A multi-scale flow fault diagnosis system for circulating fluidized bed

Abstract

The present application discloses a circulating fluidized bed multi-scale flow fault diagnosis system, which belongs to the technical field of circulating fluidized beds. The system includes: a data acquisition and preprocessing module, which obtains and preprocesses multi-source key parameter data of the circulating fluidized bed from multiple sensors and measurement points; a multi-scale feature extraction module, which uses wavelet packet decomposition algorithm and information entropy analysis method to realize adaptive multi-scale decomposition and feature extraction of multi-source key parameter data; an intelligent data fusion module, which realizes automatic weight allocation and optimization fusion of multi-source and multi-scale data based on the attention mechanism; a dynamic fault mode library, which continuously adds new abnormal modes through initialization and online learning mechanisms to realize automatic expansion and optimization of fault modes; a real-time intelligent fault diagnosis module, which realizes real-time, accurate and reliable fault diagnosis and early warning of the circulating fluidized bed system based on a multi-task deep learning model.

Description

Technical Field

[0001] The present application relates to the technical field of circulating fluidized beds, and more specifically, to a circulating fluidized bed multi-scale flow fault diagnosis system. Background Art

[0002] Circulating fluidized bed technology is widely used in the fields of chemical industry, energy, etc. Its operation complexity and high requirements for system stability make fault diagnosis and prediction a key issue. Traditional fault diagnosis methods mainly rely on empirical analysis and single sensor data, which have the problems of low diagnostic accuracy and limited ability to fuse multi-source information.

[0003] With the improvement of data acquisition technology and computing power, circulating fluidized bed fault diagnosis systems based on multi-scale flow characteristics have been developed. Such systems can achieve comprehensive perception of system status by simultaneously using multi-dimensional data obtained by multiple sensors, such as key parameters such as particle concentration, temperature, pressure, and gas composition. However, how to effectively process and analyze these multi-source, multi-scale data, and how to accurately identify and predict system operation anomalies, remain important challenges in current research.

[0004] In the existing fluidized bed fault diagnosis technology, the application of multi-scale feature extraction and intelligent data fusion technology has begun to attract attention. For example, the application of wavelet analysis in multi-scale signal processing, as well as data fusion and dynamic learning mechanisms based on deep learning models, provide new possibilities for improving diagnostic accuracy and real-time performance. However, existing methods still have shortcomings in real-time performance, adaptability and comprehensive performance, especially when dealing with complex nonlinear dynamic systems.

[0005] Therefore, the research and development of a multi-scale flow fault diagnosis system for circulating fluidized bed systems has important theoretical and application significance. The system aims to overcome the limitations of traditional methods and achieve efficient, accurate and real-time monitoring of the circulating fluidized bed system status through innovative data acquisition and preprocessing, multi-scale feature extraction, intelligent data fusion, integration of dynamic fault mode library and real-time intelligent diagnosis module, thereby improving the stability, safety and operation efficiency of the system. Summary of the invention

[0006] In order to overcome a series of defects in the prior art, the purpose of this application is to provide a circulating fluidized bed multi-scale flow pattern fault diagnosis system for the above-mentioned problems, including the following parts:

[0007] Data acquisition and preprocessing module, which acquires multi-source key parameter data of circulating fluidized bed from multiple sensors and measurement points and performs preprocessing;

[0008] The multi-scale feature extraction module uses wavelet packet decomposition algorithm and information entropy analysis method to achieve adaptive multi-scale decomposition and feature extraction of multi-source key parameter data;

[0009] Intelligent data fusion module, based on the attention mechanism, realizes automatic weight allocation and optimized fusion of multi-source and multi-scale data;

[0010] Dynamic fault mode library, through initialization and online learning mechanism, continuously adds new abnormal modes to achieve automatic expansion and optimization of fault modes;

[0011] The real-time intelligent fault diagnosis module realizes real-time, accurate and reliable fault diagnosis and early warning of the circulating fluidized bed system based on a multi-task deep learning model.

[0012] Furthermore, the multi-source key parameter data include particle concentration, temperature, pressure and gas composition; and the preprocessing operations include denoising, normalization and synchronization.

