A method and system for predicting the service life of a stainless steel pipe

Abstract

The application discloses a kind of life prediction method and system of stainless steel pipe, including the following steps: collecting sensor data of stainless steel pipe and pre-processing, generating pre-processing dataset;The pre-processing dataset is input to improved SimCLR model, and feature representation is generated;Feature representation is input to improved Mamba model, and time series modeling and residual life prediction are carried out;The contrast loss of improved SimCLR model is minimized, and the regression loss and classification loss of improved Mamba model are minimized;Real-time receive sensor data from stainless steel pipe, carry out residual life prediction, periodically update and optimize model;According to real-time prediction result, health management and maintenance decision are carried out to stainless steel pipe, and life prediction report is generated.The method can accurately predict the residual service life of stainless steel pipe, and improve the intelligent level of pipeline maintenance decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and sensor technology, and particularly relates to a service life prediction method and system for a stainless steel pipe. BACKGROUND

[0002] With the continuous improvement of industrial automation and intelligence level, the health monitoring and life prediction technology for pipeline equipment has gradually attracted widespread attention. The existing stainless steel pipe health management system mostly relies on traditional periodic detection and experience judgment method, or simple sensor data acquisition and analysis. However, in practical application, these methods generally have the following problems:

[0003] Firstly, the traditional pipeline health monitoring system often only relies on single sensor data such as pressure or temperature data, ignoring the comprehensive influence of multiple working condition factors, resulting in limited prediction accuracy; secondly, the existing life prediction model mostly adopts linear method or simple model based on statistical analysis, which cannot fully capture the complex nonlinear relationship in sensor data, resulting in inaccurate prediction results under extreme working conditions; in addition, the current prediction method is highly sensitive to data noise and abnormal values, and cannot effectively deal with the common sensor failure and data missing problems in long-term monitoring, further reducing the reliability of prediction; finally, there are still great challenges in real-time data processing and model updating, and the traditional method is difficult to realize real-time feedback mechanism, so that the model cannot be adjusted in time according to the latest data, thereby affecting the real-time and accuracy of pipeline life prediction.

[0004] Therefore, how to provide a service life prediction method and system for a stainless steel pipe is a problem that those skilled in the art need to solve. SUMMARY

[0005] One object of the present application is to provide a service life prediction method and system for a stainless steel pipe, which fully utilizes sensor data acquisition, improved SimCLR and improved Mamba model, and describes in detail a pipeline residual life prediction algorithm based on contrast learning, time series modeling and regression analysis, which has the advantages of real-time prediction, dynamic optimization and maintenance decision support.

[0006] According to the service life prediction method for a stainless steel pipe provided by the embodiment of the present application, the following steps are included:

[0007] Collecting sensor data of the stainless steel pipe and performing preprocessing to generate a preprocessed data set;

[0008] Inputting the preprocessed data set into the improved SimCLR model to generate feature representation through data enhancement, feature extraction and contrast learning;

[0009] The feature representation is input into the improved Mamba model to perform time series modeling and remaining life prediction, and a prediction result is generated.

[0010] The improved SimCLR model is minimized to contrast loss, and the improved Mamba model is minimized to regression loss and classification loss.

[0011] Real-time sensor data from the stainless steel pipe is received, remaining life prediction is performed, real-time prediction results are generated, and the improved SimCLR model and the improved Mamba model are periodically updated and optimized through real-time data feedback;

[0012] According to the real-time prediction result, the health management and maintenance decision of the stainless steel pipe is carried out, and a specific life prediction or health state evaluation report is provided.

[0013] Optionally, the sensor data of the stainless steel pipe is collected and preprocessed to generate a preprocessed data set, specifically including:

[0014] The sensor data of the stainless steel pipe is collected, and the data includes temperature data, pressure data, corrosion degree data, vibration data and image data of the surface of the stainless steel pipe, and an original data set is generated;

[0015] The original data set is preprocessed to generate a preprocessed data set, and the preprocessing includes removing noise from the temperature data, pressure data, corrosion degree data and vibration data through a filter or a smoothing algorithm to generate a denoising data set, normalizing each type of data in the denoising data set to generate a normalized data set, performing time series interpolation on missing time series data in the normalized data set to fill in missing values in the data to generate an interpolation data set, and performing size unification and color standardization on the image data to generate a standardized image data set.

[0016] The interpolation data set and the standardized image data set are aligned according to the timestamp or data correspondence, and are spliced or fused based on the corresponding features to generate a unified preprocessed data set.

[0017] Optionally, the preprocessed data set is input into the improved SimCLR model to generate a feature representation through data augmentation, feature extraction and contrast learning, specifically including:

[0018] The preprocessed data set is input into the improved SimCLR model, and the improved SimCLR model includes an adaptive multi-modal data augmentation module, a feature extraction module, a non-linear projection head and a contrast learning module.

