Automatic monitoring method and device for pipe explosion of boiler heating surface

Abstract

The invention discloses an automatic monitoring method and device for pipe explosion of a heating surface of a boiler, and relates to the technical field of monitoring of pipe explosion of the heating surface of the boiler. The monitoring method comprises a multi-source signal acquisition step, a data preprocessing and feature extraction step, a double-model parallel prediction step, a dynamic weighted fusion step, a consistency check and feedback calibration step and a risk response output step. According to the CNN-LSTM model sound wave frequency domain feature-temperature field gradient change spatio-temporal multi-source data joint identification, the limitation that a traditional model only pays attention to a single dimension is solved, feature distribution Attention and a physical guidance fusion mechanism solve the false alarm problem of a pure data driving model through a user-defined physical rule constraint function, and the accuracy and the reliability of the model are improved. A dynamic confidence SNR weighting mechanism enables the tube burst identification accuracy of the model under a strong noise working condition to be improved, and physical-data dual-drive consistency check realizes long-term maintenance of high positioning precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of boiler heating surface tube burst monitoring, in particular to an automatic monitoring method and device for boiler heating surface tube burst. BACKGROUND

[0002] The boiler is the core equipment of the thermal power plant, and its heating surface tube is operated under the combined action of high temperature, high pressure, stress and corrosion SEM for a long time. Especially under the complex variable working condition, the tube burst accident is prone to occur. According to statistics, the non-planned shutdown of the unit caused by the boiler tube burst accounts for more than 65% of the total number, which not only causes huge economic loss (such as the single maintenance cost of a 660MW unit can reach more than 8 million yuan), but also causes environmental problems due to the imbalance of combustion. Therefore, timely and accurate monitoring and early warning of the boiler heating surface tube burst are crucial to the safe operation of the power grid, the reliable operation of the equipment and the economic and environmental protection operation.

[0003] At present, the mainstream monitoring methods in the industry have obvious defects:

[0004] Acoustic monitoring method: usually a single frequency band acoustic sensor is used, and the signal is easily disturbed by the background noise generated by the mechanical vibration, steam flow and combustion in the boiler, resulting in low discrimination and high false alarm rate.

[0005] Infrared thermal imaging monitoring method: this method judges the leakage by monitoring the temperature anomaly of the tube wall. However, due to the hysteresis of heat transfer, the temperature response is not timely, and the early micro-leakage (micro-leakage) sound pressure level decays quickly, so the detection rate is low. When the system detects obvious temperature anomaly, the leakage has been expanded, and the best processing opportunity is missed.

[0006] Manual inspection method: this method depends heavily on the personal experience and responsibility of the inspection personnel, and not only has high labor intensity, but also cannot be covered all-weather and all-round, and early micro-leakage in hidden areas cannot be found.

[0007] In summary, the existing technology has defects in sensitivity, accuracy, real-time and positioning accuracy of monitoring, and cannot effectively solve the problems of early micro-leakage discovery and accurate positioning. Therefore, an advanced and reliable automatic monitoring technology is urgently needed. SUMMARY

[0008] The present application provides an automatic monitoring method and device for boiler heating surface tube burst, which aims to solve the problems of single monitoring method, high false alarm rate and low early micro-leakage detection rate in the prior art.

[0009] The automatic monitoring method for boiler heating surface tube burst according to the first aspect of the present application comprises the following steps:

[0010] S1: Multi-source signal acquisition step, through the sound wave sensor array, infrared thermal imaging module and pressure, temperature, flow sensor, the sound wave signal, infrared thermal image and operating condition parameters of the boiler heating surface are synchronously collected;

[0011] S2: Data preprocessing and feature extraction step, the collected multi-source signals are preprocessed, the spatial features of the infrared thermal image are extracted through a model based on a convolutional neural network, the time series features of the sound wave signal are extracted through a model based on a long short-term memory network, and the physical features of the operating condition parameters are extracted, and a space-time consistent data alignment matrix is constructed according to the time stamps of each signal;

[0012] S3: Double model parallel prediction step, the operating condition parameters are analyzed through a physical rule model based on the laws of thermodynamics, fluid mechanics and acoustics, and a theoretical physical prediction result is output, the spatial features and time series features are fused and analyzed through a data-driven CNN-LSTM hybrid model, and a data model prediction result is output;

[0013] S4: Dynamic weighted fusion step, the confidence is dynamically calculated based on the sound wave signal-to-noise ratio of the sound wave, and the theoretical physical prediction result of the physical rule model and the data model prediction result of the CNN-LSTM hybrid model are weighted and fused according to the confidence to obtain a preliminary fusion result;

