Intelligent leakage detection system for flue gas waste heat recovery heat exchanger

Abstract

This invention relates to the field of industrial flue gas waste heat recovery equipment technology, specifically an intelligent leak detection system for flue gas waste heat recovery heat exchangers. The system includes a multi-source data acquisition module, a data preprocessing and adaptive correction module, a leak feature identification and classification module, a leak location module, an early warning and coordinated response module, and a full lifecycle health management module. The invention acquires comprehensive core leak feature parameters and operating condition parameters through the multi-source data acquisition module. After eliminating operating condition interference through the data preprocessing and adaptive correction module, it achieves graded leak identification through the leak feature identification and classification module with a built-in CNN-LSTM hybrid deep learning model. The leak location module accurately locates the leak, the early warning and coordinated response module executes graded responses in a coordinated manner, and the full lifecycle health management module enables equipment status prediction and preventative maintenance, forming a complete closed-loop system for leak detection, location, response, and prediction.

Description

Technical Field

[0001] This invention relates to the field of industrial flue gas waste heat recovery equipment technology, specifically an intelligent detection system for leaks in flue gas waste heat recovery heat exchangers. Background Technology

[0002] In industrial sectors such as thermal power, steel, chemical, and waste incineration, flue gas waste heat recovery heat exchangers are core equipment for achieving energy conservation, emission reduction, and cascade utilization of waste heat. They facilitate heat exchange between high-temperature flue gas and media such as water, steam, and heat transfer oil through heat exchange tubes, significantly reducing exhaust heat loss and improving energy efficiency. However, industrial flue gas contains large amounts of SO2 and NO. X In harsh dust conditions, leaks such as corrosion perforation and stress cracking are highly likely to occur. Current technologies for detecting heat exchanger leaks have the following core technical deficiencies, failing to meet the requirements for safe and stable operation in industrial settings: Poor real-time detection and insufficient micro-leakage identification capabilities: Most methods rely on manual inspections, periodic hydrostatic tests, and offline non-destructive testing, which cannot achieve 24-hour online continuous detection. Micro-leakages in the early stages cannot be detected in time, and are only discovered after the leakage has expanded. This can easily lead to large-scale corrosion of heat exchange tubes, unplanned equipment shutdowns, and even safety accidents such as flue gas poisoning and boiler tube rupture.

[0003] Low detection reliability and high false alarm rate: Existing online detection solutions mostly use single parameter detection, such as only detecting flue gas humidity, tube pressure difference or water conductivity. However, industrial site conditions such as boiler load fluctuations, fuel type changes, ambient temperature and humidity changes, and flue gas original moisture content fluctuations can all cause non-leakage changes in the above parameters, which can easily trigger false alarms. This can cause on-site personnel to become desensitized to alarm signals and fail to respond in time when a real leak occurs.

[0004] Lacking precise leak location capabilities, resulting in high maintenance costs: Existing detection systems can only determine whether the heat exchanger as a whole has leaked, but cannot locate the specific tube bank or individual heat exchange tube that is leaking. During equipment maintenance, all heat exchange tubes need to be checked one by one, which significantly extends downtime for maintenance and increases maintenance costs and production losses.

[0005] Lacking predictive capabilities, the system can only passively handle faults: Existing technologies can only provide alarm response after a leak occurs, but cannot predict the corrosion and aging trend of heat exchange tubes based on the equipment's full-cycle operating data, cannot provide early warning of leak risks, and cannot perform preventative maintenance, resulting in a persistently high rate of unplanned equipment downtime.

