Method for evaluating thermal insulation performance of thermal insulation material based on intelligent sensor

Abstract

This invention discloses a method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors, belonging to the field of thermal insulation material performance testing technology. The method includes the following steps: receiving thermal insulation performance evaluation task information; calling a thermal insulation performance evaluation benchmark library to generate a thermal insulation performance evaluation constraint set; executing multi-scale bidirectional thermal excitation, collecting the thermal responses of the tested thermal path and the reference thermal path, and generating thermal response records of the thermal insulation material; generating a reliable thermal response tensor after normalization and verification; inputting an improved DeepONet model, generating an equivalent thermal parameter field of the thermal insulation material through dual-path isomorphic encoding and differential coupling with the reference thermal path; extracting abnormal locations and corresponding thermal states to generate a set of thermal anomaly spatial points; generating a cluster of thermal anomaly regions using an improved HDBSCAN algorithm; determining the basic evaluation level and executing anomaly limits to generate the thermal insulation performance evaluation result. This invention achieves reliable reconstruction of the equivalent thermal parameter field of the thermal insulation material and accurate identification of continuous thermal anomaly regions.

Description

Technical Field

[0001] This invention relates to the field of thermal insulation material performance testing technology, and in particular to a method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors. Background Technology

[0002] With the development of building energy conservation, industrial equipment insulation and low-temperature storage and transportation technology, insulation materials are widely used in building envelopes, pipelines, storage containers and thermal equipment. Their insulation performance directly affects heat loss and operating energy consumption.

[0003] Current methods for testing thermal insulation performance mainly utilize temperature sensors and heat flow sensors to collect temperature and heat flow data from both sides of the material. Then, based on steady-state heat transfer relationships, the thermal conductivity or thermal resistance is calculated, and the material's insulation status is determined based on the individual calculation results. However, the following problems commonly exist in practical applications: Existing detection methods typically employ thermal excitation in a single direction and at a fixed scale, which makes it difficult to reflect the differences in material response at different heat transfer stages and in different heat transfer directions. Sensor contact status, changes in ambient temperature, and fluctuations in thermal excitation can easily mix into the thermal response under test, making it difficult to separate the material's own thermal conductivity characteristics from external disturbances. Discrete data obtained from limited detection locations are difficult to form a continuous equivalent thermal parameter distribution, and local thermal bridges and continuous insulation abnormal areas are easily masked by the overall average result.

[0004] Existing evaluation methods mostly determine the evaluation level based on single-point thresholds or comprehensive averages, lacking the processing of spatial continuity relationships of abnormal locations and the stable state of abnormal areas, resulting in insufficient accuracy and traceability of thermal insulation performance evaluation results.

[0005] Therefore, how to provide a method for evaluating the thermal insulation performance of thermal insulation materials based on smart sensors is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an evaluation method for the thermal insulation performance of thermal insulation materials based on intelligent sensors. This invention fully utilizes intelligent sensing, multi-scale bidirectional thermal excitation, the DeepONet model, and the HDBSCAN algorithm to achieve thermal parameter reconstruction, abnormal area identification, and evaluation level determination of thermal insulation materials. It has the advantages of accurate evaluation, strong anti-interference ability, and clear anomaly location.

[0007] The method for evaluating the thermal insulation performance of thermal insulation materials based on smart sensors according to embodiments of the present invention includes the following steps: Step 1: Receive thermal insulation performance evaluation task information, parse the material to be evaluated and the evaluation conditions, call the thermal insulation performance evaluation benchmark library, and generate a thermal insulation performance evaluation constraint set; Step 2: Control the intelligent sensor group and thermal excitation device to perform multi-scale bidirectional thermal excitation according to the thermal insulation performance evaluation constraint set, and simultaneously collect the thermal response of the thermal path under test and the reference thermal path to generate thermal response records of the thermal insulation material; Step 3: Perform thermal response normalization and reliability verification on the thermal response records of the insulation material to generate a reliable thermal response tensor; Step 4: Input the reliable thermal response tensor into the improved DeepONet model, and generate the equivalent thermal parameter field of the insulation material through dual-path isomorphic encoding and differential coupling with the reference thermal path; Step 5: Process the equivalent thermal parameter field of the insulation material according to the thermal insulation performance evaluation constraint set, extract the abnormal locations and corresponding thermal states, and generate a set of thermal anomaly spatial points; Step 6: The improved HDBSCAN algorithm is used to process the spatial point set of thermal anomalies. After thermal state-constrained nearest neighbor construction and thermal bridge cluster stability screening, a cluster of thermal insulation anomaly regions is generated. Step 7: Determine the basic evaluation level according to the thermal insulation performance evaluation constraint set and the equivalent thermal parameter field of the thermal insulation material, and combine the abnormal limit of the thermal insulation abnormal area cluster to generate the thermal insulation performance evaluation result.

[0008] Optionally, step one specifically includes: Receive thermal insulation performance evaluation task information, classify the thermal insulation performance evaluation task information according to the material properties and working conditions corresponding to the record items, group the record items corresponding to the material properties into the material to be evaluated, and group the record items corresponding to the working conditions into the evaluation working conditions. Standardize the recording and expression of the evaluation materials, standardize the measurement units and recording formats of the evaluation conditions, and complete the analysis of the evaluation materials and evaluation conditions; According to the evaluation benchmark library of thermal insulation performance of the material to be evaluated, the candidate benchmark records corresponding to the material to be evaluated are screened, the evaluation conditions are compared with the working condition ranges corresponding to each candidate benchmark record, and the candidate benchmark records whose evaluation conditions do not fall into the working condition range are deleted. When one candidate benchmark record is retained, that candidate benchmark record is determined as the target benchmark record; when multiple candidate benchmark records are retained, they are sorted in ascending order of working condition range, and the candidate benchmark record ranked first is determined as the target benchmark record. Read the target baseline record, merge the target baseline record with the thermal insulation performance evaluation task information, and generate a thermal insulation performance evaluation constraint set.

[0009] Optionally, step two specifically includes: The working positions of the intelligent sensor group and thermal excitation device are determined according to the thermal insulation performance evaluation constraint set, so that the thermal path under test and the thermal insulation material form a thermal transfer connection, and the thermal path under test and the reference thermal path are collected synchronously. The thermal excitation device controls the thermal excitation to apply thermal excitation sequentially from one side of the thermal insulation material according to the thermal excitation scale specified in the thermal insulation performance evaluation constraint set, and the intelligent sensor group simultaneously collects the thermal response of the thermal path under test and the reference thermal path under each thermal excitation scale. After the acquisition of each thermal excitation scale is completed, thermal excitation is stopped. When the thermal circuit under test and the reference thermal circuit meet the start-up conditions of the next thermal excitation scale, the next thermal excitation scale is executed. After completing the acquisition in one thermal excitation direction, switch the working side of the thermal excitation device and perform thermal excitation in another thermal excitation direction in the same thermal excitation scale sequence, while simultaneously acquiring the thermal response of the thermal circuit under test and the reference thermal circuit; The collected thermal responses are recorded according to the thermal excitation scale and direction to generate thermal response records for thermal insulation materials.

[0010] Optionally, step three specifically includes: Read the thermal response record of the thermal insulation material, group the thermal response of the test thermal circuit and the reference thermal circuit according to the thermal excitation scale and thermal excitation direction, and divide the thermal response segment according to the start and end times of thermal excitation corresponding to each group of thermal response. According to the sampling time, the thermal response segments of the thermal path under test and the reference thermal path under the same thermal excitation scale and thermal excitation direction are matched, and the detection positions corresponding to the reverse thermal excitation are converted into the detection position sequence consistent with the forward thermal excitation to complete the thermal response regularization. Each thermal response segment is checked to ensure that it completely covers the corresponding thermal excitation process, that adjacent sampling times are continuous, that the thermal path under test and the reference thermal path are synchronously corresponding, and that the thermal response change is consistent with the direction of thermal excitation. Thermal response segments that fail the verification are removed to complete the reliability verification. According to the sampling length specified in the thermal insulation performance evaluation constraint set, the thermal response segments that have passed the credibility verification are resampled. The thermal responses of the tested thermal path and the reference thermal path after resampling are arranged in sequence according to the thermal excitation scale and thermal excitation direction, and tensor overlay is performed to generate a credibility thermal response tensor.