[0014] S5: Consistency inspection and feedback calibration step, the preliminary fusion result is subjected to multi-source consistency inspection, the multi-source consistency inspection includes sound wave-infrared position cross verification, thermodynamic constraint verification based on physical laws and sound-heat coupling energy verification;

[0015] If the multi-source consistency inspection passes, a final prediction score is generated;

[0016] If the multi-source consistency inspection does not pass, the weight is punished and updated, and the dynamic weighted fusion step is returned to recalculate until the inspection passes;

[0017] S6: Risk response output step, according to the final prediction score, the pipe burst risk level is determined, and the three-dimensional position information and the corresponding control instruction are output to the distributed control system.

[0018] The automatic monitoring method for the boiler heating surface pipe burst according to the embodiment of the application has at least the following beneficial effects:

[0019] 1. More comprehensive monitoring dimensions, high early detection sensitivity: The monitoring method uses "sound wave + infrared + pressure" multi-source information fusion, and the CNN-LSTM model is used for spatio-temporal joint identification of "sound wave frequency domain features-temperature field gradient change", which overcomes the limitation of traditional methods that only pay attention to a single dimension. It can effectively capture the characteristics of the early micro-cracks of the explosion tube, and the detection sensitivity of micro-cracks is improved by 3 times.

[0020] 2. Physical and data dual constraints, significantly reduced false positive rate: The Attention mechanism of feature allocation and the physical guided fusion mechanism are innovatively introduced into the CNN-LSTM model. In particular, by defining a physical rule constraint function (such as a heat conduction loss term), the "illusion" and false positive problems that may be caused by pure data-driven models that violate physical laws are effectively solved. The detection rate of micro-leakage (<0.5mm) is improved to more than 90%, and the false positive rate of leakage is controlled to less than 5%.

[0021] 3. Dynamic confidence weighting, strong adaptability and robustness: The unique dynamic confidence weighting mechanism based on sound wave signal-to-noise ratio (SNR) enables the system to intelligently judge the current signal quality. When the signal quality is good, the data model with high precision is focused on; in harsh conditions such as strong noise, the stable physical rule model is relied on more. This makes the model's recognition accuracy of the explosion tube in strong noise conditions improve to more than 90%, and the learning accuracy under good signal is as high as 97%, greatly improving the environmental adaptability and decision robustness of the system.

[0022] 4. Closed-loop consistency verification, accurate positioning and reliable system: A multi-source consistency verification closed-loop system driven by physical and data is designed. Through cross verification and feedback calibration of position, physical law and energy conversion, the internal self-consistency and correctness of the analysis results are ensured, and a high-precision positioning of ±0.5m is achieved for a long time. Even in the case of partial sensor failure, the system can still maintain more than 85% reliable decision-making ability, significantly improving the safety and reliability of the entire boiler monitoring system.

[0023] The automatic monitoring device for boiler heating surface explosion tube according to the second aspect of the present application comprises an external source sensing unit, an edge computing unit, an intelligent decision-making unit and an output control unit.

[0024] The external source sensing unit is used to collect boiler heating surface data, and the external source sensing unit comprises a sound wave sensor array arranged in a three-dimensional grid topology, an infrared thermal imaging module for acquiring the temperature field distribution of the heating surface, and pressure, temperature and flow sensors for acquiring the operating conditions.

[0025] The edge computing unit is connected with the exogenous sensing unit, and is used for pre-processing the collected multi-source signals, extracting time-frequency features of sound wave signals, spatial gradient features of infrared images and physical features of operating conditions, and constructing a time-space consistent data alignment matrix through a dynamic compensation engine to output pre-processed data.

[0026] The intelligent decision unit is connected with the edge computing unit, and the intelligent decision unit is used for making decisions on pipe explosion risks.

[0027] The physical rule model is used for analyzing operating conditions according to physical laws to generate theoretical physical prediction results.

[0028] The CNN-LSTM hybrid model is used for fusing and analyzing sound wave and infrared features to generate data model prediction results, and the CNN-LSTM hybrid model integrates an attention fusion mechanism based on physical constraints and a physical constraint regularization term.

[0029] The confidence weighting module is used for dynamically calculating confidence according to the signal-to-noise ratio of the sound wave signal, and weighting and fusing the output results of the physical rule model and the CNN-LSTM hybrid model according to the confidence.

[0030] The multi-source consistency verification module is used for verifying the consistency of position cross, physical law and energy coupling of the weighted and fused results, and triggering a feedback calibration mechanism to update the fusion weight when the verification fails.

