window duration. In this specific embodiment, the preset window duration is set to 24 hours to segment the continuously collected hardness time series data. Within each sliding window, the least squares method is used to perform a linear fitting operation on the relationship between hardness values ​​and time within the window's coverage period. This fitting process determines a fitting straight line by solving for the optimization objective of minimizing the sum of squared errors, and its mathematical expression is as follows: The slope This reflects the average rate of hardness change during the specified period. The slope value is extracted and output as the hardness change slope at the center of the corresponding time window, used in subsequent trend analysis. In other optional implementations, the sliding window duration can be adjusted to 12 hours or 48 hours depending on actual monitoring needs; the least squares method can be replaced with weighted least squares to give higher weight to recent data; or robust regression methods such as the Theil-Sen estimator can be used instead of ordinary linear regression to improve resistance to outlier data points.

[0078] Through the above scheme, this embodiment can achieve dynamic quantitative characterization of the hardness evolution trend of boiler heat-resistant alloy materials, transforming discrete hardness observation values ​​into continuous rate-of-change indicators, thereby effectively identifying periods of accelerated hardness performance degradation. Compared with existing technologies, this embodiment utilizes a fixed-duration sliding window combined with a linear fitting algorithm, enabling the capture of early degradation signals of material properties without relying on absolute hardness thresholds. This improves the sensitivity to slowly evolving faults and provides a calculable data foundation for subsequent anomaly detection based on rate-of-change characteristics.

[0079] In one specific implementation, anomaly detection logic is further applied to the slope of hardness change calculated within the sliding window. First, the slope value output by the current sliding window is... With respect to the preset normal decay rate range A comparison was made, among which HBW / hour represents the slow softening range of a material due to creep under normal high-temperature service conditions. Specifically, when the slope... And its absolute value At that time, it was determined that the period had entered the "abnormal decrease in hardness stage", in which HBW / hour is used as the rapid decline threshold. For example, if the slope of multiple consecutive sliding windows for a reheater section is below this threshold within 24 hours, the system marks this period as a high-risk evolution period. Then, the corresponding data record is locked by combining the timestamp and spatial tag, triggering the subsequent spatial correlation analysis process. In some other optional implementations, the rapid decline threshold can be dynamically adjusted according to different alloy grades; for example, for P91 steel... HBW / hour, while 15CrMo steel, due to its lower initial hardness and higher aging sensitivity, is set to... HBW / hour; or the normal decay rate range is obtained by statistical fitting of historical overhaul data, rather than a fixed constant.

[0080] Through the above scheme, this embodiment can effectively distinguish between normal long-term creep softening and sudden accelerated performance degradation of materials, avoiding misjudging reasonable aging trends as failure events. Compared with the prior art, this embodiment introduces a dual-condition joint criterion (i.e., and This improves the accuracy and robustness of anomaly detection; by setting a graded threshold system, it enhances the sensitivity to early degradation signals, providing more timely decision-making basis for operation and maintenance.

[0081] In one specific implementation, the step of constructing a spatial distribution map reflecting data density using an indexing mechanism includes: The spatial area of ​​the boiler heating surface is divided into a multi-level grid structure. Specifically, based on the physical layout of the boiler tube panel, the three-dimensional space is projected into a two-dimensional plane coordinate system, and the entire heating surface area is divided into several levels of rectangular grid units according to the principle of recursive subdivision.

[0082] Each grid cell is assigned a unique quadtree node index, which is generated by the root node path encoding in the form of "Level_X_Y", where X and Y represent the horizontal and vertical position numbers of the current node in the corresponding level, thereby enabling fast spatial location and level traversal.

[0083] The number of hardness test data points falling within each grid cell is counted, specifically including all hardness records with timestamps and spatial coordinates from periodic testing and real-time simulations; combined with the actual coverage area of ​​the grid cell (unit: m²), the data density value is calculated. It is used to characterize the density of monitoring information per unit area.

[0084] Based on the obtained data density values, a heatmap rendering algorithm is used to generate a spatial distribution map, in which different density intervals are mapped to different color intensities. Low-density areas are displayed as blue gradients, and high-density areas are displayed as red gradients, which helps maintenance personnel identify data coverage blind spots or key monitoring areas.