[0013] Furthermore, the data acquisition and preprocessing module includes:

[0014] Multi-source sensor array, using distributed multi-type sensors to achieve comprehensive perception and high-precision acquisition of key parameters of the circulating fluidized bed;

[0015] Adaptive sampling frequency regulator, dynamically adjusts the sampling frequency according to the operating status of the circulating fluidized bed and the data change rate, optimizing data collection efficiency and storage space utilization;

[0016] Multimodal data synchronization unit, using timestamp alignment and interpolation technology to ensure the consistency of data from different sources and sampling rates in the time dimension;

[0017] Self-learning denoising network, based on noise feature extraction and adaptive filtering to achieve intelligent suppression of complex background noise;

[0018] Dynamic data standardization processor, combining historical data statistics and real-time data distribution, adaptively adjusts standardization parameters to ensure data comparability;

[0019] The data quality assessment and repair module conducts real-time assessment of data quality by setting multi-dimensional evaluation indicators, and uses advanced interpolation algorithms to repair missing or damaged data.

[0020] Furthermore, the multi-scale feature extraction module includes:

[0021] Adaptive wavelet packet decomposition unit, which automatically selects the optimal wavelet basis function and decomposition layer number according to the characteristics of multi-source key parameter data, and realizes multi-resolution analysis and detail feature extraction of multi-source key parameter data;

[0022] Multi-dimensional information entropy calculator, which comprehensively calculates multiple information entropy indicators in time domain, frequency domain and time-frequency domain, and fully captures the complexity and uncertainty characteristics of multi-source key parameter data;

[0023] Nonlinear dynamic feature extractor, using phase space reconstruction and fractal theory, extracts chaotic features and fractal dimensions of multi-source key parameter data, revealing the nonlinear dynamic behavior of the circulating fluidized bed;

[0024] The time-varying mode decomposition engine, based on the empirical mode decomposition algorithm, adaptively decomposes non-stationary multi-source key parameter data and extracts time-varying features;

[0025] Multi-scale correlation analyzer, which calculates the correlation and causal relationship between multi-source key parameter data at different time scales, and identifies the key influencing factors of the state change of the circulating fluidized bed;

[0026] The sparse representation feature learner uses dictionary learning and compressed sensing technology to construct sparse representation of multi-source key parameter data and extract highly abstract and discriminative features.

[0027] Furthermore, the time domain information entropy H time It is expressed by the following formula: Among them, p i is the data value x i The probability of, N represents the type of data value;

[0028] Frequency domain information entropy H freq It is expressed by the following formula: Among them, p j is the frequency domain component X j The probability of, M represents the number of frequency domain components;

[0029] Information entropy in time-frequency domain H time-freq It is expressed by the following formula: Among them, p ij is the time domain data value x i and the frequency domain component X j The joint probability of .

[0030] Furthermore, the intelligent data fusion module includes:

[0031] The multi-source data alignment and calibration unit uses dynamic time warping algorithm and spatial registration technology to achieve accurate alignment of multi-source key parameter data in time and space dimensions;

[0032] An adaptive feature importance evaluator dynamically calculates and updates the importance weight of each feature for the characterization of the circulating fluidized bed state based on information gain and mutual information;

[0033] The multimodal feature complementary enhancer uses a cross-modal attention mechanism to capture the relationship between features of different modalities, enhance complementary information, and improve the expressiveness of fused features;

[0034] Multi-level progressive fusion network builds a hierarchical fusion structure to achieve progressive fusion from low-level perceptual features to high-level semantic features;

[0035] Uncertainty quantification and propagation unit, based on Bayesian inference and evidence theory, models and quantifies the uncertainty in the fusion process to improve the reliability of the fusion results;

[0036] The multi-objective optimization fuser is based on the Pareto optimization principle and takes into account the fusion accuracy, computational efficiency and robustness objectives to achieve dynamic optimization of the fusion strategy.

