Gas nonmetal pipeline aging degree vibration detection system and method based on deep learning

Abstract

The invention provides a fuel gas non-metal pipeline aging degree vibration detection system and method based on deep learning, and automatic evaluation of the non-metal pipeline aging degree is realized. The system comprises a vibration signal acquisition module, a vibration signal analysis module, a deep learning analysis module and an aging evaluation feedback module. The method comprises a laboratory database establishment stage and an engineering detection stage, the laboratory database establishment stage comprises a vibration signal acquisition module, a vibration signal analysis module, a deep learning analysis module and an aging evaluation feedback module, and the engineering detection stage comprises a vibration signal acquisition module and an aging evaluation feedback module. The method is high in accuracy, efficiency and non-destructive, has little influence on pipeline operation, is suitable for safety monitoring of various non-metal pipelines, and has wide application value.

Description

Technical Field

[0001] The present invention relates to the field of non-metallic pipeline detection, and in particular to a method for generating a vibration signal by a vibration exciter and using deep learning technology to determine the aging degree of a non-metallic gas pipeline. Background Art

[0002] Non-metallic pipes are widely used in the field of gas transportation due to their excellent corrosion resistance, light weight and easy installation. However, as the service life increases, non-metallic pipes, as a polymer material, face the risk of aging, which may lead to serious safety hazards such as pipeline leakage and bursting. Therefore, timely and effective assessment of the aging degree of non-metallic pipes is an important task to ensure the safe operation of non-metallic gas pipes.

[0003] At present, the aging detection of non-metallic pipelines includes macroscopic and microscopic detection methods. Macroscopic detection focuses on the overall physical state and external damage of the pipeline, and commonly used methods include visual detection and ultrasonic detection; microscopic detection focuses on the microstructure and performance of the material, and commonly used methods include infrared imaging and X-ray detection. Whether it is macroscopic or microscopic detection, traditional methods require a lot of time and manpower, especially in large-scale pipe network systems, where real-time monitoring is difficult to achieve. Among them, microscopic detection technologies such as X-ray detection and laboratory analysis equipment are expensive and have high maintenance costs, which increases the economic burden of pipeline maintenance.

[0004] The relationship between the aging degree and vibration signal of non-metallic pipelines is a complex and important research area. Over time, gas non-metallic pipelines will undergo a series of physical and chemical changes during use, which will affect the mechanical properties of the pipelines and thus cause changes in their vibration characteristics.

[0005] As non-metallic pipes age, the physical properties of their materials (such as elastic modulus, tensile strength, and toughness) change. Aging causes the material to become brittle and lose elasticity, which affects the vibration response of the pipe under external forces. The higher the degree of aging, the greater the frequency change and attenuation characteristics of the pipe's vibration characteristics. Secondly, the vibration frequency of aged non-metallic pipes may change when subjected to external stimuli (such as fluid flow, mechanical shock, etc.). Generally, the natural frequency of the pipe may decrease with increasing aging. This is because the stiffness of the material decreases, resulting in a change in the vibration mode. In addition, as the material ages, the structural integrity of the pipe may be affected, resulting in a larger vibration amplitude under the same external excitation.

[0006] As an important representation of the operating status of the pipeline, the vibration signal can reflect the subtle changes and potential damage inside the pipeline. Traditional vibration signal analysis methods often rely on manual feature extraction and empirical judgment, which are inefficient and easily affected by subjective factors. Deep learning can automatically extract representative features from the original vibration signal by building a multi-layer neural network. This automated feature extraction not only improves the accuracy of detection, but also greatly improves the detection efficiency. It can process and analyze large amounts of data in real time and adapt to different pipeline types and aging levels. Summary of the invention

[0007] In response to the above problems, the present invention provides a vibration detection system and method for the aging degree of non-metallic gas pipelines based on deep learning, which generates a vibration signal by an exciter and detects the aging degree of non-metallic gas pipelines in combination with deep learning technology.

[0008] The technical solution adopted by the present invention to solve its technical problems is: a vibration detection system and method for the aging degree of non-metallic gas pipelines based on deep learning, the system includes a vibration signal acquisition module, a vibration signal analysis module, a deep learning analysis module and an aging assessment feedback module. First, after the exciter generates a vibration signal in the pipeline, the vibration signal of the pipeline is collected through a distributed sensor network; secondly, the collected signal is denoised, and the denoised vibration signal is analyzed to extract significant features related to pipeline aging, and establish corresponding labels; then, using the extracted features, a vibration feature library of normal, slightly aged, moderately aged and severely aged pipelines is established, and a deep learning algorithm is used for model training to optimize the model and improve the recognition accuracy; finally, the vibration signal collected in real time is input into the trained deep learning model to identify and classify the degree of aging.

