Safety State Prediction Method and Device for Pressure Swing Adsorption Device Combined with Digital Twin

Abstract

The present invention discloses a method and device for predicting the safety status of a pressure swing adsorption device in combination with a digital twin, and relates to the field of equipment safety status monitoring. The method comprises: collecting operation data of the pressure swing adsorption device according to target operating conditions, using inspection defects to perform three-dimensional construction, obtaining a three-dimensional damage model, and generating a dynamic damage quantification information set; evaluating the pressure swing adsorption device and establishing a multimodal non-destructive testing database; processing the three-dimensional damage model in combination with multiple quantitative evaluation criteria to obtain a digital twin module; performing classification learning and training on the dynamic damage quantification information set to construct a service life prediction model; and performing real-time evaluation of the pressure swing adsorption device and completing the status prediction of the pressure swing adsorption device based on the device operation prediction status information set. This method solves the technical problems of insufficient real-time performance and accuracy in the existing safety status prediction of pressure swing adsorption devices, and achieves the technical effect of real-time monitoring and accurate prediction of the operating status of the pressure swing adsorption device.

Description

Technical Field

[0001] The present application relates to fields related to equipment status monitoring, and in particular to a method and device for predicting the safety status of a pressure swing adsorption device combined with digital twins. Background Art

[0002] With the rapid development of the coal chemical industry, pressure swing adsorption units, as core equipment for gas separation and purification, play a vital role in industrial production. However, under complex operating conditions, pressure swing adsorption units often face unexpected damage problems such as cracking and bulging. These damages not only shorten the service life of the units, but also seriously threaten production safety. Traditional methods for monitoring the safety status of pressure swing adsorption units mainly rely on regular offline testing and manual judgment. This method is not only time-consuming and labor-intensive, but also difficult to capture the dynamic damage changes of the unit during operation. The accumulation of these damages and the failure to detect them in time may eventually lead to serious safety accidents.

[0003] In the current related technologies, the safety status prediction of pressure swing adsorption devices has technical problems such as insufficient real-time performance and accuracy. Summary of the Invention

[0004] The present application provides a method and device for predicting the safety status of a pressure swing adsorption device in combination with a digital twin. The operation of the pressure swing adsorption device is collected according to the target operating conditions to generate a dynamic damage quantification information set. Then, based on this information set, a three-dimensional construction is performed to obtain a three-dimensional damage model to intuitively display the damage situation inside the device. Next, the three-dimensional damage model is evaluated to establish a multimodal non-destructive testing database containing multiple quantitative evaluation standards. Finally, digital twin technology is used to process the three-dimensional damage model and the quantitative evaluation standards to obtain a digital twin module. Through this module, the dynamic damage quantification information set is classified and trained to construct a service life prediction model. The operating status of the pressure swing adsorption device is evaluated in real time and the intelligent safety status prediction is predicted. The technical effect of achieving real-time monitoring and accurate prediction of the operating status of the pressure swing adsorption device is achieved.

[0005] This application provides a method for predicting the safety status of a pressure swing adsorption device in combination with digital twins, including:

[0006] The operation data of the pressure swing adsorption device is collected according to the target operating conditions to generate a dynamic damage quantification information set, and a three-dimensional construction is performed based on the dynamic damage quantification information set to obtain a three-dimensional damage model; the pressure swing adsorption device is evaluated based on the three-dimensional damage model, and a multimodal non-destructive testing database is established, and the multimodal non-destructive testing database includes multiple quantitative evaluation criteria; using digital twin technology, the three-dimensional damage model is processed in combination with the multiple quantitative evaluation criteria to obtain a digital twin module; the dynamic damage quantification information set is classified and learned through the digital twin module to construct a service life prediction model; the operating status of the pressure swing adsorption device is evaluated in real time through the service life prediction model to generate a device operation prediction status information set, and the intelligent safety status prediction of the pressure swing adsorption device is completed based on the device operation prediction status information set.

[0007] In a possible implementation, the pressure swing adsorption device is operated and collected according to the target operating conditions to generate a dynamic damage quantification information set, and the following processing is performed:

[0008] The method comprises traversing multiple operating conditions of the pressure swing adsorption device for analysis to determine the target operating condition; performing multi-dimensional damage acquisition on the pressure swing adsorption device based on the target operating condition to obtain a multimodal signal; performing a comprehensive inspection of the pressure swing adsorption device based on the multimodal signal to obtain a device inspection result; locating the damage based on the device inspection result to determine multiple damage locations; and quantifying data according to the multiple damage locations to obtain the dynamic damage quantification information set.

[0009] In a possible implementation, a three-dimensional damage model is obtained by performing three-dimensional construction based on the dynamic damage quantification information set, and performing the following processing:

[0010] Based on the dynamic damage quantification information set and the multiple damage locations, simulation verification is performed to obtain a damage verification result; based on the damage verification result, the dynamic damage quantification information set is stereo matched to obtain multiple point cloud data; based on the multimodal signal, damage evolution is performed on the multiple point cloud data to determine damage impact transfer data; using deep learning, the multimodal signal is fused according to the damage impact transfer data to generate signal fusion data; according to the signal fusion data, the multiple point cloud data are three-dimensionally combined to construct the three-dimensional damage model.