[0037] Furthermore, the formula for information gain is: Gain(X, Y) = H(Y) - H(Y|X), where X is the feature variable and Y is the target variable; H(Y) is the entropy of the target variable Y; H(Y|X) is the conditional entropy of the target variable Y given the feature X; Gain(X, Y) represents the information gain of the target variable Y under the condition of the feature variable X;

[0038] The formula for mutual information is: Where p(x, y) is the joint probability distribution of X and Y; p(x) and p(y) are the marginal probability distributions of X and Y respectively; I(X; Y) represents the mutual information between X and Y, that is, the degree of mutual dependence between them; x and y are the specific values ​​of random variables X and Y;

[0039] The importance of feature X for the characterization of the circulating fluidized bed state Weight(X) is expressed by the following formula: Weight(X)=α·Gain(X,Y)+(1-α)·I(X;Y), where α is a parameter that weighs information gain and mutual information, and its value range is [0, 1].

[0040] Furthermore, based on the Pareto optimization principle, while considering the fusion accuracy, computational efficiency and robustness goals, the dynamic optimization of the fusion strategy can be achieved through the following specific steps:

[0041] Quantify fusion accuracy, computational efficiency, and robustness into measurable indicators as the objective function of multi-objective optimization;

[0042] Identify the key parameters that affect the fusion strategy as decision variables and optimize them to find the best fusion strategy;

[0043] Set corresponding constraints according to the actual operation requirements of the system;

[0044] Use multi-objective optimization algorithms to generate Pareto optimal solution sets;

[0045] The generated Pareto optimal solution set is evaluated using the analytic hierarchy process;

[0046] A dynamic adjustment mechanism is established to periodically trigger the optimization process according to changes in the system operating status and external environment to adapt to the optimal fusion strategy in different scenarios.

[0047] Furthermore, the dynamic failure mode library includes:

[0048] Multi-level fault mode classifier, which uses a hierarchical classification structure to accurately classify fault modes from the system level to the component level, and realizes multi-dimensional location and description of faults;

[0049] The time-varying fault feature extractor combines wavelet analysis and sliding time window technology to capture the dynamic evolution of fault features and identify the development trend of faults;

[0050] Fault mode similarity calculation engine, based on deep metric learning algorithm, calculates the similarity between newly discovered faults and known fault modes, assisting in the classification and identification of new faults;

[0051] Incremental learning update mechanism, which uses online incremental learning algorithm to continuously integrate new fault data and dynamically update the feature representation and classification boundaries of fault modes;

[0052] The knowledge graph integration module integrates expert knowledge and data-driven fault modes to build a comprehensive fault knowledge graph and enhance the interpretability of fault diagnosis.

[0053] Furthermore, the real-time intelligent fault diagnosis module includes:

[0054] Real-time data stream processing engine, using efficient stream processing architecture to achieve real-time reception and preprocessing of large-scale sensor data;

[0055] Dynamic threshold adaptive regulator adjusts the fault judgment threshold in real time according to the operating status of the circulating fluidized bed to improve the accuracy and sensitivity of diagnosis;

[0056] The spatiotemporal correlation anomaly detector uses graph neural networks and time series analysis to capture spatial correlation and time series anomalies between devices, achieving early warning of system-level failures;

[0057] The diagnosis result confidence evaluator provides a confidence score for each diagnosis result based on Bayesian reasoning and model uncertainty quantification techniques.

[0058] Compared with the prior art, this application has the following beneficial effects:

[0059] This application combines advanced data processing and deep learning technologies to achieve adaptive extraction of multi-scale features from multi-source key parameter data, and achieves optimized fusion and weight distribution of multi-source and multi-scale data through intelligent data fusion. The system further innovates in the real-time update and expansion of the dynamic fault mode library, as well as the real-time fault diagnosis and early warning capabilities based on multi-task deep learning, effectively improving the fault diagnosis accuracy, reliability and real-time performance of the circulating fluidized bed system. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a schematic diagram of the structure of a circulating fluidized bed multi-scale flow fault diagnosis system disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical scheme and advantages of the implementation of the present invention clearer, the technical scheme in the embodiment of the present invention will be described in more detail below in conjunction with the drawings in the embodiment of the present invention. In the drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of the present invention, not all of the embodiments.

