Adsorption device failure prediction method and device fusing multi-modal data

Abstract

The application provides a fusion multi-modal data adsorption device failure prediction method and device, relates to the technical field of data processing, and comprises the following steps: establishing a multi-modal database, determining life-related factors of the adsorption device; introducing a detection standard, establishing a failure principle, training a prediction model using a deep learning model, and giving a failure reminder, which solves the technical problem that conventional failure prediction of the adsorption device relies on physical monitoring, lacks comprehensive analysis of the overall state of the equipment, ignores the interactive influence between different data types, and results in insufficient accuracy and reliability of failure prediction, achieves the technical effects that life-related factors are determined by fusing multi-modal data, the environment changes are dynamically adapted, the health state of the adsorption device is monitored in real time through enhanced detection test standards and failure evaluation principles, a failure reminder is given in time when potential risks occur, and comprehensive evaluation and monitoring of the long-term service state of the adsorption device are realized.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method and apparatus for predicting the failure of an adsorption device by fusing multimodal data. Background Technology

[0002] With the continuous development of industrialization and modernization, adsorption devices are widely used in many industries, especially in chemical, petroleum and energy fields, where they undertake important gas adsorption and separation tasks. These adsorption devices are often in complex working environments, especially in hydrogen-rich (H2) atmospheres, and operate under extreme conditions such as alternating loads, temperatures and pressures for a long time. Under such conditions, the materials and structures of adsorption devices are easily affected by corrosion, fatigue and damage, which can lead to a decline in equipment performance or even failure. In order to ensure the long-term safe and reliable operation of adsorption devices, it is particularly important to provide early warning and predict their failure states.

[0003] Currently, failure prediction of adsorption devices mainly relies on traditional physical monitoring, such as ultrasonic testing and magnetic testing. These methods can usually only monitor external defects or local damage to the equipment, and are difficult to fully reflect the various failure risks of the equipment in complex environments. In addition, traditional life prediction often relies on a single data type (such as residual stress or microstructure), ignoring the special effects of adsorption devices under specific working environments, such as hydrogen-rich atmospheres and alternating loads, resulting in insufficient accuracy and reliability of the prediction.

[0004] Existing technologies for predicting the conventional failure of adsorption devices rely on physical monitoring, lack comprehensive analysis of the overall state of the equipment, and ignore the interaction between different data types, resulting in insufficient accuracy and reliability of failure prediction. Summary of the Invention

[0005] This application provides a method and apparatus for predicting the failure of adsorption devices by integrating multimodal data. It solves the technical problems of conventional failure prediction of adsorption devices relying on physical monitoring, lacking a comprehensive analysis of the overall state of the equipment, ignoring the interaction between different data types, and resulting in insufficient accuracy and reliability of failure prediction. It achieves the technical effect of integrating multimodal data to determine the life-related factors, dynamically adapting to environmental changes, monitoring the health status of the adsorption device in real time through enhanced testing standards and failure evaluation principles, and issuing timely failure warnings when potential risks occur, thus realizing a comprehensive assessment and monitoring of the long-term service status of the adsorption device.

[0006] In view of the above problems, this application provides a method and apparatus for predicting the failure of adsorption devices by integrating multimodal data.

[0007] Firstly, this application provides a method for predicting the failure of an adsorption device by integrating multimodal data. The method includes: setting up a multimodal database to store microstructure data, defect feature data, and residual stress data of the adsorption device; establishing mapping relationships between the microstructure and service performance, defect features and service performance, and residual stress and service performance of the adsorption device; using the Delphi method, optimizing the arrangement and combination of characteristic parameters of microstructure, defect features, and residual stress based on the service life weighting factor of the adsorption device in an H2-rich environment to determine the life-related factors of the adsorption device; introducing a multimodal testing standard suitable for predicting the life of the adsorber under H2-rich alternating load conditions; establishing a failure evaluation principle based on multimodal testing signals, which defines the life threshold value and life evaluation criteria of the adsorption device under H2-rich alternating load conditions; and training a failure prediction model for the adsorption device based on the multimodal database, integrating the multimodal testing standard, life-related factors, life threshold value, and life evaluation criteria, and providing failure alerts.