[0019] The adaptive multi-modal data enhancement module generates a plurality of perturbation samples by adding noise to the data, time translation, time scale transformation, obtains an enhanced time series data set, performs spatial-frequency domain joint enhancement on the image data, including Fourier transform on the image data, scaling, shearing and filtering operations on the image in the frequency domain, and generates an enhanced image data set by combining geometric transformation, and combines the enhanced time series data set and the enhanced image data set to generate an enhanced data set;

[0020] The enhanced data set is input into a feature extraction module, which uses a multi-scale self-attention convolutional neural network to learn features, generates an intermediate feature representation, and the multi-scale self-attention convolutional neural network uses convolution kernels of different scales in different convolution layers, and combines a self-attention mechanism to adaptively assign attention weights;

[0021] The intermediate feature representation is input into a nonlinear projection head, which includes a multilayer perceptron and a trainable gating mechanism, maps the intermediate feature representation to a low-dimensional space through a multilayer ReLU or GELU activation function, and generates a low-dimensional feature representation, and the trainable gating mechanism is used to dynamically adjust the activation intensity of each feature channel.

[0022] An adaptive weighted contrastive loss and a dynamic temperature adjustment mechanism are used to optimize the feature representation, the adaptive weighted contrastive loss dynamically adjusts the weights of positive and negative sample pairs according to the difficulty of the samples, the difficulty of the samples is the distance or similarity of the positive and negative sample pairs in the feature space, and the dynamic temperature adjustment mechanism adjusts the temperature parameter in the loss function through an adaptive algorithm to generate the feature representation.

[0023] Optionally, the feature representation is input into an improved Mamba model to perform time series modeling and remaining life prediction, and generate a prediction result, specifically including:

[0024] The feature representation is input into an improved Mamba model, and the improved Mamba model includes a time series modeling module and a life prediction module.

[0025] The time series modeling module uses a selective state space network and a self-attention mechanism to generate a time series feature representation, the selective state space network dynamically selects the most critical time step for modeling by evaluating each time step in the feature representation, and the input of each time step is evaluated by a weighting mechanism to combine past time step information to evaluate the contribution of the current time step to the remaining life prediction as the importance of the current time step.

[0026] The self-attention mechanism is used to calculate and weight the relative importance between each time step in the time sequence feature representation, and generate a weighted time sequence feature representation. The self-attention mechanism calculates the similarity between each time step and other time steps to evaluate the contribution of each time step to the entire sequence. The attention score of each time step is used to calculate the influence of the time step on the weighted time sequence feature representation. The calculation process includes dot product operation on each pair of time steps, score normalization, and weight generation for each time step.

[0027] The weighted time sequence feature representation is input into a life prediction module. The life prediction module processes the weighted time sequence feature representation through a regression network to generate a remaining life prediction value as a prediction result output. The regression network adopts a multi-layer perceptron structure and transfers the weighted time sequence feature representation layer by layer. Each layer includes a fully connected layer and a ReLU activation function.

[0028] Optionally, the improved SimCLR model minimizes the contrastive loss, and the improved Mamba model minimizes the regression loss, specifically including:

[0029] When training the improved SimCLR model, the contrastive loss is minimized. The contrastive loss calculates the similarity between positive sample pairs and negative sample pairs, so that the distance between positive sample pairs in the feature space is as small as possible, and the distance between negative sample pairs is as large as possible.

[0030] When training the improved Mamba model, the regression loss is minimized. The regression loss calculates the difference between the remaining life prediction value predicted by the improved Mamba model and the actual remaining life value, and minimizes the difference.

[0031] The joint loss function simultaneously minimizes the regression loss and the contrastive loss by weighted sum of the regression loss and the contrastive loss. The improved SimCLR model optimizes the contrastive loss when performing feature learning, and the improved Mamba model optimizes the remaining life prediction by minimizing the regression loss.

[0032] Optionally, the real-time sensor data from the stainless steel pipe is received, the remaining life prediction is performed, the real-time prediction result is generated, and the improved SimCLR model and the improved Mamba model are periodically updated and optimized through real-time data feedback, specifically including:

[0033] Real-time sensor data from the stainless steel pipe is received, preprocessed, and a real-time data set is generated. The real-time data set is input into the improved SimCLR model for feature extraction to generate a feature representation. The feature representation is input into the improved Mamba model for remaining life prediction through a regression network to generate a real-time prediction result.

[0034] The real-time prediction result is compared with the actual monitoring value through a real-time data feedback mechanism to generate a feedback error, which is used to evaluate the accuracy of the real-time prediction result.

[0035] In the real-time data feedback mechanism, the improved SimCLR model and the improved Mamba model are periodically updated and optimized based on the feedback error, and the optimization process includes dynamically adjusting the contrast loss in the improved SimCLR model according to the feedback error, and updating the regression network and parameters in the improved Mamba model according to the real-time data.

