Methods, systems, equipment and media for assessing creep damage in plate heat exchanger welds

Abstract

This invention discloses a method, system, equipment, and medium for assessing creep damage in weld seams of plate heat exchangers, relating to the field of non-destructive testing technology for industrial equipment. The method includes: acquiring infrared thermal images of the weld seam area and equipment monitoring data; preprocessing the infrared thermal images and labeling the damage state based on the preprocessed images and equipment monitoring data to construct a training sample set; constructing an improved convolutional neural network, iteratively optimizing the parameters of the improved convolutional neural network using the training sample set, and determining the trained network as a damage assessment model; evaluating real-time monitoring data using the damage assessment model, and visualizing the evaluation results. This invention establishes the intrinsic correlation between the macroscopic temperature field and the microscopic damage of the material through non-contact thermal imaging acquisition and intelligent algorithm analysis, ultimately achieving real-time, accurate, and quantitative assessment of creep damage in weld seams of in-service plate heat exchangers.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for industrial equipment, and in particular to a method, system, equipment, and medium for assessing creep damage in weld seams of plate heat exchangers. Background Technology

[0002] Plate heat exchangers, as key equipment in industries such as petroleum, chemical, and power, operate under harsh conditions of high temperature and high pressure for extended periods, making them highly susceptible to creep damage in the welds. Creep is the phenomenon of slow plastic deformation of materials over time under constant stress and high temperature. Accumulated creep damage leads to defects such as pores and cracks in the weld microstructure, potentially causing pipeline leaks or ruptures, seriously threatening production safety. Currently, creep damage detection for this equipment mainly relies on traditional non-destructive testing, destructive metallographic analysis, and estimations based on operational experience. These methods generally suffer from drawbacks such as requiring shutdown and disassembly, inability to achieve online monitoring, or limitation to partial sampling rather than comprehensive assessment. Furthermore, destructive analysis is unsuitable for health diagnosis of in-service equipment, and experience-based estimations fail to accurately reflect the actual damage variations of individual equipment.

[0003] The core challenge in existing technologies lies in the inability to conduct large-area, rapid, and quantitative assessment and prediction of early creep damage in welds of in-service plate heat exchangers. Although non-contact infrared thermal imaging technology can acquire the temperature field distribution on the equipment surface, providing a new approach for condition monitoring, there is a complex nonlinear relationship between the microscopic creep damage inside the weld and its macroscopic surface temperature field. Conventional analysis of temperature distribution alone cannot directly and accurately infer the evolution of microscopic damage within the material, thus hindering precise diagnosis and early warning of early damage.

[0004] Therefore, there is an urgent need in this field for an intelligent assessment method that can effectively integrate macroscopic thermal imaging information with microscopic damage mechanisms. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, device and medium for assessing creep damage in plate heat exchanger welds, aiming to solve or improve at least one of the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] Methods for assessing creep damage in plate heat exchanger welds include:

[0008] Acquire infrared thermal images of the weld area and equipment monitoring data; the equipment monitoring data includes a two-dimensional temperature field data sequence and process parameters.

[0009] The infrared thermal image is preprocessed, and damage status is labeled based on the preprocessed image and equipment monitoring data to construct a training sample set;

[0010] An improved convolutional neural network is constructed, and the parameters of the improved convolutional neural network are iteratively optimized using the training sample set. The trained network is then determined as the damage assessment model.

[0011] The damage assessment model is used to evaluate real-time monitoring data to obtain an assessment result corresponding to the current state. The assessment result is then visualized on a pre-constructed three-dimensional model of the plate heat exchanger, and a multi-level early warning mechanism is activated based on preset thresholds. The assessment result includes creep damage level assessment and remaining life prediction value.

[0012] Optionally, the step of preprocessing the infrared thermal image and labeling the damage state based on the preprocessed image and device monitoring data to construct a training sample set specifically includes:

[0013] The infrared thermal image is spatially aligned with a pre-constructed 3D model of a plate heat exchanger, and the damage status of the thermal image is labeled based on finite element simulation and historical destructive detection data. Then, the original image data is rotated, translated, and scaled to perform data augmentation, resulting in a completed training sample set.

[0014] Optionally, the step of iteratively optimizing the parameters of the improved convolutional neural network using the training sample set specifically includes:

[0015] The training sample set is input into the network. First, data features are extracted from the infrared thermal image to obtain macroscopic thermal features used to characterize the damage state. Then, the macroscopic thermal features are spliced ​​and fused with process parameters to obtain fused features. The macroscopic thermal features include the highest temperature, lowest temperature, average temperature, temperature standard deviation, temperature field gradient distribution, and temperature field texture features based on the gray-level co-occurrence matrix.

