Pipeline health monitoring and temperature compensation method and system based on ultrasonic guided waves
Through rolling training of the temperature compensation algorithm and feature space modulation mechanism of the autoencoder, the problem of ultrasonic guided signal being affected by temperature is solved, and high-precision pipeline damage detection and temperature compensation are achieved, which improves detection accuracy and applicability.
Patent Information
- Application Number
- CN202510410482.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-02
- Publication Date
- 2025-08-26
AI Technical Summary
In the existing pipeline health monitoring methods, ultrasonic guided signals are severely affected by ambient temperature fluctuations, resulting in a decrease in signal distortion and damage detection accuracy. Traditional temperature compensation methods are difficult to adapt to dynamic working conditions, affecting detection accuracy and positioning accuracy.
The temperature compensation algorithm of the rolling training autoencoder is adopted to build a high-precision detection model through an unsupervised learning framework, and the temperature compensation is performed using the health status signal. Combined with the feature space modulation mechanism to eliminate the temperature influence, and realize signal reconstruction and damage detection.
It significantly improves the accuracy and robustness of pipeline damage detection under complex working conditions, enhances the applicability and reliability of the algorithm, and avoids the high cost and safety risks of destructive data acquisition.
Smart Images
Figure CN120541381A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of pipeline health monitoring, and in particular relates to a pipeline health monitoring and temperature compensation method and system based on ultrasonic guided waves. Background Art
[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.
[0003] As one of the five major modes of transportation, pipelines have become essential infrastructure for economic development and social well-being. Their structural safety not only impacts the stable operation of infrastructure but also directly impacts the protection of the ecological environment. However, over the course of long-term service, pipeline structures are inevitably affected by environmental erosion, material aging, loads, and both human and natural factors. The combined effects of these hazards lead to cumulative damage and reduced resistance, weakening the pipeline's ability to withstand natural disasters, withstand normal loads, and adapt to environmental changes. This in turn increases the risk of leakage and can even lead to catastrophic accidents. Once a pipeline leak occurs, failure to promptly detect, accurately locate, and effectively repair it will not only result in wasted resources and economic losses, but may also cause personal injury and environmental pollution, with serious consequences. Therefore, research on pipeline health monitoring and damage detection technologies is of great practical significance for improving the reliability and safety of pipeline transportation systems and ensuring sustainable economic and social development.
[0004] Current pipeline inspection methods, such as X-ray, ultrasonic phased array, and pulsed eddy current testing, have short detection distances, small inspection areas, complex operations, and manual labor, making online inspection difficult. In contrast, ultrasonic guided waves, with their low cost, long propagation distance, and high sensitivity to damage, have become a research hotspot in pipeline nondestructive testing, particularly for online monitoring. However, in actual engineering applications, fluctuations in ambient temperature can significantly affect the material's elastic modulus and guided wave propagation characteristics, leading to nonlinear distortions such as signal time-of-flight offset and amplitude attenuation. These factors can significantly interfere with damage feature extraction, reducing damage detection accuracy and positioning precision, and posing challenges to pipeline health monitoring.
[0005] Existing temperature compensation methods mainly rely on empirical formulas or static calibration. First, traditional linear models (such as the temperature-time offset coefficient method) usually assume that there is a simple proportional relationship between temperature and signal parameters. However, in reality, the dependence of material parameters on temperature has significant nonlinear characteristics. In different temperature ranges, the rate of change of the material elastic modulus is not constant, which makes the linear compensation method prone to residual error accumulation, affecting the compensation accuracy. Secondly, there is a data coupling effect in the signals of the multi-sensor array, which makes the temperature compensation process more complicated and further increases the difficulty of analysis and correction. In addition, the dynamic working conditions in the pipeline operation environment (such as diurnal temperature difference cycles and seasonal temperature mutations) require the compensation algorithm to have real-time adaptive capabilities, and the static method based on a fixed database is difficult to effectively deal with time-varying interference, which limits its applicability in actual engineering applications. Summary of the Invention
[0006] To overcome the shortcomings of the aforementioned prior art, the present invention provides a pipeline health monitoring and temperature compensation method and system based on ultrasonic guided waves. This method utilizes a rolling training autoencoder temperature compensation algorithm. Through an unsupervised learning framework, it completely eliminates reliance on damage annotation data. High-precision detection models can be constructed solely using health status signals, avoiding the high costs and safety risks associated with destructive data collection. Through a rolling training model, data scale can be exponentially expanded, improving generalization capabilities for unknown operating conditions. A feature space modulation mechanism effectively eliminates the effects of temperature on guided wave signals, significantly improving the accuracy and robustness of pipeline damage detection under complex operating conditions.