[0011] In a possible implementation, the pressure swing adsorption device is evaluated based on the three-dimensional damage model, a multimodal nondestructive testing database is established, and the following processing is performed:

[0012] The pressure swing adsorption device is traversed and traced using the three-dimensional damage model to generate damage source data; the damage impact is quantified based on the damage source data, and the multiple quantitative evaluation criteria are set; damage evaluation is performed according to the multiple quantitative evaluation criteria in combination with the three-dimensional damage model to obtain multiple damage scores; cluster analysis is performed based on the damage source data to determine multiple damage types, and feature analysis is performed based on the multiple damage types in combination with the three-dimensional damage model to obtain multiple damage features; the multimodal signals are entered in descending order according to the multiple damage scores in combination with the multiple damage features to construct the multimodal nondestructive testing database.

[0013] In a possible implementation, digital twin technology is used to process the three-dimensional damage model in combination with the multiple quantitative evaluation criteria to obtain a digital twin module, and the following processing is performed:

[0014] The digital twin technology is used in combination with the three-dimensional damage model for matching to construct a digital twin model; the multimodal signal is synchronized to the digital twin model for data-driven according to the multiple quantitative evaluation standards to obtain multimodal damage change data; dynamic damage is mapped to the three-dimensional damage model based on the multimodal damage change data to obtain dynamic damage visualization data; a multi-standard evaluation is performed according to the dynamic damage visualization data according to the multiple quantitative evaluation standards to obtain a multi-standard comprehensive score, the digital twin model is evaluated according to the multi-standard comprehensive score, and the digital twin model is added to the digital twin module according to the evaluation results.

[0015] In a possible implementation, the digital twin module performs classification learning and training on the dynamic damage quantification information set, constructs a service life prediction model, and performs the following processing:

[0016] The dynamic damage quantification information set is classified based on the multiple damage characteristics to determine multiple damage labels; classification learning is performed through the digital twin module according to the multiple damage labels combined with multiple damage characteristics to construct a gradient boosting decision tree; data training is performed on the dynamic damage quantification information set based on the gradient boosting decision tree, and the training results are cross-validated to obtain the service life prediction model.

[0017] In a possible implementation, data training is performed on the dynamic damage quantification information set based on the gradient boosting decision tree, and the training results are cross-validated to obtain the service life prediction model, and the following processing is performed:

[0018] A time series analysis algorithm is used to perform trend analysis on the dynamic damage quantification information set to determine damage evolution trend data; a regularity analysis is performed based on the gradient boosting decision tree in combination with the damage evolution trend data to draw a damage evolution curve graph; a regression calculation is performed on the pressure swing adsorption device according to the damage evolution curve graph in combination with the multiple damage characteristics to generate multiple data distribution information; supervised training is performed on the dynamic damage quantification information set based on the multiple data distribution information in combination with the multiple damage labels to generate the training results; based on the training results, first training data, second training data...Nth training data are randomly extracted, where the first training data, the second training data...Nth training data are all different training data; cross-iteration validation is performed on the first training data, the second training data...Nth training data according to an error index until the training results converge, thereby constructing the service life prediction model.

[0019] This application also provides a pressure swing adsorption device safety state prediction device combined with digital twin, including:

[0020] A device operation acquisition module, which is used to collect operation data of the pressure swing adsorption device according to the target operating conditions, generate a dynamic damage quantification information set, perform three-dimensional construction based on the dynamic damage quantification information set, and obtain a three-dimensional damage model; a damage evaluation module, which is used to evaluate the pressure swing adsorption device based on the three-dimensional damage model and establish a multimodal non-destructive testing database, and the multimodal non-destructive testing database contains multiple quantitative evaluation standards; a digital twin processing module, which is used to use digital twin technology to process the three-dimensional damage model in combination with the multiple quantitative evaluation standards to obtain a digital twin module; a service life prediction model construction module, which is used to perform classification learning and training on the dynamic damage quantification information set through the digital twin module to construct a service life prediction model; a state prediction module, which is used to perform real-time evaluation of the operating status of the pressure swing adsorption device through the service life prediction model, generate a device operation prediction status information set, and complete the intelligent safety status prediction of the pressure swing adsorption device based on the device operation prediction status information set.

[0021] The method and device for predicting the safety status of a pressure swing adsorption device combined with digital twins proposed in this application first collect operation data of the pressure swing adsorption device according to the target operating conditions to generate a dynamic damage quantification information set, perform three-dimensional construction based on the dynamic damage quantification information set, and obtain a three-dimensional damage model. Then, the pressure swing adsorption device is evaluated based on the three-dimensional damage model, and a multimodal non-destructive testing database is established. The multimodal non-destructive testing database contains multiple quantitative evaluation criteria. Then, digital twin technology is used to process the three-dimensional damage model in combination with multiple quantitative evaluation criteria to obtain a digital twin module. Then, the dynamic damage quantification information set is classified and trained through the digital twin module to construct a service life prediction model. Finally, the operating status of the pressure swing adsorption device is evaluated in real time through the service life prediction model to generate a device operation prediction status information set. Based on the device operation prediction status information set, the intelligent safety status prediction of the pressure swing adsorption device is completed, thereby achieving the technical effect of realizing real-time monitoring and accurate prediction of the operating status of the pressure swing adsorption device. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the apparatus according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0023] Figure 1 A flow chart of a method for predicting the safety status of a pressure swing adsorption device combined with digital twins provided in an embodiment of the present application.

[0024] Figure 2 A schematic diagram of the structure of a pressure swing adsorption device safety state prediction device combined with digital twin provided in an embodiment of the present application.