[0062] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0063] The embodiments and directional terms described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0064] like Figure 1 As shown, a circulating fluidized bed multi-scale flow fault diagnosis system includes the following parts:

[0065] Data acquisition and preprocessing module, which acquires multi-source key parameter data of circulating fluidized bed from multiple sensors and measurement points and performs preprocessing;

[0066] The multi-scale feature extraction module uses wavelet packet decomposition algorithm and information entropy analysis method to achieve adaptive multi-scale decomposition and feature extraction of multi-source key parameter data;

[0067] Intelligent data fusion module, based on the attention mechanism, realizes automatic weight allocation and optimized fusion of multi-source and multi-scale data;

[0068] Dynamic fault mode library, through initialization and online learning mechanism, continuously adds new abnormal modes to achieve automatic expansion and optimization of fault modes;

[0069] The real-time intelligent fault diagnosis module realizes real-time, accurate and reliable fault diagnosis and early warning of the circulating fluidized bed system based on a multi-task deep learning model.

[0070] The data acquisition and preprocessing module is responsible for acquiring the key parameter data of the circulating fluidized bed from multiple sensors and measurement points in real time and performing preprocessing. Through high-precision data acquisition and real-time monitoring, the system can timely obtain multi-source data such as bed temperature, pressure, flow rate, etc., providing a sufficient data basis for subsequent fault diagnosis. The preprocessing process includes data cleaning, filtering and denoising to ensure data quality and stability and improve the accuracy of subsequent feature extraction and analysis.

[0071] The multi-scale feature extraction module uses wavelet packet decomposition algorithm and information entropy analysis method to achieve adaptive multi-scale decomposition and feature extraction of multi-source key parameter data. This method can capture the changes and characteristics of bed operation status at different time and space scales, thereby identifying potential abnormal patterns. Through refined feature extraction, normal operation and abnormal status can be distinguished more accurately, laying a solid foundation for subsequent fault diagnosis.

[0072] Based on the attention mechanism, the intelligent data fusion module realizes automatic weight allocation and optimized fusion of multi-source and multi-scale data. Through the attention mechanism, the weight can be dynamically adjusted according to the importance and real-time nature of the data, and data of different scales and sources can be effectively fused to improve the accuracy and comprehensiveness of the comprehensive information. This intelligent fusion method can reduce information duplication and redundancy while retaining key information and improving the system's ability to understand complex operating conditions.

[0073] Through initialization and online learning mechanisms, the dynamic fault mode library continuously adds new abnormal modes to achieve automatic expansion and optimization of fault modes. As the system runs and data accumulates, the fault mode library is continuously enriched and updated, and can adapt to new operating conditions and fault modes in a timely manner, improving the system's ability to identify and respond to new faults. The establishment of a dynamic fault mode library can also provide more accurate reference and support for subsequent intelligent fault diagnosis modules.

[0074] Based on the multi-task deep learning model, the real-time intelligent fault diagnosis module realizes real-time, accurate and reliable fault diagnosis and early warning of the circulating fluidized bed system. Multi-task learning can handle multiple fault types at the same time, improving the comprehensiveness and sensitivity of diagnosis. Through real-time monitoring and data analysis, potential fault signals can be quickly identified and timely warnings can be issued to help operation and maintenance personnel respond and solve problems quickly, reducing the impact and losses caused by faults.

[0075] In summary, the circulating fluidized bed multi-scale flow state fault diagnosis system effectively improves the system's reliability, operation efficiency and fault response capability through the organic combination of data acquisition and preprocessing, multi-scale feature extraction, intelligent data fusion, dynamic fault mode library and real-time intelligent fault diagnosis module. It has important application prospects and practical significance for bed flow state problems.

[0076] Furthermore, the multi-source key parameter data include particle concentration, temperature, pressure and gas composition; and the preprocessing operations include denoising, normalization and synchronization.

[0077] Furthermore, the data acquisition and preprocessing module includes:

[0078] Multi-source sensor array, using distributed multi-type sensors to achieve comprehensive perception and high-precision acquisition of key parameters of the circulating fluidized bed;

[0079] Adaptive sampling frequency regulator, dynamically adjusts the sampling frequency according to the operating status of the circulating fluidized bed and the data change rate, optimizing data collection efficiency and storage space utilization;

[0080] Multimodal data synchronization unit, using timestamp alignment and interpolation technology to ensure the consistency of data from different sources and sampling rates in the time dimension;

[0081] Self-learning denoising network, based on noise feature extraction and adaptive filtering to achieve intelligent suppression of complex background noise;

[0082] Dynamic data standardization processor, combining historical data statistics and real-time data distribution, adaptively adjusts standardization parameters to ensure data comparability;

[0083] The data quality assessment and repair module conducts real-time assessment of data quality by setting multi-dimensional evaluation indicators, and uses advanced interpolation algorithms to repair missing or damaged data.