[0031] The output control unit is connected with the intelligent decision unit, and the output control unit is used for generating and outputting three-dimensional coordinate information of the pipe explosion position, risk level and corresponding control instructions according to the final verification result.

[0032] The automatic monitoring device for boiler heating surface pipe explosion according to the embodiment of the present application has at least the above beneficial effects because it executes the automatic monitoring method for boiler heating surface pipe explosion of the first aspect embodiment, and will not be described here.

[0033] Additional aspects and advantages of the present application will be given in part in the following description, part will become apparent from the following description, or will be understood by practicing the present application. BRIEF DESCRIPTION OF DRAWINGS

[0034] The present application will be further described below in combination with the drawings and embodiments, in which:

[0035] Figure 1 Flowchart of the automatic monitoring method for boiler heating surface pipe explosion of some embodiments of the present application;

[0036] Figure 2 The overall logic diagram of the boiler heating surface tube burst automatic monitoring method of some embodiments of the present application;

[0037] Figure 3 The flowchart of data preprocessing of the boiler heating surface tube burst automatic monitoring method of some embodiments of the present application;

[0038] Figure 4 The output diagram of the physical model of the boiler heating surface tube burst automatic monitoring method of some embodiments of the present application;

[0039] Figure 5 The output diagram of the CNN-LSTM hybrid model of the boiler heating surface tube burst automatic monitoring method of some embodiments of the present application;

[0040] Figure 6 The flowchart of confidence calculation of the boiler heating surface tube burst automatic monitoring method of some embodiments of the present application;

[0041] Figure 7 The flowchart of multi-source consistency verification of the boiler heating surface tube burst automatic monitoring method of some embodiments of the present application. DETAILED DESCRIPTION

[0042] Firstly, the embodiment of the present application provides a boiler heating surface tube burst automatic monitoring device, which comprises an external source sensing unit, an edge computing unit, an intelligent decision-making unit and an output control unit.

[0043] Referring to Figure 1 The present application also provides a boiler heating surface tube burst automatic monitoring method, which is executed by the boiler heating surface tube burst automatic monitoring device of the above-mentioned embodiment, and comprises the following steps:

[0044] S1: multi-source signal acquisition step, through the sound wave sensor array, the infrared thermal imaging module and the pressure, temperature and flow sensors, the sound wave signals, the infrared thermal image and the running condition parameters of the boiler heating surface are synchronously collected;

[0045] S2: data preprocessing and feature extraction step, the collected multi-source signals are preprocessed, the spatial features of the infrared thermal image are extracted through the model based on the convolutional neural network, the time sequence features of the sound wave signals are extracted through the model based on the long short-term memory network, and the physical features of the running condition parameters are extracted, and the time and space consistent data alignment matrix is constructed according to the time stamp of each signal;

[0046] S3: Double model parallel prediction step, through the physical rule model based on thermodynamics, fluid mechanics and acoustic law, the operating condition parameters are analyzed, the theoretical physical prediction result is output, through the data driven CNN-LSTM hybrid model, the spatial features and time sequence features are fused and analyzed, and the data model prediction result is output;

[0047] S4: Dynamic weighted fusion step, the confidence is dynamically calculated based on the sound wave signal sound wave signal-to-noise ratio, and the theoretical physical prediction result of the physical rule model and the data model prediction result of the CNN-LSTM hybrid model are weighted and fused according to the confidence to obtain the preliminary fusion result;

[0048] S5: Consistency test and feedback calibration step, the preliminary fusion result is executed multi-source consistency test, the multi-source consistency test includes sound wave-infrared position cross verification, thermodynamic constraint verification based on physical law and sound-heat coupling energy verification;

[0049] If the multi-source consistency test passes, the final prediction score is generated;

[0050] If the multi-source consistency test does not pass, the weight is updated with punishment, and the dynamic weighted fusion step is returned to recalculate until the test passes;

[0051] S6: Risk response output step, according to the final prediction score, the pipe burst risk level is determined, and the three-dimensional position information and the corresponding control instruction are output to the distributed control system.

[0052] Referring to Figures 2 to 7 The functions of each unit and the internal algorithm model are explained as follows.

[0053] The exogenous sensing unit is responsible for comprehensively and real-timely collecting multi-source heterogeneous data during the operation of the boiler, which includes an acoustic sensor array, an infrared thermal imaging module and a sensor assembly.

[0054] The acoustic sensor array is arranged in a three-dimensional grid topology to accurately capture high-frequency acoustic signals generated by pipe burst leakage and perform three-dimensional positioning. Specifically, on the front wall, rear wall and two side walls of the boiler furnace, 3 layers are arranged along the height direction. Each layer uniformly arranges 4 high-temperature resistant acoustic sensors, and the spacing between the sensors in the same layer is 1.5 meters. The array can effectively capture acoustic signals with a frequency range of 20 kHz to 100 kHz.