[0025] Explanation of the accompanying drawings: device operation acquisition module 10, damage assessment module 20, digital twin processing module 30, service life prediction model construction module 40, state prediction module 50. DETAILED DESCRIPTION

[0026] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0027] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0028] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, device, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0029] The embodiment of the present application provides a method for predicting the safety status of a pressure swing adsorption device in combination with a digital twin, such as Figure 1 As shown, the method includes:

[0030] Step S100, the operation of the pressure swing adsorption device is collected according to the target operating condition, a dynamic damage quantification information set is generated, and a three-dimensional construction is performed based on the dynamic damage quantification information set to obtain a three-dimensional damage model. Specifically, the target operating condition refers to an operating condition that has a significant impact on the damage of the pressure swing adsorption device, such as an H2-rich alternating load condition. Based on the target operating condition, multi-modal signal collection is performed on the pressure swing adsorption device, including collecting laser, electromagnetic, ultrasonic and X-ray signals, etc., to capture defects inside the pressure swing adsorption device (such as cracks, bulging, wall thickness thinning, etc.). The collected multi-modal signals are used to construct a dynamic damage quantification information set, which contains quantitative damage data of each damage location of the pressure swing adsorption device during operation. Based on the dynamic damage quantification information set, a three-dimensional reconstruction technology is used to generate a three-dimensional model of the damage site. This model intuitively displays information such as the location, shape and size of the damage.

[0031] In one possible implementation, the operation of the pressure swing adsorption device is collected according to the target operating conditions to generate a dynamic damage quantification information set. Step S100 further includes step S110, traversing multiple operating conditions of the pressure swing adsorption device for analysis to determine the target operating conditions. Specifically, the operating data of the pressure swing adsorption device under different operating conditions are collected, including key parameters such as pressure, temperature, flow, and adsorption time. These data are deeply analyzed through methods such as data statistical analysis, trend analysis, and correlation analysis to identify which operating conditions have a significant impact on the damage to the device. Based on the analysis results, a target operating condition that is most critical to the damage to the device is determined. Step S120, based on the target operating conditions, multi-dimensional damage collection is performed on the pressure swing adsorption device to obtain multi-modal signals. Specifically, non-destructive testing methods (laser, electromagnetic, ultrasonic, and X-ray, etc.) are used to collect multi-modal signals of the damage of the pressure swing adsorption device under the target operating conditions to obtain multi-dimensional information on damage characteristics such as cracks, stress, and bulging. This information has different physical properties and manifestations. In step S130, a comprehensive inspection of the pressure swing adsorption device is performed based on the multimodal signal to obtain device inspection results. Damage is located based on the device inspection results to determine multiple damage locations. Specifically, the collected multimodal signals are preprocessed, such as filtering, denoising, and feature extraction, to improve signal quality. The signal processing results are used to comprehensively inspect the pressure swing adsorption device. Based on the inspection results, the damage locations in the device are determined, including different types of damage such as cracks and bulging. In step S140, data is quantified according to the multiple damage locations to obtain the dynamic damage quantification information set. Specifically, each determined damage location is quantified based on its size, shape, depth, and other characteristics to obtain a specific numerical value of the damage. The damage quantification results are organized into a data set, including information such as damage location, damage type, and damage degree, to form a dynamic damage quantification information set. This implementation method performs multi-dimensional damage acquisition and inspection based on the target operating conditions, comprehensively understands the damage status of the pressure swing adsorption device, and improves the accuracy of state prediction.

[0032] In one possible implementation, a three-dimensional construction is performed based on the dynamic damage quantification information set to obtain a three-dimensional damage model. Step S100 further includes step S150, in which simulation verification is performed based on the dynamic damage quantification information set in combination with the multiple damage locations to obtain damage verification results. Specifically, the dynamic damage quantification information set (including quantitative data such as damage type, damage degree, and damage location) and specific information of multiple damage locations are input, and the pressure swing adsorption device is modeled using simulation software (such as ANSYS, etc.). The corresponding simulation parameters are set according to the dynamic damage quantification information set and the damage location. Through the simulation run, the actual impact of the damage in the pressure swing adsorption device is simulated, including stress distribution, deformation, etc., and the damage verification results, i.e., the damage impact data and damage expansion trend obtained by simulation, are obtained. Step S160, the dynamic damage quantification information set is stereo matched according to the damage verification results to obtain multiple point cloud data. Specifically, the damage verification results and the dynamic damage quantification information set are input, and a stereo matching algorithm (such as the ICP algorithm, etc.) is used to match the damage impact data in the damage verification results with the data in the dynamic damage quantification information set, thereby generating three-dimensional point cloud data representing the shape, position and size of the damage. These point cloud data constitute a three-dimensional representation of the damage. Step S170: Damage evolution is performed on the multiple point cloud data based on the multimodal signal to determine damage impact transfer data. Specifically, multiple point cloud data and multimodal signals are input, and a damage evolution model (such as a fracture mechanics model, a fatigue damage model, etc.) is used to process the point cloud data in combination with the multimodal signal to analyze the expansion and transfer process of the damage in the pressure swing adsorption device. The damage impact transfer data is determined by calculating parameters such as the damage rate, damage area, and damage transfer path. Step S180: Deep learning is used to fuse the multimodal signals based on the damage impact transfer data to generate signal fusion data. Specifically, a deep learning algorithm is used to extract features and perform fusion processing on multimodal signals. By training the model, the mapping relationship between damage impact transmission data and multimodal signals is learned, thereby generating signal fusion data, that is, data that fuses the features of multimodal signals. Step S190, the multiple point cloud data are three-dimensionally combined according to the signal fusion data to construct the three-dimensional damage model. Specifically, based on the feature information in the signal fusion data, the multiple point cloud data are three-dimensionally combined and optimized. By adjusting parameters such as the position, shape and size of the point cloud data, the three-dimensional model can more accurately reflect the actual situation of the damage, that is, the three-dimensional damage model is a three-dimensional representation that combines the features of multimodal signals and point cloud data. This implementation method obtains richer damage information through multimodal signal fusion processing, thereby improving the accuracy and reliability of the prediction.