[0009] The method includes two stages: laboratory database establishment and engineering inspection. The laboratory database establishment stage includes four modules: vibration signal collection, vibration signal analysis, deep learning analysis and aging assessment feedback, while the engineering inspection stage includes two modules: vibration signal collection and aging assessment feedback.

[0010] The exciter should be installed at an appropriate position on the outer wall of the pipeline, usually in the middle or at the joint of the pipeline. It is ensured that it can generate a stable excitation signal inside the pipeline. When selecting, its frequency range, volume, weight and output characteristics should be considered to ensure that it covers the vibration frequency of the pipeline and adapts to the vibration characteristics under different aging degrees.

[0011] The vibration signal acquisition module is mainly composed of a sensor, a sensor power supply, a data acquisition instrument and a data processing host.

[0012] The sensors are installed in a distributed manner and are evenly distributed over the entire length of the pipeline to comprehensively monitor the condition of the pipeline. The sensors are directly connected to a data acquisition instrument that supports multi-channel signal acquisition and simultaneously obtains vibration signals from multiple sensors.

[0013] The data acquisition instrument is installed in a data processing host, and one data processing host can be connected to multiple data acquisition instruments to support multi-channel signal acquisition.

[0014] The vibration signal analysis module mainly includes a data preprocessing module and a signal analysis module. Data preprocessing is to denoise the signal collected from the vibration signal acquisition system, and the denoising method adopted is the secondary decomposition technology.

[0015] The signal analysis module performs signal analysis on the denoised vibration signal to reveal the intrinsic connection between vibration characteristics and pipeline aging. By performing frequency domain analysis (power spectrum density analysis, harmonic analysis), time domain analysis (time domain statistical feature analysis, time series change analysis) and time-frequency analysis (wavelet transform) on the denoised vibration data, the key vibration characteristics related to the degree of aging are identified, and the extracted significant features will be combined into a mixed matrix. The features contained in the mixed matrix include time domain features (mean, variance, peak value, peak factor), frequency domain features (main frequency, harmonic component, power spectrum density), time-frequency features (wavelet transform coefficients) and pipeline inherent characteristics (pipeline natural frequency, damping ratio and vibration mode).

[0016] The deep learning analysis module mainly includes a vibration feature library and a state recognition learning module.

[0017] The extracted significant vibration features are stored in a pipeline aging degree feature library.

[0018] The vibration feature library provides data to the deep recognition learning module.

[0019] The deep recognition learning module mainly uses convolutional neural network (CNN), long short-term memory network (LSTM) and self-attention mechanism for model training. The accuracy of the model is improved by adjusting model parameters and optimizing algorithms.

[0020] The aging assessment feedback module evaluates the trained model and uses data enhancement technology (adding noise, time scaling and time shifting) to ensure accuracy under different conditions (different pipe lengths, diameters, fluid properties). The vibration signal collected in real time on site is input into the trained model, and the model will output the aging assessment result of the non-metallic pipe.

[0021] The beneficial effects of the present invention are as follows: the method for detecting the aging degree of non-metallic gas pipelines is easy to operate, does not require special equipment, and only requires conventional vibration sensors and computer hosts to complete data collection, processing and model training; the detection method does not interfere with the normal operation of the pipeline, can perform non-stop transmission detection under transmission conditions, and has no special requirements for the installation environment; the detection method is highly flexible, and can train deep learning models for different aging degrees based on the different aging degree data collected on-site, and then identify different types of non-metallic pipelines; the detection method is not only targeted, but also universal. As the database expands, the recognition ability will also be enhanced, and the aging types that can be covered will gradually increase, which is convenient for promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Attached Figure 1 Schematic diagram of vibration detection

[0023] Attached Figure 2 Algorithm flow chart Specific implementation methods

[0024] This patent aims to detect the aging degree of non-metallic gas pipelines through vibration signal analysis, and realize automatic evaluation by combining deep learning technology. The entire implementation process includes two stages: laboratory database establishment and engineering inspection. The laboratory database establishment stage includes four modules: vibration signal acquisition, vibration signal analysis, deep learning analysis and aging assessment feedback, while the engineering inspection stage includes two modules: vibration signal acquisition and aging assessment feedback.