[0008] Secondly, this application provides a failure prediction device for adsorption devices that integrates multimodal data. The device includes: a database setting unit for setting a multimodal database, which stores microstructure data, defect feature data, and residual stress data of the adsorption device, establishing mapping relationships between the microstructure and service performance, defect features and service performance, and residual stress and service performance of the adsorption device; and an optimization calculation unit for using the Delphi method to perform optimization calculations on the arrangement and combination of characteristic parameters of microstructure, defect features, and residual stress based on the service life weighting factor of the adsorption device in an H2-rich environment, to determine the failure prediction method. The system comprises: a lifespan-related factor determination unit; a standard introduction unit, which introduces multimodal testing standards applicable to adsorber lifespan prediction under H2-rich alternating load conditions; an evaluation principle establishment unit, which establishes failure evaluation principles based on multimodal testing signals, defining lifespan thresholds and lifespan evaluation criteria for adsorbers under H2-rich alternating load conditions; and a failure alert unit, which, based on the multimodal database, utilizes a deep learning model to integrate the multimodal testing standards, lifespan-related factors, lifespan thresholds, and lifespan evaluation criteria to train a failure prediction model for the adsorber and provide failure alerts.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] By establishing a multimodal database to store adsorption device data, using the Delphi method to determine lifespan-related factors, introducing multimodal testing standards, establishing failure evaluation principles, and utilizing a deep learning model to fuse multiple factors to train a failure prediction model and issue failure alerts, this application achieves the technical effect of comprehensively considering multiple factors such as microstructure, defect characteristics, and residual stress to determine lifespan-related factors. In hydrogen-rich environments, especially under extreme conditions such as alternating loads, the failure prediction model can dynamically adapt to environmental changes. Through enhanced testing standards and failure evaluation principles, the health status of the adsorption device can be monitored in real time, and failure alerts can be issued in a timely manner when potential risks occur, thus achieving a comprehensive assessment and monitoring of the long-term service status of the adsorption device. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the failure prediction method for adsorption devices that integrates multimodal data, as described in this application.

[0012] Figure 2 This is a schematic diagram of the structure of the adsorption device failure prediction device that integrates multimodal data according to this application.

[0013] Explanation of reference numerals in the attached diagram: Database setup unit 11, Optimization calculation unit 12, Standard introduction unit 13, Evaluation principle establishment unit 14, Failure reminder unit 15. Detailed Implementation

[0014] This application provides a method and apparatus for predicting the failure of adsorption devices by integrating multimodal data. It solves the technical problems of conventional failure prediction of adsorption devices relying on physical monitoring, lacking a comprehensive analysis of the overall state of the equipment, ignoring the interaction between different data types, and resulting in insufficient accuracy and reliability of failure prediction. It achieves the technical effect of integrating multimodal data to determine the life-related factors, dynamically adapting to environmental changes, monitoring the health status of the adsorption device in real time through enhanced testing standards and failure evaluation principles, and issuing timely failure warnings when potential risks occur, thus realizing a comprehensive assessment and monitoring of the long-term service status of the adsorption device.

[0015] Example 1

[0016] like Figure 1 As shown, this application provides a method for predicting the failure of an adsorption device by fusing multimodal data, wherein the method includes:

[0017] S100: Establish a multimodal database to store microstructure data, defect feature data, and residual stress data of the adsorption device, and establish mapping relationships between the microstructure and service performance, defect features and service performance, and residual stress and service performance of the adsorption device; S200: Use the Delphi method to optimize the arrangement and combination of characteristic parameters of microstructure, defect features, and residual stress according to the service life weighting factor of the adsorption device in a rich H2 environment, and determine the life-related factors of the adsorption device; S300: Introduce a multimodal testing standard suitable for predicting the life of the adsorber under rich H2 alternating load conditions.

[0018] Specifically, a multimodal database is a database system that integrates multiple types of data. It can store and manage different types of data, such as microstructure data, defect feature data, and residual stress data. These data types reflect different aspects and characteristics of the adsorption device. The purpose of the multimodal database is to provide a comprehensive data platform to facilitate the analysis and prediction of the failure risk of the adsorption device. Microstructure data refers to the microstructure information of the adsorption device material, including but not limited to grain size, phase distribution, and micro defects. This information is crucial for understanding the mechanical properties and durability of the material. Defect feature data refers to defects that may exist in the adsorption device, such as cracks, pores, or other types of defects. These defects may affect the service performance and lifespan of the device.