[0033] Step S200 evaluates the pressure swing adsorption device based on the three-dimensional damage model and establishes a multimodal nondestructive testing database, which includes multiple quantitative evaluation criteria. Specifically, the three-dimensional damage model is used to trace the source of the damage. Based on the source of the damage, the impact of the damage is quantitatively analyzed. Multiple quantitative evaluation criteria are set. Combined with the three-dimensional damage model, the quantitative evaluation criteria are used to comprehensively evaluate the damage extent, damage type, and damage location of the pressure swing adsorption device. The multimodal nondestructive testing database is established. This is a comprehensive database containing various nondestructive testing data, damage evaluation results, quantitative evaluation criteria, and other information.

[0034] In one possible implementation, the pressure swing adsorption device is evaluated based on the three-dimensional damage model, and a multimodal nondestructive testing database is established. Step S200 further includes step S210, where the three-dimensional damage model is used to traverse and trace the pressure swing adsorption device to generate damage source data. Specifically, the detailed damage information in the three-dimensional damage model is used to perform a traversal inspection of each component of the pressure swing adsorption device. By analyzing the damage's morphology, location, and size, and combining factors such as the pressure swing adsorption device's operating history and working environment, the damage formation mechanism and source are inferred to generate damage source data. Damage source data refers to information about how and where the damage originated. Step S220, based on the damage source data, the damage impact is quantified and multiple quantitative evaluation criteria are set. Specifically, based on the damage source data, the impact of the damage on the performance and safety of the pressure swing adsorption device is analyzed. These impacts are quantified and multiple quantitative evaluation criteria are set, such as damage extent, impact range, and repair difficulty. Step S230, a damage assessment is performed based on the multiple quantitative evaluation criteria combined with the three-dimensional damage model to obtain multiple damage scores. Specifically, a comprehensive evaluation is performed on each damage of the pressure swing adsorption device in combination with the three-dimensional damage model and the set quantitative evaluation criteria. Based on the evaluation results, a damage score is assigned to each damage, which reflects the severity and potential impact of the damage. Step S240: Cluster analysis is performed based on the damage source data to determine multiple damage types. Feature analysis is performed based on the multiple damage types in combination with the three-dimensional damage model to obtain multiple damage features. Specifically, cluster analysis is performed on the damage source data to group similar damages into one category and determine multiple damage types. Feature analysis is performed on each damage type in combination with the three-dimensional damage model to extract typical features of each type. Step S250: The multimodal signals are entered in descending order according to the multiple damage scores and the multiple damage features to construct the multimodal non-destructive testing database. Specifically, the multimodal signals are sorted in descending order according to the damage scores and damage features, that is, arranged according to importance or priority. The sorted multimodal signals are sequentially entered into the multimodal non-destructive testing database to construct a complete database system. This implementation method utilizes a three-dimensional damage model to traverse and trace the source, quantify damage impacts, evaluate damage, determine damage types, perform feature analysis, and enter multimodal signals in descending order. This not only ensures the accuracy and completeness of the data, but also improves the effectiveness and practicality of the database.

[0035] Step S300: Utilizing digital twin technology, the three-dimensional damage model is processed in conjunction with the multiple quantitative evaluation criteria to obtain a digital twin module. Specifically, digital twin technology is used in conjunction with the three-dimensional damage model to construct a digital twin model, multimodal signals are synchronized to the digital twin model for data-driven evaluation, and multiple quantitative evaluation criteria are used to evaluate the model to obtain the digital twin module.

[0036] In one possible implementation, digital twin technology is used to process the three-dimensional damage model in combination with the multiple quantitative evaluation criteria to obtain a digital twin module. Step S300 further includes step S310, where the digital twin technology is used to match the three-dimensional damage model to construct a digital twin model. Specifically, digital twin technology is used to match the three-dimensional damage model with the digital twin framework, that is, the pressure swing adsorption device in the physical world is associated with its counterpart in the virtual world (i.e., the digital twin model). Based on the matching results, a digital twin model is constructed that can reflect the actual operating status and damage status of the pressure swing adsorption device. Step S320, the multimodal signals are synchronized to the digital twin model according to the multiple quantitative evaluation criteria for data-driven operation, thereby obtaining multimodal damage change data. Specifically, the multimodal signals are synchronized to the digital twin model according to the quantitative evaluation criteria, and the synchronized multimodal signals are used to drive the digital twin model for simulation and analysis, including simulating the operating status and damage status of the pressure swing adsorption device under target operating conditions. Based on the simulation results, multimodal damage change data are obtained, which reflect the changes of the pressure swing adsorption device during the damage process. Step S330, based on the multimodal damage change data, dynamic damage mapping is performed to the damage three-dimensional model to obtain dynamic damage visualization data. Specifically, the multimodal damage change data is mapped to the damage three-dimensional model, including marking the location, size, shape and other information of the damage in the model. Based on the mapping results, dynamic damage visualization data are obtained, which show the damage of the pressure swing adsorption device in an intuitive manner. Step S340, a multi-standard evaluation is performed according to the multiple quantitative evaluation criteria based on the dynamic damage visualization data to obtain a multi-standard comprehensive score, the digital twin model is evaluated according to the multi-standard comprehensive score, and the digital twin model is added to the digital twin module according to the evaluation results. Specifically, the dynamic damage visualization data is used in combination with the quantitative evaluation criteria to evaluate the current health status of the pressure swing adsorption device. Based on the health status assessment results, the digital twin model is updated and adjusted as necessary. If the assessment results indicate new damage or performance changes in the PSA unit, these changes are reflected in the digital twin model to ensure consistency between the model and the physical device. The updated digital twin model will contain more accurate damage information and health status assessment results. The updated digital twin model is added to the corresponding digital twin module. By continuously updating and adding new digital twin models, the digital twin module can continuously provide accurate health status assessment capabilities. This implementation ensures that the digital twin model always remains consistent with the physical device status, providing accurate data support for intelligent safety status prediction.