[0085] In other alternative implementations, the multi-level grid structure can also be partitioned using an octree to preserve the complete three-dimensional spatial relationships; or, the index encoding method can be replaced with a spatial filling code based on the Z-order curve to improve query efficiency under large-scale data; or, the calculation of data density can introduce a time decay factor to make the weight of recently collected data points higher than that of historical data, thereby reflecting the current monitoring activity.

[0086] Through the above scheme, this embodiment can efficiently construct a hierarchical spatial index system to achieve quantitative expression and visualization of the spatial distribution characteristics of boiler heating surface hardness data; by using the quadtree node indexing mechanism, the computational complexity of spatial neighborhood retrieval is significantly reduced, and the response speed of abnormal area correlation analysis is improved; at the same time, the heat map generated based on data density can intuitively reveal the rationality of detection resource allocation and provide data support for optimizing on-site detection strategies.

[0087] In one specific implementation, based on the above embodiments, the spatial neighborhood retrieval process for the region corresponding to the abnormal trend data is refined. First, a preset search radius is determined using the grid node to which the monitoring point marked as abnormal trend data belongs as the center node. (For example This radius defines the geometrically adjacent region to be examined. Specifically, the constructed quadtree node index structure is used to quickly traverse the geographic space covered by this search radius: starting from the central node, the system recursively visits its parent node, sibling nodes, and child nodes that are at the same or adjacent levels and whose spatial location falls within the range of the central node. The system identifies all neighboring grid nodes within the specified range. Then, for each identified neighboring grid node, its unique quadtree node index is used as the database query key to extract historical hardness test data sequences associated with that grid node from the structured database. These data sequences include timestamped hardness values, temperature, and pressure records. Next, these extracted datasets are used as adjacent region data for subsequent time-series comparative analysis with the target anomaly area.

[0088] In some other alternative implementations, a preset search radius is used. It can be dynamically adjusted according to the type of boiler components, for example, using a smaller radius in the superheater area ( To improve positioning accuracy, the method extends to large-area planes of the water-cooled wall. To enhance statistical representativeness; in addition, the judgment logic of neighboring grid nodes can also introduce thermal symmetry constraints, that is, only nodes on the same heated surface (such as the front wall or side wall) and in similar heat flow distribution areas are selected for comparison, thereby eliminating mismatches caused by differences in boundary conditions.

[0089] Through the above scheme, this embodiment can efficiently and accurately locate and obtain detection data of neighboring areas that are spatially related to the abnormal area, avoiding the computational overhead of full table scanning and improving the response speed of spatial correlation analysis. At the same time, the hierarchical traversal mechanism based on quadtree index supports flexible spatial query strategies, enhances the system's adaptability to different boiler structure layouts, and provides a reliable data foundation for subsequent local anomaly identification.

[0090] In one specific implementation, based on the above embodiments, for the monitoring area marked as having an abnormal trend, a time series vector is first constructed between it and adjacent areas for quantitative comparison. Specifically, historical hardness data of the abnormal area within a preset time window is extracted from a structured database to form a time series vector. Each element corresponds to a hardness value at a sampling time; the hardness sequences of several geometrically adjacent grid nodes within its neighborhood are extracted simultaneously over the same time period to form a comparison vector. .

[0091] Then, the Pearson correlation coefficient between the two time series is calculated. This is used to measure the consistency of the changing trends of the two: in, and These are the hardness sequences of the target region. and neighborhood hardness sequence The arithmetic mean, This represents the number of sampling points in the sequence. This coefficient reflects whether the target area and the surrounding area are affected by the same operating condition fluctuations. Simultaneously, the numerical deviation between the two at corresponding time points is calculated in parallel, using the root mean square error (RMSE) as the index of difference. This value reflects the degree of dispersion in the absolute hardness levels of the two regions. Next, the calculation results are jointly judged with a set threshold: when... and When the target region and its neighborhood are confirmed to have significant non-cooperative evolutionary characteristics, it is determined to be a local independent degradation behavior, triggering the spatial difference analysis logic.

[0092] In other alternative implementations, the time series similarity metric can be replaced by the Dynamic Time Warping (DTW) distance, suitable for handling scenarios with asynchronous or nonlinear delayed responses; or the Spearman rank correlation coefficient can be used instead of the Pearson coefficient to enhance robustness to nonlinear monotonic trends; RMSE can also be replaced by the mean absolute error (MAE) or normalized mean square error (NRMSE), adjusted according to the system's error sensitivity requirements. All these alternatives maintain the basic framework of the dual-index joint criterion.