[0084] In summary, the data acquisition and preprocessing module effectively improves the circulating fluidized bed system's ability to monitor the operating status, data processing efficiency and data quality assurance through the integrated application of multi-source sensor arrays, adaptive sampling frequency regulators, multimodal data synchronization units, self-learning denoising networks, dynamic data standardization processors and data quality assessment and repair modules, providing strong technical support for the stable operation and fault diagnosis of the system.

[0085] Furthermore, the multi-scale feature extraction module includes:

[0086] Adaptive wavelet packet decomposition unit, which automatically selects the optimal wavelet basis function and decomposition layer number according to the characteristics of multi-source key parameter data, and realizes multi-resolution analysis and detail feature extraction of multi-source key parameter data;

[0087] Multi-dimensional information entropy calculator, which comprehensively calculates multiple information entropy indicators in time domain, frequency domain and time-frequency domain, and fully captures the complexity and uncertainty characteristics of multi-source key parameter data;

[0088] Nonlinear dynamic feature extractor, using phase space reconstruction and fractal theory, extracts chaotic features and fractal dimensions of multi-source key parameter data, revealing the nonlinear dynamic behavior of the circulating fluidized bed;

[0089] The time-varying mode decomposition engine, based on the empirical mode decomposition algorithm, adaptively decomposes non-stationary multi-source key parameter data and extracts time-varying features;

[0090] Multi-scale correlation analyzer, which calculates the correlation and causal relationship between multi-source key parameter data at different time scales, and identifies the key influencing factors of the state change of the circulating fluidized bed;

[0091] The sparse representation feature learner uses dictionary learning and compressed sensing technology to construct sparse representation of multi-source key parameter data and extract highly abstract and discriminative features.

[0092] In summary, the multi-scale feature extraction module effectively improves the circulating fluidized bed system's ability to capture and analyze the multi-dimensional characteristics of the bed operation status through the comprehensive application of technical means such as adaptive wavelet packet decomposition, multi-dimensional information entropy calculation, nonlinear dynamic feature extraction, time-varying pattern decomposition, multi-scale correlation analysis and sparse representation feature learning, laying a solid technical foundation for the system's intelligent monitoring and fault diagnosis.

[0093] Furthermore, the time domain information entropy H time It is expressed by the following formula: Among them, p i is the data value x i The probability of, N represents the type of data value;

[0094] Frequency domain information entropy H freq It is expressed by the following formula: Among them, p j is the probability of the frequency domain component X, and M represents the number of frequency domain components;

[0095] Information entropy in time-frequency domain H time-freq It is expressed by the following formula: Among them, p ij is the time domain data value x i and the frequency domain component X j The joint probability of .

[0096] Furthermore, the intelligent data fusion module includes:

[0097] The multi-source data alignment and calibration unit uses dynamic time warping algorithm and spatial registration technology to achieve accurate alignment of multi-source key parameter data in time and space dimensions;

[0098] An adaptive feature importance evaluator dynamically calculates and updates the importance weight of each feature for the characterization of the circulating fluidized bed state based on information gain and mutual information;

[0099] The multimodal feature complementary enhancer uses a cross-modal attention mechanism to capture the relationship between features of different modalities, enhance complementary information, and improve the expressiveness of fused features;

[0100] Multi-level progressive fusion network builds a hierarchical fusion structure to achieve progressive fusion from low-level perceptual features to high-level semantic features;

[0101] Uncertainty quantification and propagation unit, based on Bayesian inference and evidence theory, models and quantifies the uncertainty in the fusion process to improve the reliability of the fusion results;

[0102] The multi-objective optimization fuser is based on the Pareto optimization principle and takes into account the fusion accuracy, computational efficiency and robustness objectives to achieve dynamic optimization of the fusion strategy.