[0019] Residual stress data refers to stresses generated during manufacturing and processing that exist within the material and may affect the structural integrity of the device during service. Mapping relationships refer to the correlation between different data types (microstructure, defect characteristics, residual stress) established through data analysis and the service performance of the adsorption device. This relationship helps understand which factors have a significant impact on the device's lifespan and performance. The Delphi method involves collecting expert opinions through multiple rounds of anonymous surveys and feedback to reach a consensus or predict results. In this application embodiment, it is used to optimize the calculation of characteristic parameters based on the service life weighting factor under H2-rich environments. H2-rich environments refer to environments rich in hydrogen, which may have a unique impact on the material properties and service performance of the adsorption device. The service life weighting factor refers to the relative importance of various factors affecting the lifespan of the adsorption device, which may include material properties, working environment, and operating conditions. Multimodal testing standards refer to a series of standardized testing and evaluation methods used to predict the lifespan of the adsorber under H2-rich alternating load conditions.

[0020] Establishing a multimodal database is crucial for integrating and storing various data related to the performance of the adsorption device. This step provides the foundation for subsequent data analysis and model training. Comprehensive analysis of microstructure, defect characteristics, and residual stress data allows for a more complete understanding of the actual state and potential failure risks of the adsorption device. Furthermore, establishing mapping relationships helps identify key factors affecting the device's service performance, providing a scientific basis for subsequent failure prediction. The Delphi method is used to gather expert wisdom to evaluate and optimize the service life weighting factors of the adsorption device in an H2-rich environment. This process helps determine which microstructure, defect characteristics, and residual stress parameters have the greatest impact on the adsorption device's lifespan. This method quantifies the specific impact of each factor on lifespan, thus providing accurate input parameters for the prediction model.

[0021] The introduction of multimodal testing standards applicable to H2-rich alternating load conditions is to ensure that the prediction model can adapt to specific working environments. This step aims to provide a standardized framework for evaluating and predicting the performance and lifespan of adsorption devices under actual working conditions. Through these standards, the failure behavior of adsorption devices in H2-rich environments can be simulated and predicted more accurately, thereby improving the practicality and accuracy of the prediction model.

[0022] S400: Establish failure evaluation principles based on multimodal detection signals. These principles define the lifespan threshold and lifespan evaluation criteria for the adsorption device under H2-rich alternating load conditions. S500: Based on the multimodal database, utilize a deep learning model to integrate the multimodal detection test standards, lifespan-related factors, lifespan threshold, and lifespan evaluation criteria to train a failure prediction model for the adsorption device and provide failure alerts.

[0023] Specifically, the failure evaluation principle is an evaluation system based on multimodal detection signals, used to assess the performance status and failure risk of an adsorption device under specific conditions (such as H2-rich alternating load conditions). It includes a series of standards and indicators to determine whether the device has reached or exceeded a predetermined lifespan threshold, thereby judging whether it has failed. The lifespan threshold refers to a preset threshold value. When the performance parameters of the adsorption device drop below this value, it is considered that the device has reached the limit of its expected lifespan and needs to be replaced or repaired.

[0024] Lifetime assessment criteria are used to evaluate the lifetime and performance status of adsorption devices under specific conditions, including the assessment of the device's service performance, durability, and reliability, as well as the prediction of potential failure risks. Deep learning models can learn complex patterns and relationships from large amounts of data. Deep learning models are used to analyze multimodal data to predict the failure risk of adsorption devices. Failure alerts refer to the system issuing a reminder when the performance status of the adsorption device approaches or reaches the lifetime threshold, so that corresponding maintenance or replacement measures can be taken.

[0025] Establishing a failure evaluation principle based on multimodal detection signals is one of the key steps. This step involves defining and determining the life threshold value and life evaluation criteria of the adsorption device under H2-rich alternating load conditions. Through this principle, the failure risk of the device can be quantified, and an evaluation standard can be provided for subsequent prediction models. The role of this step is to provide a clear evaluation framework for the health management of the adsorption device, making the maintenance and decision-making process more scientific and systematic.

[0026] After establishing failure evaluation principles, a failure prediction model for the adsorption device is trained using a deep learning model based on the rich data in the multimodal database. This step involves integrating key information such as multimodal testing standards, life-related factors, life threshold values, and life evaluation criteria. The deep learning model can learn patterns for predicting failure risks from this complex multimodal data. In the above steps, the health status of the adsorption device is monitored in real time, and an early warning is issued when the device approaches the failure threshold, thereby achieving a comprehensive assessment and monitoring of the long-term service status of the adsorption device and improving its safety and reliability.