[0037] In step S400, the dynamic damage quantification information set is subjected to classification learning and training by the digital twin module to construct a service life prediction model. In one possible implementation, step S400 further includes step S410, which classifies the dynamic damage quantification information set based on the multiple damage features to determine multiple damage labels. Specifically, features directly related to damage are extracted from the dynamic damage quantification information set, including damage size, shape, location, and evolution rate. Based on the extracted features, a damage label is assigned to each dynamic damage quantification information set, representing different damage types or degrees. In step S420, the digital twin module performs classification learning based on the multiple damage labels combined with multiple damage features to construct a gradient boosting decision tree. Specifically, the dynamic damage quantification information set and the corresponding damage labels are input into the gradient boosting decision tree, and training is performed based on the multiple damage features and damage labels, so that the model can learn the mapping relationship between features and labels. During the training process, the model parameters and structure are continuously adjusted to improve the classification accuracy and generalization ability of the model. Step S430, data training is performed on the dynamic damage quantification information set based on the gradient boosting decision tree, and the training results are cross-validated to obtain the service life prediction model. Specifically, the dynamic damage quantification information set is trained using the gradient boosting decision tree. During the training process, the model will continuously and iteratively optimize its parameters to improve the classification accuracy. The trained model is evaluated using a cross-validation method to check its generalization ability. Cross-validation involves dividing the data into multiple subsets, and then using one part as a training set and the other part as a test set for verification in turn. Based on the results of cross-validation, the optimal model parameters and structure are selected, and the trained model is output as a service life prediction model for subsequent real-time evaluation. Among them, the service life prediction model is a model that predicts the service life of a pressure swing adsorption device based on the input dynamic damage quantification information set and damage characteristics.

[0038] In one possible implementation, data training is performed on the dynamic damage quantification information set based on the gradient boosting decision tree, and the training results are cross-validated to obtain the service life prediction model. Step S430 further includes step S431, where a time series analysis algorithm is used to perform trend analysis on the dynamic damage quantification information set to determine damage evolution trend data. Specifically, a time series analysis algorithm, such as ARIMA (autoregressive integrated moving average model), is used to predict the dynamic damage quantification information set to obtain damage evolution trend data. The prediction results are analyzed to determine the development trend and possible range of damage variation. Step S432, where a regularity analysis is performed based on the gradient boosting decision tree combined with the damage evolution trend data to plot a damage evolution curve. Specifically, the damage evolution trend data is fused with the input features of the gradient boosting decision tree to form a new training dataset. The fused dataset is used to train the gradient boosting decision tree so that it can learn the mapping relationship between damage evolution and features. By analyzing the decision paths and node splitting of the gradient boosting decision tree, the regularity of damage evolution is extracted, and a damage evolution curve is plotted to intuitively display the trends and regularities of damage variation over time. Step S433 performs a regression calculation on the pressure swing adsorption device based on the damage evolution curve and the multiple damage features, generating multiple data distribution information. Specifically, features with a strong correlation with the damage evolution trend are selected from the multiple damage features as inputs for the regression calculation. A regression model is constructed based on the input features and the damage evolution trend data. The regression model is used to predict damage to the pressure swing adsorption device, generating multiple data distribution information. The data distribution information reflects the distribution and changing trends of damage under different conditions.

[0039] Step S434: Supervised training is performed on the dynamic damage quantification information set based on the multiple data distribution information in combination with the multiple damage labels to generate the training results. Specifically, a label is assigned to each dynamic damage quantification information set based on the multiple damage labels and data distribution information. The labeled dynamic damage quantification information set is used as a training set to train the machine learning model. The machine learning model is trained using the training set so that it can learn the mapping relationship between features and labels. Step S435: Based on the training results, first training data, second training data, ... Nth training data are randomly extracted, where the first training data, the second training data, ... Nth training data are all different training data. Specifically, the training set is divided into multiple subsets, each subset containing a certain amount of training data. A portion of data is randomly extracted from each subset as a validation set or test set to evaluate the performance of the model. Ensure that the extracted data is different to avoid data duplication and overfitting. Step S436: Cross-iterative validation is performed on the first training data, the second training data, ... Nth training data according to the error index until the training results converge, thereby constructing the service life prediction model. Specifically, the mean squared error (MSE) is used to calculate the model's prediction error for each training data point. The training data is divided into multiple folds (e.g., 5-fold, 10-fold, etc.), with a different fold used each time as the validation set and the remaining folds used for training. Each fold is trained and validated, and the prediction error for each fold is recorded. Based on the cross-validation results, the model's parameters and structure are adjusted to reduce the prediction error. Cross-validation and iterative optimization are repeated until the prediction error converges. The service life prediction model is constructed using the final optimized model parameters and structure.