[0011] Optionally, the improved DeepONet model includes a thermal response coding layer, a spatiotemporal coordinate mapping layer, a thermal field feature coupling layer, and a thermal parameter field output layer, wherein a reference thermal path differential mechanism is embedded in the thermal field feature coupling layer. The thermal response coding layer reads the reliable thermal response tensor and splits the thermal response of the thermal path under test and the reference thermal path according to the thermal excitation scale and thermal excitation direction. The two thermal responses are input into a homogeneous branch network with shared coding parameters, and feature projection, stage feature extraction and dimension unification are performed sequentially to make the same coding position in the two thermal responses represent the same thermal excitation process, generating a test thermal path code and a reference thermal path code with consistent dimension and arrangement order. The spatiotemporal coordinate mapping layer reads the detection position and sampling time corresponding to each thermal response in the reliable thermal response tensor; Query coordinates are constructed based on the detection location and sampling time. The query coordinates are input into the backbone network to perform coordinate projection, position association and basis function mapping, so that the coordinate features obtained by mapping match the corresponding positions of the two codes, and generate spatiotemporal coordinate codes. The thermal field feature coupling layer receives the thermal path code to be tested, the reference thermal path code, and the spatiotemporal coordinate code, and establishes the correspondence between the two codes and the spatiotemporal coordinate code based on the thermal excitation scale, thermal excitation direction, and sampling time. The reference thermal path differential mechanism is used to differentially calibrate the reference thermal path code to be tested, thereby generating a calibration thermal response code; According to the corresponding coding position, the calibration thermal response coding and spatiotemporal coordinate coding are combined dimensionally and coupled with operator features to generate thermal field coupling features corresponding to each detection position. The thermal parameter field output layer performs equivalent thermal parameter mapping on the thermal field coupling features and arranges the mapping results according to the detection position. Merge consistent mapping results from the same detection location under different thermal excitation scales and directions, and correct inconsistent results based on the changes in the corresponding mapping results to generate an equivalent thermal parameter field for the insulation material.

[0012] Optionally, the reference thermal path differential mechanism is specifically as follows: Read the thermal path code to be tested and the reference thermal path code, and establish the corresponding code position according to the thermal excitation scale, thermal excitation direction and sampling time; Calculate the change of each corresponding coding position relative to the starting coding position of this thermal excitation scale to form the thermal path change sequence to be tested and the reference thermal path change sequence. Compare the direction of change and the range of continuous change between the thermal path change sequence under test and the reference thermal path change sequence at the corresponding coding positions; The coded changes with the same direction of change and overlapping continuous change intervals are identified as synchronous change components, and the remaining coded changes in the thermal path change sequence under test, excluding synchronous change components, are identified as independent change components of the thermal path under test. Based on the synchronously changing components, the corresponding coding position of the thermal path code to be tested is subtracted to obtain the differential calibration result; Establish the correspondence between the independent changing components of the thermal path under test under positive and reverse thermal excitation according to the detection location, and retain the independent changing components of the thermal path under test with consistent changing direction and continuous changing intervals overlapping each other. Zero-value replacement is performed on the coded values ​​with inconsistent change directions or continuous change intervals that do not overlap with each other. The differential calibration results are updated according to the original coded position to generate a calibration thermal response code that is consistent with the coded dimension of the thermal path to be tested.

[0013] Optionally, step five specifically includes: Read the equivalent thermal parameter field of the insulation material, extract the equivalent thermal parameters corresponding to each detection location according to the detection location, and match the corresponding evaluation boundary for each equivalent thermal parameter based on the thermal insulation performance evaluation constraint set; The equivalent thermal parameters corresponding to each detection location are compared with the evaluation boundary one by one. When the equivalent thermal parameters are all within the evaluation boundary, the processing of the current detection location ends. When any equivalent thermal parameter exceeds the evaluation boundary, the current detection location is determined as an abnormal location; Calculate the deviation of the equivalent thermal parameters corresponding to the abnormal location from the evaluation boundary, merge the equivalent thermal parameters and deviation results corresponding to the abnormal location according to the original detection location, and generate the corresponding thermal state of the abnormal location. The abnormal locations and their corresponding thermal states are associated with each other and encoded to generate thermal anomaly spatial points. The thermal anomaly spatial points are arranged according to the detection locations to generate a set of thermal anomaly spatial points. If no abnormal location is formed, an empty thermal anomaly spatial point set is generated.

[0014] Optionally, step six specifically includes: Read the set of thermal anomaly spatial points. If the set of thermal anomaly spatial points is empty, do not retain the thermal anomaly spatial points. If the set of thermal anomaly spatial points is not empty, establish the spatial adjacency relationship between the thermal anomaly spatial points according to the detection position. Thermal state-restricted nearest neighbor construction filters spatial adjacency relationships based on the spatial continuity relationship of adjacent thermal anomaly spatial points and the corresponding thermal state matching relationship, generating an effective nearest neighbor set; Within the effective nearest neighbor set, the distance between points is calculated based on the differences in detection location and corresponding thermal state, and the effective nearest neighbors of each thermal anomaly spatial point are arranged in ascending order of the distance between points. When the number of effective neighbors reaches the specified order of the thermal insulation performance evaluation constraint set, the effective neighbors of the corresponding order are selected, and the distance between the corresponding points is used as the thermal state core distance. Thermal anomaly spatial points whose number of effective neighbors does not reach the specified order are not included in the subsequent calculation. The maximum value between the thermal state core distance of each of two adjacent points and the distance between the two points is taken as the thermal state reach distance. A weighted graph is constructed with the thermal anomaly spatial points participating in the subsequent calculation as nodes and the thermal state reach distance as edge weights, and the minimum spanning tree is extracted. The connection relationships in the minimum spanning tree are split according to the edge weight from large to small. The splitting process of each connected component as the edge weight changes is recorded, and a hierarchical clustering tree is constructed. Based on the thermal insulation performance evaluation constraint set, branches that do not meet the cluster size requirements are deleted, the hierarchical clustering tree is compressed, and candidate abnormal clusters are generated; Thermal bridge cluster stability screening is performed on candidate anomalous clusters. The density level is represented by the reciprocal of the thermal state reach distance. The stability of candidate anomalous clusters is calculated according to the duration interval of candidate anomalous clusters in different density levels. The stability of the candidate anomaly cluster is corrected based on the continuous state of the anomaly location covered by the candidate anomaly cluster and the consistent state of the corresponding thermal state. The corrected stability of the parent cluster is compared with the sum of the corrected stability of each child cluster. If the corrected stability of the parent cluster is not less than the sum of the corrected stability of each child cluster, the parent cluster is retained; otherwise, each child cluster is retained. If no candidate anomalous cluster is formed, thermal anomaly spatial points are not retained. The retained thermal anomaly spatial points are merged according to the detection location to generate a thermal insulation anomaly region cluster.

[0015] Optionally, the thermally constrained nearest neighbor construction specifically involves: Read the set of thermal anomaly spatial points and spatial adjacency relationships, determine the thermal anomaly spatial points at both ends of each spatial adjacency relationship one by one, and verify the detection position distance between the two points according to the adjacency distance range specified in the thermal insulation performance evaluation constraint set; When the detection location spacing falls within the adjacency distance range, the corresponding spatial adjacency relationship is retained; when the detection location spacing exceeds the adjacency distance range, the corresponding spatial adjacency relationship is deleted, and candidate adjacency relationships that satisfy the spatial continuity relationship are obtained. Read the corresponding thermal states of the thermal anomaly spatial points at both ends of each candidate adjacency relationship, and compare the deviation results in the corresponding thermal states one by one according to the same evaluation criteria. When the deviation directions are consistent and the difference between the deviation results falls within the matching range specified by the thermal insulation performance evaluation constraint set, the corresponding thermal states of the two thermal anomaly spatial points at both ends are considered to match each other, and the corresponding candidate adjacency relationship is retained. If the deviation direction is inconsistent or the difference exceeds the matching range, the corresponding candidate adjacency relationship will be deleted. The candidate adjacency relationships are summarized and retained with each thermal anomaly spatial point as the center. Duplicate relationships with the same connection endpoints are deleted to form the effective nearest neighbors corresponding to each thermal anomaly spatial point. For thermal anomaly spatial points that do not retain candidate adjacency relationships, their effective nearest neighbors are recorded as empty. The effective nearest neighbors corresponding to each thermal anomaly spatial point are then summarized to generate an effective nearest neighbor set. Optionally, step seven specifically involves: Read the equivalent thermal parameter field of the insulation material, extract the equivalent thermal parameters corresponding to each detection location according to the detection location, and determine the evaluation level corresponding to each detection location based on the insulation performance evaluation constraint set; According to the evaluation level from high to low, the proportion of the current evaluation level and the detection positions with higher evaluation levels are counted in turn, and the first evaluation level that meets the conditions specified in the thermal insulation performance evaluation constraint set is determined as the basic evaluation level. Read the clusters of thermal insulation abnormal areas, extract the detection location and corresponding thermal state covered by each thermal insulation abnormal area cluster one by one, and verify each thermal insulation abnormal area cluster according to the thermal insulation performance evaluation constraint set. If the abnormality limit conditions are not met, the basic evaluation level will not be changed; if the abnormality limit conditions are met, the corresponding limit level will be determined. When multiple restriction levels exist, the restriction level with the most stringent restriction degree is selected according to the order of evaluation levels; if the basic evaluation level is not higher than the restriction level, the basic evaluation level is retained. When the basic evaluation level is higher than the limit level, the limit level shall replace the basic evaluation level; when the cluster of abnormal insulation areas is empty, the basic evaluation level shall be retained. The evaluation level after the anomaly limit is determined as the final evaluation level. The final evaluation level, the equivalent thermal parameter field of the insulation material, and the cluster of insulation anomaly areas are summarized to generate the insulation performance evaluation result.