[0093] Through the above scheme, this embodiment can effectively identify local abnormal degradation patterns that are difficult to distinguish based on single-point threshold alarms alone. By utilizing a dual verification mechanism of trend pattern and numerical offset, it significantly reduces the false alarm rate caused by sensor drift and temporary disturbance conditions. Compared with the prior art, this embodiment achieves intelligent denoising and fault location capabilities based on spatiotemporal correlation, improves the physical reliability of early warning results, and provides operators with more valuable spatial diagnostic basis for decision support.

[0094] In one specific implementation, the step of determining the safety status of the boiler heat-resistant alloy includes: The hardness test data sequences of each region of the superheater, reheater, water-cooled wall, and economizer within a preset time window are obtained. Specifically, the hardness time sequence of each region is processed by moving average filtering to eliminate the influence of instantaneous fluctuations, and the average hardness value of the current period is calculated. Meanwhile, the hardness change rate of this region within the same time window was calculated based on the least squares linear fitting method. (i.e., decay rate), serving as a dynamic indicator of material aging.

[0095] Based on the operating temperature, stress level, and severity of failure consequences of each heating surface component in the boiler system, corresponding weighting coefficients are assigned. Specifically, because the superheater operates in the highest temperature range and withstands high internal pressure for extended periods, its weight is set accordingly. The reheater is the next most important component, and its settings are as follows: The water-cooled walls and economizer operate in relatively mild environments, and are respectively set to... , All weights satisfy the normalization condition .

[0096] Multiply the current average hardness of each region by its corresponding weight and sum them to obtain the overall hardness index: in, For regional indexes, For the first The average hardness value of each region.

[0097] Will A safety threshold pre-set based on the fatigue limit and safety margin of the target heat-resistant alloy material. Comparison, among which It is derived from the creep strength curve of the corresponding steel grade in the ASME standard. If If the value is below 5%, the boiler is determined to be in an unsafe state, and a tiered alarm mechanism is triggered: a yellow warning is issued when the value is below 5%, and a red alarm is issued when the value exceeds 5%.

[0098] In some other alternative implementations, the weighting coefficients The system can be dynamically adjusted based on actual unit design parameters. For example, by inputting boiler rated load, main steam temperature and other operating conditions, the system can automatically calculate the thermal stress ratio of each region and generate adaptive weights. Alternatively, the overall hardness index can be constructed using a geometric mean or weighted moving average smoothing method to enhance the sensitivity to recent deterioration trends.

[0099] Through the above scheme, this embodiment can achieve a quantitative assessment of the overall safety of boiler heating surface components, avoiding the one-sided risk caused by relying solely on local single-point threshold judgments. Compared with the prior art, this embodiment introduces a component-based weighted fusion mechanism, making the overall safety status assessment more in line with the actual risk distribution law of engineering, and improving the scientificity and reliability of early warning decisions.

[0100] In one specific implementation, the method further includes a prediction step: First, based on historical hardness evolution data stored in a structured database, the hardness decay trajectory sequence of each monitoring area under different operating conditions is extracted. Specifically, using timestamps as indexes, calibrated Brinell hardness values ​​collected periodically within a specific grid cell are retrieved, and combined with the average temperature, pressure, and stress parameters corresponding to that period, a time series sample set of the material aging path is constructed. Then, a Monte Carlo simulation algorithm is used to iteratively simulate the sample set: In each iteration, random disturbance variables are introduced according to the coefficient distribution range in the leaf node nonlinear regression model to simulate the nonlinear hardness decay process caused by factors such as temperature fluctuations and load changes during future operation; a single simulation generates a hardness decay curve from the current moment to a preset prediction time (e.g., 5000 hours). Next, the above simulation process is repeated 10,000 times to form a statistical distribution cluster of hardness decay. Subsequently, the time point corresponding to the first drop of the hardness value to the material failure threshold (e.g., the critical HBW value determined according to ASME Code Case N-47) in each simulation trajectory is calculated, and the probability density distribution of this time point in all trajectories is statistically analyzed. Finally, based on the cumulative function of the probability distribution, the expected remaining lifespan at the preset confidence level (e.g., 90%) is determined, and a suggested maintenance schedule is generated, for example, outputting "There is a 90% probability that the reheater B zone will reach its lifespan limit within 3200±150 hours".

