A monitoring system for the thermal insulation effect of outdoor pipelines at construction sites

Abstract

This invention provides a monitoring system for the insulation effect of outdoor pipelines at construction sites, belonging to the field of outdoor pipeline insulation technology at construction sites. It includes: an ultrasonic generator, an ultrasonic sensor, a vibration generator, a vibration sensor, a monitoring camera, and a host computer. The ultrasonic generator is installed inside one end of the pipeline, and the ultrasonic sensor is installed inside the other end of the pipeline. The output end of the ultrasonic generator and the acquisition end of the ultrasonic receiver are both in contact with the inner wall of the pipeline. The monitoring camera is used to capture images of the outer wall of the pipeline. The host computer is electrically connected to the ultrasonic generator, ultrasonic sensor, vibration generator, vibration sensor, and monitoring camera. It is used to set the signal parameters and carrier signals generated by the ultrasonic generator and vibration generator, and to acquire the received ultrasonic signals collected by the ultrasonic sensor, the received vibration signals collected by the vibration sensor, and the images of the outer wall of the pipeline captured by the monitoring camera.

Description

Technical Field

[0001] This invention belongs to the field of outdoor pipeline insulation technology at construction sites, and more specifically, relates to a monitoring system for the insulation effect of outdoor pipelines at construction sites. Background Technology

[0002] Monitoring the insulation effectiveness of pipelines at construction sites is a crucial step in ensuring project quality and safe operation. In numerous sectors such as urban heating, petrochemicals, and power plants, a large number of pipelines are needed to transport hot fluids or cryogenic media. Due to significant temperature differences, condensation and ice formation on the outer walls of pipelines are highly likely, affecting the performance of the insulation layer and leading to energy loss and safety hazards. Therefore, real-time monitoring of pipeline insulation effectiveness is essential.

[0003] Currently, the following technical methods are mainly used to monitor the insulation effect of pipelines at construction sites:

[0004] 1. Infrared thermal imager temperature measurement

[0005] This method calculates the internal temperature distribution by detecting the thermal radiation from the outer wall of the pipe. However, because the outer wall temperature is affected by various factors such as ambient temperature and emissivity, this method often has a large margin of error and can only provide a rough assessment for pipes with significant temperature differences. Furthermore, infrared thermal imagers are expensive and their on-site operation is quite complex.

[0006] 2. Local temperature measurement

[0007] This method involves creating several small holes in the pipe and using temperature sensors to sample the pipe temperature from the inside. The entire pipe temperature field is then estimated using an interpolation algorithm. However, creating these holes can easily damage the pipe itself, and the limited number of measurement points restricts the interpolation accuracy. Furthermore, this method cannot detect condensation or icing inside the pipe wall.

[0008] 3. Indoor simulation test

[0009] This method constructs theoretical or numerical simulation models under laboratory conditions for specific pipe structures and operating conditions to predict the thermal insulation performance of such pipes in field use. However, due to the extreme complexity of actual operating conditions, it is difficult to accurately reflect the actual situation through simple models, resulting in a significant discrepancy between the results and the actual field conditions.

[0010] The aforementioned existing technologies all have difficulty monitoring the micro-icing phenomenon inside the pipeline without damaging the outdoor pipeline, making it difficult to monitor the pipeline's insulation effect. Summary of the Invention

[0011] In view of this, the present invention provides a monitoring system for the thermal insulation effect of outdoor pipelines at construction sites, which can solve the technical problem of existing technologies that make it difficult to monitor the micro-icing phenomenon inside the pipeline without damaging the outdoor pipeline, thus making it difficult to monitor the thermal insulation effect of the pipeline.

[0012] This invention is implemented as follows:

[0013] The first aspect of this invention provides a monitoring system for the thermal insulation effect of outdoor pipelines at construction sites, comprising an ultrasonic generator, an ultrasonic sensor, a vibration generator, a vibration sensor, a monitoring camera, and a host computer. The ultrasonic generator is installed inside one end of the pipeline, and the ultrasonic sensor is installed inside the other end of the pipeline. The output end of the ultrasonic generator and the acquisition end of the ultrasonic receiver are both in contact with the inner wall of the pipeline. The monitoring camera is used to capture images of the outer wall of the pipeline. The host computer is electrically connected to the ultrasonic generator, ultrasonic sensor, vibration generator, vibration sensor, and monitoring camera, and is used to set the signal parameters and carrier signals generated by the ultrasonic generator and vibration generator, and to acquire the received ultrasonic signals acquired by the ultrasonic sensor, the received vibration signals acquired by the vibration sensor, and the images of the outer wall of the pipeline captured by the monitoring camera. The host computer includes a pipeline thermal insulation effect monitoring module and a pipeline micro-icing monitoring module.

[0014] Based on the above technical solution, the outdoor pipeline insulation effect monitoring system for construction sites of the present invention can be further improved as follows:

[0015] The pipeline insulation effect monitoring module is used to calculate the insulation effect index based on the outer wall image, and specifically performs the following steps:

[0016] S11. Obtain multiple sets of pipeline construction parameters, including the pipeline's structure, dimensions, and materials, as parameter features;

[0017] S12. Obtain multiple sets of historical images of the pipe outer wall at different temperatures inside the pipe, and extract the texture features, brightness features, color features, and light reflection features of the pipe outer wall at different temperatures;

[0018] S13. The parameter features, texture features, brightness features, color features, and light reflection features are fused to obtain the first fused feature;

[0019] S14. Using the first fusion features corresponding to different temperatures inside the pipe as training data, train a neural network model to obtain a heat preservation effect evaluation model. The input of the training is the first fusion feature, and the output of the training is the temperature inside the pipe.

[0020] S15. Obtain the parameter features of the pipeline to be tested and the current image of the outer wall of the pipeline, perform feature fusion, and obtain the first fused feature to be tested;

[0021] S16. Using the insulation effect evaluation model, input the first fusion feature to be tested to obtain a quantitative pipe inner wall temperature range and output it to the operation and maintenance personnel.

[0022] The specific steps of S11 include: acquiring multiple sets of constructed pipe samples, measuring and recording the structural parameters, dimensional parameters, and material parameters of each set of pipe samples, such as pipe length, outer diameter, wall thickness, number of pipe segments, and material composition, and using these parameters as the parameter feature vector of that set of pipe samples; after acquiring a sufficient number of pipe samples and their corresponding parameter feature vectors, they are used for subsequent feature fusion and model training; this step aims to collect prior parameter information of the pipe itself to provide basic data support for subsequent estimation of insulation effect.

[0023] The specific steps of S12 include: first, conducting a temperature control experiment on the obtained pipe samples, that is, taking multiple images of the outer wall of the pipe using a monitoring camera at different pipe internal temperatures; then, performing image preprocessing on these outer wall images, including noise reduction and contrast enhancement, to improve image quality; next, extracting the texture features, brightness features, color features, and light reflection features of the outer wall of the pipe in each group of images; this step aims to extract visual features related to the heat preservation effect from the pipe outer wall images to provide data support for subsequent model training.

[0024] The specific steps of S13 include: performing feature-level fusion of the parameter feature vector obtained in step S11 and the multiple sets of fusion feature vectors obtained in step S12; specifically, the parameter feature vector and the fusion feature vector are first standardized, and then they are concatenated to obtain the final first fusion feature vector; this step aims to effectively fuse pipeline parameter information and visual information to provide high-quality input features for model training.