[0016] The beneficial effects of this invention are: This invention employs an improved DeepONet model to process the reliable thermal response tensor. It maintains the characteristic correspondence between the thermal path under test and the reference thermal path under the same thermal excitation process through dual-path isomorphic encoding, and uses the reference thermal path differential mechanism to calibrate the encoding of the thermal path under test. This reduces the interference of synchronous response caused by environmental changes and thermal excitation fluctuations on the equivalent thermal parameter mapping, thereby achieving reliable reconstruction of the equivalent thermal parameter field of the insulation material.

[0017] This invention employs an improved HDBSCAN algorithm to process spatial point sets of thermal anomalies. It filters effective nearest neighbors based on the spatial continuity relationship of anomaly locations and the corresponding thermal state matching relationship, so that the thermal state core distance and thermal state reach distance can reflect the actual thermal state correlation between anomaly locations. It combines the continuous state and thermal state consistency state of candidate anomaly clusters to perform stable screening of thermal bridge clusters, thereby realizing the distinction between isolated anomaly points, discrete disturbance points, and continuous insulation anomaly regions.

[0018] This invention forms a continuous processing chain from thermal response acquisition to thermal insulation performance evaluation by multi-scale bidirectional thermal excitation, synchronous acquisition of the thermal path under test and the reference thermal path, reconstruction of the equivalent thermal parameter field, clustering of abnormal regions and anomaly limits. This enables the coordinated evaluation of overall thermal insulation performance and local abnormal regions, and avoids local thermal bridges and continuous thermal insulation abnormal regions being masked by the overall average result. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of the evaluation method for the thermal insulation performance of thermal insulation materials based on intelligent sensors proposed in this invention; Figure 2 This is a schematic diagram of the structure of the improved DeepONet model proposed in this invention; Figure 3 This is a flowchart of the improved HDBSCAN algorithm proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-3 The evaluation method for the thermal insulation performance of thermal insulation materials based on smart sensors includes the following steps: Step 1: Receive thermal insulation performance evaluation task information, parse the material to be evaluated and the evaluation conditions, call the thermal insulation performance evaluation benchmark library, and generate a thermal insulation performance evaluation constraint set; Step 2: Control the intelligent sensor group and thermal excitation device to perform multi-scale bidirectional thermal excitation according to the thermal insulation performance evaluation constraint set, and simultaneously collect the thermal response of the thermal path under test and the reference thermal path to generate thermal response records of the thermal insulation material; Step 3: Perform thermal response normalization and reliability verification on the thermal response records of the insulation material to generate a reliable thermal response tensor; Step 4: Input the reliable thermal response tensor into the improved DeepONet model, and generate the equivalent thermal parameter field of the insulation material through dual-path isomorphic encoding and differential coupling with the reference thermal path; Step 5: Process the equivalent thermal parameter field of the insulation material according to the thermal insulation performance evaluation constraint set, extract the abnormal locations and corresponding thermal states, and generate a set of thermal anomaly spatial points; Step 6: The improved HDBSCAN algorithm is used to process the spatial point set of thermal anomalies. After thermal state-constrained nearest neighbor construction and thermal bridge cluster stability screening, a cluster of thermal insulation anomaly regions is generated. Step 7: Determine the basic evaluation level according to the thermal insulation performance evaluation constraint set and the equivalent thermal parameter field of the thermal insulation material, and combine the abnormal limit of the thermal insulation abnormal area cluster to generate the thermal insulation performance evaluation result.

[0022] In this embodiment, step one specifically includes: The thermal insulation performance evaluation platform receives thermal insulation performance evaluation task information, verifies the completeness of the records of the material to be evaluated and the evaluation conditions, and terminates the current evaluation task if the material to be evaluated lacks material category, nominal thickness or apparent density, or if the evaluation conditions lack ambient temperature, relative humidity or boundary heat transfer state. According to the record caliber of the thermal insulation performance evaluation benchmark library, the materials to be evaluated and the evaluation conditions are unified, the material categories are converted into standard material codes, the nominal thickness is unified to millimeters and retained to 0.1 millimeters, the apparent density is unified to kilograms per cubic meter and rounded to the nearest integer, the ambient temperature is unified to degrees Celsius and retained to 0.1 degrees Celsius, the relative humidity is unified to percentage and rounded to the nearest integer, and the boundary heat transfer state is converted into the corresponding state code. Search the thermal insulation performance evaluation benchmark library for the material to be evaluated, retain benchmark records with consistent material category and nominal thickness and apparent density falling within the corresponding range, and then filter out benchmark records with mismatched ambient temperature, relative humidity or boundary heat transfer state based on the evaluation conditions. When retaining one baseline record, it is determined as the target baseline record. When retaining multiple baseline records, the sum of the ratio of the width of the ambient temperature range to the total span of the ambient temperature in the warehouse and the ratio of the width of the relative humidity range to the total span of the relative humidity in the warehouse is used as the matching range value. Sort the benchmark records in ascending order of matching range values. If the matching range values ​​are the same, select the benchmark record with the latest effective time as the target benchmark record. If no benchmark record is retained, terminate the current evaluation task. The target benchmark records and thermal insulation performance evaluation task information are merged to generate a thermal insulation performance evaluation constraint set, providing a unified execution basis for thermal excitation control, thermal anomaly judgment and evaluation level determination.

[0023] In this embodiment, step two specifically involves: The thermal insulation performance evaluation platform determines the detection positions of the intelligent sensor group and the thermal excitation device according to the thermal insulation performance evaluation constraint set. The intelligent sensor group is arranged on both sides of the thermal insulation material. The thermal path under test adopts a constant pressure clamping structure to form a thermal transfer connection with the thermal insulation material. The contact pressure is maintained at 2.0±0.2kPa. The reference thermal path is set in the same detection head and uses a unified sampling clock. The thermal excitation device continuously applies a heat flux density of 150W / m² from the insulation material side for 60s, 180s and 360s respectively, forming a three-level thermal excitation scale. The intelligent sensor group synchronously collects the temperature response and heat flux response of the thermal path under test and the reference thermal path at a sampling frequency of 2Hz. The time difference between the two recordings at the same sampling moment does not exceed 20ms. After each thermal excitation scale is completed, thermal excitation is stopped. When the temperature response of the tested thermal circuit and the reference thermal circuit is less than 0.3℃ and the heat flow response is less than 2W / m², and this condition is maintained for 30 seconds, the next thermal excitation scale is started. If the start conditions are not met, the natural decay thermal response is recorded. After completing the acquisition in one thermal excitation direction, the working side of the thermal excitation device is switched. Under the condition that the detection position remains unchanged, the acquisition in another thermal excitation direction is completed in the same thermal excitation scale sequence. The thermal responses of the thermal path under test and the reference thermal path are arranged according to the detection position, thermal excitation scale, thermal excitation direction and sampling time to generate thermal response records of the insulation material, providing a complete bidirectional thermal response basis for subsequent thermal response normalization and reliable verification.