[0103] In summary, the intelligent data fusion module effectively improves the circulating fluidized bed system's comprehensive perception and accurate diagnosis capabilities of complex states through the comprehensive application of technical means such as multi-source data alignment and calibration, adaptive feature importance evaluation, multi-modal feature complementary enhancement, multi-level progressive fusion, uncertainty quantification and propagation, and multi-objective optimization, providing strong technical support for the intelligent monitoring and management of system operation.

[0104] Furthermore, the formula for information gain is: Gain(X, Y) = H(Y) - H(Y|X), where X is the feature variable and Y is the target variable; H(Y) is the entropy of the target variable Y; H(Y|X) is the conditional entropy of the target variable Y given the feature X; Gain(X, Y) represents the information gain of the target variable Y under the condition of the feature variable X;

[0105] The formula for mutual information is: Where p(x, y) is the joint probability distribution of X and Y; p(x) and p(y) are the marginal probability distributions of X and Y respectively; I(X; Y) represents the mutual information between X and Y, that is, the degree of mutual dependence between them; x and y are the specific values ​​of random variables X and Y;

[0106] The importance of feature X for the characterization of the circulating fluidized bed state Weight(X) is expressed by the following formula: Weight(X)=α·Gain(X,Y)+(1-α)·I(X;Y), where α is a parameter that weighs information gain and mutual information, and its value range is [0, 1].

[0107] Furthermore, based on the Pareto optimization principle, while considering the fusion accuracy, computational efficiency and robustness goals, the dynamic optimization of the fusion strategy can be achieved through the following specific steps:

[0108] Quantify fusion accuracy, computational efficiency, and robustness into measurable indicators as the objective function of multi-objective optimization;

[0109] Identify the key parameters that affect the fusion strategy as decision variables and optimize them to find the best fusion strategy;

[0110] Set corresponding constraints according to the actual operation requirements of the system;

[0111] Use multi-objective optimization algorithms to generate Pareto optimal solution sets;

[0112] The generated Pareto optimal solution set is evaluated using the analytic hierarchy process;

[0113] A dynamic adjustment mechanism is established to periodically trigger the optimization process according to changes in the system operating status and external environment to adapt to the optimal fusion strategy in different scenarios.

[0114] Furthermore, the dynamic failure mode library includes:

[0115] Multi-level fault mode classifier, which uses a hierarchical classification structure to accurately classify fault modes from the system level to the component level, and realizes multi-dimensional location and description of faults;

[0116] The time-varying fault feature extractor combines wavelet analysis and sliding time window technology to capture the dynamic evolution of fault features and identify the development trend of faults;

[0117] Fault mode similarity calculation engine, based on deep metric learning algorithm, calculates the similarity between newly discovered faults and known fault modes, assisting in the classification and identification of new faults;

[0118] Incremental learning update mechanism, which uses online incremental learning algorithm to continuously integrate new fault data and dynamically update the feature representation and classification boundaries of fault modes;

[0119] The knowledge graph integration module integrates expert knowledge and data-driven fault modes to build a comprehensive fault knowledge graph and enhance the interpretability of fault diagnosis.

[0120] In summary, the dynamic fault mode library module effectively improves the diagnostic capability and processing efficiency of the circulating fluidized bed system for various fault conditions through the comprehensive application of technical means such as multi-level classification, time-varying feature extraction, similarity calculation, incremental learning and knowledge graph integration, providing important guarantees for the safe and stable operation of the system.

[0121] Furthermore, the real-time intelligent fault diagnosis module includes:

[0122] Real-time data stream processing engine, using efficient stream processing architecture to achieve real-time reception and preprocessing of large-scale sensor data;

[0123] Dynamic threshold adaptive regulator adjusts the fault judgment threshold in real time according to the operating status of the circulating fluidized bed to improve the accuracy and sensitivity of diagnosis;

[0124] The spatiotemporal correlation anomaly detector uses graph neural networks and time series analysis to capture spatial correlation and time series anomalies between devices, achieving early warning of system-level failures;

[0125] The diagnosis result confidence evaluator provides a confidence score for each diagnosis result based on Bayesian reasoning and model uncertainty quantification techniques.