[0055] The infrared thermal imaging module is used to monitor the temperature field changes of the heating surface pipe wall. The module is installed on the fire observation hole of the boiler through a specially designed water-cooled protective sleeve. It has a scanning function, and the scanning range can cover the entire water-cooled wall heating surface, so as to obtain the two-dimensional temperature field distribution map of the boiler heating surface in real time.

[0056] In order to obtain the real-time operation conditions of the boiler, the sensor assembly is also provided with other key physical sensors, including but not limited to differential pressure sensors, temperature sensors, flow sensors installed in the furnace or the steam-water system, etc. These sensors are used to obtain the operating pressure of the boiler, the medium temperature, the flow, the flue gas dust concentration and the pipe wall material and other related operating condition parameters, providing a data basis for subsequent physical model analysis and dynamic compensation.

[0057] The edge computing unit is responsible for preprocessing, feature extraction and space-time alignment of the raw signals collected from the external sensing unit, providing high-quality data input for intelligent decision-making. Its function flow includes signal preprocessing and feature extraction, and space-time data alignment.

[0058] Signal preprocessing and feature extraction include acoustic signal processing, infrared image processing, pressure signal processing and operating condition parameter acquisition.

[0059] Acoustic signal processing: first, wavelet transform (Wavelet Transform) is used to denoise the original acoustic signal, effectively filtering out background noise such as mechanical vibration and combustion during boiler operation. Then, short-time Fourier transform (STFT) is used for time-frequency analysis to generate a time-frequency spectrum of the acoustic signal. Based on the time-frequency spectrum, high-dimensional features are extracted through a convolutional neural network (CNN) model, including high-frequency impulse transient features, wide-frequency energy distribution patterns, voiceprint geometric structure features, high / low frequency energy ratios and spectral entropy, etc.

[0060] Infrared image processing: through spatial gradient calculation and time series change analysis, dynamic features in the infrared thermal image sequence are extracted by a long short-term memory network (LSTM) model, including hotspot coordinates, hotspot shape, hotspot diffusion direction and diffusion acceleration, etc.

[0061] Pressure signal processing: the signal collected by the pressure sensor is subjected to sliding difference processing to extract the main frequency feature signal when the pressure oscillates.

[0062] Operating condition parameter acquisition: directly acquire the physical quantities output by other sensors as operating condition parameters.

[0063] Space-time data alignment includes spatial positioning and dynamic compensation and alignment.

[0064] Spatial positioning: using an acoustic sensor array, the three-dimensional spatial coordinates of the burst pipe point are calculated by calculating the time difference of arrival (TDOA) of the burst pipe acoustic signal to different sensors.

[0065] Dynamic compensation and alignment: Due to the different response delays of different sensors, different sampling frequencies, and the fact that the propagation speed of sound waves in the furnace is affected by real-time changes in temperature and pressure, a dynamic compensation engine is introduced. Based on real-time acquired temperature, pressure and other working condition parameters, the engine dynamically corrects the sound speed model to improve the TDOA positioning accuracy, and calculates the physical response time difference between each signal source (sound wave, infrared, pressure). Based on this time difference, the engine constructs a time and space consistent data alignment matrix, aligning the original data of different sources and different time points to a unified time axis to form a one-to-one data pair for subsequent models.

[0066] The engine is an intelligent software module that internally solidifies the physical models of heat conduction, fluid mechanics and acoustic propagation. It receives real-time temperature and pressure data from the working condition sensor components, dynamically calculates the sound speed distribution map in different areas of the furnace. When the TDOA algorithm performs sound source positioning calculation, a fixed sound speed value is no longer called, but this real-time updated sound speed field, thus realizing dynamic compensation of spatial positioning. At the same time, the engine estimates the theoretical response delay of infrared, pressure and other signals relative to the sound wave signal based on the physical model.

[0067] Based on the calculated dynamic sound speed and time delay, the engine performs accurate time translation and interpolation resampling on each data stream, and finally outputs a time and space fully aligned data alignment matrix. In each row of this matrix (representing a time slice), there are sound wave characteristics, infrared characteristics and working condition parameters that describe the same physical event at the same time. This "clean, synchronized and consistent" data set is the fundamental guarantee for the subsequent intelligent decision unit to make accurate judgments. As shown in Figure 3 After processing by the unit, the output is normalized preprocessed data.