[0027] Furthermore, the multimodal database is used to store microstructure data, defect feature data, and residual stress data of the adsorption device, and the method further includes:

[0028] The microstructure data, defect feature data, and residual stress data are preprocessed to obtain multimodal preprocessed data; the multimodal preprocessed data is enhanced to select the optimal split point; based on the environmental change law in the H2-rich environment and the optimal split point, a multi-type damage mode interaction mechanism model is constructed.

[0029] Specifically, preprocessing refers to processing directly acquired microstructure data, defect feature data, and residual stress data to improve data quality and make it more suitable for subsequent analysis and model training. Preprocessing steps include data cleaning, normalization, and denoising. Multimodal preprocessed data refers to preprocessed datasets from different data sources (such as microstructure data, defect feature data, and residual stress data), which are integrated together to facilitate multimodal analysis. Data augmentation is a technique that increases the size of a dataset by creating data variants, with the aim of improving the model's generalization ability and reducing overfitting. In this scenario, augmentation may involve applying specific transformations to the preprocessed data to simulate different damage modes.

[0030] The optimal split point refers to the best position in the model when building a tree, under which features or conditions the dataset is divided into subsets. Choosing the optimal split point is to maximize the performance of the model, such as classification accuracy. The multi-class damage mode interaction mechanism model aims to understand and simulate the interaction and transformation mechanism between various damage modes that may occur in the adsorption device in the H2-rich environment. This model helps to deeply understand how different damage modes affect the overall performance and lifespan of the device.

[0031] Preprocessing is a crucial step in the data preparation stage. By preprocessing microstructure data, defect feature data, and residual stress data, cleaner and more consistent multimodal preprocessed data can be obtained. This step improves data quality, reduces the impact of noise and outliers, and makes the data more suitable for subsequent analysis and model training. Preprocessed data can more accurately reflect the actual state of the adsorption device and provide reliable input for failure prediction.

[0032] The multimodal preprocessed data is augmented, and the optimal split point is selected. Furthermore, in order to simulate the different damage modes that the adsorption device may experience in an H2-rich environment, this step increases the diversity and complexity of the dataset, improves the model's ability to identify and predict different damage modes, and through data augmentation, the model can learn a wider range of damage features, thereby improving its generalization ability and prediction accuracy.

[0033] A multi-damage mode interaction mechanism model is constructed. Specifically, based on the environmental change law in H2-rich environments, the optimal split point is used to construct the multi-damage mode interaction mechanism model. This model further deepens the understanding and simulation of the interaction and transformation mechanism between different damage modes, accurately predicts the behavior and lifespan of the adsorption device in specific environments, and provides a scientific basis for maintenance and replacement decisions.

[0034] Furthermore, based on the optimal split point and considering the environmental change patterns in H2-rich environments, a multi-type damage mode interaction mechanism model is constructed. The method includes:

[0035] Train a KNN classifier and set a distance metric; based on the distance metric, formulate a distance distribution map; analyze the clustering characteristics under different damage patterns according to the distance distribution map, and determine multiple damage patterns.

[0036] Specifically, the KNN classifier is an instance-based machine learning algorithm used for classification and regression. It predicts the class of a sample based on the classes of its K nearest neighbors through a voting process. In the KNN algorithm, distance metrics are methods used to calculate the similarity between samples. Common distance metrics include Euclidean distance, Manhattan distance, and Minkowski distance.

[0037] Distance distribution maps are used to show the distance distribution of different samples in the feature space. Through distance distribution maps, the similarity and differences between samples can be observed intuitively. Clustering characteristics refer to the clustering of data points in the feature space. By analyzing clustering characteristics, natural grouping or categories in the data can be identified. Multiple damage modes refer to the different types of damage that the adsorption device may suffer, such as cracks, corrosion, fatigue, etc. Each damage mode has its specific characteristics and effects.

[0038] Training a KNN classifier involves selecting an appropriate distance metric and using training data to train the KNN model. The role of the KNN classifier is to classify new samples based on known damage samples and predict their damage type. By setting an appropriate distance metric, the accuracy and robustness of the classifier can be improved, making it better adaptable to the characteristics of multimodal data.

[0039] Based on the selected distance metric, a distance distribution map is constructed. This step involves calculating the distances between samples and displaying this distance information on the map. The purpose of the distance distribution map is to provide an intuitive way to observe and analyze the similarities and differences between samples, which is crucial for understanding the structure and characteristics of the data. Through the distance distribution map, we can better understand the distribution of different damage patterns in the feature space, providing a basis for subsequent clustering analysis and damage pattern recognition.