[0016] Then, a recurrent neural network is used to process the time series data to capture the dynamic evolution of damage over time. The Softmax function is used to output a discrete probability distribution of damage level, and a linear activation function is used to output a continuous prediction of remaining life.

[0017] Optionally, the parameter iterative optimization process employs a cross-entropy loss function or a mean squared error loss function, and utilizes the Adam optimizer for parameter iterative optimization.

[0018] Optionally, the improved convolutional neural network is based on the traditional architecture with the following optimizations:

[0019] First, for the multi-scale feature fusion module:

[0020] Parallel convolution kernel design is adopted: three different sizes of convolution kernels, 3×3, 5×5 and 7×7, are set to extract local subtle features, mesoscale texture features and global distribution features of the weld temperature field, respectively;

[0021] Introducing a feature pyramid network: Through top-down feature fusion and lateral connections, feature maps at different levels are weighted and fused to enhance the sensitivity to early, weak damage thermal features.

[0022] Secondly, an attention mechanism is embedded:

[0023] Add a channel attention module after the key convolutional layer of the network: by adaptively learning the weights of each temperature feature channel, highlight the thermal features that are strongly correlated with creep damage and suppress background noise;

[0024] Combined with the spatial attention module: spatial weights are assigned to the weld area, focusing on the heat-affected zone of the weld that is prone to creep damage, and reducing the interference of non-critical areas on the model;

[0025] Add residual connections and dropout layers: The residual blocks alleviate the gradient vanishing problem in deep networks, while a regularization layer with a dropout rate of 0.3 is set before the fully connected layers to prevent the model from overfitting.

[0026] This invention also provides a system for assessing creep damage in plate heat exchanger welds, comprising:

[0027] The data acquisition unit is used to acquire infrared thermal images of the weld area and equipment monitoring data; the equipment monitoring data includes a two-dimensional temperature field data sequence and process parameters.

[0028] The sample construction unit is used to preprocess the infrared thermal image and label the damage status based on the preprocessed image and equipment monitoring data to construct a training sample set.

[0029] The model training unit is used to construct an improved convolutional neural network, perform iterative parameter optimization on the improved convolutional neural network using the training sample set, and determine the trained network as a damage assessment model.

[0030] The model evaluation unit is used to evaluate real-time monitoring data using the damage evaluation model, obtain evaluation results corresponding to the current state, visualize the evaluation results on a pre-constructed three-dimensional model of the plate heat exchanger, and activate a multi-level early warning mechanism based on preset thresholds; wherein, the evaluation results include creep damage level assessment and remaining life prediction value.

[0031] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-described plate heat exchanger weld creep damage assessment method.

[0032] The present invention also provides a computer-readable storage medium, characterized in that it stores a computer program, which, when executed by a processor, implements the method for assessing creep damage of plate heat exchanger welds as described above.

[0033] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0034] This invention discloses a method, system, equipment, and medium for assessing creep damage in weld seams of plate heat exchangers. The method includes acquiring infrared thermal images of the weld seam area and equipment monitoring data; preprocessing the infrared thermal images and labeling the damage state based on the preprocessed images and equipment monitoring data to construct a training sample set; constructing an improved convolutional neural network, using the training sample set to iteratively optimize the parameters of the improved convolutional neural network, and determining the trained network as the damage assessment model; using the damage assessment model to evaluate real-time monitoring data, and visualizing the evaluation results. This invention establishes the intrinsic correlation between the macroscopic temperature field and the microscopic damage of the material through non-contact thermal imaging acquisition and intelligent algorithm analysis, ultimately achieving real-time, accurate, and quantitative assessment of creep damage in weld seams of in-service plate heat exchangers. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a flowchart illustrating the method for assessing weld creep damage in a plate heat exchanger in this embodiment.

[0037] Figure 2 This is a schematic diagram of the plate heat exchanger weld creep damage assessment system in this embodiment. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] The purpose of this invention is to provide a method, system, device and medium for assessing creep damage in plate heat exchanger welds, aiming to solve or improve at least one of the above-mentioned technical problems.