[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:
[0008] The first aspect of the present invention provides a pipeline health monitoring and temperature compensation method based on ultrasonic guided waves;
[0009] A pipeline health monitoring and temperature compensation method based on ultrasonic guided waves, comprising:
[0010] Obtain the guided wave signals of the pipeline under test in the intact state and the guided wave signals in the lossy state respectively, and perform preprocessing;
[0011] Constructing a temperature embedding autoencoder neural network model, and using the waveguide signal in the lossless state and the waveguide signal in the lossy state to train the temperature embedding autoencoder neural network model through a reverse gradient fusion strategy, a composite loss function, and a dynamic learning rate scheduling;
[0012] The pre-processed waveguide signal is input into the trained temperature embedding autoencoder neural network model to generate a temperature-compensated waveguide signal.
[0013] The temperature embedded autoencoder neural network model encodes the waveguide signal through a multi-channel encoder to obtain a characteristic tensor of the waveguide signal;
[0014] The real-time temperature of the current waveguide signal and the baseline temperature corresponding to the expected output signal are obtained through the bidirectional temperature modulation module, and the temperature difference is calculated;
[0015] Performing affine transformation on the temperature difference and the characteristic tensor of the waveguide signal through a fully connected layer to obtain a scaling coefficient and an offset coefficient, using the scaling coefficient and the offset coefficient to compensate for the temperature dependence of the material elastic modulus and propagation attenuation, and obtaining the signal characteristics after temperature modulation through feature space modulation;
[0016] The temperature-modulated signal features are input into the residual signal decoder, and the original waveguide signal is combined with the temperature features through two deconvolution operations to output a compensation signal.
[0017] As a further technical solution, the process of obtaining the guided wave signal of the pipeline to be tested is:
[0018] Build an ultrasonic guided wave test platform and symmetrically arrange a piezoelectric sensor array on the pipeline to be tested. Use a Hanning window modulated sinusoidal signal as the excitation signal, and set the signal modulation parameters.
[0019] The guided wave signal of the pipeline to be tested is obtained according to the set signal modulation parameters and the piezoelectric sensor array.
[0020] As a further technical solution, the preprocessing process includes crosstalk suppression, normalization and elimination of abnormal data points on the waveguide signal.
[0021] As a further technical solution, the preprocessing process also includes expanding the data set size through a rolling matching mechanism and allocating training weights according to the data quality of different channels.
[0022] As a further technical solution, the process of expanding the dataset size through the rolling matching mechanism and allocating training weights according to the data quality of different channels is as follows:
[0023] For each waveguide signal in a lossless state, traverse all waveguide signals in a lossless state to generate input target pairs:
[0024] For each channel of the collected guided wave signal, the data reliability of the sensor channel is quantified based on the temperature and peak time, and the correlation coefficient and significance index are calculated as shown in the following formula:
[0025]
[0026] In the formula, r is the correlation coefficient; p is the significance index; T k is temperature; tpk is the peak time, and T k and t pk The mean of , n is the number of samples in a healthy state;
[0027] The channel weight w is defined according to the correlation coefficient and significance index, as shown in the following formula:
[0028] w=|r|·(1-p);
[0029] Where w is the weight, w∈[0,1].