[0101] In other alternative implementations, the underlying model upon which the Monte Carlo simulation relies can be replaced with a dynamic degradation model based on Bayesian updates, which uses newly observed hardness data to correct prior distributions in real time; or, the sources of uncertainty in the simulation input can include not only environmental parameter disturbances but also measurement error terms and model parameter drift terms; or, the presentation of the prediction results can be further extended to a risk heat map, mapping the failure probability of different spatial regions to a three-dimensional boiler model to assist in the formulation of regional maintenance strategies.

[0102] Through the above-described scheme, this embodiment can achieve probabilistic prediction of the remaining life of boiler heat-resistant alloy components, overcoming the limitations of traditional periodic maintenance or threshold alarms. Compared with the prior art, this embodiment can quantitatively assess the potential risk path of material performance degradation under complex operating conditions with multiple uncertainties, provide statistically significant maintenance decision support, and significantly improve the foresight and economy of equipment operation and maintenance.

[0103] This invention provides a database-based system for testing the hardness of boiler heat-resistant alloys. For example... Figure 2 As shown, the system includes: The multi-source data aggregation module M100 is configured to acquire data separately and perform correlation and integration. Specifically, this module is configured to acquire offline hardness test data from the maintenance record database or input terminal, and acquire temperature load data and working pressure data within the corresponding operating cycle from the sensor network distributed in the boiler superheater, reheater, and water-cooled wall areas. This module is further configured to logically correlate the physical parameters from the above different sources according to the service cycle, integrate them into a multi-dimensional feature vector in a unified format, and execute the K-means clustering algorithm to automatically divide them into independent datasets corresponding to different heated surface structural parts based on the distribution characteristics of each data point in the hardness, temperature coefficient, and pressure parameter space.

[0104] In one embodiment, the multi-source data aggregation module M100 uses a standardized interface protocol (such as OPC UA) to access the DCS system to obtain real-time operating condition data streams such as temperature and pressure, and achieves high-concurrency data throughput processing through a memory buffer. Simultaneously, this module periodically synchronizes historical hardness testing data through a database interface (such as JDBC or ODBC). In other optional implementations, this module can use message middleware (such as asynchronous communication mechanisms like Kafka or RabbitMQ) to aggregate data with the front-end sensor network (for operating condition data) and handheld testing terminals (for hardness data), thereby improving the system's scalability and fault tolerance in large-scale deployment scenarios.

[0105] The structured modeling engine M200 is connected to the multi-source data aggregation module M100. It is configured to extract working condition characteristic variables from the separated independent datasets, run decision tree algorithms to model the evolution of material properties in different regions, and train a region-specific hardness model that matches a specific region. The modeling engine is also configured to use the trained model to perform nonlinear mapping processing on historical test data, and write the mapping results, along with timestamps and spatial identification information, into a relational database that supports transaction management, thereby realizing structured data storage and version traceability.

[0106] In one embodiment, the structured modeling engine M200 uses the Classification and Regression Tree (CART) framework to construct the model's logical structure, selects a composite index of temperature and load as the splitting criterion, and recursively divides the input space to capture nonlinear response characteristics. In other optional embodiments, the modeling engine may also use random forests, gradient boosting trees, or neural networks to replace the single decision tree structure in order to adapt to more complex scenarios with multiple coupled factors.

[0107] The M300 time-series trend analyzer is coupled to a structured database and configured to periodically read hardness sequence data from each monitoring point sorted by time. It uses a sliding window mechanism to dynamically slice continuous observations. Within each time window, it applies the least squares method to fit a linear trend term and calculates its slope as the hardness change rate for the current period. When the change rate exceeds a preset anomaly threshold, it generates an anomaly trend object with a time range marker and injects it into the subsequent analysis process.

[0108] In one embodiment, the time series trend analyzer M300 sets the sliding window length to a preset window duration. In this specific embodiment, the preset window duration is set to 24 hours to ensure coverage of typical daily load fluctuation cycles and avoid misjudgments caused by short-term disturbances. In other optional embodiments, the window length can be adaptively adjusted according to actual operating conditions, such as switching to a 12-hour or 72-hour window based on seasonal load patterns, or using an exponentially weighted moving average (EWMA) instead of a linear fitting method to enhance sensitivity to recent changes.