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] like Figure 1 As shown, the present invention provides a method for assessing creep damage in weld seams of plate heat exchangers, comprising:

[0042] Step 100: Acquire infrared thermal images of the weld area and equipment monitoring data; the equipment monitoring data includes a two-dimensional temperature field data sequence and process parameters. Specifically:

[0043] Under standard operating conditions, the plate heat exchanger is scanned using a high-resolution infrared thermal imager to obtain a two-dimensional temperature field data sequence covering the entire weld surface; key process parameters, including medium pressure and temperature, are collected simultaneously, as shown in Table 1.

[0044] Table 1 Key Process Parameters

[0045]

[0046] Step 200: Preprocess the infrared thermal image, and label the damage state based on the preprocessed image and device monitoring data to construct a training sample set. Specifically:

[0047] The acquired infrared thermal images are preprocessed, including: spatially aligning the images with the three-dimensional model of the plate heat exchanger; labeling the damage status of the thermal images based on finite element simulation, historical destructive detection data, or domain expert experience; and performing operations such as rotation, translation, and scaling on the image data to perform data augmentation and expand the training sample set. Specific features and dimensions are shown in Table 2.

[0048] Table 2 Specific Feature Items and Dimensions

[0049]

[0050] Step 300: Construct an improved convolutional neural network, iteratively optimize the parameters of the improved convolutional neural network using the training sample set, and determine the trained network as the damage assessment model. Specifically:

[0051] After applying the training sample set to the improved convolutional neural network, macroscopic thermal features characterizing the damage state are extracted from the preprocessed infrared thermal images. These features include, but are not limited to, maximum temperature, minimum temperature, average temperature, temperature standard deviation, temperature field gradient distribution, and temperature field texture features based on the gray-level co-occurrence matrix. Subsequently, the extracted macroscopic thermal features are fused with synchronously acquired operating process parameters in a multimodal manner to form a feature tensor that comprehensively describes the equipment state.

[0052] During training, either the cross-entropy loss function or the mean squared error loss function is used, and the Adam optimizer is employed for iterative parameter optimization. For the loss function:

[0053] Damage level classification loss (cross-entropy loss):

[0054] For discrete damage levels (0-4), weighted cross-entropy loss is used to address the sample imbalance problem (few samples are available for early damage):

[0055] ;

[0056] in, For cross-entropy loss, N For the sample size, c Damage level, y ic For the sample i The true label (unique hot encoding). p ic The predicted rank probability by the model. w c For weights.

[0057] Remaining lifetime prediction loss (mean squared error loss):

[0058] For continuous remaining lifetime values, mean squared error (MSE) combined with L1 regularization is used:

[0059] ;

[0060] in, For mean square error loss, y i For the sample i The actual remaining lifespan (hours). For predicted values, λ This is the regularization coefficient (taken as 0.001). θk For the first k Each model weight parameter, M This represents the total number of parameters.

[0061] The specific training steps are as follows:

[0062] A dedicated deep learning network model is constructed, employing an improved convolutional neural network architecture to automatically learn deep spatial damage features from infrared thermal images. The high-level feature vectors extracted by the convolutional neural network are concatenated and fused with process parameter features. For time-series data processing, a recurrent neural network module is further integrated to capture the dynamic evolution of damage over time. The model's output layer is configured according to task requirements: either a softmax function is used to output a discrete damage level probability distribution, or a linear activation function is used to output continuous remaining life predictions. Finally, the network is trained using the labeled dataset processed in step 200, employing either cross-entropy loss or mean squared error loss, and iteratively optimizing the model parameters using the Adam optimizer.

[0063] For the assessment of weld creep damage in plate heat exchangers, the improved convolutional neural network (CNN) has been optimized based on the traditional architecture as follows:

[0064] ① Multi-scale feature fusion module

[0065] Parallel convolutional kernel design is adopted: three different sizes of convolutional kernels, 3×3, 5×5 and 7×7, are set in the shallow layer of the network to extract local subtle features, mesoscale texture features and global distribution features of the weld temperature field, respectively, to solve the problem that a single convolutional kernel is not enough to capture complex thermal features.

[0066] Introducing Feature Pyramid Network (FPN): Through top-down feature fusion and lateral connections, feature maps at different levels (such as shallow edge features and deep semantic features) are weighted and fused to enhance the sensitivity to early weak damage thermal features.

[0067] ②Attention mechanism embedding

[0068] A channel attention module (SENet) is added after the key convolutional layer of the CNN: by adaptively learning the weights of each temperature feature channel, thermal features strongly correlated with creep damage are highlighted and background noise is suppressed.