[0024] In this embodiment, step three specifically includes: The thermal insulation performance evaluation platform uses the detection location as an index, reads the thermal response record of the thermal insulation material according to the three-level thermal excitation scale and two thermal excitation directions, extracts the baseline record 10 seconds before the start and end time of the thermal excitation, and extracts the record until the difference between the temperature response of the tested thermal path and the reference thermal path relative to the current baseline does not exceed 0.3℃, and the difference between the heat flow response of the tested thermal path and the reference thermal path does not exceed 2W / m² and is maintained continuously for 30 seconds, thus forming the thermal response segments of the tested thermal path and the reference thermal path. A sampling time sequence was established with a uniform time interval of 0.5s. The most recent sampling time was used to complete the correspondence between the two thermal responses. Records with a time difference of more than 20ms were treated as missing records. When the duration of continuous missing data does not exceed 1 second, piecewise linear interpolation is used to fill in the missing data; when it exceeds 1 second, the missing data marker is retained. The detection position corresponding to the reverse thermal excitation is converted according to the mirror order of the detection area to complete the thermal response regularization. The sampling coverage, continuous missing duration, dual-path synchronization correspondence rate, and heat flow direction consistency of each thermal response segment are verified one by one. The sampling coverage and dual-path synchronization correspondence rate are not less than 98%, the continuous missing duration is not more than 1 second, and the proportion of sampling points whose heat flow response direction is consistent with the heat excitation direction within the thermal excitation duration is not less than 90%. If the trust verification fails, the corresponding thermal response segment is re-acquired once. If it fails again, all thermal response segments at the corresponding detection location are removed. When all detection locations are eliminated, the current evaluation task is terminated and the thermal response reliability verification failure status is recorded; when there are still valid detection locations, subsequent normalization and tensor overlay are performed on the retained thermal response segments. The temperature response was subtracted from the corresponding baseline mean and then divided by 10℃, the heat flux response was subtracted from the corresponding baseline mean and then divided by 150W / m², the time segment was mapped to [0,1] and piecewise linear interpolation was used to unify it into 256 sampling points; Tensor overlay is performed according to the detection location, thermal excitation scale, thermal excitation direction, thermal path type, response type, and sampling point order to generate a reliable thermal response tensor, providing a fixed-dimensional dual-path thermal response input for improving the DeepONet model.

[0025] In this embodiment, the improved DeepONet model includes a thermal response coding layer, a spatiotemporal coordinate mapping layer, a thermal field feature coupling layer, and a thermal parameter field output layer. The thermal field feature coupling layer embeds a reference thermal path differential mechanism, specifically: The thermal insulation performance evaluation platform calls the improved DeepONet model with fixed parameters. Before deploying the improved DeepONet model, training is completed using standard samples with known equivalent thermal parameters and corresponding heat conduction simulation samples. The training is terminated and the network parameters are fixed when the loss does not decrease for 10 consecutive rounds. During online evaluation, the reliable thermal response tensor is read according to the detection location. Under each thermal excitation scale and thermal excitation direction, the temperature response and heat flow response at the same sampling time are combined into a two-dimensional thermal response vector in the order of temperature response first and heat flow response second. The vector is arranged according to 256 sampling times to form the thermal response sequence of the thermal path under test and the thermal response sequence of the reference thermal path, respectively. The thermal response coding layer uses a homogeneous branch network with shared coding parameters to process the thermal response sequence of the thermal path under test and the thermal response sequence of the reference thermal path. Each two-dimensional thermal response vector is projected by a 2×64 linear projection matrix and the GELU function. The projection result is subjected to a one-dimensional convolution with a kernel length of 5, a stride of 1 and two zero values ​​padded at both ends along the sampling time direction to generate 128-dimensional stage features. Then, the dimensions are unified by layer normalization. The three-level thermal excitation scale and two thermal excitation directions corresponding to the same detection position are arranged in a fixed order to generate a 6×256×128 thermal path code to be tested and a 6×256×128 reference thermal path code, so that the two codes remain corresponding under the same thermal excitation scale, thermal excitation direction and sampling time. The spatiotemporal coordinate mapping layer reads the detection position and the normalized sampling time. The horizontal and vertical coordinates of the detection position are normalized to [0,1] using minimum-maximum normalization. When the upper and lower limits of the coordinates are the same, the normalization result is recorded as zero. A 10^-6 numerical stability term is added to the normalization denominator. The horizontal coordinate, vertical coordinate, and normalized sampling time are arranged in a fixed order to form the three-dimensional query coordinates. The three-dimensional query coordinates are processed sequentially by a 3×128 linear projection matrix, a hyperbolic tangent activation function, and a 128×128 basis function mapping matrix to generate a 128-dimensional spatiotemporal coordinate code for each query coordinate. The thermal field feature coupling layer establishes a corresponding index of the thermal path code to be tested, the reference thermal path code, and the spatiotemporal coordinate code based on the thermal excitation scale, thermal excitation direction, and sampling time. The reference thermal path differential mechanism uses the reference thermal path code as a reference under the same operating condition to perform differential calibration on the thermal path code to be tested, generating a 6×256×128 calibration thermal response code. The calibration thermal response code is multiplied dimension-wise with the corresponding 128-dimensional spatiotemporal coordinate code at each thermal excitation scale, thermal excitation direction and sampling time. The product result is coupled with the operator feature through a 128×4 linear mapping matrix to generate the thermal field coupling feature corresponding to each query coordinate. The thermal parameter field output layer is set with four output channels, which correspond to the equivalent local thermal resistance state, the equivalent thermal diffusion state, the heat penetration state, and the reliable thermal parameter state in sequence. The linear outputs corresponding to the equivalent local thermal resistance state and the equivalent thermal diffusion state are denormalized according to the mean and standard deviation saved during the training phase. When the denormalization result is less than zero, it is treated as zero. The heat penetration state and the reliable thermal parameter state are restricted to the [0,1] interval by the Sigmoid function. Read the last 20% of the output of each thermal excitation duration interval and calculate the average value by channel to form the mapping result corresponding to the current thermal excitation scale and thermal excitation direction; When comparing the forward and reverse mapping results corresponding to the same detection location and the same thermal excitation scale channel by channel, and when the relative deviation between the equivalent local thermal resistance state and the equivalent thermal diffusion state does not exceed 8%, and the absolute deviation between the heat penetration state and the reliable thermal parameter state does not exceed 0.08, the average value of the mapping results for the two thermal excitation directions is taken. When any deviation exceeds the corresponding threshold, normalized weighting is performed according to the reliability states of the thermal parameters in the two thermal excitation directions. If the sum of the two reliability states of thermal parameters is less than 0.1, the current thermal excitation scale is marked as invalid. The remaining effective thermal excitation scales are normalized and weighted according to the reliability of thermal parameters to generate the equivalent thermal parameter mapping result corresponding to the current detection position; When all three thermal excitation scales are invalid, the current detection location is marked as an invalid detection location and is not written into the equivalent thermal parameter field of the insulation material; When all detection locations are marked as invalid detection locations, the current evaluation task is terminated and the thermal parameter mapping failure status is recorded. The remaining valid detection locations are merged to generate the equivalent thermal parameter field of the insulation material. Both the original DeepONet and the improved DeepONet models retain the basic structure of branch network encoding input function, backbone network encoding query coordinate, and the coupling of branch encoding and backbone encoding execution operators; The improved DeepONet model replaces the original single-path branch input with a dual-path isomorphic encoding of the thermal path under test and the reference thermal path using shared encoding parameters. A differential mechanism of the reference thermal path is embedded at the position where the original branch encoding and the backbone encoding are directly coupled, so that the thermal path encoding under test can complete differential calibration under the same operating conditions before participating in operator coupling. At the same time, the original single-value operator output is expanded into a four-channel mapping for equivalent local thermal resistance state, equivalent thermal diffusion state, heat penetration state, and thermal parameter reliability state. The improved positions correspond to the branch network input end, the branch network and backbone network coupling end, and the output end of the original DeepONet, respectively. Therefore, the improved DeepONet model can separate the synchronous changes of the thermal path under test and the reference thermal path while retaining the original DeepONet operator learning structure, forming an equivalent thermal parameter field of the insulation material with directional consistency, scale consistency and reliable state labeling, providing a stable equivalent thermal parameter basis for the extraction of abnormal locations and corresponding thermal states.

[0026] In this embodiment, the reference thermal differential mechanism is specifically as follows: The reference thermal path differential mechanism receives the thermal path code to be tested and the reference thermal path code. Both codes are arranged according to three thermal excitation scales, two thermal excitation directions, 256 sampling times and 128-dimensional features. A one-to-one correspondence is established based on the thermal excitation scale, thermal excitation direction, sampling time and feature dimensions. Using the code of the first sampling moment of each thermal excitation process as a benchmark, the code change of subsequent sampling moments relative to the corresponding benchmark is calculated to form the thermal path change sequence to be tested and the reference thermal path change sequence. When the absolute value of the encoding change is less than 0.02, it is recorded as zero change. When the same feature dimension maintains the same direction of change for 4 consecutive sampling times, it is determined as a continuous change interval. The two change sequences are compared item by item in terms of their change direction and continuous change interval under the same feature dimension. If the change direction is consistent, the overlap ratio of continuous change intervals is not less than 0.60, and the ratio of the difference in the amount of code change to the larger absolute value of the two types of code change does not exceed 0.15, they are identified as synchronous change components. When calculating the ratio of changes, a 10^-6 numerical stability term is added to the denominator. The smaller absolute value of the two coded changes is taken as the absolute value of the synchronous change component. The remaining change after deducting the synchronous change component in the thermal path change sequence under test is determined as the independent change component of the thermal path under test. The differential calibration result is obtained by subtracting the coding values ​​of the thermal path code corresponding to the sampling time and feature dimension based on the synchronous change components. According to the independent changing components of the thermal path under test under forward and reverse thermal excitation corresponding to the detection position, if the changing direction is consistent and the overlap ratio of continuous changing intervals is not less than 0.60, the corresponding differential calibration result is retained; if the changing direction is inconsistent or the overlap ratio of continuous changing intervals is less than 0.60, the corresponding code value is set to zero. The processing results are arranged in the order of original thermal excitation scale, thermal excitation direction, sampling time and feature dimension to generate calibration thermal response codes. After the reference thermal path code completes differential calibration, it no longer participates in operator feature coupling. The calibration thermal response code is used to weaken common thermal disturbances and retain the independent thermal response of the material that has obtained bidirectional thermal excitation support, providing calibrated branch codes for reconstructing the equivalent thermal parameter field of the insulation material in the thermal field feature coupling layer.