[0036] Optionally, the health management and maintenance decision of the stainless steel pipe according to the real-time prediction result provides a specific life prediction or health status evaluation report, which specifically includes:

[0037] According to the real-time prediction result, a life prediction report is generated, and the life prediction report includes the remaining life of the pipeline, the health status evaluation and the potential risk, and the current prediction is compared with the historical data to evaluate the health status of the stainless steel pipe and whether there is a potential risk.

[0038] Based on the life prediction report and the health status evaluation report, whether preventive maintenance or equipment replacement is needed is determined according to the remaining life prediction value, if the predicted remaining life is lower than a set threshold, the health management plan prompts maintenance or replacement, and if the health status evaluation shows that the pipeline is in a normal working state, the maintenance is delayed.

[0039] In the real-time data feedback mechanism, the health status evaluation report is dynamically adjusted in combination with new real-time data and real-time prediction results, the remaining life and the health status are re-evaluated whenever new sensor data is input into the system, and the life prediction report is updated according to the latest evaluation results, and the health status evaluation report is adjusted.

[0040] The life prediction report or the health status evaluation report is presented in a digital form.

[0041] According to the life prediction system of the stainless steel pipe, the following modules are included:

[0042] The data acquisition and preprocessing module is used for acquiring sensor data of the stainless steel pipe and performing data preprocessing to generate a preprocessed data set.

[0043] The feature extraction and generation module is used for inputting the preprocessed data set into the improved SimCLR model, generating feature representation through data enhancement, feature extraction and contrast learning.

[0044] The remaining life prediction module is used for inputting the feature representation into the improved Mamba model to perform time series modeling and remaining life prediction.

[0045] a loss optimization module for optimizing the improved SimCLR model and the improved Mamba model;

[0046] a feedback optimization module for receiving sensor data from the stainless steel pipe in real time, periodically updating and optimizing the improved SimCLR model and the improved Mamba model through a real-time data feedback mechanism;

[0047] a health management and maintenance decision module for generating a life prediction report and a health status evaluation report according to the real-time prediction results.

[0048] The beneficial effects of the present application are:

[0049] The present application can effectively improve the accuracy of the remaining life prediction of the stainless steel pipe by combining multi-source sensor data and introducing the improved SimCLR model and the improved Mamba model. First, by preprocessing the sensor data, eliminating noise and filling missing values, the data quality is higher, thereby providing a reliable foundation for subsequent analysis. Second, the improved SimCLR model is used to generate feature representations through data augmentation, feature extraction and contrastive learning, which can deeply mine the state features of the stainless steel pipe under various working conditions and improve the prediction accuracy. Third, the improved Mamba model is combined for time series modeling and remaining life prediction, which can dynamically evaluate the health status of the pipeline based on historical data and provide more accurate prediction results. Further, through the real-time data feedback mechanism, the system can periodically optimize and update the model to ensure that the prediction results always reflect the current pipeline state, greatly improving the real-time performance and robustness of the model. Finally, the life prediction report and the health status evaluation report generated based on the real-time prediction results can provide a scientific basis for maintenance decisions, avoiding the drawbacks of relying on experience in traditional methods, helping users to timely maintain or replace the pipeline, effectively prolonging the service life of the equipment and reducing maintenance costs. In summary, the present application has significant advantages in improving prediction accuracy, real-time performance and reliability, and can be widely applied to health management and maintenance decision-making of stainless steel pipes, with high engineering application value and social benefits. BRIEF DESCRIPTION OF DRAWINGS

[0050] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0051] Figure 1 a flowchart of a life prediction method for a stainless steel pipe according to the present application;

[0052] Figure 2 a schematic diagram of an improved SimCLR model in a life prediction method for a stainless steel pipe according to the present application;

[0053] Figure 3 A framework diagram of a stainless steel pipe service life prediction system according to the present application. DETAILED DESCRIPTION

[0054] The application will now be described in further detail with reference to the drawings. These drawings are simplified schematic diagrams which show the basic structure of the application in a diagrammatic manner only, and therefore only show the components relevant to the application.

[0055] REFERENCE Figures 1-2 A stainless steel pipe service life prediction method, comprising the following steps:

[0056] Collecting sensor data of the stainless steel pipe and preprocessing to generate a preprocessed dataset;

[0057] Inputting the preprocessed dataset into an improved SimCLR model, generating feature representations through data augmentation, feature extraction and contrastive learning;

[0058] Inputting the feature representations into an improved Mamba model, performing time series modeling and residual life prediction to generate prediction results;

[0059] Minimizing the contrastive loss for the improved SimCLR model and minimizing the regression loss and classification loss for the improved Mamba model;

[0060] Real-time receiving of sensor data from the stainless steel pipe, residual life prediction, generation of real-time prediction results, and periodic updating and optimization of the improved SimCLR model and the improved Mamba model through real-time data feedback;

[0061] Health management and maintenance decision-making of the stainless steel pipe according to the real-time prediction results, providing specific life prediction or health status evaluation reports.