[0069] Combined with the spatial attention module: spatial weights are allocated to the weld area, focusing on the heat-affected zone of the weld that is prone to creep damage, and reducing the interference of non-critical areas on the model.

[0070] Add residual connections and dropout layers: The residual blocks alleviate the gradient vanishing problem in deep networks, while a regularization layer with a dropout rate of 0.3 is set before the fully connected layers to prevent the model from overfitting.

[0071] Step 400: The real-time monitoring data is evaluated using the damage assessment model to obtain an assessment result corresponding to the current state. This assessment result is then visualized on a pre-constructed 3D model of the plate heat exchanger, and a multi-level early warning mechanism is activated based on preset thresholds. The assessment result includes creep damage level assessment and remaining life prediction. Specifically:

[0072] By inputting preprocessed data obtained from online monitoring into a trained deep learning model, the model can output in real time the creep damage level assessment or remaining life prediction corresponding to the current state. Finally, the above assessment and prediction results are displayed intuitively on a 3D model of the plate heat exchanger, and a multi-level early warning mechanism is activated based on preset thresholds to support equipment maintenance decisions.

[0073] like Figure 2 As shown, this invention also provides a plate heat exchanger creep damage assessment system based on infrared thermal imaging and deep learning. It adopts a layered architecture design, sequentially including a data acquisition layer, a data preprocessing and communication layer, an intelligent analysis layer, and an application display layer. The data acquisition layer consists of an infrared thermal imaging module and a sensor module. The infrared thermal imaging module automatically scans and acquires the temperature field distribution on the weld surface using a high-resolution infrared thermal imager, a controllable gimbal, and a protective cooling cover. The sensor module collects medium process parameters in real time using pressure and temperature sensors. The data preprocessing (including a data acquisition unit, a sample construction unit, a model training unit, and a model evaluation unit) and communication layer includes a data preprocessing server and an industrial network. It is responsible for spatial alignment, enhancement, and primary feature extraction of the raw thermal image data, and achieves stable and real-time data transmission between layers via wired Ethernet or wireless 5G networks.

[0074] The intelligent analysis layer includes a model server and a system database. The model server deploys and runs trained deep learning evaluation models, performing intelligent diagnosis and analysis of online data. The system database centrally stores historical detection data, model parameters, and real-time evaluation results. The application presentation layer includes a monitoring terminal and an alarm module. The monitoring terminal provides a human-machine interactive visual interface to display damage assessment results, equipment 3D models, and historical status trends. The alarm module initiates multi-level early warnings based on the diagnostic results, promptly notifying relevant personnel through various means such as sound and light, SMS, and email, thus forming a complete closed-loop system from data acquisition to status early warning.

[0075] In summary, during creep damage in plate heat exchanger welds, changes in the material's microstructure (such as pore formation and crack propagation) cause a regular evolution in thermophysical properties such as thermal conductivity and heat capacity, as well as mechanical properties. These changes directly affect the heat conduction and heat dissipation characteristics of the equipment under constant operating loads, resulting in highly specific "thermal signatures" on its surface temperature field that differ significantly from those in its healthy state. These signatures and the degree of creep damage constitute a complex and deep nonlinear mapping relationship, which is difficult to accurately establish using traditional physical models or empirical analytical methods.

[0076] However, traditional image processing or temperature threshold alarm methods can only capture apparent temperature anomalies and cannot effectively interpret the complex "thermal feature" patterns that contain damage information. Therefore, this invention introduces a deep learning model, particularly a convolutional neural network (CNN), which excels at automatically learning hierarchical features from large-scale data. Through end-to-end training, the CNN model can adaptively extract features at different scales from the input infrared thermal image, ranging from shallow edges and textures to deep semantics, and ultimately accurately fit a high-order nonlinear correlation between the "thermal image" and the "damage state."

[0077] The intrinsic physical relationship between the evolution of microscopic damage, changes in macroscopic thermophysical properties, and variations in surface temperature field is transformed into a mathematical mapping problem that can be approximated with high precision through deep learning. Using a trained convolutional neural network, this invention achieves indirect, intelligent quantitative measurement and state recognition of invisible microscopic creep damage, thereby constructing an end-to-end intelligent diagnostic model that directly assesses microscopic damage from macroscopic thermal imaging.