[0030] As a further technical solution, the process of training the temperature embedded autoencoder neural network model through reverse gradient fusion strategy, composite loss function and dynamic learning rate scheduling is as follows:
[0031] The reverse gradient fusion strategy calculates the loss function gradient of the model parameters through conventional back propagation. The calculation process is as follows:
[0032]
[0033] Where, is the gradient of the loss function; L t and L w Indicates the difference between two signals of the same state but different temperatures in the time domain and waveform domain respectively, w is the channel weight; μ and δ are the coefficients of the loss function in the time domain and waveform domain respectively; Representative according to L t Calculate the gradient of the loss function;
[0034] The constructed composite loss function jointly optimizes the time domain characteristics and waveform physical consistency, L t and L w The expression is as follows:
[0035] L t =MSE(S i (t|T i ),S j (t|T j ));
[0036] L w =1-Corr(S i (t|T i ),S j (t|T j ));
[0037] Where MSE represents the point-by-point mean square error between the predicted signal and the true signal, and Corr represents the correlation coefficient between the predicted signal and the true signal. While MSE focuses on point-by-point error, Corr focuses on overall waveform similarity and aligns signal time shifts, eliminating time delay deviations caused by temperature.
[0038] The dynamic learning rate scheduling adopts cosine annealing dynamic learning rate, specifically:
[0039]
[0040] Among them, μ t Represents the learning rate of the current cycle, μ max and μ min are the upper and lower limits of the learning rate, and T is a complete annealing cycle.
[0041] As a further technical solution, the process of mapping the temperature difference to the scaling coefficient and the offset coefficient through the fully connected layer is shown in the following formula:
[0042] γ = W2(ReLU(W1(ΔT)));
[0043] β = V2(ReLU(V1(ΔT)));
[0044] Where W1, W2, V1, and V2 are the parameters of the fully connected layer; ΔT is the temperature difference; γ is the scaling factor; β is the offset factor; and ReLU is the activation function.
[0045] A second aspect of the present invention provides a pipeline health monitoring and temperature compensation system based on ultrasonic guided waves.
[0046] A pipeline health monitoring and temperature compensation system based on ultrasonic guided waves, comprising:
[0047] The waveguide signal acquisition module is configured to respectively acquire the waveguide signal of the pipeline under test in the intact state and the waveguide signal in the lossy state, and perform preprocessing;
[0048] The model training module is configured to: construct a temperature embedding autoencoder neural network model, and train the temperature embedding autoencoder neural network model using the waveguide signal in the lossless state and the waveguide signal in the lossy state through a reverse gradient fusion strategy, a composite loss function, and a dynamic learning rate scheduling;
[0049] The temperature difference calculation module is configured to: input the preprocessed waveguide signal into the trained temperature embedding autoencoder neural network model, encode the waveguide signal through a multi-channel encoder to obtain the characteristic tensor of the waveguide signal; obtain the real-time temperature of the current waveguide signal and the baseline temperature corresponding to the expected output signal through the bidirectional temperature modulation module, and calculate the temperature difference;
[0050] a feature space modulation module configured to: perform affine transformation on the feature tensors of the temperature difference and the waveguide signal through a fully connected layer to obtain a scaling factor and an offset factor, use the scaling factor and the offset factor to compensate for the temperature dependence of the material elastic modulus and propagation attenuation, and obtain the signal characteristics after temperature modulation through feature space modulation;
[0051] The temperature compensation module is configured to: input the signal characteristics after temperature modulation into the residual signal decoder, combine the original waveguide signal with the temperature characteristics through two deconvolution operations, and output a compensation signal.
[0052] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the pipeline health monitoring and temperature compensation method based on ultrasonic guided waves as described in the first aspect of the present invention.
[0053] A fourth aspect of the present invention provides an electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the pipeline health monitoring and temperature compensation method based on ultrasonic guided waves as described in the first aspect of the present invention are implemented.
[0054] One or more of the above technical solutions have the following beneficial effects:
[0055] (1) This invention addresses the issue of signal drift caused by environmental factors in ultrasonic guided waves in pipelines. It proposes a pipeline health monitoring and temperature compensation method and system based on ultrasonic guided waves. This method, through an unsupervised learning framework, completely eliminates the reliance on damage annotation data and constructs a high-precision detection model using only health status signals, avoiding the high costs and safety risks associated with destructive data collection. Through a rolling training model, the data scale is exponentially expanded, improving the generalization capability for unknown working conditions.