[0025] The specific steps of S14 include: using the first fused feature vector obtained in the previous step as the input feature and the temperature inside the pipe as the output label, constructing a neural network model for supervised learning; the neural network model adopts a feedforward network, convolutional network, or other deep learning structure, and during the training process, optimization algorithms such as stochastic gradient descent are used to minimize the loss function between the network output and the real temperature label, and the network parameters are continuously adjusted; an early stopping strategy is set during the training process to prevent the model from overfitting, and the model is cross-validated; this step aims to learn a neural network model that can accurately evaluate the insulation effect of the pipe based on the first fused feature vector.

[0026] The specific steps of S15 include: obtaining the parameter feature vector of the pipeline to be tested, acquiring the current image of the outer wall of the pipeline, extracting the corresponding visual feature vector, and then performing feature-level fusion of the parameter feature vector and the visual feature vector in the manner of step S13 to obtain the first fused feature vector to be tested; this step aims to extract and fuse features of the pipeline to be tested in order to prepare for subsequent model prediction.

[0027] The specific steps of S16 include: inputting the first fused feature vector to be tested obtained in step S15 into the heat insulation effect evaluation model trained in step S14, i.e., the neural network model; the neural network model outputs a temperature value or temperature range as a quantitative evaluation result of the heat insulation effect of the pipeline; if the difference between the temperature value or temperature range and the actual inner wall temperature of the pipeline exceeds a preset threshold, the heat insulation effect of the pipeline is determined to be poor; this step aims to use the trained neural network model to intelligently evaluate the heat insulation effect of the pipeline under test, providing important decision-making basis for operation and maintenance personnel.

[0028] The pipeline micro-crystal monitoring module is used to calculate the location and size of micro-ice formations inside the pipeline based on the outer wall image, the received ultrasonic signal, and the received vibration signal. Specifically, it performs the following steps:

[0029] S21. Construct a finite element model of the pipeline, calculate the propagation attenuation of the ultrasonic signal and the vibration signal at different frequencies and amplitudes, and obtain the ultrasonic signal parameters and vibration signal parameters with the minimum propagation attenuation.

[0030] S22. Using the ultrasonic signal parameters and vibration signal parameters with the minimum propagation attenuation obtained, set the ultrasonic generator and the vibration generator, and acquire the parameter characteristics of the pipe under test and the current image of the outer wall of the pipe, as well as the received signal and the received vibration signal.

[0031] S23. The received ultrasonic signal is preprocessed to obtain the target ultrasonic signal; after the vibration signal is preprocessed, the vibration caused by the fluid flow in the pipe is deleted, and only the portion of the vibration signal directly emitted by the vibration generator is retained to obtain the target vibration signal.

[0032] S24. Extract the time-frequency features of the target ultrasonic signal and the target vibration signal, and fuse them into a second fusion feature to be tested;

[0033] S25. The first fusion feature to be tested and the second fusion feature to be tested are fused with parameter constraints to obtain the feature to be tested;

[0034] S26. Using a pre-trained micro-icing detection model, input the features to be tested to obtain a feature vector representing the location and size of micro-icing in the pipeline, and output it to the maintenance personnel.

[0035] The specific steps of S21 include: first, establishing a refined finite element model of the pipeline, which can accurately describe the pipeline's geometric structure, material properties, and other parameters; then, based on the finite element model, calculating the propagation characteristics of ultrasonic and vibration signals in the pipeline at different frequencies and amplitudes using numerical simulation methods; and finding the optimal combination of ultrasonic and vibration signal parameters that minimizes propagation attenuation by comparing the simulation results under different parameter combinations. This step aims to determine the ultrasonic and vibration signal parameters that can obtain the best propagation characteristics, providing theoretical guidance for subsequent signal detection.

[0036] The specific steps of S22 include: setting the parameters of the actual ultrasonic generator and vibration generator using the optimal ultrasonic and vibration signal parameters obtained in step S21; simultaneously acquiring the structural, dimensional, and material parameters of the pipe under test as feature vector inputs; on the other hand, acquiring the current image of the outer wall of the pipe through a monitoring camera; and then acquiring the received ultrasonic signal and received vibration signal after propagation through the pipe from the receiving ends of the ultrasonic sensor and vibration sensor, respectively. This step aims to set the actual system according to the theoretically optimal parameters and acquire on-site image data and sensor signal data to prepare for subsequent processing.

[0037] The specific steps of S23 include: preprocessing the received ultrasonic signal obtained in step S22 to extract a pure target ultrasonic signal; the preprocessing may include noise reduction, filtering, background signal removal, etc.; similarly, the received vibration signal is also preprocessed to remove noise vibration caused by factors such as fluid flow in the pipe, and only retain the target vibration signal emitted by the vibration generator; this step aims to extract the effective signal components that truly reflect the internal state of the pipe from the original received signal, laying the foundation for subsequent feature extraction and pattern recognition.

[0038] The specific steps of S24 include: extracting time-frequency features from the target ultrasonic signal and the target vibration signal obtained in step S23; for the ultrasonic signal, using time-frequency analysis methods such as short-time Fourier transform and wavelet transform, extracting the time-frequency energy distribution map of the signal as a time-frequency feature vector; for the vibration signal, using similar time-frequency analysis methods to obtain the corresponding time-frequency feature vector; in addition, some statistical features are extracted; the multiple time-frequency feature vectors of the ultrasonic and vibration signals are fused to obtain the fused second fused feature vector; this step aims to convert the received ultrasonic and vibration signals into feature vector forms that are easy for the model to recognize, preparing for subsequent model prediction.

[0039] The specific steps of S25 include: performing parameter-constrained fusion of the first fusion feature vector to be tested obtained in step S15 and the second fusion feature vector to be tested obtained in step S24; specifically, using the pipeline parameter feature vector obtained in step S11, a feature fusion function based on a physical model is designed according to the acoustic and thermal properties of the pipeline material, and the function is used to perform a nonlinear transformation on the first and second fusion feature vectors to be tested to obtain the final feature vector to be tested; this step aims to effectively fuse the pipeline structural parameters and their physical properties with image and signal features to improve the accuracy of subsequent model predictions.

[0040] The specific steps for training the micro-icing detection model include: constructing a simulated environment under laboratory conditions to create pipe icing phenomena of different shapes and sizes, and using professional detection equipment to accurately measure the location and size of the icing as annotation data; collecting corresponding image data and sensor signal data on these pipe sections according to the method of the invention, and constructing an initial training dataset after associating it with the icing annotation data; preprocessing the training data to extract input feature vectors and output labels; designing a deep neural network model, adopting an encoder-decoder framework, and through iterative learning of a large number of training samples, the model can automatically learn the key patterns of micro-icing from the input features and map them to reasonable output predictions; the trained model can achieve accurate detection of micro-icing conditions in pipes.