[0040] Based on the analysis of the distance distribution map, the clustering characteristics under different damage modes are analyzed to determine multiple damage modes. Specifically, the data aggregation in the feature space is identified and analyzed to discover different damage modes, identify different types of damage that the adsorption device may suffer, and establish a feature model for each damage mode. By analyzing the clustering characteristics, new damage samples can be predicted and identified more accurately, thereby improving the accuracy and practicality of the failure prediction model.

[0041] Furthermore, based on the distance distribution map, the clustering characteristics under different damage modes are analyzed to determine multiple damage modes. The method includes:

[0042] An incremental learning mechanism is introduced to allow the distance distribution map to be updated synchronously; a periodic calibration mechanism is established based on the distance distribution map.

[0043] Specifically, incremental learning is a machine learning paradigm that allows learning algorithms to learn incrementally. This means that the model can update its knowledge as it receives new data without needing to retrain from scratch. This mechanism is particularly useful for handling continuously arriving data streams. In incremental learning, synchronous updates refer to the model immediately updating its parameters or knowledge base upon receiving new data to reflect the latest data features. Periodic calibration is a periodic verification and adjustment process used to ensure that the performance of a model or system remains at the expected level. During calibration, the model may be retrained using new data or its parameters may be adjusted to adapt to changes in data distribution.

[0044] Introducing an incremental learning mechanism into a failure prediction model enables the model to adapt to data distribution and environmental conditions that change over time. It allows the model to dynamically update its parameters and knowledge base, including distance distribution maps, when it receives new multimodal data. The role of the incremental learning mechanism is to improve the model's adaptability and flexibility, enabling it to capture the latest damage patterns and trends, thereby maintaining prediction accuracy and relevance.

[0045] Establishing a periodic calibration mechanism to periodically evaluate the model's performance and adjust it based on the latest data is crucial for ensuring the model's long-term stability and accuracy. Through periodic calibration, it's possible to detect whether the model is still suitable for the current data distribution, and whether parameters need to be adjusted or retrained to adapt to new data features. This maintains the model's predictive performance and reduces potential risks caused by outdated models.

[0046] Furthermore, based on the distance distribution map, a periodic calibration mechanism is established, the method of which includes:

[0047] Obtain the calibration cycle of the periodic calibration mechanism; within the calibration cycle, collect time-stamped microstructure data, time-stamped defect feature data, and time-stamped residual stress data as calibration samples; use the calibration samples to retrain the KNN classifier and obtain the calibrated distance distribution map.

[0048] Specifically, a calibration cycle refers to the time interval set in a periodic calibration mechanism. At the end of this interval, a calibration process is performed to update and adjust the model. Timestamped data refers to each data record being accompanied by a timestamp, recording the specific time the data was generated. This is crucial for tracking data changes over time and for trend analysis. In machine learning, calibration samples refer to the dataset used for model calibration. This data is used to adjust model parameters to ensure that the model accurately reflects the latest data features. Retraining refers to performing additional training on an already trained model using new data or calibration samples to update model parameters and improve the model's accuracy and adaptability.

[0049] Establishing a regular calibration cycle is crucial to ensure the model can continuously adapt to new data. Specifically, a suitable time interval is determined, at the end of which the model is calibrated. The calibration cycle setting needs to consider the rate of data change and business requirements to ensure the model can be updated in a timely manner while avoiding resource waste caused by excessively frequent calibrations. Determining the calibration cycle helps maintain the long-term performance and accuracy of the model.

[0050] During each calibration cycle, timestamped microstructure data, defect feature data, and residual stress data are collected. These data serve as calibration samples for updating and adjusting the model. Collecting these data captures the state changes of the adsorption device at different points in time, ensuring that the model can reflect the latest equipment status and environmental conditions. Using timestamped data helps analyze the trends and patterns of the data over time, which is crucial for predicting the accuracy of the model.

[0051] Retraining the KNN classifier using collected calibration samples is the core step in model calibration. Specifically, the model parameters are updated using the latest calibration samples to ensure that the model can accurately reflect the latest data features and trends. The purpose of retraining is to improve the model's adaptability and accuracy, enabling it to better predict the failure risk of the adsorption device. Through retraining, the model can learn new damage patterns and trends, thereby improving the accuracy and reliability of failure prediction. After retraining, a calibrated distance distribution map can be obtained, which will provide a more accurate basis for subsequent analysis and decision-making.