[0056] (2) When the temperature in the pipeline monitoring environment fluctuates randomly, the current temperature parameter is input into the feature extraction layer of the autoencoder in real time, automatically adjusting the signal reconstruction constraints to ensure the accuracy of temperature compensation. In practical applications, any temperature signal can be flexibly selected as a reference, thereby improving the algorithm's temperature compensation capability and enhancing its applicability and reliability in complex working conditions.
[0057] (3) Through the characteristic space modulation mechanism, the influence of temperature on the guided wave signal is effectively eliminated, and the accuracy and robustness of pipeline damage detection under complex working conditions are significantly improved.
[0058] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0060] Figure 1 This is a flow chart of the method of the first embodiment.
[0061] Figure 2 This is a schematic diagram of the detection system simulation in the first embodiment.
[0062] FIG3 is a diagram showing a visualization entity of an excitation signal in the first embodiment; Figure 3a is the time domain signal diagram, Figure 3b It is a frequency domain signal diagram.
[0063] Figure 4 Schematic diagram of structural damage on the pipeline introduced in the first embodiment.
[0064] Figure 5 Schematic diagram of the model training process in the first embodiment.
[0065] Figure 6 This is a schematic diagram of the model architecture constructed in the first embodiment.
[0066] Figure 7 This is a system structure diagram of the second embodiment. DETAILED DESCRIPTION
[0067] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.
[0068] It should be noted that the terms used herein are for describing particular embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.
[0069] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.
[0070] Example 1
[0071] This embodiment discloses a pipeline health monitoring and temperature compensation method based on ultrasonic guided waves;
[0072] like Figure 1As shown, a pipeline health monitoring and temperature compensation method based on ultrasonic guided waves includes:
[0073] Step S1, respectively obtaining the guided wave signal of the pipeline to be tested in the intact state and the guided wave signal in the lossy state, and performing preprocessing;
[0074] Combine Figure 2 To obtain guided wave signals, a pipeline ultrasonic guided wave testing platform is first constructed. It primarily consists of a DC power supply, a PC, a pipeline piezoelectric ultrasonic guided wave monitor, a coaxial shielded cable, a piezoelectric sensor ring, and the pipeline to be tested. In this example, the pipeline to be tested is a custom-made 9.2-meter carbon steel pipeline with NPS6, Schedule 40 dimensions. The pipeline has an outer diameter of 0.84 meters. The piezoelectric sensor ring is installed 0.5 meters from one end of the pipeline, with the sensor facing away from the end as the positive direction. The sensor position is set as the zero point for the axial position. A weld is located at 3.7 meters and 6.9 meters along the pipeline. The piezoelectric sensor ring consists of 16 symmetrical circular piezoelectric ceramic sensors.
[0075] The test uses a pipeline piezoelectric ultrasonic guided wave monitoring system with a signal sampling frequency of 3 MHz. The excitation signal is an N-cycle sinusoidal signal modulated by a Hanning window. The formula is:
[0076]
[0077] Among them, f c is the excitation signal frequency, and A is the amplitude of the sinusoidal signal.
[0078] When selecting the number of cycles and the excitation frequency, for the number of cycles, the larger the N, the narrower the excitation signal frequency band, the more concentrated the response signal energy, the larger the amplitude of the weld reflection echo signal and the end face echo signal, but the wider the time domain width of the excitation signal and the response signal, the more serious the waveform superposition; for the excitation frequency, the piezoelectric sensor responds better to the guided wave signal at a lower frequency, but as the excitation frequency increases, the length of the wave packet will become shorter and the reflected wave packet will become more concentrated, which is conducive to improving the defect location accuracy.
[0079] Taking the above factors into consideration, an 8-cycle sinusoidal signal modulated by a Hanning window is selected in this embodiment. The excitation signal frequency is 40kHz, and the excited waveguide mode is the T(0,1) mode. The time domain and frequency domain diagrams of the excitation signal are shown in Figure 2. Figure 3a and Figure 3b shown.
[0080] According to the obtained optimal signal modulation parameters, the signal excitation and acquisition of each sensor path in the entire sensor network are performed in sequence to obtain the guided wave signal in a lossless state, which is recorded as the reference signal.