[0062] In this embodiment, the collecting of sensor data of the stainless steel pipe and the preprocessing to generate a preprocessed dataset specifically includes:

[0063] Collecting sensor data of the stainless steel pipe, the data including temperature data, pressure data, corrosion degree data, vibration data and image data of the surface of the stainless steel pipe, generating an original dataset;

[0064] The original data set is preprocessed to generate a preprocessed data set, the preprocessing including removing noise from temperature data, pressure data, corrosion degree data, vibration data through a filter or a smoothing algorithm to generate a denoised data set, normalizing each type of data in the denoised data set to ensure that data of different dimensions have the same weight, generating a normalized data set, performing time series interpolation on missing time series data in the normalized data set to fill in missing values in the data, generating an interpolated data set, and performing size unification and color standardization on image data to generate a standardized image data set;

[0065] The interpolated data set and the standardized image data set are aligned according to the timestamp or data correspondence, and are spliced or fused based on the corresponding features to generate a unified preprocessed data set.

[0066] In this embodiment, the preprocessed data set is input into the improved SimCLR model to generate feature representations through data augmentation, feature extraction and contrastive learning, specifically including:

[0067] The preprocessed data set is input into the improved SimCLR model, and the improved SimCLR model includes an adaptive multi-modal data augmentation module, a feature extraction module, a nonlinear projection head and a contrastive learning module;

[0068] The adaptive multi-modal data augmentation module generates multiple perturbation samples by adding noise to the data, time translation and time scale transformation to obtain an enhanced time series data set, performs spatial-frequency domain joint enhancement on image data, including Fourier transform on image data, scaling, shearing and filtering operations on images in the frequency domain, and generates an enhanced image data set in combination with geometric transformation, and combines the enhanced time series data set and the enhanced image data set to generate an enhanced data set;

[0069] The enhanced data set is input into the feature extraction module, and the feature extraction module uses a multi-scale self-attention convolutional neural network to learn features to generate intermediate feature representations, and the multi-scale self-attention convolutional neural network can effectively capture local and global information by using convolution kernels of different scales in different convolution layers, and combines a self-attention mechanism to adaptively assign attention weights, so that the network can focus on key features in the data and improve the expression ability of the improved SimCLR model for complex information;

[0070] The intermediate feature representation is input into the nonlinear projection head, and the nonlinear projection head includes a multi-layer perceptron and a trainable gating mechanism, which maps the intermediate feature representation to a low-dimensional space through a multi-layer ReLU or GELU activation function to generate a low-dimensional feature representation, and the trainable gating mechanism is used to dynamically adjust the activation intensity of each feature channel to selectively highlight important features and improve the sensitivity of the improved SimCLR model to key features.

[0071] The adaptive weighted contrastive loss adjusts the weights of positive and negative sample pairs according to the difficulty of the samples, that is, the distance or similarity of the positive and negative sample pairs in the feature space, so that the difficult-to-distinguish negative sample pairs obtain a higher weight, thereby improving the learning ability of the model for difficult samples, and the dynamic temperature adjustment mechanism adjusts the temperature parameter in the loss function through an adaptive algorithm, so that the improved SimCLR model can be flexibly adjusted at different training stages, thereby improving the optimization efficiency and effect, and generating the feature representation.

[0072] In the embodiment, the feature representation is input into the improved Mamba model to perform time series modeling and residual life prediction, and generate a prediction result, specifically including:

[0073] The feature representation is input into the improved Mamba model, and the improved Mamba model includes a time series modeling module and a life prediction module.

[0074] The time series modeling module adopts a selective state space network and a self-attention mechanism to capture the time series dependency in the feature representation and generate a time series feature representation. The selective state space network dynamically selects the most critical time step for modeling by evaluating each time step in the feature representation. The input of each time step is combined with the information of past time steps through a weighting mechanism to evaluate the contribution of the current time step to the residual life prediction as the importance of the current time step. This evaluation process enables the improved Mamba model to automatically select the most relevant time step for modeling, thereby reducing computational complexity and retaining the most influential time series data.

[0075] The self-attention mechanism is used to calculate and weight the relative importance between each time step in the time series feature representation to generate a weighted time series feature representation. The self-attention mechanism evaluates the contribution of each time step to the entire sequence by calculating the similarity between each time step and other time steps. The attention score of each time step is used to calculate the influence of the time step on the weighted time series feature representation. The calculation process includes dot product operation on each pair of time steps, score normalization, and use to generate the weight of each time step. The self-attention mechanism enables the improved Mamba model to assign higher weights to time steps that have strong dependencies with other time steps, thereby strengthening the feature expression of key time steps and ensuring that the improved Mamba model focuses on the most important time for life prediction.