[0052] Furthermore, the method further includes retraining the KNN classifier using the calibration samples.

[0053] Set performance evaluation metrics, including accuracy, recall, and F1 score; evaluate the performance of the distance distribution maps before and after calibration using these metrics, and verify the calibration by comparing the changes in performance evaluation before and after calibration.

[0054] Specifically, performance evaluation metrics are standards used to measure model performance, including accuracy, recall, and F1 score, which quantify the quality of the model's prediction results. Accuracy refers to the proportion of samples correctly predicted by the model out of the total samples, reflecting the accuracy of the model's predictions. Recall, also known as true positive rate or sensitivity, refers to the proportion of positive samples successfully identified by the model out of all actual positive samples, reflecting the model's ability to identify positive samples. The F1 score is the harmonic mean of accuracy and recall, used to measure the overall performance of the model, especially in cases of class imbalance. Calibration validation refers to the process of verifying the effectiveness of calibration by comparing the performance changes of the model before and after calibration. If the model's performance improves after calibration, it indicates that the calibration is effective.

[0055] Setting performance evaluation metrics is an important step in measuring and monitoring model performance during model calibration. By setting metrics such as accuracy, recall, and F1 score, the predictive performance of the model can be quantified and optimized accordingly. The role of these metrics is to provide an objective standard to evaluate the predictive ability of the model, understand the model's performance in identifying different damage modes, and make necessary adjustments.

[0056] Before and after calibration, evaluating the performance of the distance distribution map using the set performance evaluation indicators is a key step in verifying the calibration effect. Specifically, the model prediction results before calibration are compared with the actual results, and then this process is repeated to evaluate the model after calibration. The purpose of the evaluation is to determine whether the calibration has improved the model's prediction accuracy and reliability. By comparing the performance indicators before and after calibration, we can understand the specific impact of calibration on model performance, thereby determining whether the calibration was successful.

[0057] Analyze the changes in performance evaluation indicators to determine whether the calibration is effective. If the calibrated model shows improvement in indicators such as accuracy, recall, and F1 score, it indicates that the calibration is effective and the model's predictive performance has been improved. This ensures that the model can continuously adapt to new data and environmental conditions and maintain its ability to predict failure risks. Through calibration verification, the long-term performance and reliability of the model are guaranteed, ensuring that the failure prediction of the adsorption device is more accurate and timely.

[0058] Furthermore, the method also includes:

[0059] Based on the integrated sensing and monitoring module, a data transmission network is determined. The integrated sensing and monitoring module is used to collect microstructure data, defect feature data, and residual stress data of the adsorption device. An encrypted communication protocol is used to protect the data transmission network for data security.

[0060] Specifically, the integrated sensing and monitoring module is used to collect key data from the adsorption device in real time, such as microstructure data, defect feature data, and residual stress data; the data transmission network refers to the network that transmits the collected data from the monitoring module to the data processing center or storage system. This network can be wireless or wired and is responsible for data transmission and communication; the encrypted communication protocol is used to protect the data from unauthorized access or tampering during data transmission. The encrypted communication protocol encrypts the data through encryption algorithms to ensure the security and integrity of data transmission.

[0061] An integrated sensing and monitoring module is deployed to collect microstructure data, defect characteristic data, and residual stress data of the adsorption device. A data transmission network is established to ensure the secure and efficient transmission of this critical data to the data processing center. The data transmission network provides a reliable data flow channel, enabling timely processing and analysis of the collected data, thereby achieving real-time monitoring of the adsorption device's status. An encrypted communication protocol is employed to protect the data transmission network, preventing data interception or tampering during transmission. This protocol ensures that only authorized users or systems can access and interpret the transmitted data. This data security protection enhances system security and prevents the leakage of sensitive data.