[0027] In this embodiment, step five specifically includes: The thermal insulation performance evaluation platform reads the equivalent thermal parameter field of the thermal insulation material, extracts the equivalent local thermal resistance state, equivalent thermal diffusion state, heat penetration state and thermal parameter reliability state according to the detection location, and matches the corresponding evaluation boundary according to the thermal insulation performance evaluation constraint set. The equivalent local thermal resistance state is evaluated using the lower limit, while the equivalent thermal diffusion state and the thermal penetration state are evaluated using the upper limit, in order to avoid misjudging changes in parameters with good thermal insulation performance as abnormal. When the equivalent local thermal resistance is not lower than the lower limit of the evaluation, and the equivalent thermal diffusion and heat penetration are not higher than the upper limit of the evaluation, the processing of the current detection location ends; when any of the states exceeds the corresponding evaluation boundary, the current detection location is determined as an abnormal location. When the equivalent local thermal resistance state is lower than the evaluation lower limit, the difference between the state value and the evaluation lower limit is divided by the evaluation lower limit to generate a negative normalized deviation. When the equivalent thermal diffusion state or thermal penetration state is higher than the evaluation upper limit, the difference between the state value and the evaluation upper limit is divided by the evaluation upper limit to generate a positive normalized deviation; the normalized deviation corresponding to the state that does not exceed the evaluation boundary is recorded as zero. Following a fixed order of equivalent local thermal resistance state, equivalent thermal diffusion state, and heat penetration state, the state values, corresponding normalized deviations, and reliable thermal parameter states are merged to generate the corresponding thermal state for the abnormal location. The horizontal and vertical coordinates of the abnormal location are associated with the corresponding thermal state to generate thermal anomaly spatial points. The thermal anomaly spatial point set is generated by summarizing the detection locations in order. If no abnormal location is formed, an empty thermal anomaly spatial point set is generated. Therefore, the spatial point set of thermal anomalies retains the anomaly location and actual thermal parameter state, while unifying the anomaly scale of different thermal parameters with normalized deviation, and carries the reliable state of thermal parameters, providing a consistent input for improving the HDBSCAN algorithm to perform thermal state matching, inter-point distance calculation and stable screening of thermal bridge clusters.

[0028] In this embodiment, step six specifically includes: The thermal insulation performance evaluation platform reads the set of thermal anomaly spatial points using a single evaluation task as the processing unit. When the set of thermal anomaly spatial points is empty, it generates a cluster of empty thermal insulation anomaly regions and ends the clustering process. When the set of thermal anomaly spatial points is not empty, spatial adjacency relationships are established according to the detection location. Thermal state-constrained nearest neighbors are used to construct and filter spatial adjacency relationships to generate an effective nearest neighbor set. The constraint set for thermal insulation performance evaluation specifies that the effective nearest neighbor order is 4 and the cluster size requirement is 6 thermal anomaly spatial points. The horizontal and vertical coordinates are normalized by minimum and maximum values ​​according to the upper and lower limits of the detection area. When the coordinate span is zero, the corresponding normalization result is recorded as zero. The three normalization deviations are divided by one and the sum of their absolute values, keeping the original positive and negative directions unchanged, so that the processing result is limited to between negative one and one. For two thermal anomaly spatial points in the effective nearest neighbor set, calculate the normalized horizontal coordinate difference and vertical coordinate difference respectively. Take the square root of the average of the squares of the two differences to obtain the spatial distance. Calculate the corresponding differences of the three processed deviations respectively. Take the square root of the average of the squares of the three differences and divide by 2 to obtain the thermal state distance. Subtract the smaller value of the two points' thermal parameter confidence states from one to obtain the confidence penalty term; multiply the spatial location distance, thermal state distance, and the square of the confidence penalty term by 0.35, 0.55, and 0.10 respectively, sum them, and take the square root to generate the distance between points; The effective neighbors of each thermal anomaly spatial point are sorted in ascending order of their inter-point distance. When the number of effective neighbors reaches 4, the 4th effective neighbor is selected, and the corresponding inter-point distance is used as the thermal state core distance. When the number of effective neighbors does not reach 4, the current thermal anomaly spatial point is marked as a non-core point and stops participating in hierarchical clustering. For two spatial points of thermal anomalies that have an effective nearest neighbor relationship, the maximum value between the thermal state core distance of each point and the distance between the two points is taken to generate the thermal state reach distance. A weighted graph is constructed using spatial points of thermal anomalies participating in hierarchical clustering as nodes, effective nearest neighbor relationships as connecting edges, and the distance between thermal states as edge weights. Check each edge in ascending order of edge weight, select the edge that can connect two nodes that are not yet in the same connected component, until all nodes in each connected component are connected, forming a minimum spanning tree; split the edges in the minimum spanning tree in descending order of edge weight, record the generation, splitting and disappearance process of each connected component, and construct a hierarchical clustering tree. When the thermal state reach distance is not zero, its reciprocal is taken as the density level; when the thermal state reach distance is zero, it is calculated as 0.000001 and its reciprocal is taken. When a branch in the hierarchical clustering tree contains no less than 6 thermal anomaly spatial points, the corresponding branch is retained; when it contains less than 6 thermal anomaly spatial points, no candidate anomaly cluster is generated, and the corresponding thermal anomaly spatial point is recorded as exiting the current density level. After all branches are compressed, candidate anomaly clusters are generated. The generation density level of each candidate anomaly cluster and the exit density level of each thermal anomaly spatial point within the cluster are recorded one by one. The difference between each exit density level and the generation density level is summed to obtain the stability of the candidate anomaly cluster. Determine the minimum outer detection area of ​​the candidate anomaly cluster covering the anomaly location, calculate the ratio of the number of anomaly locations contained in the current candidate anomaly cluster to the number of effective detection locations located within the minimum outer detection area in the equivalent thermal parameter field of the insulation material, and generate the continuous state coefficient. Calculate the absolute difference between the three processed deviations within the candidate anomaly cluster and their corresponding medians. Take the average of all absolute differences and use the natural exponential function to calculate the function value corresponding to the opposite of the average value to obtain the thermal state consistency value. The average value of the reliable thermal parameter states within the cluster is calculated, and the product of the thermal state consistency value and the average value of the reliable thermal parameter states is used as the consistency state coefficient. The continuous state coefficient is multiplied by 0.60 and the consistency state coefficient is multiplied by 0.40 and then summed to obtain the stability correction coefficient. The product of the candidate anomalous cluster stability and the stability correction coefficient is used as the corrected stability. Compare the corrected stability of the parent cluster with the sum of the corrected stability of each child cluster from the bottom up of the hierarchical clustering tree. If the corrected stability of the parent cluster is not less than the sum of the corrected stability of each child cluster, retain the parent cluster and delete the corresponding child cluster. If the corrected stability of the parent cluster is less than the sum of the corrected stability of each child cluster, retain each child cluster. According to the detection location, the thermal anomaly spatial points within the cluster are merged and retained. The cluster number is determined by the detection position of the first one in the cluster, and the thermal anomaly area cluster is generated. Both the original HDBSCAN algorithm and the improved HDBSCAN algorithm retain the core distance calculation, cross-distance construction, minimum spanning tree splitting, hierarchical clustering tree compression, and density clustering structure for cluster stability selection; The original HDBSCAN algorithm searches for nearest neighbors using a unified distance metric and determines the clustering results based on the persistent state of candidate clusters across density levels. The improved HDBSCAN algorithm adds thermal state-restricted nearest neighbor construction in the original nearest neighbor search stage, allowing only spatially continuous and correspondingly thermally matched hot anomaly spatial points to participate in the core distance calculation. In the original cluster stability selection stage, continuous state coefficients and consistent state coefficients are introduced to correct the stability of candidate anomaly clusters. The improved positions correspond to the nearest neighbor construction end and cluster selection end of the original HDBSCAN algorithm, respectively. Therefore, the improved HDBSCAN algorithm can prevent the formation of clusters of thermal anomaly spatial points that are close in location but do not match in thermal state. It can also screen continuous insulation anomaly regions from four aspects: density stability, anomaly location continuity, thermal state consistency and thermal parameter reliability. This provides a cluster of insulation anomaly regions with clear coverage and thermal state basis for the anomaly limit of the basic evaluation level.