[0068] The intelligent decision unit is the core of the method of the present application, which adopts a strategy of dual driving by physical model and data model, parallel decision, weighted fusion and consistency verification to realize high-precision and high-robustness diagnosis. As shown in Figure 3 The internal logic includes physical rule model output, CNN-LSTM hybrid model prediction, confidence weighted fusion and multi-source consistency verification.

[0069] Among them, for the physical rule model, the model is based on the basic physical laws of thermodynamics, fluid mechanics and acoustics. It receives real-time temperature, pressure, flow rate, acoustic energy and other physical data provided by the edge computing unit, and performs calculations, finally outputting a set of prediction results based on physical theory, including: theoretical acoustic energy value, physical score, physical boundary for constraining data model, and position verification value for cross-validation.

[0070] As shown in Figure 5As shown, the CNN-LSTM hybrid model is a purely data-driven deep learning model. The model receives spatial features from infrared images extracted by CNN and temporal features from acoustic signal sequences extracted by LSTM. It employs a physically constrained attention fusion mechanism to weightedly fuse feature vectors from different time steps or spatial regions. This mechanism allows the model to adaptively focus more on physically important features (such as concentrated energy or dramatic gradient changes), improving the model's analytical efficiency and accuracy. This mechanism gives the model a "smart focus." It doesn't treat all spatiotemporal features equally, but rather, through a learnable weight network, it automatically allocates more computational resources and attention to physically more "suspicious" regions when fusing acoustic and infrared features. For example, when considering the temperature gradient of a region in an infrared image... When the abnormality increases, the Attention mechanism will increase the weight of the acoustic features of that region and the corresponding time period in the fusion process.

[0071] Regarding physical constraint regularization, in order to address the problem of "illusions" or false alarms that may arise from pure data models that violate physical laws, the total loss function ζ of the model is... total In this paper, a physical constraint regularization term ζ is introduced. reg Total loss function ζ total Defined as:

[0072] ζ total =ζ data +λζ reg ;

[0073] Where, ζ data The basic loss term can be the root mean square error or average variance between the data model predictions and the physical observations; λ is the regularization coefficient used to control the strength of the physical constraints; ζ reg This is a physical constraint regularization term, which can also be understood as a temperature conduction loss term. Its specific form is:

[0074]

[0075] This formula forces the model to predict a temperature field T that conforms to the physical laws of heat conduction. Where, T j It is the temperature predicted by the model at the j-th sampling point; It is the rate of change of temperature at that point over time; ζ is the Laplace operator, representing the curvature of the temperature at that point in space; α is the thermal diffusivity of the pipe wall material. When the model predicts a severe violation of the law of heat conduction, ζ... reg The value of will become very large, thus generating a huge gradient to correct the model parameters, guide the model to output a solution that conforms to the laws of physics, and effectively suppress false alarms.

[0076] Finally, the trained CNN-LSTM hybrid model outputs a set of data-based prediction results, including burst probability, burst location prediction, and weight of each feature (datascore).

[0077] As Figure 6 shown, for confidence weighted fusion, in order to dynamically balance the contribution of physical model and data model, a confidence weighted mechanism based on the signal-to-noise ratio (SNR) of acoustic wave signal is introduced, and the confidence is calculated by Sigmoid function:

[0078] confidence = 1 / (1+exp(-0.5*(SNR-SNR0)));

[0079] Wherein, the slope coefficient k = 0.5 is set, and the reference signal-to-noise ratio SNR0 = 10 dB. When the on-site signal quality is good (SNR > 10 dB), the confidence is close to 1, and the system trusts the data model more; when the signal quality is poor and the noise is large (SNR < 5 dB), the confidence is close to 0.3 or below, and the system turns to rely more on the robust physical rule model.

[0080] The fusion calculation formula is:

[0081] Final confidence = confidence*data score +(1-confidence)*phy score ;

[0082] The confidence is also dynamically updated. When the subsequent consistency check fails, the SNR will be adjusted penalized according to the error type, and the confidence value will be updated, and the weighted fusion will be performed again. For example:

[0083] Position error penalty: Penalati = 10 x (Pos error -0.5) (i.e. for each 0.1 mm of position error, the SNR increases by 1 dB, and the maximum is 20 dB), Pos error is the position error.

[0084] Physical constraint error penalty: Penalati = 15 x Phy error (i.e. for each 0.01 of physical constraint error, the SNR increases by 0.15 dB), Phy error is the physical constraint error.

[0085] Energy loss error penalty: Penalati = 0.2 x Energ error(i.e. energy loss error exceeds 10, SNR increases 2dB), Energ error is the energy loss error.