[0078] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0079] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for evaluating the creep damage of a welded joint of a plate heat exchanger, characterized in that, The method comprises the following steps: acquiring an infrared thermal image of a weld area and equipment monitoring data; the equipment monitoring data comprises a two-dimensional temperature field data sequence and process parameters; preprocessing the infrared thermal image, and labeling a damage state according to the preprocessed image and the equipment monitoring data to construct a training sample set; constructing an improved convolutional neural network, iteratively optimizing parameters of the improved convolutional neural network by using the training sample set, and determining the trained network as a damage evaluation model; evaluating real-time monitoring data by using the damage evaluation model to obtain an evaluation result corresponding to a current state, visualizing the evaluation result on a pre-constructed plate heat exchanger three-dimensional model, and starting a multi-level early warning mechanism according to a preset threshold; wherein the evaluation result comprises a creep damage grade evaluation and a residual life prediction value; the iteratively optimizing parameters of the improved convolutional neural network by using the training sample set comprises the following steps: inputting the training sample set into the network, first extracting data features of the infrared thermal image to obtain macro thermal features for representing the damage state, and then fusing the macro thermal features and the process parameters to obtain fused features; the macro thermal features comprise a maximum temperature, a minimum temperature, an average temperature, a temperature standard deviation, a temperature field gradient distribution, and a temperature field texture feature based on a gray level co-occurrence matrix; then, processing time series data by using a recurrent neural network to capture the dynamic evolution law of the damage over time, and outputting a discrete damage grade probability distribution by using a Softmax function and a continuous residual life prediction value by using a linear activation function; the improved convolutional neural network is optimized on the basis of a traditional architecture as follows: firstly, for a multi-scale feature fusion module: parallel convolution kernel design is adopted: three different size convolution kernels of 3x3, 5x5 and 7x7 are set to extract local fine features, medium-scale texture features and global distribution features of the weld temperature field respectively; a feature pyramid network is introduced: features of different levels are weighted and fused through top-down feature fusion and horizontal connection to enhance the sensitivity to early weak damage thermal features; secondly, an attention mechanism is embedded: a channel attention module is added after a key convolution layer of the network: the weights of each temperature feature channel are adaptively learned to highlight the thermal features strongly related to the creep damage and suppress the background noise; a spatial attention module is combined: spatial weights are allocated to the weld area to focus on the weld heat affected zone prone to creep damage and reduce the interference of non-key areas on the model; a residual connection and a dropout layer are added: the residual block is used to relieve the gradient vanishing problem of the deep network, and a dropout rate of 0.3 is set in the regularization layer before the full connection layer to prevent model overfitting.

2. The plate heat exchanger weld creep damage assessment method of claim 1, wherein, the preprocessing the infrared thermal image, and labeling a damage state according to the preprocessed image and the equipment monitoring data to construct a training sample set comprises the following steps: The infrared thermal image is spatially aligned with a pre-constructed plate heat exchanger three-dimensional model, and the damage state of the thermal image is labeled based on finite element simulation and historical destructive detection data, and then the original image data is rotated, translated and scaled for data enhancement to obtain a constructed training sample set.

3. The plate heat exchanger weld creep damage assessment method of claim 1, wherein, The cross-entropy loss function or the mean square error loss function is used in the parameter iterative optimization process, and the Adam optimizer is used for parameter iterative optimization.

4. A system for evaluating the creep damage of a welded joint of a plate heat exchanger, applying the method according to any one of claims 1 to 3, characterized in that The method comprises the following steps: a data acquisition unit configured to acquire infrared thermal images of a weld area and equipment monitoring data; the equipment monitoring data comprises a two-dimensional temperature field data sequence and process parameters; a sample construction unit configured to preprocess the infrared thermal images and label the damage state according to the preprocessed images and the equipment monitoring data to construct a training sample set; a model training unit configured to construct an improved convolutional neural network, iteratively optimize parameters of the improved convolutional neural network using the training sample set, and determine the trained network as a damage assessment model; a model evaluation unit configured to evaluate real-time monitoring data using the damage assessment model to obtain an evaluation result corresponding to a current state, visualize the evaluation result on a pre-constructed plate heat exchanger three-dimensional model, and start a multi-level early warning mechanism according to a pre-set threshold; wherein the evaluation result comprises a creep damage grade assessment and a residual life prediction value.

5. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the plate heat exchanger weld creep damage assessment method according to any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, The computer program is stored in the memory and is executed by the processor to implement the plate heat exchanger weld creep damage assessment method according to any one of claims 1-3.

Images