[0041] Compared with existing technologies, the beneficial effects of the outdoor pipeline insulation effect monitoring system provided by this invention are as follows: The technical solution of this invention constructs a neural network model by fusing multi-source heterogeneous data such as pipeline structural parameters, image visual information, and ultrasonic / vibration signals, achieving simultaneous detection and evaluation of the insulation effect of the pipeline insulation layer and the micro-icing situation inside the pipeline. This method utilizes visual sensors (monitoring cameras) to capture visual features such as image texture and color of the insulation layer from the pipeline exterior; it uses acoustic sensors (ultrasonic transmitting and receiving devices, vibration generators, and vibration sensors) to obtain the signal propagation attenuation characteristics within the pipeline, indirectly reflecting the performance of the insulation material; simultaneously, it utilizes prior information such as the pipeline's structural parameters and material, and inputs the multi-source data fusion into the neural network model. Through training with a large amount of labeled data, the model can automatically mine the inherent mapping relationship between multi-modal data, intelligently evaluate the pipeline insulation effect, and detect whether there are micro-icing areas inside the pipeline. It has the following characteristics:

[0042] 1. Diversified detection methods and data fusion

[0043] This invention employs multiple detection methods, including visual and acoustic sensors, to acquire heterogeneous data from various sources, such as appearance images and ultrasonic / vibration sensing signals. Through effective fusion, it comprehensively reflects various conditions inside and outside the pipeline, enabling a more accurate assessment of insulation performance. Visual data directly reflects visual features such as texture and color of the outer wall, while acoustic data indirectly detects signal attenuation within the pipeline. These two types of data complement each other, jointly revealing the thermal insulation performance of the insulation layer. Furthermore, by incorporating prior structural parameters of the pipeline itself, the fused data is enriched, broadening its applicability. The comprehensive utilization of heterogeneous data from multiple sources avoids the limitations of a single data source, significantly improving the comprehensiveness and accuracy of the detection.

[0044] 2. High detection resolution and high level of intelligence.

[0045] This method utilizes deep learning technology to train a neural network model, enabling it to automatically learn complex intrinsic feature mapping relationships from multi-source data. This allows for a refined evaluation of insulation performance and achieves intelligent detection of micro-icing areas within pipelines at the microscale. The neural network possesses powerful feature representation and pattern recognition capabilities, fully leveraging the inherent meaning of fused data to significantly improve detection resolution and intelligence. Traditional methods based on experience or manual rules often suffer from complex feature engineering and poor generalization, while this solution completely avoids these problems.

[0046] In summary, the present invention solves the technical problem in the prior art that it is difficult to monitor the micro-icing phenomenon inside the pipeline without damaging the outdoor pipeline, and thus difficult to monitor the pipeline insulation effect. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention 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.

[0048] Figure 1 A schematic diagram of the system provided by the present invention;

[0049] Figure 2 A flowchart illustrating the execution steps of the pipeline insulation effect monitoring module;

[0050] Figure 3 A flowchart showing the steps performed by the pipeline micro-knot monitoring module; Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0052] like Figure 1 The diagram shows a flowchart of an outdoor pipeline insulation effect monitoring system provided by this invention. The system includes the following steps: It comprises an ultrasonic generator, an ultrasonic sensor, a vibration generator, a vibration sensor, a monitoring camera, and a host computer. The ultrasonic generator is installed inside one end of the pipeline, and the ultrasonic sensor is installed inside the other end of the pipeline. The output end of the ultrasonic generator and the acquisition end of the ultrasonic receiver are both in contact with the inner wall of the pipeline. The monitoring camera is used to capture images of the outer wall of the pipeline. The host computer is electrically connected to the ultrasonic generator, ultrasonic sensor, vibration generator, vibration sensor, and monitoring camera. It is used to set the signal parameters and carrier signals generated by the ultrasonic generator and vibration generator, and to acquire the received ultrasonic signals collected by the ultrasonic sensor, the received vibration signals collected by the vibration sensor, and the images of the outer wall of the pipeline captured by the monitoring camera. The host computer includes a pipeline insulation effect monitoring module and a pipeline micro-icing monitoring module.

[0053] The pipeline insulation effect monitoring module is used to calculate the insulation effect index based on the outer wall image, and specifically performs the following steps:

[0054] S11. Obtain multiple sets of pipeline construction parameters, including the pipeline's structure, dimensions, and materials, as parameter features;

[0055] S12. Obtain multiple sets of historical images of the pipe outer wall at different temperatures inside the pipe, and extract the texture features, brightness features, color features, and light reflection features of the pipe outer wall at different temperatures;

[0056] S13. The parameter features, texture features, brightness features, color features, and light reflection features are fused to obtain the first fused feature;

[0057] S14. Using the first fusion features corresponding to different temperatures inside the pipe as training data, train a neural network model to obtain a heat preservation effect evaluation model. The input of the training is the first fusion feature, and the output of the training is the temperature inside the pipe.

[0058] S15. Obtain the parameter features of the pipeline to be tested and the current image of the outer wall of the pipeline, perform feature fusion, and obtain the first fused feature to be tested;

[0059] S16. Using the insulation effect evaluation model, input the first fusion feature to be tested to obtain a quantitative pipe inner wall temperature range and output it to the operation and maintenance personnel.

[0060] The pipeline micro-crystal monitoring module is used to calculate the location and size of micro-ice formations inside the pipeline based on the outer wall image, the received ultrasonic signal, and the received vibration signal. Specifically, it performs the following steps:

[0061] S21. Construct a finite element model of the pipeline, calculate the propagation attenuation of the ultrasonic signal and the vibration signal at different frequencies and amplitudes, and obtain the ultrasonic signal parameters and vibration signal parameters with the minimum propagation attenuation.

[0062] S22. Using the ultrasonic signal parameters and vibration signal parameters with the minimum propagation attenuation obtained, set the ultrasonic generator and the vibration generator, and acquire the parameter characteristics of the pipe under test and the current image of the outer wall of the pipe, as well as the received signal and the received vibration signal.

[0063] S23. The received ultrasonic signal is preprocessed to obtain the target ultrasonic signal; after the vibration signal is preprocessed, the vibration caused by the fluid flow in the pipe is deleted, and only the portion of the vibration signal directly emitted by the vibration generator is retained to obtain the target vibration signal.

[0064] S24. Extract the time-frequency features of the target ultrasonic signal and the target vibration signal, and fuse them into a second fusion feature to be tested;

[0065] S25. The first fusion feature to be tested and the second fusion feature to be tested are fused with parameter constraints to obtain the feature to be tested;

[0066] S26. Using a pre-trained micro-icing detection model, input the features to be tested to obtain a feature vector representing the location and size of micro-icing in the pipeline, and output it to the maintenance personnel.

[0067] The specific implementation methods of the above steps are described in detail below:

[0068] The specific implementation of step S11 is as follows: First, acquire multiple sets of constructed pipe samples. Measure and record the structural, dimensional, and material parameters of each set of pipe samples, such as pipe length, outer diameter, wall thickness, number of pipe segments, and material composition. Use these parameters as the parameter feature vector for that set of pipe samples. After acquiring a sufficient number of pipe samples and their corresponding parameter feature vectors, they can be used for subsequent feature fusion and model training. The purpose of this step is to collect prior parameter information about the pipes themselves, providing basic data support for subsequent estimation of insulation performance.

[0069] The specific implementation of step S12 is as follows: First, a temperature control experiment is conducted on the obtained pipe samples, that is, multiple images of the outer wall of the pipe are taken using a monitoring camera at different internal pipe temperatures. Then, these outer wall images are preprocessed, including noise reduction and contrast enhancement, to improve image quality. Next, the texture features, brightness features, color features, and light reflection features of the outer wall of the pipe in each set of images are extracted. Texture features can be extracted using the gray-level co-occurrence matrix method, by calculating the gray-level relationship between pixels in the image to obtain a feature vector describing texture details; brightness features can be obtained by calculating the gray-level histogram of the image; color features can be extracted using the color histogram; and light reflection features can be extracted by detecting highlight areas in the image. By fusing the above multiple features, the fused feature vector corresponding to the set of images can be obtained. Repeating the above process, multiple sets of fused feature vectors corresponding to different temperatures can be obtained for images obtained at different temperatures. The purpose of this step is to extract visual features related to the heat preservation effect from the pipe outer wall images, providing data support for subsequent model training.