[0081] At an axial distance of 7.55m, the pipeline is damaged, and a damage of 80mm×25mm×2mm is introduced. Figure 4 As shown, repeat the previous step to obtain the waveguide signal in the lossy state, which is recorded as the damage signal.
[0082] Furthermore, the acquired waveguide signals are subjected to crosstalk suppression and normalization, and all collected data are analyzed as a whole. Abnormal data points are removed to construct a high-quality training set for subsequent model training and optimize model learning efficiency.
[0083] Specifically, during the duration of the excitation signal, there is crosstalk. Therefore, it is necessary to eliminate noise interference and invalid data caused by temperature anomalies in the signal to ensure the high quality of the training set. The theoretical duration of the signal is calculated based on the expression of the excitation signal, which is as follows:
[0084]
[0085] Where N represents the N-cycle modulation of the excitation signal, f c is the frequency of the excitation signal. The time window is appropriately stretched to filter the crosstalk and normalize to eliminate amplitude fluctuations. The normalized expression is:
[0086]
[0087] Where S norm is the normalized signal; S is the signal to be normalized; is the average value of S; S max is the maximum value of S; S min is the minimum value of S;
[0088] After the data is normalized, the peak position corresponds to the boundary echo. Since temperature changes will cause the material to expand, the flight time is basically linearly related to temperature. Therefore, it is considered to identify and eliminate invalid data based on physical laws. First, the measured data of each channel {T k ,t pk} Perform linear fitting:
[0089] t pk =aT k +b;
[0090] Next, plot t pk -T k Image, remove the data points that are obviously deviated and eliminate invalid data.
[0091] Finally, in order to maximize the use of limited data, the temperature difference-signal mapping relationship is used to build training pairs to enhance the generalization ability of the model. i (t|Ti ), traverse other health signals H in the dataset j (t|T j ), thus generating input-target pairs:
[0092] Input={H i (t|T i ),T i ,T j};
[0093] Output={H j (t|T j )};
[0094] Through the rolling matching mechanism, the data set size of n collected data is expanded to n 2 , data utilization is greatly improved. Each signal is associated with data from multiple temperature points, forming a one-to-many training relationship.
[0095] For each channel in the collected data, according to the temperature T k and peak time t pk To quantify the data reliability of the sensor channel, the correlation coefficient r and significance index p are calculated, and the expressions are as follows:
[0096]
[0097] Where, T k and t pk is the data point of the kth sample, and T k and t pk The mean of n is the number of samples in the healthy state.
[0098] Since the larger the absolute value of the correlation coefficient, the stronger the correlation between the data; the smaller the significance index, the more significant the correlation. Therefore, the channel weight w is defined according to the calculated correlation coefficient r and significance index p, and the expression is as follows:
[0099] w=|r|·(1-p);
[0100] The adaptive channel weight w∈[0,1] is calculated to tilt more training intensity towards data with higher reliability.
[0101] Step S2: constructing a temperature embedding autoencoder neural network model, and using the waveguide signal in the lossless state and the waveguide signal in the lossy state to train the temperature embedding autoencoder neural network model through a reverse gradient fusion strategy, a composite loss function, and a dynamic learning rate scheduling;
[0102] Combine Figure 5In step S2, the acquired reference signal is used to construct a healthy state data set, and the damaged signal is used to construct a damaged state data set. The healthy state data set is used to train the model, and the damaged state data set is directly used to test the temperature compensation effect of the model.
[0103] The constructed temperature embedding autoencoder neural network model consists of three components: a multi-channel encoder, a temperature modulation module, and a decoder. Training only requires healthy signals. Once the model implements temperature compensation, it can be actively expanded to compensate for damaged signals.
[0104] (1) Calculate the gradient of the loss function of the network parameters through conventional back propagation. The gradient is calculated as follows:
[0105]
[0106] Where, is the gradient of the loss function; L t and L w Indicates the difference between two signals of the same state but different temperatures in the time domain and waveform domain respectively, w is the channel weight; μ and δ are the coefficients of the loss function in the time domain and waveform domain respectively; Representative according to L t Compute the gradient of the loss function.