[0040] Step S500: The operating status of the pressure swing adsorption device is evaluated in real time using the service life prediction model to generate a device operation prediction status information set. Based on the device operation prediction status information set, an intelligent safety status prediction of the pressure swing adsorption device is completed. Specifically, data is acquired from the pressure swing adsorption device in real time and input into the constructed service life prediction model. The service life prediction model performs calculations based on the input data, evaluates the current operating status of the pressure swing adsorption device, and predicts its future service life. After the model evaluation, a series of predictive information about the operating status of the pressure swing adsorption device is generated, including possible damage locations, damage levels, remaining service life, etc. This information is integrated into the device operation prediction status information set. Based on the device operation prediction status information set, an intelligent safety status prediction is performed on the pressure swing adsorption device to determine whether it is in a normal, abnormal, or dangerous state, and corresponding maintenance recommendations or warning information are given to ensure the safety of production operations and prevent possible failures or accidents. The embodiment of the present application collects information on the operation of the pressure swing adsorption device according to the target operating conditions to generate a dynamic damage quantification information set. Then, based on this information set, a three-dimensional construction is performed to obtain a three-dimensional damage model to intuitively display the damage situation inside the device. Next, the three-dimensional damage model is evaluated to establish a multimodal non-destructive testing database containing multiple quantitative evaluation criteria. Finally, digital twin technology is used to process the three-dimensional damage model and the quantitative evaluation criteria to obtain a digital twin module. Through this module, the dynamic damage quantification information set is classified and trained to construct a service life prediction model. The operating status of the pressure swing adsorption device is evaluated in real time and the intelligent safety status is predicted. These technical means achieve the technical effect of realizing real-time monitoring and accurate prediction of the operating status of the pressure swing adsorption device.

[0041] In the above, refer to Figure 1 The safety state prediction method of the pressure swing adsorption device combined with digital twin according to the embodiment of the present invention is described in detail. Figure 2 A safety state prediction device for a pressure swing adsorption device combined with digital twin according to an embodiment of the present invention is described.

[0042] The digital twin-based safety state prediction device for a pressure swing adsorption device, according to an embodiment of the present invention, addresses the technical issues of insufficient real-time and accuracy in existing pressure swing adsorption device safety state prediction, achieving the technical effect of real-time monitoring and accurate prediction of the operating state of the pressure swing adsorption device. The digital twin-based safety state prediction device for a pressure swing adsorption device comprises: a device operation acquisition module 10, a damage assessment module 20, a digital twin processing module 30, a service life prediction model construction module 40, and a state prediction module 50.

[0043] The device operation acquisition module 10 is used to collect the operation of the pressure swing adsorption device according to the target operating conditions, generate a dynamic damage quantification information set, perform three-dimensional construction based on the dynamic damage quantification information set, and obtain a three-dimensional damage model; the damage evaluation module 20 is used to evaluate the pressure swing adsorption device based on the three-dimensional damage model, and establish a multimodal non-destructive testing database, which contains multiple quantitative evaluation standards; the digital twin processing module 30 is used to use digital twin technology to process the three-dimensional damage model in combination with the multiple quantitative evaluation standards to obtain a digital twin module; the service life prediction model construction module 40 is used to perform classification learning and training on the dynamic damage quantification information set through the digital twin module to construct a service life prediction model; the state prediction module 50 is used to perform real-time evaluation of the operating state of the pressure swing adsorption device through the service life prediction model, generate a device operation prediction state information set, and complete the intelligent safety state prediction of the pressure swing adsorption device based on the device operation prediction state information set.

[0044] The specific configuration of the device operation acquisition module 10 will be described in detail below. As described above, the operation of the pressure swing adsorption device is acquired according to the target operating condition to generate a dynamic damage quantification information set. The device operation acquisition module 10 may further include: a target operating condition determination unit for traversing multiple operating conditions of the pressure swing adsorption device for analysis to determine the target operating condition; a multidimensional damage acquisition unit for performing multidimensional damage acquisition on the pressure swing adsorption device based on the target operating condition to obtain a multimodal signal; a damage location unit for performing a comprehensive inspection of the pressure swing adsorption device based on the multimodal signal to obtain a device inspection result, and performing damage location based on the device inspection result to determine multiple damage locations; and a data quantification unit for performing data quantification according to the multiple damage locations to obtain the dynamic damage quantification information set.

[0045] Among them, three-dimensional construction is performed based on the dynamic damage quantification information set to obtain a three-dimensional damage model, and the device operation acquisition module 10 may further include: a simulation verification unit for performing simulation verification based on the dynamic damage quantification information set in combination with the multiple damage positions to obtain a damage verification result; a stereo matching unit for performing stereo matching on the dynamic damage quantification information set according to the damage verification result to obtain multiple point cloud data; a damage evolution unit for performing damage evolution on the multiple point cloud data based on the multimodal signal to determine the damage impact transfer data; a multimodal signal fusion unit for using deep learning to fuse the multimodal signal according to the damage impact transfer data to generate signal fusion data; a point cloud data three-dimensional combination unit for performing three-dimensional combination on the multiple point cloud data according to the signal fusion data to construct the three-dimensional damage model.