[0086] As Figure 7 shown in the flow chart of multi-source consistency verification, the preliminary result after weighted fusion is introduced into multi-source cross verification based on physical law for final consistency verification. This step builds a closed-loop system of detection-diagnosis-calibration-verification.

[0087] Thermodynamic physical constraint verification: verify whether the change gradient of pressure and temperature satisfies a certain relationship, which should satisfy: Phy error = |dp / dt + 2.7·dT / dt| < 0.1. dp / dt and dT / dt are the change gradients of pressure and temperature in t time, and the physical constraint error Phy error can be understood as a linear combination of the change gradients of pressure and temperature.

[0088] Acoustic-infrared position cross verification: verify whether the acoustic positioning coordinates Pa and infrared positioning coordinates Pt point to the same area, which should satisfy: Pos error = ||Pa-Pt|| < 0.5m. Pa is the acoustic positioning coordinates; Pt is the infrared positioning coordinates, and the position excess error Pos error can be understood as the Euclidean distance between the acoustic positioning coordinates and the infrared positioning coordinates of the infrared image.

[0089] Acoustic-thermal coupling energy verification: verify whether the acoustic energy E pred predicted by the model matches the theoretically calculated acoustic energy E theory , which should satisfy: Energ error = |E pred -E theory | < 50dB. E pred is the model predicted acoustic energy, E pred = K·(dT / dt) 2 ; E theory is the theoretical acoustic energy, and the energy loss error Energ error can be understood as the difference between the model predicted acoustic energy and the theoretically calculated acoustic energy.

[0090] Feedback calibration: when any of the verifications fails, the system will determine that the current result is not reliable, and will perform weight punishment on the corresponding constraint function. After returning to the previous step to update the confidence and weight, the fusion output is re-performed until all verifications pass finally.

[0091] The output control unit is responsible for converting the final verified diagnosis result into operable information and actions.

[0092] Information output: After the consistency check passes, the system generates the final prediction score (Final confidence ). At the same time, the three-dimensional coordinate information of the burst pipe position is output, and visualized on the human-computer interaction interface.

[0093] Risk classification and linkage control: According to the size of the final prediction score, the burst pipe risk is classified, and the distributed control system (DCS) is linked to perform the corresponding response action. The risk level and the corresponding response action are as follows:

[0094] [0, 0.3): low risk, the system continues to monitor, and the state indication is green.

[0095] [0.3, 0.6): medium risk, the system issues a yellow warning, prompting the operator to pay attention and adjust the parameters.

[0096] [0.6, 0.85): high risk, the system issues an orange alarm, automatically or prompts to perform a load reduction operation, and starts a backup pump.

[0097] [0.85, 1]: critical risk, the system issues a red alarm, and the DCS is linked to execute an emergency shutdown program.

[0098] Through the above embodiments, the present application constructs a complete automatic monitoring system from multi-source data acquisition, intelligent preprocessing, material-number dual model decision, dynamic weighted fusion to closed-loop verification and response control, which can accurately and reliably realize early warning and positioning of boiler heating surface burst pipe.

[0099] The technical scheme according to the embodiments of the present application has the following beneficial effects:

[0100] 1. More comprehensive monitoring dimension, high early detection sensitivity: The monitoring method of "sound wave + infrared + pressure" multi-source information fusion is adopted, and the CNN-LSTM model is used to identify the "sound wave frequency domain feature-temperature field gradient change" in space-time joint, which overcomes the limitation of traditional methods that only pay attention to a single dimension. It can effectively capture the characteristics of the early micro-cracks of the burst pipe, and the detection sensitivity of the micro-cracks is improved by 3 times.

[0101] 2. Physical and data dual constraints, significantly reduced false positive rate: The Attention mechanism of feature allocation and the fusion mechanism of physical guidance are innovatively introduced into the CNN-LSTM model. Especially by defining a physical rule constraint function (such as a heat conduction loss term), the "hallucinations" and false positives that may be produced by pure data-driven models that violate physical laws are effectively solved. The detection rate of micro-leakage (<0.5mm) is improved to more than 90%, and the false positive rate of leakage is controlled to be less than 5%.

[0102] 3. Dynamic confidence weighting, strong adaptability and robustness: The dynamic confidence weighting mechanism based on the sound wave signal-to-noise ratio (SNR) is unique, which enables the system to intelligently judge the current signal quality. When the signal quality is good, the high-precision data model is focused on; in the harsh working conditions such as strong noise, the stable physical rule model is relied on more. This makes the model's recognition accuracy of the burst pipe in strong noise conditions improve to more than 90%, and the learning accuracy under high-quality signals is as high as 97%, greatly improving the environmental adaptability and decision robustness of the system.