[0062] In summary, the failure prediction method and apparatus for adsorption devices that integrate multimodal data provided in this application have the following technical advantages:

[0063] By employing a multimodal database to store microstructure data, defect characteristic data, and residual stress data of the adsorption device, a mapping relationship is established between the adsorption device's microstructure and service performance, defect characteristics and service performance, and residual stress and service performance. Using the Delphi method, based on the service life weighting factor of the adsorption device in a H2-rich environment, the characteristic parameters of microstructure, defect characteristics, and residual stress are optimized through arrangement and combination calculations to determine the life-related factors of the adsorption device. A multimodal testing standard suitable for adsorber life prediction under H2-rich alternating load conditions is introduced; and a failure evaluation based on multimodal testing signals is established. The principles and failure evaluation principles are used to define the life threshold and life evaluation criteria of the adsorption device under H2-rich alternating load conditions. Based on a multimodal database, a deep learning model is used to integrate multimodal testing standards, life-related factors, life thresholds, and life evaluation criteria to train a failure prediction model for the adsorption device and issue failure alerts. This application achieves the technical effect of integrating multimodal data to determine life-related factors, dynamically adapting to environmental changes, and monitoring the health status of the adsorption device in real time through enhanced testing standards and failure evaluation principles, and issuing timely failure alerts when potential risks occur, thus realizing a comprehensive assessment and monitoring of the long-term service status of the adsorption device.

[0064] Example 2

[0065] Based on the same inventive concept as the adsorption device failure prediction method that integrates multimodal data in the foregoing embodiments, such as Figure 2 As shown, this application provides a failure prediction device for adsorption devices that integrates multimodal data, wherein the device includes:

[0066] Database setting unit 11 is used to set up a multimodal database. The multimodal database is used to store microstructure data, defect feature data, and residual stress data of the adsorption device, and to establish the mapping relationship between the microstructure and service performance of the adsorption device, the defect feature and service performance, and the residual stress and service performance.

[0067] The optimization calculation unit 12 is used to perform optimization calculations on the characteristic parameters of microstructure, defect features, and residual stress by arranging and combining them according to the service life weight factor of the adsorption device in a rich H2 environment using the Delphi method, so as to determine the life-related factors of the adsorption device.

[0068] Standard introduction unit 13 is used to introduce a multimodal testing standard applicable to the prediction of adsorber life under H2-rich alternating load conditions.

[0069] Evaluation principle establishment unit 14 is used to establish failure evaluation principles based on multimodal detection signals. The failure evaluation principles are used to define the life threshold value and life evaluation criteria of the adsorption device under H2-rich alternating load conditions.

[0070] The failure reminder unit 15 is used to train the failure prediction model of the adsorption device based on the multimodal database, using a deep learning model, and integrating the multimodal detection test standards, lifespan-related factors, lifespan threshold values, and lifespan evaluation criteria, and to provide failure reminders.

[0071] Furthermore, the database setting unit 11 is also used to perform the following method:

[0072] The microstructure data, defect feature data, and residual stress data are preprocessed to obtain multimodal preprocessed data; the multimodal preprocessed data is enhanced to select the optimal split point; based on the environmental change law in the H2-rich environment and the optimal split point, a multi-type damage mode interaction mechanism model is constructed.

[0073] Furthermore, the database setting unit 11 is also used to perform the following method:

[0074] Train a KNN classifier and set a distance metric; based on the distance metric, formulate a distance distribution map; analyze the clustering characteristics under different damage patterns according to the distance distribution map, and determine multiple damage patterns.

[0075] Furthermore, the database setting unit 11 is also used to perform the following method:

[0076] An incremental learning mechanism is introduced to allow the distance distribution map to be updated synchronously; a periodic calibration mechanism is established based on the distance distribution map.

[0077] Furthermore, the database setting unit 11 is also used to perform the following method:

[0078] Obtain the calibration cycle of the periodic calibration mechanism; within the calibration cycle, collect time-stamped microstructure data, time-stamped defect feature data, and time-stamped residual stress data as calibration samples; use the calibration samples to retrain the KNN classifier and obtain the calibrated distance distribution map.

[0079] Furthermore, the database setting unit 11 is also used to perform the following method:

[0080] Set performance evaluation metrics, including accuracy, recall, and F1 score; evaluate the performance of the distance distribution maps before and after calibration using these metrics, and verify the calibration by comparing the changes in performance evaluation before and after calibration.

[0081] Furthermore, the adsorption device failure prediction device that integrates multimodal data is also used to perform the following method:

[0082] Based on the integrated sensing and monitoring module, a data transmission network is determined. The integrated sensing and monitoring module is used to collect microstructure data, defect feature data, and residual stress data of the adsorption device. An encrypted communication protocol is used to protect the data transmission network for data security.

[0083] This specification and accompanying drawings are merely illustrative examples of this application. Various modifications and combinations may be made to them without departing from the spirit and scope of this application. Such modifications and variations fall within the scope of the claims of this application and their equivalents, and this application intends to include such modifications and variations.