[0107] (2) The constructed composite loss function jointly optimizes the time domain characteristics and the physical consistency of the waveform, L t and L w The expression is as follows:
[0108]
[0109] MSE represents the point-by-point mean square error between the predicted signal and the true signal, and Corr represents the correlation coefficient between the predicted signal and the true signal. While MSE focuses on point-by-point error, Corr focuses on overall waveform similarity and aligns signal time shifts, eliminating temperature-induced delay variations.
[0110] (3) Finally, a cosine annealing dynamic learning rate is introduced into the training process to prevent overfitting. The expression is:
[0111]
[0112] Among them, μ t represents the learning rate of the current cycle, and μ max and μ min are the upper and lower limits of the learning rate, and T is a complete annealing cycle.
[0113] Step S3, inputting the preprocessed waveguide signal into the trained temperature embedded autoencoder neural network model to generate a temperature compensated waveguide signal;
[0114] Among them, combined Figure 6 , the temperature embedded autoencoder neural network model encodes the waveguide signal through a multi-channel encoder to obtain a feature tensor of the waveguide signal;
[0115] Let the input dimension be X∈R B×32×L (B: batch size, L is signal length), encodes the 32-channel waveguide signal, and converts the original signal into a 128-dimensional feature tensor through two convolutional layers with a stride of 2. The output feature tensor dimension is X∈R B×128×L / 4 .
[0116] The temperature T1 of the input signal and the baseline temperature T2 corresponding to the expected output signal are obtained through the bidirectional temperature modulation module, and the temperature difference ΔT = T1-T2 is calculated, where the expected output signal refers to the waveguide signal at the same state and baseline temperature. For example, the input signal is the signal of a damaged structure at 25°C, but the database only contains healthy structure signals at 20°C. In order to ensure the accuracy of subsequent calculations, temperature compensation is required, and the input signal needs to be converted into the signal of a damaged structure at 20°C. The temperature difference is then affine transformed through two fully connected layers to obtain the scaling factor γ and the offset factor β, which are expressed as follows:
[0117] γ = W2(ReLU(W1(ΔT)));
[0118] β = V2(ReLU(V1(ΔT)));
[0119] Where W1, W2, V1, V2 are the parameters of the fully connected layer; ΔT is the temperature difference; γ is the scaling factor; β is the offset factor; ReLU is the activation function. Output γ,β∈R B×128 .
[0120] Perform feature space modulation, the expression is:
[0121]
[0122] Where, represents element-by-element multiplication, while F and F′ represent the signal feature space before and after affine transformation mapping, respectively. After feature modulation, the scaling factor γ can compensate for the change in material elastic modulus caused by temperature, while the offset coefficient β corrects the temperature dependence of propagation attenuation.
[0123] The temperature modulated feature F′ is input into the constructed residual signal decoder, and the original signal is combined with the temperature feature through two deconvolution operations to output the compensated signal In addition, the decoder also contains residual connections to implement the identity mapping constraint, which is expressed as follows:
[0124]
[0125] Where X is the feature tensor.
[0126] Example 2
[0127] This embodiment discloses a pipeline health monitoring and temperature compensation system based on ultrasonic guided waves;
[0128] like Figure 7 As shown, a pipeline health monitoring and temperature compensation system based on ultrasonic guided waves includes:
[0129] The waveguide signal acquisition module is configured to respectively acquire the waveguide signal of the pipeline under test in the intact state and the waveguide signal in the lossy state, and perform preprocessing;
[0130] The model training module is configured to: construct a temperature embedding autoencoder neural network model, and train the temperature embedding autoencoder neural network model using the waveguide signal in the lossless state and the waveguide signal in the lossy state through a reverse gradient fusion strategy, a composite loss function, and a dynamic learning rate scheduling;
[0131] The temperature difference calculation module is configured to: input the preprocessed waveguide signal into the trained temperature embedding autoencoder neural network model, encode the waveguide signal through a multi-channel encoder to obtain the characteristic tensor of the waveguide signal; obtain the real-time temperature of the current waveguide signal and the baseline temperature corresponding to the expected output signal through the bidirectional temperature modulation module, and calculate the temperature difference;
[0132] a feature space modulation module configured to: perform affine transformation on the feature tensors of the temperature difference and the waveguide signal through a fully connected layer to obtain a scaling factor and an offset factor, use the scaling factor and the offset factor to compensate for the temperature dependence of the material elastic modulus and propagation attenuation, and obtain the signal characteristics after temperature modulation through feature space modulation;
[0133] The temperature compensation module is configured to: input the signal characteristics after temperature modulation into the residual signal decoder, combine the original waveguide signal with the temperature characteristics through two deconvolution operations, and output a compensation signal.