[0103] 4. Closed-loop consistency check, accurate positioning and reliable system: A multi-source consistency check closed-loop system driven by physical and data is designed. Through cross verification and feedback calibration of position, physical law and energy conversion, the internal self-consistency and correctness of the analysis result are ensured, and high-precision positioning of ±0.5m is achieved for a long time. Even in the case of partial sensor failure, the system can still maintain more than 85% reliable decision-making ability, significantly improving the safety and reliability of the entire boiler monitoring system.

[0104] Examples of the above-described embodiments are shown in the accompanying drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described above by referring to the drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation on the present application.

[0105] In the description of the present application, it should be understood that the orientation description, such as the orientation or position relationship indicated by up, down, front, back, left, right, etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0106] In the description of the present application, the meaning of several is one or more, the meaning of multiple is more than two, greater than, less than, more than, etc. are not included in the number, above, below, etc. are understood to include the number. If it is described as first, second, it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the order of indicated technical features.

[0107] In the description of the present application, unless otherwise explicitly limited, the words such as setting, installing, connecting, etc. should be broadly understood, and those skilled in the art can reasonably determine the specific meaning of the above words in the present application in combination with the specific content of the technical solution.

[0108] The embodiments of the present application are described in detail above with reference to the drawings, but the present application is not limited to the above-described embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present application.

Claims

1. An automated monitoring method for boiler heating surface tube rupture, characterized in that, Includes the following steps: S1: Multi-source signal acquisition step, which uses an acoustic sensor array, an infrared thermal imaging module, and pressure, temperature, and flow sensors to simultaneously acquire acoustic signals, infrared thermal images, and operating parameters of the boiler heating surface. S2: Data preprocessing and feature extraction steps: preprocess the collected multi-source signals, extract the spatial features of the infrared thermal image using a convolutional neural network-based model, extract the time series features of the acoustic signal using a long short-term memory network-based model, extract the physical features of the operating condition parameters, and construct a spatiotemporally consistent data alignment matrix based on the timestamps of each signal. S3: The dual-model parallel prediction step analyzes the operating condition parameters using a physical rule model based on the laws of thermodynamics, fluid mechanics, and acoustics, and outputs theoretical physical prediction results. It also uses a data-driven CNN-LSTM hybrid model to fuse and analyze the spatial and time series features, and outputs data model prediction results. S4: Dynamic weighted fusion step, based on the acoustic signal-to-noise ratio of the acoustic signal, the confidence level is dynamically calculated, and the theoretical physical prediction results of the physical rule model and the data model prediction results of the CNN-LSTM hybrid model are weighted and fused according to the confidence level to obtain a preliminary fusion result; S5: Consistency verification and feedback calibration step, performing multi-source consistency verification on the preliminary fusion results, including acoustic-infrared position cross-verification, thermodynamic constraint verification based on physical laws, and acoustic-thermal coupling energy verification; If the multi-source consistency test passes, a final prediction score is generated; If the multi-source consistency test fails, the weights are updated penalized, and the process returns to the dynamic weighted fusion step for recalculation until the test passes. S6: Risk response output step: Based on the final prediction score, determine the risk level of pipe burst, and output three-dimensional position information and corresponding control commands to the distributed control system.

2. The automated monitoring method for boiler heating surface tube rupture according to claim 1, characterized in that, In the dynamic weighted fusion step, the confidence level is calculated based on the acoustic signal-to-noise ratio (SNR) of the acoustic signal using the Sigmoid function. When the SNR is high, the weight of the CNN-LSTM hybrid model is higher; when the SNR is low, the weight of the physical rule model is higher.

3. The automated monitoring method for boiler heating surface tube rupture according to claim 2, characterized in that, The Sigmoid function is: confidence = 1 / (1 + exp(-0.5*(SNR - SNR0))), where confidence is the confidence level, SNR is the acoustic signal-to-noise ratio, and SNR0 is the reference signal-to-noise ratio; the dynamic weighted fusion step follows the following formula: Final confidence =confidence*data score +(1-confidence)*phy score , wherein the Final confidence The data is the output of the weighted fusion. score The data model prediction result of the CNN-LSTM hybrid model, the phy score The theoretical physical prediction result of the physical rule model is given, and the confidence is the confidence level.

4. The automated monitoring method for boiler heating surface tube rupture according to claim 1, characterized in that, The CNN-LSTM hybrid model includes a loss function, which comprises a basic loss term and a physical constraint regularization term. The physical constraint regularization term is used to penalize prediction results that do not conform to the physical laws of heat conduction.