[0070] The specific implementation of step S13 is as follows: The parameter feature vector obtained in step S11 and the multiple sets of fused feature vectors obtained in step S12 are fused at the feature level. Specifically, the parameter feature vector and the fused feature vector can be standardized first, and then they are concatenated to obtain the final first fused feature vector. This feature fusion method fully utilizes the parameter information of the pipeline itself and the image visual information, which can improve the discrimination ability of the subsequent model. The purpose of this step is to effectively fuse the pipeline parameter information and visual information to provide high-quality input features for model training.

[0071] The specific implementation of step S14 is as follows: Using the first fused feature vector obtained in the previous step as the input feature and the temperature inside the pipe as the output label, a neural network model is constructed for supervised learning. The neural network model can employ a feedforward network, convolutional network, or other deep learning structures. The specific network structure used can be selected based on the characteristics of the actual dataset. During training, optimization algorithms such as stochastic gradient descent can be used to minimize the loss function between the network output and the true temperature label, continuously adjusting the network parameters to improve the model's predictive performance on the training set. An early stopping strategy can be set during training to prevent overfitting. Furthermore, cross-validation of the model is required, selecting the model that performs best on the validation set. The purpose of this step is to learn a neural network model capable of accurately evaluating the pipe's insulation effect based on the first fused feature vector.

[0072] The specific implementation of step S15 is as follows: Obtain the parameter feature vector of the pipe to be tested, acquire the current image of the outer wall of the pipe, and extract the corresponding visual feature vector. Then, perform feature-level fusion of the parameter feature vector and the visual feature vector in the manner described in step S13 to obtain the first fused feature vector to be tested. The purpose of this step is to extract and fuse features of the pipe to be tested, preparing it for subsequent model prediction.

[0073] The specific implementation of step S16 is as follows: The first fused feature vector to be tested obtained in step S15 is input into the insulation effect evaluation model trained in step S14, i.e., the neural network model. Since the neural network has learned the mapping relationship between feature vectors and temperature during training, after inputting the first fused feature vector to be tested, the model can output a temperature value or temperature range as a quantitative evaluation result of the insulation effect of the pipeline. This temperature value or temperature range can be compared with the actual inner wall temperature of the pipeline. If the difference exceeds a preset threshold (e.g., 5 degrees Celsius), it can be determined that the insulation effect of the pipeline is poor, and measures such as repair or replacement of insulation materials are required. The purpose of this step is to use the trained neural network model to intelligently evaluate the insulation effect of the pipeline under test, providing important decision-making basis for operation and maintenance personnel.

[0074] The advantage of this module lies in its ability to comprehensively and accurately evaluate the insulation effect of pipes by fusing pipe parameter information and image visual information, avoiding misjudgments caused by relying on a single feature. Furthermore, the model trained using deep learning technology can automatically discover high-dimensional mapping relationships between features and insulation effects without requiring the manual design of complex rules, thus exhibiting strong robustness and applicability.

[0075] The specific implementation of step S21 is as follows: First, a refined finite element model of the pipeline is established, which can accurately describe the pipeline's geometric structure, material properties, and other parameters. Then, based on this finite element model, the propagation characteristics of ultrasonic and vibration signals within the pipeline at different frequencies and amplitudes are calculated using numerical simulation methods, including wave propagation attenuation, reflection, and refraction. Specifically, the finite element time-domain analysis method can be used to discretize the pipeline into a large number of small elements, calculate the time-series response of the field quantity within each element, and thus obtain the spatiotemporal distribution of the field quantity throughout the pipeline. By comparing the simulation results at different frequencies and amplitudes, the optimal combination of ultrasonic and vibration signal parameters that minimizes attenuation can be found. The purpose of this step is to determine which ultrasonic and vibration signal parameters can achieve the best propagation characteristics in this type of pipeline, providing theoretical guidance for subsequent signal detection.

[0076] The specific implementation of step S22 is as follows: Using the optimal ultrasonic and vibration signal parameters obtained in step S21, the parameters of the actual ultrasonic generator and vibration generator are set so that they emit ultrasonic and vibration signals of corresponding frequencies and amplitudes, respectively. Simultaneously, the structural, dimensional, and material parameters of the pipe under test are acquired as feature vector inputs. On the other hand, an image of the outer wall of the pipe is captured by a monitoring camera. Then, the received ultrasonic and vibration signals propagating through the pipe are acquired from the receiving ends of the ultrasonic and vibration sensors, respectively. The purpose of this step is to set the actual system according to the theoretically optimal parameters and acquire on-site image data and sensor signal data, preparing for subsequent signal processing and feature extraction.

[0077] The specific implementation of step S23 is as follows: The received ultrasonic signal acquired in step S22 is preprocessed to extract a clean target ultrasonic signal. Preprocessing may include denoising, filtering, and background signal removal. Denoising can employ time-frequency analysis methods such as wavelet transform; filtering can use digital filtering algorithms such as bandpass filtering; background signal removal can be achieved by modeling the reference signal (i.e., the received signal in an empty pipe) and then subtracting the background signal from the actual received signal to obtain the target ultrasonic signal. Similarly, the received vibration signal is also preprocessed to remove noise vibrations caused by factors such as fluid flow within the pipe, retaining only the target vibration signal emitted by the vibration generator. The purpose of this step is to extract effective signal components that truly reflect the internal state of the pipe from the original received signal, laying the foundation for subsequent feature extraction and pattern recognition.

[0078] The specific implementation of step S24 is as follows: Time-frequency features are extracted from the target ultrasonic signal and target vibration signal obtained in step S23. For the ultrasonic signal, time-frequency analysis methods such as short-time Fourier transform and wavelet transform can be used to extract the time-frequency energy distribution map of the signal as a time-frequency feature vector. A similar time-frequency analysis method can be used for the vibration signal to obtain the corresponding time-frequency feature vector. In addition, some statistical features, such as signal energy and peak factor, can be extracted. Multiple time-frequency feature vectors of the ultrasonic and vibration signals are concatenated or fused to obtain the fused second feature vector to be measured. The purpose of this step is to transform the received ultrasonic and vibration signals into feature vector forms that are easily recognized by the model, preparing for subsequent model prediction.

[0079] The specific implementation of step S25 is as follows: The first fused feature vector to be tested obtained in step S15 and the second fused feature vector to be tested obtained in step S24 are subjected to parameter-constrained fusion. This fusion process is not a simple concatenation or weighted summation, but rather a nonlinear fusion of the two feature vectors based on a specific fusion function by introducing pipeline parameters as constraints. Specifically, using the pipeline parameter feature vector obtained in step S11, a feature fusion function based on a physical model can be designed according to the acoustic and thermal properties of the pipeline material. This function is then used to perform a nonlinear transformation on the first and second fused feature vectors to be tested, yielding the final feature vector to be tested. The purpose of this step is to effectively fuse the pipeline structural parameters and their physical properties with image and signal features, thereby improving the accuracy of subsequent model predictions.