[0109] The spatial topology association unit M400 is connected to the time series trend analyzer M300 and is configured to construct a spatial index structure reflecting the data density distribution based on the spatial coordinate information of the boiler heating surface. It can quickly locate the geographic neighborhood of the abnormal trend object in the physical layout according to the spatial location tag carried by the abnormal trend object. The unit is further configured to retrieve the historical hardness evolution data of the adjacent area, compare the consistency of hardness change between the target area and the neighboring area through statistical methods, and trigger the generation instruction of the difference analysis report when the similarity between the two is lower than the set standard.

[0110] In one embodiment, the spatial topology association unit M400 organizes spatial data using a hierarchical grid index based on geometric coordinates, supporting efficient spatial range queries and neighborhood retrieval operations. In other optional embodiments, this unit can employ R-trees, quadtrees, or other spatial database indexing techniques to achieve equivalent functionality, with the specific choice depending on the boiler structural complexity and data granularity requirements. The spatial topology association unit M400 also incorporates a multi-level grid indexer, which specifically uses a quadtree node structure to discretize the physical space of the boiler's heating surface. The holographic visualization interface M500 establishes a data channel with the spatial topology association unit M400 and the structured database. It is configured to parse the content of the received difference analysis report, integrate the current status and evolution trend of each region, and calculate the overall safety index that represents the overall equipment safety level. The interface is further configured to drive the three-dimensional graphics rendering component to generate a digital twin view of the boiler containing spatial positioning information, and to graphically identify the location of areas with abnormal behavior and their risk level on the display terminal.

[0111] In one embodiment, the holographic visualization interface M500 constructs a 3D rendering environment based on WebGL or Unity3D engines, reconstructing the boiler's heating surfaces into an interactive digital model according to their actual layout. In other optional embodiments, the interface can integrate VR / AR display devices to output immersive monitoring views, or present macroscopic distribution characteristics through a two-dimensional planar heat map combined with a GIS base map. The holographic visualization interface M500 is further configured with a lifespan prediction algorithm unit, used to perform Monte Carlo iterative calculations based on historical data in a structured database, simulating the hardness decay trajectory within a preset future time period, and mapping the calculated failure probability to a color gradient in the 3D view.

[0112] The aforementioned modules are interconnected via industrial Ethernet, forming a closed-loop data acquisition-modeling-analysis-feedback link. The multi-source data aggregation module M100 cleans and classifies the raw data, providing high-quality input for subsequent modeling; the structured modeling engine M200 establishes physically meaningful region-specific prediction models based on the classification results, improving the accuracy of soft measurement; the time-series trend analyzer M300 focuses on evolutionary behavior over time, identifying potential degradation trends; the spatial topology association unit M400 introduces spatial constraints, eliminating isolated noise interference through neighborhood comparison, improving diagnostic reliability; and finally, the holographic visualization interface M500 presents the analysis conclusions uniformly, realizing the information transformation from data to decision. This collaborative analysis mechanism across time and space effectively solves the problems of delayed early warnings and frequent false alarms caused by data fragmentation, lack of dynamic modeling capabilities, and spatial correlation judgment in traditional detection.

[0113] Through the above scheme, this embodiment can realize continuous monitoring of the performance of boiler heat-resistant alloy materials throughout the entire life cycle, deduce the hardness evolution trend under non-stop conditions, significantly improve the early anomaly identification capability, and reduce the false alarm rate by integrating spatiotemporal dual-dimensional analysis, providing reliable technical support for the safe operation and planned maintenance of generator units.

[0114] In one specific implementation, the precision calibration terminal includes a physical coupling device and a metallographic correction unit, both of which establish a data connection with the multi-source data aggregation module M100 via a wired or wireless communication interface. The physical coupling device includes a rigid support platform with a mass greater than a preset reference mass, such as 5 kg. This platform is made of cast iron or high-strength carbon steel, with a surface flatness controlled within 0.02 mm and a roughness Ra not exceeding 1.6 μm. An acoustic coupling layer is formed on the upper surface of the rigid support platform by coating with Vaseline, silicone grease, or a special ultrasonic coupling agent, with a thickness controlled between 10 and 50 μm, to achieve a tight fit between the tube wall sample and the platform. The small-diameter tube sample to be tested (outer diameter 38–60 mm, wall thickness 4–8 mm) is fixed above the coupling layer by a magnetic clamp or mechanical clamping mechanism to ensure no relative displacement or vibration energy leakage during the impact of the Leeb hardness tester. The Leeb hardness tester probe acts perpendicularly on the sample surface to acquire the original Leeb hardness signal.