5. The automated monitoring method for boiler heating surface tube rupture according to claim 4, characterized in that, ζ total =ζ data +λζ reg , where the ζ total Let ζ be the loss function. data The basic loss term is the mean squared error or average variance of the features in the data model; λ is the introduced physical regularization term, and ζ is the base loss term. reg Let be the regularization coefficient, which satisfies: Where M represents the total number of sampling points selected on the boiler heating surface for calculation, and T is the temperature of the j-th point on the boiler heating surface.

6. The automated monitoring method for boiler heating surface tube rupture according to claim 1, characterized in that, In the multi-source consistency test, the Euclidean distance between the acoustic positioning coordinates of the acoustic signal and the infrared positioning coordinates of the infrared thermogram is the position error; the linear combination of the pressure change gradient of the operating condition parameter and the temperature change gradient of the operating condition parameter is the physical constraint error; and the difference between the acoustic energy predicted by the model and the acoustic energy calculated by theory is the energy loss error. The acoustic-infrared position cross-validation requires that the position deviation error be less than a preset position error threshold. The thermodynamic constraint verification requirement is less than a preset physical error threshold; The acoustic-thermal coupling energy verification requires that the energy loss error be less than a preset energy error threshold.

7. The automated monitoring method for boiler heating surface tube rupture according to claim 6, characterized in that, The position error threshold is 0.5m, the physical error threshold is 0.1m, and the energy error threshold is 50dB.

8. The automated monitoring method for boiler heating surface tube rupture according to claim 6, characterized in that, The punitive update includes: For every 0.1 mm of positional error exceeding the tolerance, the acoustic signal-to-noise ratio increases by 1 dB, with a maximum increase of 20 dB. For every 0.01 ohm increase in the physical constraint error, the acoustic signal-to-noise ratio increases by 0.15 dB. For every 10 units of energy error, the acoustic signal-to-noise ratio increases by 2 dB.

9. The automated monitoring method for boiler heating surface tube rupture according to claim 1, characterized in that, In the data preprocessing and feature extraction steps, constructing a spatiotemporally consistent data alignment matrix based on the timestamps of each signal includes: Wavelet transform is used to filter out background noise from the acoustic signal, the spatial coordinates of the burst point are calculated using a positioning method based on the time difference of arrival, and the data alignment matrix is ​​constructed by a dynamic compensation engine.

10. An automated monitoring device for boiler heating surface tube rupture, characterized in that, It includes an external sensing unit, an edge computing unit, an intelligent decision-making unit, and an output control unit; The external sensing unit is used to collect data of the boiler heating surface. The external sensing unit includes an acoustic sensor array arranged in a three-dimensional mesh topology, an infrared thermal imaging module for acquiring the temperature field distribution of the heating surface, and pressure, temperature and flow sensors for acquiring operating conditions. The edge computing unit is connected to the external sensing unit and is used to preprocess the acquired multi-source signals, extract the time-frequency features of the acoustic signal, the spatial gradient features of the infrared image, and the physical features of the operating conditions, and construct a spatiotemporally consistent data alignment matrix through a dynamic compensation engine to output the preprocessed data. The intelligent decision-making unit is connected to the edge computing unit. The intelligent decision-making unit is used to make decisions on the risk of pipe bursting. The intelligent decision-making unit includes a physical rule model, a CNN-LSTM hybrid model, a confidence weighting module, and a multi-source consistency verification module. The physical rule model is used to analyze operating conditions based on physical laws and generate theoretical physical prediction results. The CNN-LSTM hybrid model is used to fuse and analyze acoustic and infrared features to generate data model prediction results. The CNN-LSTM hybrid model integrates a physical constraint-based attention fusion mechanism and a physical constraint regularization term. The confidence weighting module is used to dynamically calculate the confidence level based on the signal-to-noise ratio of the acoustic signal, and to perform weighted fusion of the output results of the physical rule model and the CNN-LSTM hybrid model based on the confidence level. The multi-source consistency verification module is used to perform consistency verification on the weighted fusion results in terms of positional intersection, physical laws and energy coupling, and triggers a feedback calibration mechanism to update the fusion weights when the verification fails. The output control unit is connected to the intelligent decision-making unit. The output control unit is used to generate and output the three-dimensional coordinate information of the burst pipe location, the risk level, and the corresponding control commands based on the final inspection results. Furthermore, the automated monitoring device for boiler heating surface tube rupture is configured to perform the automated monitoring method for boiler heating surface tube rupture according to any one of claims 1 to 9.

Images