[0134] Example 3
[0135] The purpose of this embodiment is to provide a computer-readable storage medium.
[0136] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a pipeline health monitoring and temperature compensation method based on ultrasonic guided waves as described in Example 1.
[0137] Example 4
[0138] The purpose of this embodiment is to provide an electronic device.
[0139] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the pipeline health monitoring and temperature compensation method based on ultrasonic guided waves as described in Example 1 are implemented.
[0140] The steps involved in the apparatuses of Examples 2, 3, and 4 above correspond to those of Method Example 1. For detailed implementations, please refer to the relevant description of Example 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and causing the processor to perform any method of the present invention.
[0141] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.
[0142] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.
Claims
1. A pipeline health monitoring and temperature compensation method based on ultrasonic guided waves, characterized in that: include: Obtain the guided wave signals of the pipeline under test in the intact state and the guided wave signals in the lossy state respectively, and perform preprocessing; Constructing a temperature embedding autoencoder neural network model, and using the waveguide signal in the lossless state and the waveguide signal in the lossy state to train the temperature embedding autoencoder neural network model through a reverse gradient fusion strategy, a composite loss function, and a dynamic learning rate scheduling; The pre-processed waveguide signal is input into the trained temperature embedding autoencoder neural network model to generate a temperature-compensated waveguide signal. The temperature embedded autoencoder neural network model encodes the waveguide signal through a multi-channel encoder to obtain a characteristic tensor of the waveguide signal; The real-time temperature of the current waveguide signal and the baseline temperature corresponding to the expected output signal are obtained through the bidirectional temperature modulation module, and the temperature difference is calculated; Performing affine transformation on the temperature difference and the characteristic tensor of the waveguide signal through a fully connected layer to obtain a scaling coefficient and an offset coefficient, using the scaling coefficient and the offset coefficient to compensate for the temperature dependence of the material elastic modulus and propagation attenuation, and obtaining the signal characteristics after temperature modulation through feature space modulation; The temperature-modulated signal features are input into the residual signal decoder, and the original waveguide signal is combined with the temperature features through two deconvolution operations to output a compensation signal.
2. The pipeline health monitoring and temperature compensation method based on ultrasonic guided waves according to claim 1, characterized in that: The process of obtaining the guided wave signal of the pipeline to be tested is as follows: Build an ultrasonic guided wave test platform and symmetrically arrange a piezoelectric sensor array on the pipeline to be tested. Use a Hanning window modulated sinusoidal signal as the excitation signal, and set the signal modulation parameters. The guided wave signal of the pipeline to be tested is obtained according to the set signal modulation parameters and the piezoelectric sensor array.
3. The pipeline health monitoring and temperature compensation method based on ultrasonic guided waves according to claim 1, characterized in that: The preprocessing process includes crosstalk suppression, normalization and elimination of abnormal data points on the waveguide signal.
4. The pipeline health monitoring and temperature compensation method based on ultrasonic guided waves according to claim 1, characterized in that: The preprocessing process also includes expanding the data set size through a rolling matching mechanism and allocating training weights according to the data quality of different channels.
5. The pipeline health monitoring and temperature compensation method based on ultrasonic guided waves according to claim 4, characterized in that: The process of expanding the dataset size through the rolling matching mechanism and allocating training weights according to the data quality of different channels is as follows: For each waveguide signal in a lossless state, traverse all waveguide signals in a lossless state to generate input target pairs: For each channel of the collected guided wave signal, the data reliability of the sensor channel is quantified based on the temperature and peak time, and the correlation coefficient and significance index are calculated as shown in the following formula: In the formula, r is the correlation coefficient; p is the significance index; T k is temperature; t pk is the peak time, and T k and t pk The mean of , n is the number of samples in a healthy state; The channel weight w is defined according to the correlation coefficient and significance index, as shown in the following formula: w=|r|·(1-p); Where w is the weight, w∈[0,1].