Claims

1. A method of adsorber failure prediction fusing multi-modal data, characterized in that, The method comprises: Setting up a multi-modal database for storing microstructure data, defect feature data, and residual stress data of the adsorption device, and establishing mapping relationships between microstructure and service performance, defect feature and service performance, and residual stress and service performance of the adsorption device; Using the Delphi method, according to the service life weight factor of the adsorption device in the H2-rich environment, performing optimization calculation on the arrangement and combination of characteristic parameters of the microstructure, defect feature, and residual stress, and determining the life-related factors of the adsorption device; Introducing a multi-modal detection test standard suitable for life prediction of the adsorber under the H2-rich alternating load condition; Establishing a failure evaluation principle based on the multi-modal detection signal, which is used to define the life threshold and life evaluation criterion of the adsorption device under the H2-rich alternating load condition; Based on the multi-modal database, using a deep learning model, integrating the multi-modal detection test standard, life-related factors, life threshold, and life evaluation criterion, training a failure prediction model of the adsorption device, and providing failure warning.

2. The method of claim 1, wherein, The multi-modal database is used to store microstructure data, defect feature data, and residual stress data of the adsorption device, and the method further comprises: Preprocessing the microstructure data, defect feature data, and residual stress data to obtain multi-modal preprocessed data; Enhancing the multi-modal preprocessed data and selecting an optimal split point; According to the optimal split point, a multi-class damage mode mutual construction mechanism model is constructed according to the environmental change law under the H2-rich environment.

3. The method of claim 2, wherein, According to the optimal split point, a multi-class damage mode mutual construction mechanism model is constructed according to the environmental change law under the H2-rich environment, and the method comprises: Training a KNN classifier and setting a distance measurement method; Based on the distance measurement method, a distance distribution map is drawn; According to the distance distribution map, the clustering characteristics under different damage modes are analyzed, and a multi-class damage mode is determined.

4. The method of claim 3, wherein, According to the distance distribution map, the clustering characteristics under different damage modes are analyzed, and a multi-class damage mode is determined, and the method comprises: Introducing an incremental learning mechanism to allow the distance distribution map to be updated synchronously; Based on the distance distribution map, a periodic calibration mechanism is established.

5. The method of claim 4, wherein, Based on the distance distribution map, a periodic calibration mechanism is established, and the method comprises: Obtaining a calibration period of the periodic calibration mechanism; In the calibration period, collect time-stamped microstructure data, time-stamped defect feature data, and time-stamped residual stress data as calibration samples; Using the calibration samples, the KNN classifier is retrained to obtain a calibrated distance distribution map.

6. The method of claim 5, wherein, Using the calibration samples, the KNN classifier is retrained, and the method further comprises: Setting performance evaluation indicators, including accuracy, recall rate, and F1 score; Through the performance evaluation indicators, the performance of the distance distribution map before and after calibration is evaluated, and the performance evaluation change before and after calibration is compared for calibration verification.

7. The method of claim 1, wherein, The method further comprises: Based on the integrated sensing monitoring module, a data transmission network is determined, the integrated sensing monitoring module being used to collect microstructure data, defect feature data, and residual stress data of the adsorption device; An encryption communication protocol is used to perform data security protection on the data transmission network.

8. An adsorber device failure prediction apparatus fusing multi-modal data, characterized in that, The device comprises: A database setting unit, which is used to set a multi-modal database, the multi-modal database being used to store the microstructure data, defect feature data, and residual stress data of the adsorption device, and to establish a mapping relationship between the microstructure of the adsorption device and service performance, between the defect feature and service performance, and between the residual stress and service performance; An optimization calculation unit, which is used to perform optimization calculation on the characteristic parameters of the microstructure, defect feature, and residual stress by using a Delphi method and according to a service life weight factor of the adsorption device under a H2-rich environment, to determine life-related factors of the adsorption device; A standard introduction unit, which is used to introduce multi-modal detection test standards suitable for life prediction of the adsorber under H2-rich alternating load conditions; An evaluation principle establishment unit, which is used to establish a failure evaluation principle based on multi-modal detection signals, the failure evaluation principle being used to define a life threshold value and a life evaluation criterion of the adsorption device under H2-rich alternating load conditions; A failure reminding unit, which is used to train a failure prediction model of the adsorption device by using a deep learning model, and to fuse the multi-modal detection test standards, life-related factors, life threshold value, and life evaluation criterion, based on the multi-modal database, to perform failure reminding.

Images