[0006] Poor adaptability to operating conditions and sharp drop in accuracy under complex conditions: The detection thresholds of existing detection systems are mostly fixed and cannot be adaptively adjusted according to real-time operating conditions on site. Under complex operating conditions such as boiler load changes, start-up and shutdown, and fuel switching, the detection accuracy drops significantly and cannot adapt to the changing operating environment of industrial sites. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent detection system for leaks in flue gas waste heat recovery heat exchangers, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A smart detection system for leaks in a flue gas waste heat recovery heat exchanger includes a multi-source data acquisition module, a data preprocessing and adaptive correction module, a leak feature identification and classification judgment module, a leak location module, an early warning and linkage response module, and a full life cycle health management module. The multi-source data acquisition module is used to collect in real time the pressure, flow rate, conductivity, and pH value of the heat exchanger tube-side medium, the humidity, SO2 concentration, temperature, and pressure parameters of the shell-side flue gas, as well as the equipment operating condition parameters such as the boiler real-time load, induced draft fan frequency, and flue gas damper opening. The data preprocessing and adaptive correction module is connected to the multi-source data acquisition module. It is used to filter and denoise the acquired multi-source data, remove outliers, and has a built-in working condition-parameter mapping model. It calculates the benchmark values ​​of each detection parameter under the current working condition based on the real-time working condition parameters, performs dynamic compensation and correction on the preprocessed detection data, eliminates parameter deviations caused by non-leakage working condition fluctuations such as boiler load fluctuations, fuel changes, and ambient temperature and humidity changes, and outputs a standardized detection dataset. The leakage feature identification and classification module is connected to the data preprocessing and adaptive correction module. It has a built-in CNN-LSTM hybrid deep learning model trained on multi-source data. It is used to extract the spatiotemporal features of leakage based on the standardized detection dataset, identify the leakage conditions and complete the three-level classification of leakage risk, and output the leakage classification results and leakage feature data. The leak location module is connected to the data preprocessing and adaptive correction module and the leak feature recognition and classification judgment module, respectively. It is used to match the spatiotemporal distribution characteristics of the leak based on the standardized detection dataset and the leak feature data to locate the specific zone where the leak occurs and the location of the heat exchange tube. The early warning and linkage response module is communicatively connected to the leakage feature identification and classification judgment module and the leakage location module, respectively. It is used to output the corresponding level of early warning alarm signal according to the leakage classification result and the leakage location result, and to link and execute the corresponding operating condition adjustment and safety protection actions. The full life cycle health management module is connected to the leak feature identification and classification module. It is used to build a corrosion and aging model of the heat exchanger tubes based on historical operating data and real-time detection data, predict the remaining service life, and output preventive maintenance suggestions.

[0009] In a preferred embodiment of the present invention, the multi-source data acquisition module includes a tube-side medium detection unit, a shell-side flue gas detection unit, and an operating condition parameter acquisition unit; the tube-side medium detection unit is further used to acquire the inlet and outlet temperature parameters of the heat exchanger tube-side medium; the shell-side flue gas detection unit is further used to acquire the inlet and outlet flow rates, oxygen content, and NO of the heat exchanger shell-side flue gas. x Concentration and dust concentration parameters; the operating condition parameter acquisition unit is also used to acquire fuel type parameters.

[0010] In a preferred embodiment of the present invention, the data preprocessing and adaptive correction module includes a data filtering unit, an outlier removal unit, and an operating condition adaptive correction unit; the data filtering unit uses a Kalman filter algorithm to filter and denoise the collected data; the outlier removal unit completes outlier identification, removal, and data completion based on the 3σ criterion; and the operating condition adaptive correction unit has a built-in operating condition-parameter mapping model that is a parameter benchmark calculation model constructed based on historical operating data using a multiple linear regression algorithm.

[0011] In a preferred embodiment of the present invention, the leakage feature identification and classification judgment module sets three judgment thresholds based on historical leakage operating data, namely, a micro-leakage early warning threshold corresponding to a leakage probability of 30%-60%, a medium-leakage alarm threshold corresponding to a leakage probability of 60%-90%, and a large-leakage emergency shutdown threshold corresponding to a leakage probability ≥90%. The leakage risk classification judgment can be completed according to the threshold range in which the leakage probability value is located.

[0012] In a preferred embodiment of the present invention, the leakage feature identification and classification judgment module is equipped with a CNN-LSTM hybrid deep learning model, wherein the CNN network extracts the spatial correlation features of multi-source detection data, the LSTM network extracts the time series change features of the detection data, and the model is pre-trained based on historical leakage condition data, normal condition data, and various interference condition data, and can distinguish between leakage conditions and non-leakage interference conditions such as normal condition fluctuations, fuel changes, load adjustments, and environmental temperature and humidity changes.

[0013] In a preferred embodiment of the present invention, the leak location module includes a heat exchange tube partition management unit and a spatiotemporal feature matching unit; the heat exchange tube partition management unit is used to divide the heat exchange tubes of the heat exchanger into several independent detection partitions with corresponding independent parameter monitoring points according to the tube row and the process; the spatiotemporal feature matching unit has a built-in leak feature library, which can extract the spatiotemporal change features of the detection parameters of each partition, match them with the leak feature library, and locate the partition and single heat exchange tube where the leak is located.

[0014] In a preferred embodiment of the present invention, the early warning and linkage response module includes a graded early warning unit and a linkage control unit; the graded early warning unit can output graded early warning signals such as on-site audible and visual early warning, remote platform early warning, and emergency shutdown alarm according to the leakage classification result; the linkage control unit can communicate with the boiler DCS system and the on-site actuator, and adjust the boiler load, flue gas damper opening, and pipe-side medium valve opening according to the leakage classification result, and trigger the shutdown protection program in emergency conditions.