6. The pipeline health monitoring and temperature compensation method based on ultrasonic guided waves according to claim 1, characterized in that: The process of training the temperature embedding autoencoder neural network model through reverse gradient fusion strategy, composite loss function and dynamic learning rate scheduling is as follows: The reverse gradient fusion strategy calculates the loss function gradient of the network parameters through conventional back propagation. The calculation process is as follows: Where, is the gradient of the loss function; L t and L w Indicates the difference between two signals of the same state but different temperatures in the time domain and waveform domain respectively, w is the channel weight; μ and δ are the coefficients of the loss function in the time domain and waveform domain respectively; Representative according to L t Calculate the gradient of the loss function; The constructed composite loss function jointly optimizes the time domain characteristics and waveform physical consistency, L t and L w The expression is as follows: L t =MSE(S i (t|T i ),S j (t|T j )); L w =1-Corr(S i (t|T i ),S j (t|T j )); Where MSE represents the point-by-point mean square error between the predicted signal and the true signal, and Corr represents the correlation coefficient between the predicted signal and the true signal. While MSE focuses on point-by-point error, Corr focuses on overall waveform similarity and aligns signal time shifts, eliminating time delay deviations caused by temperature. The dynamic learning rate scheduling adopts cosine annealing dynamic learning rate, specifically: Among them, μ t Represents the learning rate of the current cycle, μ max and μ min are the upper and lower limits of the learning rate, and T is a complete annealing cycle.
7. The pipeline health monitoring and temperature compensation method based on ultrasonic guided waves according to claim 1, characterized in that: The characteristic tensors of the temperature difference and the waveguide signal are affine transformed through the fully connected layer to obtain the scaling coefficient and the offset coefficient as shown in the following formula: γ=W2(ReLU(W1(ΔT)00); β = V2(ReLU(V1(ΔT))); Where W1, W2, V1, and V2 are the parameters of the fully connected layer; ΔT is the temperature difference; γ is the scaling factor; β is the offset factor; and ReLU is the activation function.
8. A pipeline health monitoring and temperature compensation system based on ultrasonic guided waves, characterized by: include: The waveguide signal acquisition module is configured to respectively acquire the waveguide signal of the pipeline under test in an intact state and in a lossy state, and perform preprocessing; The model training module is configured to: construct a temperature embedding autoencoder neural network model, and train the temperature embedding autoencoder neural network model using the waveguide signal in the lossless state and the waveguide signal in the lossy state through a reverse gradient fusion strategy, a composite loss function, and a dynamic learning rate scheduling; The temperature difference calculation module is configured to: input the preprocessed waveguide signal into the trained temperature embedding autoencoder neural network model, encode the waveguide signal through a multi-channel encoder, and obtain a feature tensor of the waveguide signal; The real-time temperature of the current waveguide signal and the baseline temperature corresponding to the expected output signal are obtained through the bidirectional temperature modulation module, and the temperature difference is calculated; a feature space modulation module configured to: perform affine transformation on the feature tensors of the temperature difference and the waveguide signal through a fully connected layer to obtain a scaling factor and an offset factor, use the scaling factor and the offset factor to compensate for the temperature dependence of the material elastic modulus and propagation attenuation, and obtain the signal characteristics after temperature modulation through feature space modulation; The temperature compensation module is configured to: input the signal characteristics after temperature modulation into the residual signal decoder, combine the original waveguide signal with the temperature characteristics through two deconvolution operations, and output a compensation signal.
9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the pipeline health monitoring and temperature compensation method based on ultrasonic guided waves as described in any one of claims 1 to 7 are implemented.
10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the pipeline health monitoring and temperature compensation method based on ultrasonic guided waves as described in any one of claims 1 to 7 are implemented.