[0080] The specific implementation of step S26 is as follows: The feature vector to be tested obtained in step S25 is input into a pre-trained micro-icing detection model. This model adopts a deep neural network structure, which can automatically learn the representation of the location and size of micro-icing from the feature vector. Specifically, the deep neural network consists of multiple convolutional layers, pooling layers, and fully connected layers. The output layer is a feature vector of fixed length, where the value of each dimension corresponds to the icing state of a certain section of the pipeline. For example, if the value of a certain dimension is large, it indicates that there may be a large ice block at that location. By decoding the output vector, the size of the ice at each location in the pipeline can be reconstructed, and the results can be presented to maintenance personnel in the form of images or reports. The purpose of this step is to use a data-driven deep learning model to intelligently detect and identify micro-icing in the pipeline, providing important decision-making basis for pipeline operation and maintenance.

[0081] The specific implementation method for training the micro-icing detection model is described in detail below:

[0082] Obtaining the training dataset is a crucial step in model training. Because micro-icing is a hidden physical process, it is difficult to directly observe and label the precise location and size of ice formation in actual engineering pipelines. Therefore, it is necessary to obtain labeled training data by constructing a simulated environment under laboratory conditions. Specifically, the following steps can be followed:

[0083] First, multiple pipes of different materials (such as steel pipes and plastic pipes) and specifications are prepared in the laboratory and installed on supports to form multiple detachable pipe sections. Second, icing phenomena of different shapes and sizes are artificially created on some pipe sections, for example, by introducing cryogenic gas and controlling the icing rate, or by injecting cryogenic liquid and then heating it to partially freeze it. Simultaneously, professional detection equipment (such as endoscopes and X-ray imaging) is used to accurately measure the location and size of the icing, which serves as annotation data. Then, following the method of the invention, ultrasonic generators, sensors, vibration generators, and cameras are deployed on these pipe sections to collect corresponding image data and sensor signal data. Correlating this data with the icing annotation data constructs the initial training dataset.

[0084] After obtaining a sufficient number of training samples, the data needs to be preprocessed, including data cleaning and standardization, to ensure data consistency and quality. Then, following the method in step S24, time-frequency features are extracted from the ultrasonic and vibration signals; following the method in step S15, pipe parameter features and image visual features are extracted; finally, they are fused with parameter constraints as in step S25 to obtain the input feature vector for each training sample. As for the output labels, multi-hot encoding can be used to divide the pipe into multiple equal-length segments, each segment corresponding to a label dimension, using 0 / 1 values ​​to indicate whether the segment is icy.

[0085] After obtaining the input features and output labels, the specific structure of the deep neural network model can be designed. This model can adopt an encoder-decoder framework, where the encoder consists of multiple layers of convolutional neural networks to automatically learn high-level representations of the input features; the decoder consists of multiple layers of upsampling convolutional layers and deconvolutional layers, progressively decoding the encoder's output into feature maps of the same length as the output labels. Specifically:

[0086] The encoder can employ classic convolutional network structures such as VGG and ResNet, encoding input features through multiple convolutional and pooling layers to obtain high-dimensional feature maps. The decoder can use upsampling network structures such as UNet, upsampling and decoding the encoder's output through transposed convolutional layers to ultimately output a prediction result of the same length as the label. During training, a binary cross-entropy loss function can be used, supplemented by data augmentation and regularization strategies to prevent overfitting.

[0087] Through iterative learning using a large number of training samples, this deep neural network model can automatically extract key patterns of micro-icing from input image, parameter, and signal features, and map them onto reasonable output predictions. The trained model can effectively decode the feature vector to be tested in step S25 into feature maps representing the location and size of icing, thereby achieving accurate detection of micro-icing conditions inside the pipeline.

[0088] In practical applications, it is necessary to evaluate the model's detection performance under different operating conditions and continuously optimize it based on actual errors, such as adjusting the network structure, loss function form, and regularization strategies. Furthermore, due to the high real-time requirements of micro-icing detection, the model also needs to be compressed and accelerated to ensure rapid deployment and efficient inference on edge devices. In summary, through the rational design of experimental schemes, feature extraction methods, and deep learning models, combined with the accumulation of a large amount of training data, this micro-icing detection model can significantly improve the accuracy and intelligence level of detection, providing strong technical support for pipeline operation and maintenance.

[0089] Furthermore, the implementation process requires determining the values ​​of several key parameters, such as frequency, amplitude, and other ultrasonic and vibration signal parameters; signal preprocessing parameters like denoising and filtering; the specific form of the feature fusion function; architectural parameters like the number of layers and nodes in the neural network; and the threshold for determining icing. The proper setting of these parameters directly impacts the system's detection performance, thus requiring optimization based on extensive experimental data and practical engineering experience. Simultaneously, in practical applications, a comprehensive evaluation of the system's real-time performance, stability, and reliability is necessary to ensure its efficient and continuous provision of monitoring services to the construction site.

[0090] In summary, this invention proposes a monitoring method that comprehensively utilizes pipeline structural parameters, appearance images, and signal sensing data. Through effective feature fusion and intelligent modeling, it can simultaneously assess the insulation effect and micro-icing of pipelines in real time.

[0091] The following is a detailed description of the calculation process involved in specific embodiments of the present invention:

[0092] Detailed implementation of step S11:

[0093] Suppose there are N sets of constructed pipeline samples. For the i-th (i = 1, 2, ..., N) set of samples, let its structural parameter vector be... The dimension parameter vector is Material parameter vector is Then the parameter feature vector of this set of samples It can be represented as:

[0094]

[0095] Among them, f p (·) represents a feature fusion function, which can be a simple concatenation or a nonlinear transformation based on domain knowledge.

[0096] For all N groups of samples, the parametric feature matrix can be obtained. This data will be used as the basis for subsequent feature fusion and model training.

[0097] Detailed implementation of step S12:

[0098] Suppose that the temperature inside the pipe has M different values ​​{T1, T2, ..., T...} M For the j-th (j = 1, 2, ..., M) temperature value T j Obtain K images of the outer wall of the i-th pipe sample at this temperature {I i，j，1 I i，j，2 , ..., I i，j，K}

[0099] For each image I i，j，k (k = 1, 2, ..., K), extract the following visual features:

[0100] Texture feature vector It can be calculated using the gray-level co-occurrence matrix method:

[0101]

[0102]

[0103] Where p(m, n, θ, d) is the gray-level co-occurrence matrix, representing the frequency of occurrence of pixel pairs (m, n) with a distance of d and a direction of θ in the image; L is the number of gray levels.

[0104] Brightness feature vector Histogram statistics can be used for calculation:

[0105]

[0106]

[0107] Where, n l is the number of pixels that appear at gray level l, and N is the total number of pixels.

[0108] Color feature vector It can be calculated using a color histogram:

[0109]

[0110]

[0111] Where, r m g m b m R, G, and B represent the frequencies of gray levels m in the red, green, and blue channels, respectively, and R, G, and B are the gray level numbers of the three channels.

[0112] Light reflection feature vector Calculations can be made through highlight area detection:

[0113]

[0114]

[0115] σ g , σ b Similar calculations

[0116] Where H is the set of pixel coordinates of the detected highlight region in the image, and N h For its number of pixels, I r (x, y), I g (x, y), I b (x, y) are the RGB three-channel values ​​at pixel (x, y), respectively.

[0117] By concatenating the above four types of feature vectors, the fused visual feature vector of the image is obtained:

[0118]

[0119] For the j-th temperature value T j The feature vectors of K images can be fused to obtain the fused feature vector at that temperature.