[0115] The metallographic correction unit is an embedded processing module integrated into the portable calibration terminal. Its built-in non-volatile memory pre-stores a Leeb-Britt hardness conversion table based on different pearlite spheroidization levels. This conversion table is constructed based on experimental calibration data, classifying the pearlite spheroidization level G into 1 to 5 levels for typical heat-resistant alloys such as 12CrMoV and P91, and correspondingly setting a metallographic correction coefficient δ(G). For example, when G=3, δ(G)= 8 HBW, when G=4, δ(G)= 15 HBW. The metallographic correction unit receives the raw HL value from the physical coupling device, combines it with the material grade and spheroidization level information input by the user, calls the corresponding correction model to calculate and output the calibrated Brinell hardness digital signal HBWcalc = A×HL + B + δ(G), and transmits it to the multi-source data aggregation module M100 for further processing via RS-485 or Wi-Fi protocol.

[0116] In some other alternative implementations, the acoustic coupling layer can be replaced with a solid gel membrane or a pre-cured silicone rubber gasket to improve the ease of reuse in the field; the rigid support platform can be fixed by a vacuum adsorption structure instead of a magnetic clamp, which is suitable for calibration of non-ferromagnetic alloy samples; the conversion relationship in the metallographic correction unit can also be dynamically updated through a cloud database, supporting remote loading of correction parameter tables for new alloy materials.

[0117] Through the above-described scheme, this embodiment effectively eliminates the Leeb hardness measurement deviation caused by elastic deformation of small-diameter samples. Furthermore, by introducing a correction mechanism based on metallographic aging, it significantly improves the accuracy of converting on-site hardness test results to Brinell hardness standard values. Compared with existing technologies, this embodiment achieves integrated processing of standardized physical measurement conditions and material condition compensation, ensuring long-term comparability and modeling usability of the hardness data acquired on-site, and providing a highly reliable data input foundation for subsequent database-driven predictive analysis.

[0118] In one specific implementation, the spatial topology association unit M400 incorporates a multi-level grid indexer configured to hierarchically divide the physical space of the boiler heating surface. Specifically, the two-dimensional unfolded plane or three-dimensional geometric model of the boiler heating surface is recursively divided into four sub-regions, forming a quadtree node structure. Each level of nodes corresponds to a specific spatial range, until the physical area covered by a leaf node is less than a preset threshold (e.g., 1 square meter). Each quadtree node stores the hardness statistical characteristics of all monitoring points within that spatial region, including but not limited to average hardness, average hardness change slope, and data acquisition density. Simultaneously, each node also contains a set of physical neighborhood pointers, pointing to its directly adjacent grid nodes (up, down, left, and right) and the opposite tube row nodes with symmetrical thermal distribution characteristics, thereby establishing a complete spatial connection relationship. When the system receives a query request for an object exhibiting an abnormal trend, the multi-level grid indexer locates the object's position in the quadtree based on its grid node. It then directly jumps to the corresponding neighboring nodes and thermo-symmetric nodes via physical neighborhood pointers, quickly extracting historical hardness data from adjacent areas for comparative analysis without global traversal. In other optional implementations, the multi-level grid indexer employs an R-tree or octree structure to adapt to different boiler layouts; the connection rules for physical neighborhood pointers can be dynamically generated based on pre-imported CAD topology information from the actual piping layout diagram.

[0119] Compared with existing technologies, this embodiment achieves efficient modeling of the spatial relationship of boiler heating surfaces by constructing a quadtree node structure with physical neighborhood pointers. This reduces the time complexity of neighborhood retrieval operations from linear to logarithmic, significantly improving the response speed of spatial correlation analysis. At the same time, the introduction of thermo-symmetric node pointers supports intelligent comparison of symmetric regions of the flow field, enhancing the system's ability to identify regional overheating faults and effectively avoiding false alarms caused by judgments based on a single numerical threshold.

[0120] In one specific implementation, the holographic visualization interface M500 integrates a dynamic risk rendering engine. This engine is configured to receive hardness change slope data of each monitoring area output from the time-series trend analyzer M300, and calculate the remaining life prediction value of each area based on the material fatigue evolution model. The dynamic risk rendering engine further establishes a mapping relationship with the spatial location tags (Grid ID) in the structured database, and renders the remaining life prediction value to the corresponding area of ​​the boiler's three-dimensional geometric model after color gradient encoding. The color gradually changes from green to yellow to red, representing sufficient remaining life, critical life, and severely insufficient life, respectively. For areas where the current absolute value of hardness is still within the safe threshold range but its change slope exceeds the preset rapid decline threshold, the dynamic risk rendering engine enables a high-frequency flashing texture overlay display mechanism to highlight such potential early deterioration areas in a visual flashing manner. The flashing frequency is set to 3 Hz to 5 Hz to ensure significant visual recognition under the lighting environment of the control room.