[0076] The weighted time sequence feature representation is input into a life prediction module, the life prediction module processes the weighted time sequence feature representation through a regression network, generates a residual life prediction value as a prediction result output, the regression network adopts a multilayer perceptron structure, and the weighted time sequence feature representation is transmitted layer by layer, each layer includes a fully connected layer and a ReLU activation function, in this way, the regression network can capture the complex nonlinear relationship in the weighted time sequence feature representation, and finally output the residual life prediction value of the stainless steel pipe, indicating the remaining time that the pipeline can continue to use.

[0077] In the embodiment, the improved SimCLR model minimizes the contrast loss, and the improved Mamba model minimizes the regression loss, specifically including:

[0078] When the improved SimCLR model is trained, the contrast loss is minimized, the contrast loss is calculated by calculating the similarity between the positive sample pair and the negative sample pair, so that the distance of the positive sample pair in the feature space is as small as possible, and the distance of the negative sample pair is as large as possible, through this process, the improved SimCLR model can optimize the learned feature representation, thereby improving the distinguishing ability of the model to different samples and improving the expression effect of the features;

[0079] When the improved Mamba model is trained, the regression loss is minimized, the regression loss is calculated by calculating the difference value between the residual life prediction value predicted by the improved Mamba model and the actual residual life value, and the difference value is minimized, by minimizing the regression loss, the improved Mamba model can improve the accuracy of the residual life prediction, so that the improved Mamba model can accurately predict the remaining use time of the stainless steel pipe, thereby providing an effective basis for maintenance decision;

[0080] The regression loss and the contrast loss are simultaneously minimized by the joint loss function to optimize the overall performance of the improved SimCLR model and the improved Mamba model, the joint loss function performs weighted summation on the regression loss and the contrast loss, so that the improved SimCLR model optimizes the contrast loss when performing feature learning, and the improved Mamba model optimizes the residual life prediction by minimizing the regression loss, through this joint optimization strategy, the feature learning and the residual life prediction task can be processed at the same time, thereby achieving effective balance between the two tasks and improving the overall prediction ability.

[0081] In the embodiment, the sensor data from the stainless steel pipe is received in real time, the residual life prediction is performed, the real-time prediction result is generated, and the improved SimCLR model and the improved Mamba model are periodically updated and optimized through real-time data feedback, specifically including:

[0082] Real-time sensor data from the stainless steel pipe is received, pre-processed, and used to generate a real-time dataset, which is then input into the improved SimCLR model for feature extraction and generation of feature representations. These feature representations are input into the improved Mamba model, which uses a regression network to predict the remaining useful life and generate real-time prediction results.

[0083] The real-time prediction results are compared with actual monitoring values through a real-time data feedback mechanism, which generates feedback errors used to evaluate the accuracy of the real-time prediction results. This mechanism ensures that the improved SimCLR model and the improved Mamba model can adjust in real-time and maintain high prediction accuracy as data continues to flow in and the state of the equipment changes.

[0084] In the real-time data feedback mechanism, the improved SimCLR model and the improved Mamba model are periodically updated and optimized based on feedback errors. The optimization process includes dynamically adjusting the contrast loss in the improved SimCLR model to improve the accuracy of feature representations, and updating the regression network and parameters in the improved Mamba model based on real-time data to more accurately predict the remaining useful life. In this way, the model can adjust and optimize itself after receiving new sensor data to adapt to changes in data.

[0085] In this embodiment, the health management and maintenance decision-making of the stainless steel pipe based on the real-time prediction results provide specific life prediction or health status assessment reports, which include:

[0086] Based on the real-time prediction results, a life prediction report is generated, which includes the remaining useful life of the pipe, health status assessment, and potential risks. By comparing the current prediction with historical data, the health status of the stainless steel pipe and potential risks are assessed.

[0087] Based on the life prediction report and health status assessment report, it is determined whether preventive maintenance or equipment replacement is needed based on the predicted remaining useful life. If the predicted remaining useful life is below a certain threshold, the health management plan suggests maintenance or replacement. If the health status assessment shows that the pipe is in normal working condition, maintenance is delayed. This health management plan can serve as a basis for daily maintenance and emergency repair decisions.

[0088] In the real-time data feedback mechanism, the health status assessment report is dynamically adjusted based on new real-time data and real-time prediction results. Whenever new sensor data is input into the system, the remaining useful life and health status are re-evaluated, and the life prediction report is updated based on the latest assessment results.

[0089] The life prediction report or health status evaluation report is presented in a digital form, which can be updated in real time and adjusted according to the changes of the sensor data, ensuring that the maintenance team always makes decisions based on the latest pipeline status, thereby improving the management and maintenance efficiency of the equipment and prolonging the service life of the stainless steel pipe.