[0046] The specific configuration of the damage assessment module 20 will be described in detail below. As described above, based on the three-dimensional damage model, the pressure swing adsorption device is evaluated and a multimodal non-destructive testing database is established. The damage assessment module 20 may further include: a traversal and tracing unit for traversing the pressure swing adsorption device using the three-dimensional damage model to generate damage source data; a damage impact quantification unit for quantifying the damage impact according to the damage source data and setting the multiple quantitative evaluation criteria; a damage assessment unit for performing damage assessment in accordance with the multiple quantitative evaluation criteria in combination with the three-dimensional damage model to obtain multiple damage scores; a damage feature acquisition unit for performing cluster analysis based on the damage source data to determine multiple damage types, performing feature analysis based on the multiple damage types in combination with the three-dimensional damage model to obtain multiple damage features; and a multimodal non-destructive testing database construction unit for entering the multimodal signals in descending order according to the multiple damage scores in combination with the multiple damage features to construct the multimodal non-destructive testing database.

[0047] The specific configuration of the digital twin processing module 30 will be described in detail below. As described above, the digital twin technology is used to process the three-dimensional damage model in combination with the multiple quantitative evaluation criteria to obtain a digital twin module. The digital twin processing module 30 may further include: a digital twin model construction unit for using the digital twin technology in combination with the three-dimensional damage model for matching to construct a digital twin model; a data driving unit for synchronizing the multimodal signal to the digital twin model for data driving according to the multiple quantitative evaluation criteria to obtain multimodal damage change data; a dynamic damage mapping unit for performing dynamic damage mapping to the three-dimensional damage model based on the multimodal damage change data to obtain dynamic damage visualization data; a multi-standard evaluation unit for performing multi-standard evaluation according to the dynamic damage visualization data in accordance with the multiple quantitative evaluation criteria to obtain a multi-standard comprehensive score, evaluating the digital twin model according to the multi-standard comprehensive score, and adding the digital twin model to the digital twin module according to the evaluation result.

[0048] The specific configuration of the service life prediction model construction module 40 will be described in detail below. As described above, the dynamic damage quantification information set is classified and trained through the digital twin module to construct a service life prediction model. The service life prediction model construction module 40 may further include: a damage label determination unit for classifying the dynamic damage quantification information set based on the multiple damage features and determining multiple damage labels; a gradient boosting decision tree construction unit for performing classification learning based on the multiple damage labels combined with multiple damage features through the digital twin module to construct a gradient boosting decision tree; and a service life prediction model acquisition unit for performing data training on the dynamic damage quantification information set based on the gradient boosting decision tree, cross-validating the training results, and obtaining the service life prediction model.

[0049] Among them, data training is performed on the dynamic damage quantification information set based on the gradient boosting decision tree, and the training results are cross-validated to obtain the service life prediction model. The service life prediction model acquisition unit may further include: a trend analysis subunit for performing trend analysis on the dynamic damage quantification information set using a time series analysis algorithm to determine the damage evolution trend data; a regularity analysis subunit for performing regularity analysis based on the gradient boosting decision tree combined with the damage evolution trend data to draw a damage evolution curve; a regression calculation subunit for performing regression calculation on the pressure swing adsorption device according to the damage evolution curve combined with the multiple damage characteristics to generate multiple data distributions information; a supervised training subunit is used to perform supervised training on the dynamic damage quantification information set according to the multiple data distribution information in combination with the multiple damage labels to generate the training result; a training data random extraction subunit is used to randomly extract first training data, second training data...Nth training data based on the training result, wherein the first training data, the second training data...the Nth training data are all different training data; a cross-iteration validation subunit is used to perform cross-iteration validation on the first training data, the second training data...the Nth training data according to the error index until the training result converges, thereby constructing the service life prediction model.

[0050] The safety state prediction device for a pressure swing adsorption device combined with a digital twin provided in an embodiment of the present invention can execute the safety state prediction method for a pressure swing adsorption device combined with a digital twin provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0051] Although the present application makes various references to certain modules in the apparatus according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.

[0052] The above specific embodiments do not constitute a limitation to the scope of protection of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application should be included in the scope of protection of this application. In some cases, the actions or steps recorded in this application can be performed in an order different from that in the embodiments and can still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A safety status prediction method for a pressure swing adsorption device combined with digital twins is characterized by: The method comprises: Performing operation data collection on the pressure swing adsorption device according to target operating conditions to generate a dynamic damage quantification information set, and performing three-dimensional construction based on the dynamic damage quantification information set to obtain a three-dimensional damage model; Evaluating the pressure swing adsorption device based on the three-dimensional damage model and establishing a multimodal nondestructive testing database, wherein the multimodal nondestructive testing database includes multiple quantitative evaluation criteria; Using digital twin technology, processing is performed based on the three-dimensional damage model in combination with the multiple quantitative evaluation criteria to obtain a digital twin module; Performing classification learning and training on the dynamic damage quantification information set through the digital twin module to construct a service life prediction model; The operating state of the pressure swing adsorption device is evaluated in real time by using the service life prediction model to generate a device operation prediction state information set, and an intelligent safety state prediction of the pressure swing adsorption device is completed based on the device operation prediction state information set; The pressure swing adsorption device is evaluated based on the three-dimensional damage model, and a multimodal nondestructive testing database is established. The method includes: Using the three-dimensional damage model to traverse and trace the source of the pressure swing adsorption device to generate damage source data; quantifying damage impact according to the damage source data and setting the plurality of quantitative evaluation criteria; Performing damage evaluation according to the multiple quantitative evaluation criteria in combination with the three-dimensional damage model to obtain multiple damage scores; Performing cluster analysis based on the damage source data to determine multiple damage types, and performing feature analysis based on the multiple damage types in combination with the three-dimensional damage model to obtain multiple damage features; Entering the multimodal signals in descending order according to the multiple damage scores combined with the multiple damage features to construct the multimodal nondestructive testing database; Using digital twin technology, processing is performed based on the three-dimensional damage model in combination with the multiple quantitative evaluation criteria to obtain a digital twin module, the method comprising: The digital twin technology is combined with the three-dimensional damage model for matching to construct a digital twin model; Synchronizing the multimodal signals to the digital twin model for data-driven operation according to the multiple quantitative evaluation criteria to obtain multimodal damage change data, including: synchronizing the multimodal signals to the digital twin model according to the quantitative evaluation criteria, and using the synchronized multimodal signals to drive the digital twin model for simulation and analysis, including simulating the operating state and damage condition of the pressure swing adsorption device under target operating conditions; and obtaining multimodal damage change data based on the simulation results, wherein the data reflects changes in the pressure swing adsorption device during the damage process; Performing dynamic damage mapping on the damage three-dimensional model based on the multimodal damage change data to obtain dynamic damage visualization data; A multi-standard evaluation is performed according to the dynamic damage visualization data in accordance with the multiple quantitative evaluation criteria to obtain a multi-standard comprehensive score, the digital twin model is evaluated according to the multi-standard comprehensive score, and the digital twin model is added to the digital twin module according to the evaluation result.