[0025] During the database establishment stage of the experiment, several non-metallic gas pipelines with at least four different degrees of aging are first selected as experimental samples, namely: new pipelines, slightly aged pipelines, moderately aged pipelines and severely aged pipelines.

[0026] First, if Figure 1 As shown in the figure, the exciter is installed on the outer wall of the pipe in the middle, and 12 sensors are evenly arranged on the outer surface of the pipe, with a symmetrical distribution to cover the vibration response of the entire pipe. Each sensor needs to be firmly fixed with a clamp and connected to a multi-channel data acquisition instrument through a data cable to ensure the stability of signal transmission.

[0027] In terms of data acquisition system settings, a 16-channel data acquisition instrument is used, and the sampling frequency (5000Hz) is set to collect data from all sensors at the same time. At this stage, the exciter is preliminarily configured, and the frequency is set between 400Hz and 1000Hz, and the amplitude is between 0.5mm and 1mm. After starting the exciter, the frequency is gradually adjusted to observe the response of the acceleration sensor, and a frequency sweep experiment is performed to record the vibration response at each frequency. At the same time, the amplitude of the exciter is appropriately adjusted according to the preliminary experimental results to determine the optimal excitation frequency and amplitude of the exciter.

[0028] During the data acquisition process, the output signal of the acceleration sensor is recorded in real time by the data acquisition instrument while the exciter is working, and the data is stored in the computer for subsequent analysis.

[0029] After data collection, the collected data is first analyzed, mainly including signal preprocessing, signal analysis and labeling. Signal preprocessing includes secondary decomposition method denoising, filtering and standardization.

[0030] Next, the denoised vibration signal is analyzed to extract the significant features related to pipeline aging. The time domain features include mean, variance, peak and crest factor; the frequency domain features are extracted by Fourier transform method, including main frequency components, harmonic characteristics and power spectrum density. The extracted significant features are combined into a mixed matrix. The features contained in the mixed matrix include time domain features (mean, variance, peak and crest factor), frequency domain features (main frequency, harmonic components and power spectrum density), time-frequency features (wavelet transform coefficients) and pipeline inherent characteristics (pipeline natural frequency, damping ratio and vibration mode). These features together provide important information for the assessment and monitoring of pipeline aging.

[0031] Finally, create labels. A label dataset is constructed based on the actual aging degree of the pipeline. The labels are usually divided into four levels: 0 (normal state), 1 (slight aging, service life within 15 years), 2 (moderate aging, service life between 15 and 30 years), and 3 (severe aging, service life over 30 years). These labels are paired with the feature vector of each experimental sample to form a standard training set and test set, thereby providing support for subsequent model training.

[0032] During the deep learning analysis process, the vibration signal features are stored in the pipeline aging feature database, and the convolutional neural network (CNN), long short-term memory network (LSTM) and self-attention mechanism are used for modeling, and the model is trained using the training set. The PyTorch deep learning framework is used for model construction and training. After training, the test set is used to evaluate the model performance, and indicators such as accuracy, recall, and F1 score are calculated to ensure that the model can effectively identify the aging degree of the pipeline.

[0033] In the aging assessment feedback module, in order to ensure that the model can adapt to different conditions (such as different pipeline materials and fluid properties), evaluations are performed under different fluid pressures and pipeline materials and diameters, and data enhancement techniques (adding noise, time scaling, and time shifting) are used to improve the robustness of the model.

[0034] During the engineering inspection phase, there are two modules: vibration signal collection and aging assessment feedback. Vibration signal collection should be carried out as follows: Figure 1 The laboratory collection method shown in the figure is carried out, including the arrangement and parameter selection of the exciter, sensor and data acquisition instrument. After the data collection is completed, the data is transmitted and stored in the computer. After the output results of the evaluation database model are evaluated, the aging degree of the pipeline is automatically evaluated and an evaluation report is generated.

[0035] In a certain inspection project case, a section of a 26-year-old gas polyethylene pipeline was excavated. After collecting vibration signals from the excavated gas PE pipeline, these signals were input into the optimized and evaluated real-time monitoring model. After model analysis, a classification label "2" was generated, which indicates the degree of aging of the pipeline. The final aging assessment results showed that the pipeline was in a moderately aged state and regular annual inspections were required to ensure safe transportation of the pipeline.