[0090] Reference Figure 3 A service life prediction system of a stainless steel pipe, comprising the following modules:

[0091] A data acquisition and preprocessing module for acquiring sensor data of the stainless steel pipe and preprocessing the data to generate a preprocessed data set;

[0092] A feature extraction and generation module for inputting the preprocessed data set into an improved SimCLR model to generate feature representations through data augmentation, feature extraction and contrastive learning;

[0093] A residual life prediction module for inputting the feature representations into an improved Mamba model to perform time series modeling and residual life prediction;

[0094] A loss optimization module for optimizing the improved SimCLR model and the improved Mamba model;

[0095] A feedback optimization module for receiving sensor data from the stainless steel pipe in real time, periodically updating and optimizing the improved SimCLR model and the improved Mamba model through a real-time data feedback mechanism;

[0096] A health management and maintenance decision module for generating a life prediction report and a health status evaluation report based on real-time prediction results.

[0097] Embodiment 1

[0098] In order to verify the feasibility of the application in implementation, the application is applied to the pipeline health management system of a certain petrochemical enterprise, specifically to the residual service life prediction and maintenance decision of the stainless steel pipe in the enterprise. A large number of stainless steel pipes are used in the production process of the enterprise, and these pipes are subjected to the long-term action of high temperature, high pressure and corrosive substances, so it is very important to accurately predict their service life and health status. The traditional pipeline management method relies on periodic inspection and manual judgment, but this method often relies on experience and has great subjectivity, making it difficult to ensure accurate health assessment and timely maintenance of the pipeline.

[0099] In the pipeline health management system of the enterprise, various sensors are installed to monitor the working condition of the pipeline in real time, including temperature sensors, pressure sensors, vibration sensors, corrosion sensors and cameras for monitoring the corrosion condition of the pipeline surface. Through these sensors, the system will collect a large amount of real-time data every day, which will be used as the basis for the remaining life prediction and health status evaluation of the stainless steel pipeline. In order to verify the effectiveness of the method, a number of stainless steel pipelines that have been running for more than 5 years are selected for data collection and analysis.

[0100] These data mainly include temperature data (unit: ℃), pressure data (unit: MPa), vibration data (unit: m / s²), corrosion degree data (unit: mm) and surface image data. The system collects data from these sensors every hour, and the entire experimental period is 30 days. Table 1 below is a part of the collected data samples:

[0101] Table 1 Stainless steel pipeline sensor data collection sample table

[0102] Date Temperature (°C) Pressure (MPa) Vibration (m / s²) Corrosion degree (mm) Surface image data (encoded) 2025 / 9 / 1 95.2 8.5 0.12 0.15 Image_001 2025 / 9 / 2 96.1 8.7 0.13 0.16 Image_002 2025 / 9 / 3 94.8 8.4 0.14 0.18 Image_003 2025 / 9 / 4 97 8.9 0.16 0.2 Image_004 2025 / 9 / 5 95.5 8.6 0.15 0.22 Image_005

[0103] In order to ensure the quality of the data, preprocessing is carried out after data collection, including denoising, standardization and time series interpolation. Temperature, pressure, vibration and corrosion degree data are filtered and denoised to eliminate abnormal data caused by sensor noise or external interference. For image data, size unification and color standardization are performed to ensure that image data collected at different times have consistent formats, facilitating subsequent processing.

[0104] Subsequently, these preprocessed data are input into the improved SimCLR model to generate feature representations through data augmentation, feature extraction and contrastive learning. Data augmentation mainly includes time translation, time scale transformation and spatial-frequency domain joint enhancement after Fourier transform of image data. Through these processes, an enhanced data set containing rich features is generated, and feature representations representing the health status of the pipeline are extracted through the adaptive multi-modal data augmentation module and the feature extraction module.

[0105] After the generation of feature representations, these features are input into the improved Mamba model for time series modeling and remaining life prediction. The improved Mamba model uses a selective state space network and a self-attention mechanism to evaluate the importance of each time step through a weighting mechanism and combine past time step information to generate weighted time series feature representations. Finally, the remaining useful life of the pipeline is predicted through a regression network.

[0106] For example, on September 5, 2025, the data collected was processed by the model to predict the remaining life of the stainless steel pipe as 2.3 years, with an error within the 5% error range, while the actual life prediction value of the pipeline was 2.4 years. This accuracy verifies the efficiency and accuracy of the method of the present application.

[0107] Since the working environment and state of the pipeline may change in actual production, real-time data feedback and model updating are crucial. The system of the present application can receive sensor data from the stainless steel pipe in real time and periodically update and optimize the improved SimCLR and improved Mamba model according to the real-time data feedback mechanism. For example, when the pressure of a certain pipeline is monitored to exceed the normal working range, the system will automatically adjust the model weight and update the feature representation to improve the accuracy of the remaining life prediction.