[0120]

[0121] The fused feature vector corresponding to all M temperature values ​​is denoted as

[0122] Specific implementation of step S13:

[0123] The parameter feature matrix P obtained in step S11 and the fusion feature matrix V obtained in step S12 are combined. i (i = 1, 2, ..., N) undergo feature-level fusion. Since the dimensions and numerical ranges of these two types of features may differ, standardization is required first.

[0124]

[0125]

[0126] Where, μ P and σ P Here, P represents the mean and standard deviation, respectively. and V i The mean and standard deviation.

[0127] After standardization, the two types of features are concatenated:

[0128] i = 1, 2, ..., N

[0129] The first fusion feature matrix F corresponding to each group of samples can then be obtained. i This is used as input to the neural network model.

[0130] Specific implementation of step S14:

[0131] Construct a deep feedforward neural network model, with the first fusion feature matrix F obtained in step S13 as the input. i The output is the corresponding pipe temperature value T. i (i = 1, 2, ..., N). This neural network may include L hidden layers, and the output of the l-th (l = 1, 2, ..., L) layer is:

[0132]

[0133] in, The input is σ(·), where σ is the activation function (such as the ReLU function), and W is the input. (l) and These are the weight matrix and bias vector for this layer, respectively.

[0134] The last layer is the output layer, and its output is:

[0135]

[0136] During training, mean squared error is used as the loss function:

[0137]

[0138] Here, Θ contains all trainable parameters in the network. The stochastic gradient descent algorithm is used to iteratively optimize Θ:

[0139]

[0140] η is the learning rate, which can be adaptively adjusted using an exponential decay method. Additionally, strategies such as L2 regularization and dropout can be introduced to prevent overfitting.

[0141] Through extensive iterative training, a neural network model capable of accurately evaluating the insulation effect of pipes can be obtained.

[0142] Specific implementation of step S15:

[0143] Obtain the structural parameter vector of the pipeline under test Size parameter vector and material parameter vector Through the feature fusion function f in step S11 p (·), to obtain the corresponding parameter feature vector:

[0144]

[0145] Acquire the current set of images of the outer wall of the pipe {I q，1 I q，2 , ..., I q，K′ Repeating the image feature extraction process in step S12, the fused visual feature vector of the pipeline can be obtained.

[0146] eigenvectors of parameters and visual feature vectors Standardize the data, then perform feature-level fusion as described in step S13:

[0147]

[0148]

[0149] The first fusion feature vector of the pipeline can then be obtained.

[0150] Detailed implementation of step S16:

[0151] The first fused feature vector obtained in step S15 The data is input into the neural network model trained in step S14, and following the feedforward calculation process, the model's predicted temperature inside the pipe can be obtained.

[0152]

[0153] if Therefore, the insulation effect of the pipe is judged to be poor, where T q The actual temperature inside the pipe, ∈ T The preset temperature difference threshold can be set to around 5 degrees Celsius, based on engineering experience.

[0154] This step allows for the automatic and intelligent assessment of the insulation performance of the pipeline under test, providing a basis for decision-making in on-site operation and maintenance.

[0155] Detailed implementation of step S21:

[0156] For the specific parameters of the pipeline, a detailed three-dimensional finite element model is constructed, including the pipeline's geometric dimensions, material anisotropy properties, and medium parameters. For acoustic problems, the pipeline field Ω can be divided into a large number of element Ω using finite element discretization. e(e = 1, 2, ..., N) e ), where N e This represents the total number of units.

[0157] Based on the shape function N of the discrete unit e (x, y, z) and nodal degrees of freedom u e The displacement field at any point (x, y, z) inside the pipe can be expressed as:

[0158]

[0159] Substituting the strong form of the propagation equation into the finite element discretization equation, we obtain the algebraic equation for each element with respect to the nodal degrees of freedom ue:

[0160] K e u e =f e

[0161] Among them, K e f is the element stiffness matrix. e Given the load vector, by assembling all the element equations, we can obtain the system of algebraic equations for the entire pipe field:

[0162] Ku = f

[0163] K is the global stiffness matrix, u is the unknown nodal displacement, and f is the known global load.

[0164] By applying acoustic excitation boundary conditions to the above equation and introducing a time factor, the acoustic response at any time and space point inside the pipe can be solved, thereby analyzing the propagation behavior of acoustic signals inside the pipe.

[0165] For excitation signals with different frequencies f and amplitudes A, the corresponding propagation attenuation coefficients α(f, A) can be calculated, and the optimal parameter combination {f0, A0} corresponding to the minimum value can be found:

[0166]

[0167] Detailed implementation of step S22:

[0168] Using the optimal parameter combination {f0, A0} obtained in step S21, the actual ultrasonic generator and vibration generator are set accordingly so that they emit ultrasonic and vibration signals with frequency f0 and amplitude A0.

[0169] Detailed implementation of step S23:

[0170] The raw received ultrasonic signal s acquired from the ultrasonic sensor receiver in step S22 r (t) is preprocessed to extract the pure target ultrasonic signal sd (t).

[0171] First, regarding s r (t) is subjected to wavelet transform to obtain its scaling coefficient c. j (k) and wavelet coefficients d j (k):

[0172]

[0173] Where, φ J，k (t) is the scaling function, ψ j，k (t) is the wavelet function, and J is the maximum scale layer.

[0174] Using wavelet thresholding denoising algorithm, the wavelet coefficients d j (k) Perform thresholding to remove noise components and obtain the denoised wavelet coefficients.

[0175]

[0176] λ j The threshold value for the j-th layer can be determined using methods such as empirical Bayesian thresholding.

[0177] Using the reconstruction formula, we can start from c J (k) and Reconstructed denoised target ultrasonic signal s d (t).

[0178] In addition, background signal removal is required. The received ultrasonic signal (s) in an empty pipe can be measured in advance. b (t), and fit it using the autoregressive moving average modeling method:

[0179]

[0180] Where, φ i Here are the autoregressive coefficients, p is the model order, and ∈ b (t) represents the white noise residual.

[0181] Substituting the above equation into s d In (t), the background signal can be removed to obtain the final target ultrasound signal:

[0182]

[0183] For receiving vibration signal v r (t) can also be obtained by similar wavelet thresholding denoising and background signal removal methods to obtain the target vibration signal v. t(t). The purpose of this step is to separate the effective signal components that truly reflect the internal state of the pipeline from the original received signal.

[0184] Detailed implementation of step S24:

[0185] The target ultrasonic signal s obtained in step S23 t (t) and the target vibration signal v t (t) Time-frequency features are extracted separately.

[0186] For s t (t) can be analyzed in time and frequency using Short Time Fourier Transform (STFT):

[0187]

[0188] Where g(t) is a window function, which measures s t By taking the inner product of a local slice of g(t) at time τ and g(t), the time-frequency coefficients at time τ and frequency ω can be obtained.

[0189] The matrix S formed by the STFT coefficients of all times τ and frequencies ω is s. t The time-frequency energy distribution of (t). S can be expanded into a one-dimensional time-frequency eigenvector. As a time-frequency characteristic descriptor for ultrasonic signals.

[0190] For v t If (t), then wavelet transform can be used for time-frequency analysis:

[0191]

[0192] Where ψ(t) is the wavelet basis function, a is the scale parameter, and b is the time displacement parameter.