2. The method for predicting the safety status of a pressure swing adsorption device combined with digital twin according to claim 1, characterized in that: The operation of the pressure swing adsorption device is collected according to the target operating conditions to generate a dynamic damage quantification information set, and the method includes: traversing multiple operating conditions of the pressure swing adsorption device for analysis to determine the target operating condition; Performing multi-dimensional damage acquisition on the pressure swing adsorption device based on the target operating condition to obtain a multi-modal signal; Performing a comprehensive inspection of the pressure swing adsorption device based on the multimodal signal to obtain device inspection results, and locating damage based on the device inspection results to determine multiple damage locations; Data quantification is performed according to the multiple damage positions to obtain the dynamic damage quantification information set.

3. The method for predicting the safety status of a pressure swing adsorption device combined with digital twins according to claim 2, characterized in that: Performing a three-dimensional construction based on the dynamic damage quantification information set to obtain a three-dimensional damage model includes: Performing simulation verification based on the dynamic damage quantification information set and the multiple damage locations to obtain a damage verification result; Performing stereo matching on the dynamic damage quantification information set according to the damage verification result to obtain a plurality of point cloud data; Performing damage evolution on the plurality of point cloud data based on the multimodal signal to determine damage impact transfer data; Using deep learning, the multimodal signals are fused according to the damage impact transfer data to generate signal fusion data; The plurality of point cloud data are three-dimensionally combined according to the signal fusion data to construct the three-dimensional damage model.

4. The method for predicting the safety status of a pressure swing adsorption device combined with digital twins according to claim 1, characterized in that: The dynamic damage quantification information set is subjected to classification learning and training by the digital twin module to construct a service life prediction model, the method comprising: classifying the dynamic damage quantification information set based on the multiple damage features to determine multiple damage labels; The digital twin module performs classification learning according to the multiple damage labels combined with multiple damage features to construct a gradient boosting decision tree; The dynamic damage quantification information set is trained with data based on the gradient boosting decision tree, and the training results are cross-validated to obtain the service life prediction model.

5. The method for predicting the safety status of a pressure swing adsorption device combined with digital twins according to claim 4 is characterized in that: Data training is performed on the dynamic damage quantification information set based on the gradient boosting decision tree, and the training results are cross-validated to obtain the service life prediction model. The method includes: Performing trend analysis on the dynamic damage quantification information set using a time series analysis algorithm to determine damage evolution trend data; Perform regularity analysis based on the gradient boosting decision tree and the damage evolution trend data to draw a damage evolution curve; Performing regression calculation on the pressure swing adsorption device according to the damage evolution curve diagram and the multiple damage characteristics to generate multiple data distribution information; Performing supervised training on the dynamic damage quantification information set according to the multiple data distribution information combined with the multiple damage labels to generate the training result; randomly extracting first training data, second training data, ..., Nth training data based on the training result, wherein the first training data, the second training data, ..., the Nth training data are all different training data; Cross-iteration validation is performed on the first training data, the second training data, ... the Nth training data according to an error index until the training results converge, thereby constructing the service life prediction model.

6. The safety status prediction device of the pressure swing adsorption device combined with digital twin is characterized by: The device is used to implement the method for predicting the safety state of a pressure swing adsorption device combined with digital twin according to any one of claims 1 to 5, and the device includes: a device operation acquisition module, the device operation acquisition module being used to collect operation data of the pressure swing adsorption device according to target operating conditions, generate a dynamic damage quantification information set, and perform three-dimensional construction based on the dynamic damage quantification information set to obtain a three-dimensional damage model; a damage assessment module, the damage assessment module being used to evaluate the pressure swing adsorption device based on the three-dimensional damage model and establish a multimodal nondestructive testing database, the multimodal nondestructive testing database including a plurality of quantitative evaluation criteria; A digital twin processing module, configured to utilize digital twin technology to process the three-dimensional damage model in combination with the multiple quantitative evaluation criteria to obtain a digital twin module; A service life prediction model construction module, wherein the service life prediction model construction module is used to perform classification learning training on the dynamic damage quantification information set through the digital twin module to construct a service life prediction model; A state prediction module is used to evaluate the operating state of the pressure swing adsorption device in real time through the service life prediction model, generate a device operation prediction state information set, and complete the intelligent safety state prediction of the pressure swing adsorption device based on the device operation prediction state information set.

Images