[0121] In some other alternative implementations, the color gradient mapping employs a nonlinear interpolation algorithm to reflect the impact of hardness decay acceleration on remaining lifetime; the enabling conditions for high-frequency flicker textures also include a Pearson correlation coefficient of less than 0.75 and a root mean square error greater than 15 HBW between the region and its neighboring regions; in other implementations, the dynamic risk rendering engine supports interactive switching of view modes, including rendering by the current hardness value, rendering by decay rate, or rendering by Monte Carlo simulation of failure probability, with all modes sharing the same set of spatial indexes and graphics rendering pipeline.

[0122] Through the above scheme, this embodiment can realize multi-dimensional and highly sensitive visualization of the state of boiler heat-resistant alloy components. In particular, it can identify areas with abnormal evolution trends when the physical properties have not exceeded the standards, thereby improving the early fault warning capability. At the same time, it uses spatial correlation information to filter false alarms, thereby enhancing the accuracy and response efficiency of operation and maintenance decisions.

[0123] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A database-based method for boiler heat-resistant alloy hardness detection, characterized in that, The method comprises the following steps: Obtain hardness detection data of the superheater, reheater and water wall area of the boiler, obtain the operating condition data of the area in the corresponding operating cycle, and the operating condition data comprises temperature load data and working pressure data, integrate the hardness detection data and the operating condition data, and generate a multi-dimensional feature vector comprising hardness value, temperature coefficient and pressure parameter; Use K-means clustering algorithm to perform iterative operation on the multi-dimensional feature vector, and separate the data into classified data sets according to the distribution characteristics of the hardness value, temperature coefficient and pressure parameter contained in the multi-dimensional feature vector; Extract temperature load characteristics and working condition characteristics according to the area label in the classified data set, train a region-specific hardness model by using a decision tree algorithm, and perform feature mapping on the historical detection data by using the region-specific hardness model and store the historical detection data in a structured database; Extract hardness time series data arranged in time sequence from the structured database, calculate the hardness change slope by using a sliding window, and mark the time period in which the hardness change slope exceeds a preset abnormal threshold as abnormal trend data; Based on the spatial position information in the abnormal trend data, use an index mechanism to construct a spatial distribution map reflecting data density, and retrieve adjacent area data of the area corresponding to the abnormal trend data in the spatial distribution map; Calculate the hardness value similarity between the area corresponding to the abnormal trend data and the area corresponding to the adjacent area data, and generate an analysis report comprising difference characteristics to determine the safety state of the boiler heat-resistant alloy when the hardness value similarity is lower than a preset similarity threshold.

2. The database-based boiler heat-resistant alloy hardness detection method according to claim 1, characterized in that, The hardness detection data is obtained through the following calibration process: For different grades of heat-resistant alloy, a Rockwell-Brinell hardness conversion relationship model based on different heat treatment states and different metallographic structures is established in advance; For a pipe wall sample with a mass less than a preset mass threshold, the pipe wall sample is coupled to a rigid support table with a mass greater than a preset reference mass through a coupling medium to perform Rockwell hardness measurement and obtain a Rockwell hardness value; Input the Rockwell hardness value into the Rockwell-Brinell hardness conversion relationship model to calculate and obtain a Brinell hardness value, and use the Brinell hardness value as the hardness detection data.

3. The database-based boiler heat-resistant alloy hardness detection method according to claim 2, characterized by, The step of coupling the pipe wall sample to the rigid support table with a mass greater than a preset reference mass through a coupling medium to perform Rockwell hardness measurement further comprises: Use vaseline as the coupling medium to coat the contact surface between the pipe wall sample and the rigid support table to form a coupling layer; Press the pipe wall sample on the rigid support table until the thickness of the coupling layer is uniform and free of bubbles; Apply an impact force in a direction perpendicular to the surface of the pipe wall sample to collect a Rockwell hardness signal.