[0108] According to the real-time prediction results, the system generates a pipeline life prediction report and a health status evaluation report. The report content includes the remaining life of the pipeline, health status evaluation, potential risks, and whether preventive maintenance or equipment replacement is needed. Assuming that in the report on September 7, 2025, the system predicts that a certain pipeline has a remaining life of 1.5 years and the corrosion degree is rising, showing a high potential risk. According to this result, the system recommends preventive maintenance to avoid future equipment failure or safety accidents.

[0109] Table 2. Stainless steel pipe remaining life prediction report and health status evaluation table

[0110] Date Remaining life (years) Health status assessment Potential risks Recommended maintenance measures 2025 / 9 / 7 1.5 Healthy High corrosion risk Preventive maintenance 2025 / 9 / 8 1.3 Healthy Medium corrosion risk Deferred maintenance, monitor 2025 / 9 / 9 1 Minor damage High corrosion risk Replace immediately

[0111] As can be seen from Table 2, over time, the remaining life of the pipeline gradually decreases, and the accuracy of the health status evaluation also continuously improves. Through real-time monitoring and prediction, the system can timely discover potential risks and take corresponding maintenance measures to avoid production downtime and safety accidents caused by pipeline aging or damage. In addition, the introduction of the data feedback mechanism enables the model to be dynamically optimized and adjusted in actual application, improving the real-time and accuracy of the prediction results.

[0112] By applying the present application to the service life prediction system of the stainless steel pipe, the remaining life of the pipeline can be accurately predicted and dynamically optimized, significantly improving the efficiency and accuracy of pipeline health management. Compared with traditional periodic detection and manual judgment methods, the method of the present application not only reduces labor costs, but also improves the timeliness of pipeline maintenance and replacement, avoiding safety risks and production downtime caused by pipeline failure. At the same time, the real-time data feedback mechanism and the optimization and updating process make the prediction results more in line with the actual working conditions, providing scientific and accurate decision-making basis for the production management of enterprises.

[0113] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, makes equivalent replacement or change within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method of predicting the service life of a stainless steel pipe, characterized by, The method comprises the following steps: Collecting sensor data of the stainless steel pipe and preprocessing to generate a preprocessed dataset; Inputting the preprocessed dataset into an improved SimCLR model to generate feature representations through data augmentation, feature extraction, and contrastive learning; Inputting the feature representations into an improved Mamba model to perform time series modeling and residual life prediction, and generating prediction results; Minimizing the contrastive loss of the improved SimCLR model and minimizing the regression loss and classification loss of the improved Mamba model; Real-time receiving of sensor data from the stainless steel pipe, residual life prediction, real-time prediction results, and periodic updating and optimization of the improved SimCLR model and the improved Mamba model through real-time data feedback; Health management and maintenance decision-making of the stainless steel pipe according to the real-time prediction results, and providing specific life prediction or health status evaluation report; The improved SimCLR model comprises an adaptive multi-modal data augmentation module, a feature extraction module, a nonlinear projection head, and a contrastive learning module; The adaptive multi-modal data augmentation module generates multiple perturbation samples by adding noise, time translation, and time scale transformation to the data, obtains an enhanced time series dataset, performs spatial-frequency domain joint enhancement on image data, including Fourier transform on image data, scaling, shearing, and filtering operations on images in the frequency domain, and generates an enhanced image dataset by combining geometric transformation, and combines the enhanced time series dataset and the enhanced image dataset to generate an enhanced dataset; The feature extraction module uses a multi-scale self-attention convolutional neural network to learn features and generate intermediate feature representations, and the multi-scale self-attention convolutional neural network uses convolution kernels of different scales in different convolution layers and combines a self-attention mechanism to adaptively assign attention weights; The nonlinear projection head comprises a multilayer perceptron and a trainable gating mechanism, maps the intermediate feature representations to a low-dimensional space through multilayer ReLU or GELU activation functions, generates low-dimensional feature representations, and the trainable gating mechanism is used to dynamically adjust the activation intensity of each feature channel; An adaptive weighted contrastive loss and a dynamic temperature adjustment mechanism are used to optimize the feature representations, the adaptive weighted contrastive loss dynamically adjusts the weights of positive and negative sample pairs according to the difficulty of the samples, the difficulty of the samples is the distance or similarity of the positive and negative sample pairs in the feature space, and the dynamic temperature adjustment mechanism adjusts the temperature parameter in the loss function through an adaptive algorithm to generate the feature representations; The improved Mamba model comprises a time series modeling module and a life prediction module; The time series modeling module uses a selective state space network and a self-attention mechanism to generate time series feature representations, the selective state space network dynamically selects the most critical time steps for modeling by evaluating each time step in the feature representations, and the input of each time step is evaluated by a weighting mechanism to evaluate the contribution of the current time step to the residual life prediction, and the current time step importance is used as the input of the next time step. The self-attention mechanism is used to calculate and weight the relative importance between each time step in the time sequence feature representation, generating a weighted time sequence feature representation. The self-attention mechanism calculates the similarity between each time step and other time steps to evaluate the contribution of each time step to the entire sequence. The attention score of each time step is used to calculate the influence of the time step on the weighted time sequence feature representation. The calculation process includes dot product operation on each pair of time steps, score normalization, and weight generation for each time step. The weighted time sequence feature representation is input into the life prediction module, which processes the weighted time sequence feature representation through a regression network to generate a remaining life prediction value as a prediction result output. The regression network uses a multi-layer perceptron structure to pass the weighted time sequence feature representation layer by layer, with each layer including a fully connected layer and a ReLU activation function.