[0029] In this embodiment, the thermally confined nearest neighbor construction specifically involves: The thermal insulation performance evaluation platform reads the spatial point set and spatial adjacency relationship of thermal anomalies, and the detection position is recorded using a unified planar coordinate system and millimeters as the unit of measurement; In this embodiment, the detection positions are arranged at equal intervals of 50 mm. The thermal insulation performance evaluation constraint set sets the upper limit of the adjacent distance to 75 mm, so that the detection positions that are horizontally adjacent, vertically adjacent, and diagonally adjacent are included in the spatial continuity verification. Read the horizontal and vertical coordinates of the thermal anomaly spatial points at both ends of each spatial adjacency relationship one by one, calculate the difference between the horizontal and vertical coordinates of the two points respectively, and take the square root of the sum of the squares of the two coordinate differences to obtain the detection position spacing; when the detection position spacing does not exceed 75 mm, the corresponding spatial adjacency relationship is retained, and when the detection position spacing exceeds 75 mm, the corresponding spatial adjacency relationship is deleted to form candidate adjacency relationships; Following a fixed order of equivalent local thermal resistance state, equivalent thermal diffusion state, and thermal penetration state, the normalized deviation corresponding to the thermal anomaly spatial points at both ends of the candidate adjacency relationship is read; when the absolute value of the normalized deviation is less than 0.02, it is treated as zero. When both normalized deviations corresponding to the same equivalent thermal parameter are zero, they are determined to be matched. When the two normalized deviations are in the same positive and negative direction and the absolute difference does not exceed 0.20, they are also determined to be matched. When one end is zero and the other end is non-zero, the positive and negative directions are inconsistent, or the absolute difference exceeds 0.20, they are determined to be mismatched. When the normalized deviations of the three equivalent thermal parameters match each other, the corresponding candidate adjacency relationship is retained; when any normalized deviation does not match, the corresponding candidate adjacency relationship is deleted. Relationship identifiers are generated in ascending order of the sequence numbers of the detection positions at both ends, and only one candidate adjacency relationship with the same relationship identifier is retained. For each thermal anomaly spatial point, other thermal anomaly spatial points that have a reserved candidate adjacency relationship with it are summarized to form the effective nearest neighbors corresponding to the current thermal anomaly spatial point. The effective nearest neighbors corresponding to the hot anomaly spatial points that do not retain candidate adjacency relationships are recorded as empty. The effective nearest neighbors corresponding to all hot anomaly spatial points are summarized to generate an effective nearest neighbor set. Therefore, the thermal state-restricted nearest neighbor construction uses the detection location spacing to limit the spatial connection range between anomalies, and uses three normalized deviations to block the connection relationship that does not match the anomaly type or degree, providing a spatially continuous and thermally state-matched nearest neighbor basis for the construction of thermal state core distance, thermal state mutual distance and hierarchical clustering tree.

[0030] In this embodiment, step seven specifically includes: The thermal insulation performance evaluation platform reads the equivalent thermal parameter field of the thermal insulation material and extracts the equivalent local thermal resistance state, equivalent thermal diffusion state, and heat penetration state according to the detection location. The thermal insulation performance evaluation constraint set specifies the evaluation levels as Level 1 to Level 4, with Level 1 being the highest and Level 4 being the lowest, and preserves the level boundaries of the three equivalent thermal parameters; the individual level of each of the three equivalent thermal parameters is determined one by one, and the lowest individual level is taken as the evaluation level of the current detection position; In order from Level 1 to Level 4, the proportion of the number of valid detection locations at the current evaluation level and those at or above the current evaluation level to the total number of valid detection locations in the equivalent thermal parameter field of the insulation material is counted. When the first proportion is not less than 80%, the corresponding evaluation level is determined as the basic evaluation level. Read the clusters of thermal insulation anomaly areas, calculate the proportion of the number of anomaly locations contained in each thermal insulation anomaly area cluster to the total number of effective detection locations in the equivalent thermal parameter field of the thermal insulation material, and generate the anomaly coverage ratio. For each thermal anomaly spatial point within the cluster, three normalized deviations are read, the maximum absolute value is selected, and then the median of all the maximum values ​​is calculated to generate the cluster anomaly degree. When the abnormal coverage ratio is less than 5% and the cluster abnormality degree is less than 0.20, no restriction level is generated; when the abnormal coverage ratio is not less than 10% or the cluster abnormality degree is not less than 0.50, level four is determined as the restriction level; for other abnormal insulation areas, clusters are determined as level three as the restriction level. When multiple restriction levels exist, the restriction level with the strictest restriction is selected. If the basic evaluation level is not higher than the restriction level, the basic evaluation level is retained. If it is higher than the restriction level, the restriction level is used to replace the basic evaluation level. If no restriction level is generated or the cluster of abnormal insulation areas is empty, the basic evaluation level is retained. The processed evaluation level is determined as the final evaluation level. The final evaluation level, the equivalent thermal parameter field of the insulation material and the cluster of insulation anomalies are summarized to generate the insulation performance evaluation result. Therefore, the thermal insulation performance evaluation results can take into account both the overall grade distribution and continuous abnormal areas, avoid local minor anomalies from changing the overall evaluation, and prevent stable thermal bridge areas from being masked by the overall grade.

[0031] Example 1: To verify the feasibility of the present invention in practice, the present invention was applied to the factory sampling inspection scenario of rigid polyurethane insulation boards for building exterior walls; 16 insulation boards with dimensions of 600 mm × 600 mm × 50 mm and apparent density of 34 ± 2 kg / m³ were selected, of which 4 boards were defect-free insulation boards, and the remaining 12 boards were marked with local thickness reduction areas, low density filling areas and interface debonding areas. Among them, 4 boards each had 2 known abnormal areas, and 8 boards each had 3 known abnormal areas, forming a total of 32 known abnormal areas. The abnormality locations and boundary ranges were determined based on manufacturing records and cross-sectional inspection results.

[0032] Each insulation board has a detection position set at a 50 mm interval, forming 169 detection positions per insulation board, for a total of 2704 detection positions. After thermal response normalization, reliability verification and re-acquisition, 2698 valid detection positions are retained. Each valid detection position includes three levels of thermal excitation scale, two thermal excitation directions, and the temperature and heat flow responses corresponding to the thermal path under test and the reference thermal path.

[0033] According to the method of this invention, the construction of the reliable thermal response tensor, the reconstruction of the equivalent thermal parameter field of the insulation material, and the generation of the thermal anomaly spatial point set are completed. The improved DeepONet model adopts 128-dimensional feature encoding and completes the coding calibration of the thermal path to be tested by referring to the thermal path difference mechanism, generating a total of 486 thermal anomaly spatial points. The improved HDBSCAN algorithm sets the effective nearest neighbor order to 4 and the cluster size requirement to 6 thermal anomaly spatial points. The thermal state-constrained nearest neighbor construction and thermal bridge cluster stability screening are used to generate the insulation anomaly region cluster, and the final evaluation level is determined by combining the equivalent thermal parameter field of the insulation material.

[0034] Ninety-six detection locations were selected for verification using calibrated heat flow detection devices. Thirty-two known abnormal areas were used as the basis for anomaly identification and verification. The actual evaluation level was determined by combining the cross-sectional inspection results of 16 insulation boards and the thermal performance verification results. The traditional unidirectional steady-state threshold evaluation method, the original DeepONet model combined with the original HDBSCAN algorithm, the improved DeepONet model combined with the original HDBSCAN algorithm, and the method of this invention were set up for comparison. Each method used the same insulation board, detection location, thermal response record, and evaluation boundary. The average relative error of equivalent local thermal resistance state, abnormal area identification rate, false alarm rate of abnormal location, average deviation of abnormal boundary, and consistency rate of evaluation level were statistically analyzed. The specific comparison results are shown in Table 1.