4. The database-based boiler heat-resistant alloy hardness detection method according to claim 2, characterized by, The step of pre-establishing a Rockwell-Brinell hardness conversion relationship model based on different heat treatment states and different metallographic structures further comprises: Collect metallographic pictures of heat-resistant alloy with different aging degrees, and extract pearlite spheroidization level characteristics; According to the pearlite spheroidization level characteristics, modify the basic conversion formula to generate the Rockwell-Brinell hardness conversion relationship model comprising a metallographic correction coefficient.

5. The database-based boiler heat-resistant alloy hardness detection method according to claim 1, characterized in that, The step of iteratively operating the multi-dimensional feature vector by using the K-means clustering algorithm comprises: Setting the number of clustering clusters as 3, respectively corresponding to the superheater, the reheater and the water wall; Calculating the Euclidean distance between each data point and the initial clustering center, and distributing each data point to the cluster where the initial clustering center is located closest to the data point; In each iteration, the centroid of each cluster is recalculated until the moving distance of the clustering center is less than a preset convergence threshold.

6. The database-based boiler heat-resistant alloy hardness detection method according to claim 5, characterized in that, The step of separating data into a classified data set according to the distribution characteristics of the hardness value, the temperature coefficient and the pressure parameter contained in the multi-dimensional feature vector comprises: When the average hardness value of a cluster in the clustering result deviates from the theoretical hardness range of the boiler component corresponding to the cluster and exceeds a preset standard deviation, adjusting the weight of the temperature coefficient in the multi-dimensional feature vector and re-executing the step of iteratively operating the multi-dimensional feature vector by using the K-means clustering algorithm.

7. The database-based boiler heat-resistant alloy hardness detection method according to claim 1, characterized by, The step of training the region-specific hardness model by using the decision tree algorithm comprises: Using a classification and regression tree algorithm to construct a region-specific hardness model structure; Selecting the ratio of temperature to load as the splitting attribute of the tree node; Calculating the information gain ratio of different splitting points, selecting the point with the maximum information gain ratio as the splitting threshold, and recursively dividing the classified data set into a high-temperature high-load subset and a low-temperature low-load subset.

8. The database-based boiler heat-resistant alloy hardness detection method according to claim 7, characterized by, The step of training the region-specific hardness model by using the decision tree algorithm further comprises: At the leaf node of the decision tree, a multivariate nonlinear regression equation is used to fit the hardness prediction value; The multivariate nonlinear regression equation contains a temperature coefficient term, a load coefficient term and a material composition coefficient term, wherein the temperature coefficient term is in a proportional relationship with the negative exponent of temperature, and the load coefficient term is in a proportional relationship with the positive exponent of pressure.

9. The database-based boiler heat-resistant alloy hardness detection method according to claim 1, characterized by, The step of calculating the hardness change slope by using a sliding window comprises: Setting the time length of the sliding window as a preset window duration; In each sliding window, a least squares method is used to linearly fit the change of the hardness time series data over time; The slope of the fitting straight line of the linear fitting is obtained as the hardness change slope at the time of the sliding window.

10. A database-based boiler heat-resistant alloy hardness detection system, characterized in that, The system for implementing the database-based boiler heat-resistant alloy hardness detection method according to any one of claims 1 to 9 comprises: A multi-source data aggregation module configured to obtain hardness detection data of each heated surface region of a boiler, obtain operating condition data of the region in a corresponding operating cycle, the operating condition data including temperature load data and working pressure data, and integrate the hardness detection data and the operating condition data to generate a multi-dimensional feature vector, and execute a K-means clustering algorithm to separate the multi-dimensional feature vector into independent data sets respectively corresponding to a superheater, a reheater and a water wall according to physical parameter distribution characteristics; A structured modeling engine configured to extract operating condition characteristics in the independent data sets, train a region-specific hardness model by using a decision tree algorithm, and complete feature mapping and structured storage of historical data by using the region-specific hardness model. The time series trend analyzer is configured to perform a sliding window operation on the stored time series data, calculate a hardness change slope, and mark a time period in which the hardness change slope exceeds a preset abnormal threshold as an abnormal trend object; The spatial topology correlation unit is configured to establish a spatial distribution map index reflecting data density, retrieve spatial adjacent area data of the abnormal trend object based on the spatial distribution map index, and trigger a difference analysis report when a hardness similarity between the abnormal trend object and the spatial adjacent area data is lower than a preset standard; The holographic visualization interface is configured to analyze the difference analysis report, calculate an overall safety index, and render a three-dimensional boiler safety state view with an abnormal area location marked.

Images