2. The method of predicting the service life of a stainless steel pipe according to claim 1, characterized by, The sensor data of the stainless steel pipe is collected and preprocessed to generate a preprocessed data set, which specifically includes: Collecting sensor data of the stainless steel pipe, including temperature data, pressure data, corrosion degree data, vibration data, and image data of the surface of the stainless steel pipe, to generate an original data set; Preprocessing the original data set to generate a preprocessed data set, which includes removing noise from the temperature data, pressure data, corrosion degree data, and vibration data through filters or smoothing algorithms to generate a denoised data set, normalizing each type of data in the denoised data set to generate a normalized data set, performing time series interpolation on missing time series data in the normalized data set to fill in missing values in the data, generating an interpolated data set, and performing size unification and color standardization on the image data to generate a standardized image data set; Aligning the interpolated data set and the standardized image data set by timestamp or data correspondence, and concatenating or fusing based on the corresponding features to generate a unified preprocessed data set.

3. The method of predicting the service life of a stainless steel pipe according to claim 1, characterized by, The improved SimCLR model minimizes the contrast loss, and the improved Mamba model minimizes the regression loss, specifically including: When training the improved SimCLR model, the contrast loss is minimized by calculating the similarity between positive sample pairs and negative sample pairs, so that the distance between positive sample pairs in the feature space is as small as possible, and the distance between negative sample pairs is as large as possible; When training the improved Mamba model, the regression loss is minimized by calculating the difference between the remaining life prediction value predicted by the improved Mamba model and the actual remaining life value, minimizing the difference; The joint loss function simultaneously minimizes the regression loss and the contrast loss by weighting and summing the regression loss and the contrast loss, so that the improved SimCLR model optimizes the contrast loss when learning features, and the improved Mamba model optimizes the remaining life prediction by minimizing the regression loss.

4. The method of predicting the service life of a stainless steel pipe according to Claim 1, wherein The real-time receiving of sensor data from the stainless steel pipe, the generation of real-time prediction results, and the periodic updating and optimization of the improved SimCLR model and the improved Mamba model through real-time data feedback, specifically include: Real-time sensor data from the stainless steel pipe is received, preprocessed, and used to generate a real-time dataset, which is input into an improved SimCLR model for feature extraction and generation of feature representations. The real-time prediction results are compared with actual monitoring values through a real-time data feedback mechanism to generate feedback errors, which are used to evaluate the accuracy of the real-time prediction results. In the real-time data feedback mechanism, the improved SimCLR model and the improved Mamba model are periodically updated and optimized based on the feedback errors.

5. The method of predicting the service life of a stainless steel pipe according to Claim 1, wherein The optimization process includes dynamically adjusting the contrast loss in the improved SimCLR model according to the feedback errors and updating the regression network and parameters in the improved Mamba model based on real-time data. Based on the real-time prediction results, health management and maintenance decisions for the stainless steel pipe are made, and specific life prediction or health status assessment reports are provided, which include: Based on the real-time prediction results, a life prediction report is generated, which includes the remaining life of the pipe, health status assessment, and potential risks. Based on the life prediction report and health status assessment report, it is determined whether preventive maintenance or equipment replacement is needed based on the remaining life prediction value. If the predicted remaining life is below a certain threshold, the health management plan prompts maintenance or replacement.

6. A system for predicting the service life of a stainless steel pipe, which executes the method for predicting the service life of a stainless steel pipe according to any one of claims 1 to 5, characterized by If the health status assessment shows that the pipe is in a normal working state, maintenance is delayed. The life prediction report or health status assessment report is presented in a digital form. The system includes the following modules: A data collection and preprocessing module for collecting sensor data from the stainless steel pipe and preprocessing the data to generate a preprocessed dataset. A feature extraction and generation module for inputting the preprocessed dataset into an improved SimCLR model to generate feature representations through data augmentation, feature extraction, and contrast learning. A remaining life prediction module for inputting the feature representations into an improved Mamba model for time series modeling and remaining life prediction. A loss optimization module for optimizing the improved SimCLR model and the improved Mamba model. A feedback optimization module for receiving real-time sensor data from the stainless steel pipe and periodically updating and optimizing the improved SimCLR model and the improved Mamba model through a real-time data feedback mechanism. A health management and maintenance decision module for generating life prediction reports and health status assessment reports based on real-time prediction results.

Images