[0126] In summary, the real-time intelligent fault diagnosis module effectively improves the real-time monitoring, accurate diagnosis and early warning capabilities of the circulating fluidized bed system through its advanced data stream processing capabilities, dynamic threshold adjustment, spatiotemporal correlation anomaly detection and confidence assessment technology. The combination of these technologies enables the system to quickly respond to and handle various fault conditions in a complex industrial environment, ensuring the safe operation and stability of the system to the greatest extent, while optimizing maintenance and management efficiency.

[0127] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A circulating fluidized bed multi-scale flow fault diagnosis system, characterized in that: It includes the following parts: Data acquisition and preprocessing module, which acquires multi-source key parameter data of circulating fluidized bed from multiple sensors and measurement points and performs preprocessing; The multi-scale feature extraction module uses wavelet packet decomposition algorithm and information entropy analysis method to achieve adaptive multi-scale decomposition and feature extraction of multi-source key parameter data; Intelligent data fusion module, based on the attention mechanism, realizes automatic weight allocation and optimized fusion of multi-source and multi-scale data; Dynamic fault mode library, through initialization and online learning mechanism, continuously adds new abnormal modes to achieve automatic expansion and optimization of fault modes; The real-time intelligent fault diagnosis module realizes real-time, accurate and reliable fault diagnosis and early warning of the circulating fluidized bed system based on a multi-task deep learning model.

2. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 1, characterized in that: Multi-source key parameter data include particle concentration, temperature, pressure and gas composition; pre-processing operations include denoising, normalization and synchronization.

3. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 1, characterized in that: The data acquisition and preprocessing modules include: Multi-source sensor array, using distributed multi-type sensors to achieve comprehensive perception and high-precision acquisition of key parameters of the circulating fluidized bed; Adaptive sampling frequency regulator, dynamically adjusts the sampling frequency according to the operating status of the circulating fluidized bed and the data change rate, optimizing data collection efficiency and storage space utilization; Multimodal data synchronization unit, using timestamp alignment and interpolation technology to ensure the consistency of data from different sources and sampling rates in the time dimension; Self-learning denoising network, based on noise feature extraction and adaptive filtering to achieve intelligent suppression of complex background noise; Dynamic data standardization processor, combining historical data statistics and real-time data distribution, adaptively adjusts standardization parameters to ensure data comparability; The data quality assessment and repair module conducts real-time assessment of data quality by setting multi-dimensional evaluation indicators, and uses advanced interpolation algorithms to repair missing or damaged data.

4. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 1, characterized in that: The multi-scale feature extraction module includes: Adaptive wavelet packet decomposition unit, which automatically selects the optimal wavelet basis function and decomposition layer number according to the characteristics of multi-source key parameter data, and realizes multi-resolution analysis and detail feature extraction of multi-source key parameter data; Multi-dimensional information entropy calculator, which comprehensively calculates multiple information entropy indicators in time domain, frequency domain and time-frequency domain, and fully captures the complexity and uncertainty characteristics of multi-source key parameter data; Nonlinear dynamic feature extractor, using phase space reconstruction and fractal theory, extracts chaotic features and fractal dimensions of multi-source key parameter data, revealing the nonlinear dynamic behavior of the circulating fluidized bed; The time-varying mode decomposition engine, based on the empirical mode decomposition algorithm, adaptively decomposes non-stationary multi-source key parameter data and extracts time-varying features; Multi-scale correlation analyzer, which calculates the correlation and causal relationship between multi-source key parameter data at different time scales, and identifies the key influencing factors of the state change of the circulating fluidized bed; The sparse representation feature learner uses dictionary learning and compressed sensing technology to construct sparse representation of multi-source key parameter data and extract highly abstract and discriminative features.

5. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 4, characterized in that: Time domain information entropy H time It is expressed by the following formula: Among them, p i is the data value x i The probability of, N represents the type of data value; Frequency domain information entropy H freq It is expressed by the following formula: Among them, p j is the frequency domain component X j The probability of, M represents the number of frequency domain components; Information entropy in time-frequency domain H time-freq It is expressed by the following formula: Among them, p ij is the time domain data value x i and the frequency domain component X j The joint probability of .

6. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 1, characterized in that: The intelligent data fusion module includes: The multi-source data alignment and calibration unit uses dynamic time warping algorithm and spatial registration technology to achieve accurate alignment of multi-source key parameter data in time and space dimensions; An adaptive feature importance evaluator dynamically calculates and updates the importance weight of each feature for the characterization of the circulating fluidized bed state based on information gain and mutual information; The multimodal feature complementary enhancer uses a cross-modal attention mechanism to capture the relationship between features of different modalities, enhance complementary information, and improve the expressiveness of fused features; Multi-level progressive fusion network builds a hierarchical fusion structure to achieve progressive fusion from low-level perceptual features to high-level semantic features; Uncertainty quantification and propagation unit, based on Bayesian inference and evidence theory, models and quantifies the uncertainty in the fusion process to improve the reliability of the fusion results; The multi-objective optimization fuser is based on the Pareto optimization principle and takes into account the fusion accuracy, computational efficiency and robustness objectives to achieve dynamic optimization of the fusion strategy.

7. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 6, characterized in that: The formula for information gain is: Gain(X,Y)=H(Y)-H(Y|X), where X is the feature variable and Y is the target variable; H(Y) is the entropy of the target variable Y; H(Y|X) is the conditional entropy of the target variable Y given the feature X; Gain(X,Y) represents the information gain of the target variable Y under the condition of the feature variable X; The formula for mutual information is: Where p(x, y) is the joint probability distribution of X and Y; p(x) and p(y) are the marginal probability distributions of X and Y respectively; I(X; Y) represents the mutual information between X and Y, that is, the degree of mutual dependence between them; x and y are the specific values ​​of random variables X and Y; The importance of feature X for the characterization of the circulating fluidized bed state Weight(X) is expressed by the following formula: Weight(X) = α·Gain(X,Y)+(1-α)·I(X;Y), where α is a parameter that weighs information gain and mutual information, and its value range is [0,1].

8. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 6, characterized in that: Based on the Pareto optimization principle, while considering the fusion accuracy, computational efficiency and robustness goals, the dynamic optimization of the fusion strategy can be achieved through the following specific steps: Quantify the fusion accuracy, computational efficiency and robustness into measurable indicators as the objective function of multi-objective optimization; Identify the key parameters that affect the fusion strategy as decision variables and optimize them to find the best fusion strategy; Set corresponding constraints according to the actual operation requirements of the system; Use multi-objective optimization algorithms to generate Pareto optimal solution sets; The generated Pareto optimal solution set is evaluated using the analytic hierarchy process; A dynamic adjustment mechanism is established to periodically trigger the optimization process according to changes in the system operating status and external environment to adapt to the optimal fusion strategy in different scenarios.

9. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 1, characterized in that: The dynamic failure mode library includes: Multi-level fault mode classifier, which uses a hierarchical classification structure to accurately classify fault modes from the system level to the component level, and realizes multi-dimensional location and description of faults; The time-varying fault feature extractor combines wavelet analysis and sliding time window technology to capture the dynamic evolution of fault features and identify the development trend of faults; Fault mode similarity calculation engine, based on deep metric learning algorithm, calculates the similarity between newly discovered faults and known fault modes, assisting in the classification and identification of new faults; Incremental learning update mechanism, which uses online incremental learning algorithm to continuously integrate new fault data and dynamically update the feature representation and classification boundaries of fault modes; The knowledge graph integration module integrates expert knowledge and data-driven fault modes to build a comprehensive fault knowledge graph and enhance the interpretability of fault diagnosis.

10. A circulating fluidized bed multi-scale flow pattern fault diagnosis system according to claim 1, characterized in that: The real-time intelligent fault diagnosis module includes: Real-time data stream processing engine, using efficient stream processing architecture to achieve real-time reception and preprocessing of large-scale sensor data; Dynamic threshold adaptive regulator, which adjusts the fault judgment threshold in real time according to the operating status of the circulating fluidized bed to improve the accuracy and sensitivity of diagnosis; The spatiotemporal correlation anomaly detector uses graph neural networks and time series analysis to capture spatial correlation and time series anomalies between devices, achieving early warning of system-level failures; The diagnosis result confidence evaluator provides a confidence score for each diagnosis result based on Bayesian reasoning and model uncertainty quantification techniques.

Images