[0035] Table 1. Validation Results of Different Thermal Insulation Performance Evaluation Methods

[0036] As shown in Table 1, the average relative error of the equivalent local thermal resistance state of the method of the present invention is 4.8%, which is 4.1 percentage points lower than the original DeepONet model and the original HDBSCAN algorithm combination method. This indicates that the dual-path isomorphic coding and the reference thermal path differential mechanism can reduce the synchronous thermal response interference caused by thermal excitation fluctuations and environmental changes, and improve the accuracy of equivalent thermal parameter field reconstruction.

[0037] The improved DeepONet model combined with the original HDBSCAN algorithm achieved an anomaly region identification rate of 87.5% and an average deviation of 27.1 mm for anomaly boundaries. The corresponding indicators of the method in this invention reached 93.8% and 18.9 mm, respectively, indicating that thermal state-constrained nearest neighbor construction can limit the formation of erroneous connections between anomalies that are spatially close but have inconsistent thermal states, and thermal bridge cluster stable screening can retain spatially continuous and consistent anomaly regions.

[0038] Traditional unidirectional steady-state threshold evaluation methods are easily affected by local temperature fluctuations, with a false alarm rate of 12.1% for abnormal locations. The method of this invention uses thermal parameter deviation, spatial continuity, and cluster stability to jointly determine the abnormal region, reducing the false alarm rate to 3.7%. The method of this invention achieves a consistency rate of 93.8% in evaluation levels. Among 16 insulation boards, the evaluation results of 15 boards are consistent with the cross-sectional inspection results and thermal performance verification results. The evaluation level of 1 board deviates due to the presence of a complex abnormal state in the boundary area. This shows that the abnormality limit treatment can prevent local thermal anomalies from being masked by the overall evaluation results.

[0039] In summary, this embodiment forms a complete processing flow from thermal response acquisition, thermal parameter field reconstruction, abnormal region identification to the generation of thermal insulation performance evaluation results through multi-scale bidirectional thermal excitation, synchronous acquisition of the thermal path under test and the reference thermal path, construction of reliable thermal response tensor, reconstruction of thermal parameters of the improved DeepONet model, identification of abnormal regions in the improved HDBSCAN algorithm, and abnormal limit processing. This achieves reliable reconstruction of the equivalent thermal parameter field of the insulation material, accurate identification of continuous insulation abnormal regions, and reliable output of the thermal insulation performance evaluation level.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors, characterized in that, Includes the following steps: Step 1: Receive thermal insulation performance evaluation task information, parse the material to be evaluated and the evaluation conditions, call the thermal insulation performance evaluation benchmark library, and generate a thermal insulation performance evaluation constraint set; Step 2: Control the intelligent sensor group and thermal excitation device to perform multi-scale bidirectional thermal excitation according to the thermal insulation performance evaluation constraint set, and simultaneously collect the thermal response of the thermal path under test and the reference thermal path to generate thermal response records of the thermal insulation material; Step 3: Perform thermal response normalization and reliability verification on the thermal response records of the insulation material to generate a reliable thermal response tensor; Step 4: Input the reliable thermal response tensor into the improved DeepONet model, and generate the equivalent thermal parameter field of the insulation material through dual-path isomorphic encoding and differential coupling with the reference thermal path; Step 5: Process the equivalent thermal parameter field of the insulation material according to the thermal insulation performance evaluation constraint set, extract the abnormal locations and corresponding thermal states, and generate a set of thermal anomaly spatial points; Step 6: The improved HDBSCAN algorithm is used to process the spatial point set of thermal anomalies. After thermal state-constrained nearest neighbor construction and thermal bridge cluster stability screening, a cluster of thermal insulation anomaly regions is generated. Step 7: Determine the basic evaluation level according to the thermal insulation performance evaluation constraint set and the equivalent thermal parameter field of the thermal insulation material, and combine the abnormal limit of the thermal insulation abnormal area cluster to generate the thermal insulation performance evaluation result.

2. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 1, characterized in that, Step one specifically involves: Receive thermal insulation performance evaluation task information, classify the thermal insulation performance evaluation task information according to the material properties and working conditions corresponding to the record items, group the record items corresponding to the material properties into the material to be evaluated, and group the record items corresponding to the working conditions into the evaluation working conditions. Standardize the recording and expression of the evaluation materials, standardize the measurement units and recording formats of the evaluation conditions, and complete the analysis of the evaluation materials and evaluation conditions; According to the evaluation benchmark library of thermal insulation performance of the material to be evaluated, the candidate benchmark records corresponding to the material to be evaluated are screened, the evaluation conditions are compared with the working condition ranges corresponding to each candidate benchmark record, and the candidate benchmark records whose evaluation conditions do not fall into the working condition range are deleted. When a candidate benchmark record is retained, that candidate benchmark record is determined as the target benchmark record; When retaining multiple candidate benchmark records, sort them in ascending order of working condition range, and determine the candidate benchmark record that ranks first in the sorting as the target benchmark record; Read the target baseline record, merge the target baseline record with the thermal insulation performance evaluation task information, and generate a thermal insulation performance evaluation constraint set.

3. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 1, characterized in that, Step two specifically involves: The working positions of the intelligent sensor group and thermal excitation device are determined according to the thermal insulation performance evaluation constraint set, so that the thermal path under test and the thermal insulation material form a thermal transfer connection, and the thermal path under test and the reference thermal path are collected synchronously. The thermal excitation device controls the thermal excitation to apply thermal excitation sequentially from one side of the thermal insulation material according to the thermal excitation scale specified in the thermal insulation performance evaluation constraint set, and the intelligent sensor group simultaneously collects the thermal response of the thermal path under test and the reference thermal path under each thermal excitation scale. After the acquisition of each thermal excitation scale is completed, thermal excitation is stopped. When the thermal circuit under test and the reference thermal circuit meet the start-up conditions of the next thermal excitation scale, the next thermal excitation scale is executed. After completing the acquisition in one thermal excitation direction, switch the working side of the thermal excitation device and perform thermal excitation in another thermal excitation direction in the same thermal excitation scale sequence, while simultaneously acquiring the thermal response of the thermal circuit under test and the reference thermal circuit; The collected thermal responses are recorded according to the thermal excitation scale and direction to generate thermal response records for thermal insulation materials.

4. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 1, characterized in that, Step three specifically involves: Read the thermal response record of the thermal insulation material, group the thermal response of the test thermal circuit and the reference thermal circuit according to the thermal excitation scale and thermal excitation direction, and divide the thermal response segment according to the start and end times of thermal excitation corresponding to each group of thermal response. According to the sampling time, the thermal response segments of the thermal path under test and the reference thermal path under the same thermal excitation scale and thermal excitation direction are matched, and the detection positions corresponding to the reverse thermal excitation are converted into the detection position sequence consistent with the forward thermal excitation to complete the thermal response regularization. Each thermal response segment is checked to ensure that it completely covers the corresponding thermal excitation process, that adjacent sampling times are continuous, that the thermal path under test and the reference thermal path are synchronously corresponding, and that the thermal response change is consistent with the direction of thermal excitation. Thermal response segments that fail the verification are removed to complete the reliability verification. According to the sampling length specified in the thermal insulation performance evaluation constraint set, the thermal response segments that have passed the credibility verification are resampled. The thermal responses of the tested thermal path and the reference thermal path after resampling are arranged in sequence according to the thermal excitation scale and thermal excitation direction, and tensor overlay is performed to generate a credibility thermal response tensor.

5. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 1, characterized in that, The improved DeepONet model includes a thermal response coding layer, a spatiotemporal coordinate mapping layer, a thermal field feature coupling layer, and a thermal parameter field output layer. The thermal field feature coupling layer embeds a reference thermal path differential mechanism. The thermal response coding layer reads the reliable thermal response tensor and splits the thermal response of the thermal path under test and the reference thermal path according to the thermal excitation scale and thermal excitation direction. The two thermal responses are input into a homogeneous branch network with shared coding parameters, and feature projection, stage feature extraction and dimension unification are performed sequentially to make the same coding position in the two thermal responses represent the same thermal excitation process, generating a test thermal path code and a reference thermal path code with consistent dimension and arrangement order. The spatiotemporal coordinate mapping layer reads the detection position and sampling time corresponding to each thermal response in the reliable thermal response tensor; Query coordinates are constructed based on the detection location and sampling time. The query coordinates are input into the backbone network to perform coordinate projection, position association and basis function mapping, so that the coordinate features obtained by mapping match the corresponding positions of the two codes, and generate spatiotemporal coordinate codes. The thermal field feature coupling layer receives the thermal path code to be tested, the reference thermal path code, and the spatiotemporal coordinate code, and establishes the correspondence between the two codes and the spatiotemporal coordinate code based on the thermal excitation scale, thermal excitation direction, and sampling time. The reference thermal path differential mechanism is used to differentially calibrate the reference thermal path code to be tested, thereby generating a calibration thermal response code; According to the corresponding coding position, the calibration thermal response coding and spatiotemporal coordinate coding are combined dimensionally and coupled with operator features to generate thermal field coupling features corresponding to each detection position. The thermal parameter field output layer performs equivalent thermal parameter mapping on the thermal field coupling features and arranges the mapping results according to the detection position. Merge consistent mapping results from the same detection location under different thermal excitation scales and directions, and correct inconsistent results based on the changes in the corresponding mapping results to generate an equivalent thermal parameter field for the insulation material.

6. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 5, characterized in that, The reference thermal differential mechanism is specifically as follows: Read the thermal path code to be tested and the reference thermal path code, and establish the corresponding code position according to the thermal excitation scale, thermal excitation direction and sampling time; Calculate the change of each corresponding coding position relative to the starting coding position of this thermal excitation scale to form the thermal path change sequence to be tested and the reference thermal path change sequence. Compare the direction of change and the range of continuous change between the thermal path change sequence under test and the reference thermal path change sequence at the corresponding coding positions; The coded changes with the same direction of change and overlapping continuous change intervals are identified as synchronous change components, and the remaining coded changes in the thermal path change sequence under test, excluding synchronous change components, are identified as independent change components of the thermal path under test. Based on the synchronously changing components, the corresponding coding position of the thermal path code to be tested is subtracted to obtain the differential calibration result; Establish the correspondence between the independent changing components of the thermal path under test under positive and reverse thermal excitation according to the detection location, and retain the independent changing components of the thermal path under test with consistent changing direction and continuous changing intervals overlapping each other. Zero-value replacement is performed on the coded values ​​with inconsistent change directions or continuous change intervals that do not overlap with each other. The differential calibration results are updated according to the original coded position to generate a calibration thermal response code that is consistent with the coded dimension of the thermal path to be tested.

7. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 1, characterized in that, Step five specifically involves: Read the equivalent thermal parameter field of the insulation material, extract the equivalent thermal parameters corresponding to each detection location according to the detection location, and match the corresponding evaluation boundary for each equivalent thermal parameter based on the thermal insulation performance evaluation constraint set; The equivalent thermal parameters corresponding to each detection location are compared with the evaluation boundary one by one. When the equivalent thermal parameters are all within the evaluation boundary, the processing of the current detection location ends. When any equivalent thermal parameter exceeds the evaluation boundary, the current detection location is determined as an abnormal location; Calculate the deviation of the equivalent thermal parameters corresponding to the abnormal location from the evaluation boundary, merge the equivalent thermal parameters and deviation results corresponding to the abnormal location according to the original detection location, and generate the corresponding thermal state of the abnormal location. The abnormal locations and their corresponding thermal states are associated with each other and encoded to generate thermal anomaly spatial points. The thermal anomaly spatial points are arranged according to the detection locations to generate a set of thermal anomaly spatial points. If no abnormal location is formed, an empty thermal anomaly spatial point set is generated.

8. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 1, characterized in that, Step six specifically involves: Read the set of thermal anomaly spatial points. If the set of thermal anomaly spatial points is empty, do not retain the thermal anomaly spatial points. If the set of thermal anomaly spatial points is not empty, establish the spatial adjacency relationship between the thermal anomaly spatial points according to the detection position. Thermal state-restricted nearest neighbor construction filters spatial adjacency relationships based on the spatial continuity relationship of adjacent thermal anomaly spatial points and the corresponding thermal state matching relationship, generating an effective nearest neighbor set; Within the effective nearest neighbor set, the distance between points is calculated based on the differences in detection location and corresponding thermal state, and the effective nearest neighbors of each thermal anomaly spatial point are arranged in ascending order of the distance between points. When the number of effective neighbors reaches the specified order of the thermal insulation performance evaluation constraint set, the effective neighbors of the corresponding order are selected, and the distance between the corresponding points is used as the thermal state core distance. Thermal anomaly spatial points whose number of effective neighbors does not reach the specified order are not included in the subsequent calculation. The maximum value between the thermal state core distance of each of two adjacent points and the distance between the two points is taken as the thermal state reach distance. A weighted graph is constructed with the thermal anomaly spatial points participating in the subsequent calculation as nodes and the thermal state reach distance as edge weights, and the minimum spanning tree is extracted. The connection relationships in the minimum spanning tree are split according to the edge weight from large to small. The splitting process of each connected component as the edge weight changes is recorded, and a hierarchical clustering tree is constructed. Based on the thermal insulation performance evaluation constraint set, branches that do not meet the cluster size requirements are deleted, the hierarchical clustering tree is compressed, and candidate abnormal clusters are generated; Thermal bridge cluster stability screening is performed on candidate anomalous clusters. The density level is represented by the reciprocal of the thermal state reach distance. The stability of candidate anomalous clusters is calculated according to the duration interval of candidate anomalous clusters in different density levels. The stability of the candidate anomaly cluster is corrected based on the continuous state of the anomaly location covered by the candidate anomaly cluster and the consistent state of the corresponding thermal state. The corrected stability of the parent cluster is compared with the sum of the corrected stability of each child cluster. If the corrected stability of the parent cluster is not less than the sum of the corrected stability of each child cluster, the parent cluster is retained; otherwise, each child cluster is retained. If no candidate anomalous cluster is formed, thermal anomaly spatial points are not retained. The retained thermal anomaly spatial points are merged according to the detection location to generate a thermal insulation anomaly region cluster.

9. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 8, characterized in that, The thermally constrained nearest neighbor construction specifically refers to: Read the set of thermal anomaly spatial points and spatial adjacency relationships, determine the thermal anomaly spatial points at both ends of each spatial adjacency relationship one by one, and verify the detection position distance between the two points according to the adjacency distance range specified in the thermal insulation performance evaluation constraint set; When the detection location spacing falls within the adjacency distance range, the corresponding spatial adjacency relationship is retained; when the detection location spacing exceeds the adjacency distance range, the corresponding spatial adjacency relationship is deleted, and candidate adjacency relationships that satisfy the spatial continuity relationship are obtained. Read the corresponding thermal states of the thermal anomaly spatial points at both ends of each candidate adjacency relationship, and compare the deviation results in the corresponding thermal states one by one according to the same evaluation criteria. When the deviation directions are consistent and the difference between the deviation results falls within the matching range specified by the thermal insulation performance evaluation constraint set, the corresponding thermal states of the two thermal anomaly spatial points at both ends are considered to match each other, and the corresponding candidate adjacency relationship is retained. If the deviation direction is inconsistent or the difference exceeds the matching range, the corresponding candidate adjacency relationship will be deleted. The candidate adjacency relationships are summarized and retained with each thermal anomaly spatial point as the center. Duplicate relationships with the same connection endpoints are deleted to form the effective nearest neighbors corresponding to each thermal anomaly spatial point. For thermal anomaly spatial points that do not retain candidate adjacency relationships, their effective nearest neighbors are recorded as empty. The effective nearest neighbors corresponding to each thermal anomaly spatial point are summarized to generate an effective nearest neighbor set.

10. The method for evaluating the thermal insulation performance of thermal insulation materials based on intelligent sensors according to claim 1, characterized in that, Step seven specifically involves: Read the equivalent thermal parameter field of the insulation material, extract the equivalent thermal parameters corresponding to each detection location according to the detection location, and determine the evaluation level corresponding to each detection location based on the insulation performance evaluation constraint set; According to the evaluation level from high to low, the proportion of the current evaluation level and the detection positions with higher evaluation levels are counted in turn, and the first evaluation level that meets the conditions specified in the thermal insulation performance evaluation constraint set is determined as the basic evaluation level. Read the clusters of thermal insulation abnormal areas, extract the detection location and corresponding thermal state covered by each thermal insulation abnormal area cluster one by one, and verify each thermal insulation abnormal area cluster according to the thermal insulation performance evaluation constraint set. If the abnormality limit conditions are not met, the basic evaluation level will not be changed; if the abnormality limit conditions are met, the corresponding limit level will be determined. When multiple restriction levels exist, the restriction level with the most stringent restriction degree is selected according to the order of evaluation levels; if the basic evaluation level is not higher than the restriction level, the basic evaluation level is retained. When the basic evaluation level is higher than the limit level, the limit level is used instead of the basic evaluation level. When the cluster of areas with abnormal insulation is empty, the basic evaluation level is retained; The evaluation level after the anomaly limit is determined as the final evaluation level. The final evaluation level, the equivalent thermal parameter field of the insulation material, and the cluster of insulation anomaly areas are summarized to generate the insulation performance evaluation result.

Images