[0193] By calculating the wavelet coefficients WT at different scales a and displacements b, v can be obtained. t The scale-time coefficient matrix V of (t) is expanded into a one-dimensional wavelet time-frequency eigenvector. As a time-frequency characteristic descriptor of vibration signals.

[0194] In addition, it is also possible to target s t (t) and v t (t) Extract statistical features, such as energy and peak factor, and concatenate them to the time-frequency feature vector.

[0195] Finally, the time-frequency feature vector of the ultrasonic signal is... and the time-frequency eigenvector of the vibration signal By performing concatenation or nonlinear fusion, the second fused feature vector to be tested can be obtained.

[0196] Specific implementation of step S25:

[0197] The first fused feature vector obtained in step S15 and the second fused feature vector obtained in step S24 Parameter constraint fusion is performed to obtain the final feature vector to be tested.

[0198] This fusion process involves knowledge of physical constraints and cannot be simply spliced ​​or weighted summed; it needs to follow a specific fusion function. Let this function be f. c (·), whose inputs include and pipeline parameter feature vectors The final feature vector to be tested is:

[0199]

[0200] Wherein, the fusion function f c The specific form of (·) can be an analytical solution based on a physical model, or a nonlinear function approximation obtained through data-driven learning.

[0201] For example, finite element simulation can be used to construct theoretical propagation response models of ultrasonic and vibration signals in pipelines under different temperature fields and icing distributions. Based on this model, given pipeline parameters... By observing the temperature field distribution and its corresponding characteristics, the theoretical acoustic response features can be estimated. Comparing these theoretical features with the measured values... By performing residual analysis and using nonlinear least squares fitting, f can be learned. c The parameterized form of (·).

[0202] In practical applications, f can be obtained through pre-training offline. c The model (·) can also be continuously optimized online to adapt to diverse on-site conditions. By fusing visual, acoustic, and structural parameter features with physical constraints, the complementary relationships between various modal data can be maximized, improving the accuracy and robustness of subsequent model predictions.

[0203] Detailed implementation of step S26:

[0204] The feature vector to be tested obtained in step S25 The input is fed into a pre-trained micro-icing detection model, whose output is a fixed-length feature mapping vector. Used to indicate the icing status of different sections within a pipe.

[0205] This model employs a deep encoder-decoder network structure. The encoder part consists of a convolutional neural network that processes the input... Perform automatic encoding to extract its high-level semantic feature representation.

[0206]

[0207] Where fenc(·) is the nonlinear mapping function of the encoder network, and Θenc represents all trainable parameters of the network.

[0208] The decoder consists of multiple upsampling convolutional and deconvolutional layers, which convert the encoder output into a decoder. Stepwise decoding into feature maps of the same length as the output target

[0209]

[0210] fdec(·) is the mapping function of the decoder network, and Θdec is its trainable parameter.

[0211] During training, given the labeled pipe icing states, they are discretized into... As training labels. By minimizing the loss function between the labels and the network output:

[0212]

[0213] Where N is the number of training samples and λ is the regularization coefficient. The parameters of the encoder and decoder are continuously optimized through backpropagation algorithm.

[0214] After training convergence, for the pipeline under test, it is... Input the model, and the corresponding icing mapping vector will be output. Through the Decoding allows us to obtain the location and size information of ice formation in each section of the pipeline, providing a basis for on-site operation and maintenance.

[0215] This encoder-decoder model framework can predict the output target directly from the input features end-to-end, without the need for manually designing complex feature engineering or rules, and has powerful representation learning capabilities. By training on a large amount of labeled sample data, the model can automatically discover the high-dimensional mapping relationship between heterogeneous fused features and icing distribution.

[0216] It should be noted that the variables involved in the above calculation process are explained in the following table:

[0217]

[0218]

[0219]

[0220]

[0221] Specifically, the principle of this invention is heterogeneous data fusion and deep learning modeling. By fusing multi-source heterogeneous data such as visual, acoustic, and structural parameters at the feature level, and then inputting the data into a deep neural network model, the automatic mining and modeling of the data's meaning can be achieved, thereby accurately evaluating the pipe insulation effect.

[0222] 1. Mapping between visual features and heat insulation effect

[0223] Visual features such as the texture, color, and brightness distribution of a pipe's outer wall can effectively reflect its thermal insulation performance. For example, a rough and uneven outer surface indicates poor quality insulation material; large areas of dark or dampness on the outer wall suggest severe condensation or icing; and uneven brightness distribution indicates uneven internal temperature distribution. These visual features can be extracted by using computer vision techniques such as texture analysis, brightness analysis, and color analysis on pipe appearance images. However, relying solely on visual features is insufficient for accurately and quantitatively assessing insulation effectiveness; integration with other information is necessary.

[0224] 2. Mapping between acoustic characteristics and thermal insulation effect

[0225] The propagation of acoustic signals within pipelines is influenced by factors such as the pipeline's structure, materials, and the medium, with the thermal insulation performance of the insulation layer being a major contributing factor. Poor insulation quality leads to a significant temperature difference between the pipeline's interior and the surrounding environment, resulting in a marked difference in temperature field distribution and consequently, a significant increase in acoustic signal attenuation. Detailed finite element numerical simulations can accurately depict the intrinsic mapping relationship between acoustic response and insulation performance. By monitoring acoustic signal attenuation in actual pipelines, the thermal insulation performance of the insulation layer can be indirectly assessed. Furthermore, acoustic signals can also detect condensation / icing within the pipeline, as these conditions alter the signal propagation path and cause abnormal attenuation.

[0226] 3. Mapping between pipe structural parameters and insulation effect

[0227] The structural parameters of the pipeline itself, such as length, diameter, and material, also affect its thermal insulation performance. For example, for the same insulation material, a larger diameter pipeline results in less heat loss per unit area; a longer pipeline increases the overall difficulty of insulation. Introducing these prior parameter information helps the model to assess the adaptability of pipelines under different operating conditions.

[0228] 4. Heterogeneous data fusion

[0229] While the aforementioned visual, acoustic, and structural parameters can each reflect different aspects of the pipe insulation effect, each has its own limitations and shortcomings, making accurate evaluation difficult when used alone. Therefore, it is necessary to effectively fuse them to leverage their respective advantages and extract the maximum amount of information about the insulation effect. Feature fusion cannot be achieved by simply concatenating or weighted summation; instead, it requires incorporating prior physical constraints and designing a fusion function based on a physical model. This function performs a nonlinear transformation on the heterogeneous features, resulting in a more compact and information-rich fused feature vector.

[0230] 5. Deep Neural Network Modeling

[0231] The fused feature vector is used as input, and an end-to-end deep neural network model can be trained using a large amount of labeled training data (generated using a simulated environment). Because deep neural networks possess powerful automatic feature learning capabilities, capable of mapping from low-level features to high-level semantic concepts, they can automatically uncover the intrinsic mapping relationships between complex, uninterpretable heterogeneous data. Specifically, this neural network can adopt an encoder-decoder structure: the encoder part consists of multi-layer convolutional networks that automatically encode the fused input feature vector, extracting its high-level semantic representation; the decoder part consists of upsampling convolutional networks and deconvolutional networks that progressively decode the encoder's output into feature maps of the same length as the target output.

[0232] Through supervised training with a large amount of labeled data, this neural network can learn to map the heterogeneous fused feature vectors from the input to output targets such as the actual temperature field distribution, micro-icing locations, and sizes within the pipe. This data-driven automatic feature extraction and pattern recognition process avoids the complex manual feature design and rule formulation required in traditional methods, significantly improving the intelligence and generalization ability of the detection.