[0036] Through the above steps, this patent establishes a vibration detection system and method for the aging degree of gas non-metallic pipelines based on deep learning. This method can not only automatically detect and evaluate the aging degree of pipelines efficiently and accurately, but also provide reliable technical support for practical applications. Through continuous optimization and adjustment, future research can further improve the accuracy and stability of the model and provide more effective solutions for the safety monitoring and maintenance of non-metallic pipelines.

Claims

1. A gas non-metallic pipeline aging degree vibration detection system and method based on deep learning, characterized by: The system includes a vibration signal acquisition module, a vibration signal analysis module, a deep learning analysis module and an aging assessment feedback module; first, a vibration signal of a specific frequency is generated in the pipeline by an exciter to stimulate the pipeline vibration response; then, the distributed vibration signal acquisition system captures the vibration signals at various positions of the pipeline in real time and transmits the data to a data processing host; secondly, the collected vibration signal is denoised, and the key vibration features related to the aging degree are extracted by analyzing the denoised vibration signal and the aging degree; then, the extracted significant vibration features are input into the deep learning analysis module, and the recognition accuracy of the model is improved by adjusting the model parameters and optimizing the algorithm; finally, the accuracy of the aging assessment feedback model is determined, and the vibration signal collected on site is input into the model to obtain the assessment result of the pipeline aging degree.

2. According to the deep learning-based gas non-metallic pipeline aging degree vibration detection system and method according to claim 1, it is characterized by: The method includes two stages: laboratory database establishment and engineering inspection. The laboratory database establishment stage includes four modules: vibration signal collection, vibration signal analysis, deep learning analysis and aging assessment feedback, while the engineering inspection stage includes two modules: vibration signal collection and aging assessment feedback.

3. The system and method for vibration detection of aging degree of non-metallic gas pipelines based on deep learning according to claim 1 is characterized in that: The vibration signal acquisition module is composed of an exciter, a distributed sensor array, a multi-channel data acquisition instrument and a data processing host. The exciter should be installed at an appropriate position on the outer wall of the pipeline, usually in the middle or at the joint of the pipeline. The output signal needs to be adjusted to cover the vibration frequency range of the pipeline, so as to effectively stimulate the characteristic vibration of the aging pipeline. The distributed sensor array is composed of multiple acceleration sensors, which are evenly arranged on the outer surface of the pipeline. The multi-channel data acquisition instrument is connected to the distributed sensor array using a signal line to collect the vibration signals output by each sensor in real time and convert them into digital signals. The data processing host is connected to the multi-channel data acquisition instrument to pre-process and analyze the collected vibration signals to complete the monitoring and evaluation of the aging status of the pipeline.

4. The system and method for vibration detection of aging degree of non-metallic gas pipelines based on deep learning according to claim 1 is characterized in that: The vibration signal analysis module is composed of a signal preprocessing module and a signal analysis module; the signal preprocessing module performs a denoising operation on the vibration signal collected from the vibration signal acquisition module, and uses a secondary decomposition technique to process the noise in the signal to retain useful vibration information; the signal analysis module performs an in-depth analysis on the denoised vibration signal to reveal the intrinsic connection between the vibration characteristics and pipeline aging; by performing frequency domain analysis (such as power spectrum density analysis, harmonic analysis, etc.), time domain analysis (such as time domain statistical feature analysis, time series change analysis, etc.) and time-frequency analysis (such as wavelet transform, etc.) on the collected vibration data, it is possible to identify key vibration characteristics related to the aging process, and combine the extracted significant features into a mixed matrix to provide a comprehensive feature description for subsequent analysis.

5. The system and method for vibration detection of aging degree of non-metallic gas pipelines based on deep learning according to claim 1 is characterized in that: The deep learning analysis module consists of a vibration feature library and a state recognition learning module; the vibration feature library is used to store the extracted significant vibration features related to aging, and classify the features of different aging states to ensure real-time updating of the feature library; the state recognition learning module uses the classified data in the vibration feature library to train a deep learning model, mainly using convolutional neural networks (CNN), long short-term memory networks (LSTM) and self-attention mechanisms to obtain a deep learning model that can be used to identify the aging degree of non-metallic gas pipelines.

6. The system and method for vibration detection of aging degree of non-metallic gas pipelines based on deep learning according to claim 1 is characterized in that: The aging assessment feedback module is used to evaluate the model, using data enhancement technology (adding noise, time scaling and time shifting) to ensure the accuracy of the model under conditions of different pipeline lengths, diameters, fluid properties, etc.

Images