[0233] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A construction site outdoor pipe insulation effect monitoring system, characterized in that, The system includes an ultrasonic generator, an ultrasonic sensor, a vibration generator, a vibration sensor, a monitoring camera, and a host computer. The ultrasonic generator is installed inside one end of the pipe, and the ultrasonic sensor is installed inside the other end of the pipe. The output end of the ultrasonic generator and the acquisition end of the ultrasonic sensor are both in contact with the inner wall of the pipe. The monitoring camera is used to capture images of the outer wall of the pipe. The host computer is electrically connected to the ultrasonic generator, ultrasonic sensor, vibration generator, vibration sensor, and monitoring camera. It is used to set the signal parameters and carrier signals generated by the ultrasonic generator and vibration generator, and to acquire the received ultrasonic signals acquired by the ultrasonic sensor, the received vibration signals acquired by the vibration sensor, and the images of the outer wall of the pipe captured by the monitoring camera. The host computer includes a pipe insulation effect monitoring module and a pipe micro-icing monitoring module. The pipeline insulation effect monitoring module is used to calculate the insulation effect index based on the image of the outer wall of the pipeline, and specifically performs the following steps: S11. Obtain multiple sets of structural parameters, dimensional parameters, and material parameters of the pipeline as parameter feature vectors; S12. Obtain multiple sets of historical images of the outer wall of the pipe at different temperatures inside the pipe, extract the texture features, brightness features, color features and light reflection features of the outer wall of the pipe at different temperatures, fuse the above features to obtain multiple sets of fused feature vectors; S13. The parameter feature vector and multiple sets of fused feature vectors are fused to obtain the first fused feature vector; S14. Using the first fusion feature vectors corresponding to different temperatures inside the pipe as training data, a neural network model is trained to obtain a thermal insulation effect evaluation model. The input of the training is the first fusion feature vector, and the output of the training is the temperature inside the pipe. S15. Obtain the parameter feature vector of the pipeline to be tested and the current image of the outer wall of the pipeline, perform feature vector fusion, and obtain the first fused feature vector to be tested. S16. Using the insulation effect evaluation model, input the first fused feature vector to be tested to obtain a quantitative pipe inner wall temperature range and output it to the operation and maintenance personnel. The pipeline micro-icing monitoring module is used to calculate the location and size of micro-icing inside the pipeline based on the pipeline's outer wall image, the received ultrasonic signal, and the received vibration signal, specifically by performing the following steps: S21. Construct a finite element model of the pipeline, calculate the propagation attenuation of ultrasonic and vibration signals at different frequencies and amplitudes, and obtain the ultrasonic signal parameters and vibration signal parameters with the minimum propagation attenuation. S22. The ultrasonic generator and the vibration generator are set using the ultrasonic signal parameters and vibration signal parameters with the minimum propagation attenuation obtained, and the parameter feature vector of the pipe under test and the current image of the outer wall of the pipe are obtained, as well as the received ultrasonic signal and the received vibration signal. S23. The received ultrasonic signal is preprocessed to obtain the target ultrasonic signal; after the received vibration signal is preprocessed, the vibration caused by the fluid flow in the pipe is deleted, and only the received part of the vibration signal emitted by the vibration generator is retained to obtain the target vibration signal. S24. Extract the time-frequency features of the target ultrasonic signal and the target vibration signal, and fuse them into a second fused feature vector to be tested; S25. The first fused feature vector to be tested and the second fused feature vector to be tested are fused with parameter constraints to obtain the feature vector to be tested; S26. Using a pre-trained micro-icing detection model, input the feature vector to be tested to obtain the location and size of micro-icing in the pipeline, and output it to the operation and maintenance personnel.

2. The construction site outdoor pipe heat preservation effect monitoring system according to claim 1, characterized in that, The specific steps of S13 include: first, standardizing the parameter feature vector and multiple sets of fused feature vectors respectively, and then concatenating them to obtain the first fused feature vector.

3. The construction site outdoor pipe heat preservation effect monitoring system according to claim 2, characterized in that, The specific steps of S16 include: inputting the first fused feature vector to be tested obtained in step S15 into the heat preservation effect evaluation model trained in step S14, and outputting a temperature value or temperature range as a quantitative evaluation result of the pipe heat preservation effect; if the difference between the temperature value or temperature range and the actual pipe inner wall temperature exceeds a preset threshold, the heat preservation effect of the pipe is determined to be poor.

4. The outdoor pipeline insulation effect monitoring system at construction sites according to claim 3, characterized in that, The specific steps of S22 include: setting parameters for the actual ultrasonic generator and vibration generator using the ultrasonic signal parameters and vibration signal parameters obtained in step S21; simultaneously acquiring the structural parameters, dimensional parameters, and material parameters of the pipe under test as input parameter feature vectors; on the other hand, acquiring the current image of the outer wall of the pipe through a monitoring camera; and then acquiring the received ultrasonic signal and received vibration signal after propagation through the pipe from the acquisition ends of the ultrasonic sensor and vibration sensor, respectively.

5. A monitoring system for the thermal insulation effect of outdoor pipelines at construction sites according to claim 4, characterized in that, The specific steps of S23 include: preprocessing the received ultrasonic signal obtained in step S22 to extract the pure target ultrasonic signal; the preprocessing includes noise reduction, filtering, and background signal removal; similarly, the received vibration signal is also preprocessed to remove noise vibration caused by fluid flow factors in the pipe, and only the received part of the vibration signal emitted by the vibration generator is retained to obtain the target vibration signal.

6. The outdoor pipeline insulation effect monitoring system at construction sites according to claim 5, characterized in that, The specific steps of S24 include: extracting time-frequency features from the target ultrasonic signal and the target vibration signal obtained in step S23; for the target ultrasonic signal, using the short-time Fourier transform time-frequency analysis method to extract the time-frequency energy distribution map of the signal as a time-frequency feature vector; for the target vibration signal, using the wavelet transform time-frequency analysis method to obtain the corresponding time-frequency feature vector; and fusing the time-frequency feature vectors of the target ultrasonic signal and the target vibration signal to obtain the fused second fused feature vector to be measured.

7. The outdoor pipeline insulation effect monitoring system at construction sites according to claim 6, characterized in that, The specific steps of S25 include: using the pipeline parameter feature vector obtained in step S11, designing a feature fusion function based on a physical model according to the acoustic and thermal properties of the pipeline material, and performing a nonlinear transformation on the first and second fusion feature vectors to be tested to obtain the feature vector to be tested.

8. A monitoring system for the thermal insulation effect of outdoor pipelines at construction sites according to claim 7, characterized in that, The specific steps for training the micro-icing detection model include: constructing a simulated environment under laboratory conditions to create icing phenomena in pipes of different shapes and sizes, and using professional detection equipment to accurately measure the location and size of the icing as annotation data; collecting corresponding image data and sensor signal data on these pipe sections, and constructing an initial training dataset after associating it with the icing annotation data; preprocessing the training dataset to extract input feature vectors and output labels; designing a deep neural network model, using an encoder-decoder framework, and through iterative learning with a large number of training samples, enabling the model to automatically learn the key patterns of micro-icing from the input feature vectors and map them to reasonable output predictions; and using the trained model for accurate detection of micro-icing conditions in pipes.

Images