[0015] In a preferred embodiment of the present invention, the full life cycle health management module includes a corrosion rate calculation unit, a remaining life prediction unit, and a maintenance suggestion generation unit. The corrosion rate calculation unit is used to calculate the real-time corrosion rate and cumulative corrosion amount of the heat exchange tube based on historical and real-time detected tube-side medium pH value, conductivity, and flue gas corrosivity parameters, using Faraday's law and the corrosion weight loss method. The remaining life prediction unit has a built-in LSTM corrosion aging prediction model, which can predict the remaining service life of the heat exchange tube based on the real-time corrosion rate and historical corrosion trends, combined with the design wall thickness and minimum allowable wall thickness of the heat exchange tube. The maintenance suggestion generation unit can output corresponding preventive maintenance and replacement suggestions based on the prediction results.

[0016] In a preferred embodiment of the present invention, a data storage and visualization module is further included. The data storage and visualization module is communicatively connected to a multi-source data acquisition module, a data preprocessing and adaptive correction module, a leak feature identification and classification judgment module, a leak location module, an early warning and linkage response module, and a full life cycle health management module. It is used to store full operating data, detection data, judgment results, handling records and maintenance data, and at the same time, it displays the heat exchanger operating status, leak risk level, location results and remaining life prediction data in real time through a visualization interface.

[0017] In a preferred embodiment of the present invention, a remote interaction module is further included. The remote interaction module is communicatively connected to the data storage and visualization module and is used to synchronize detection data, alarm information, and maintenance suggestions to the remote operation and maintenance platform. At the same time, it receives remotely issued control commands and model update parameters to realize remote operation and maintenance and system iteration.

[0018] One technical advantage of this invention is that: The core technical solution of this invention is the system architecture described in the claims. It acquires full-dimensional core leakage characteristic parameters and operating condition parameters through a multi-source data acquisition module. After eliminating operating condition interference through a data preprocessing and adaptive correction module, it achieves graded leakage identification through a leakage feature identification and graded judgment module with a built-in CNN-LSTM hybrid deep learning model. Simultaneously, it completes accurate leakage location through a leakage location module. It performs graded handling through an early warning and linkage handling module. At the same time, it achieves equipment status prediction and preventive maintenance through a full life cycle health management module, forming a complete closed-loop system for leakage detection-location-handling-prediction. To address the issues of poor real-time detection and insufficient micro-leakage identification capabilities in existing technologies, this invention collects core leakage characteristic parameters such as tube conductivity, pH value, and shell flue gas humidity through a multi-source data acquisition module. Combined with a CNN-LSTM hybrid deep learning model, it can capture subtle changes in multiple parameters caused by micro-leakage, enabling 24-hour online continuous detection. It can identify micro-leakage faults in the early stages of heat exchange tubes, address them in advance to prevent the leakage fault from escalating, and ensure the safe and stable operation of the equipment.

[0019] To address the problems of "low detection reliability and high false alarm rate" in existing technologies, this invention uses a data preprocessing and adaptive correction module with a built-in operating condition-parameter mapping model to dynamically compensate for detection data based on real-time operating conditions, eliminating non-leakage parameter deviations caused by operating condition fluctuations. At the same time, through a pre-trained multi-class deep learning model, it effectively distinguishes between leakage operating conditions and non-leakage operating condition fluctuations, significantly improving the accuracy and reliability of detection.

[0020] To address the problems of existing technologies such as "lack of accurate leak location capability and high maintenance costs": This invention uses a partitioned leak location module to divide the heat exchange tube into multiple independent detection zones with monitoring points. By matching the leak feature library with a spatiotemporal feature matching unit, the leak can be accurately located in the tube row zone or even a single heat exchange tube, which can significantly reduce maintenance costs and production downtime losses.

[0021] To address the problem that existing technologies "lack predictive capabilities and can only passively handle faults": This invention uses a full life cycle health management module, based on real-time corrosion rate calculation and LSTM corrosion aging prediction model, to accurately predict the remaining service life of heat exchange tubes and output targeted preventive maintenance suggestions, thereby realizing the transformation from passive fault handling to proactive preventive maintenance and significantly reducing the rate of unplanned equipment downtime.

[0022] To address the problems of poor adaptability to operating conditions and sharp drop in accuracy under complex operating conditions in existing technologies, this invention uses an adaptive correction module to dynamically compensate the detection data based on operating parameters such as real-time boiler load and flue gas damper opening. Even under complex operating conditions such as boiler load changes, start-up and shutdown, and fuel switching, it can still maintain high detection accuracy and perfectly adapt to the complex operating environment of industrial sites. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the module connection of the intelligent detection system for leaks in the flue gas waste heat recovery heat exchanger described in this invention. Detailed Implementation

[0024] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present application.

[0025] The embodiments of this application will now be described in detail. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0026] The terms "first" and "second" in the specification and claims of this application may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise stated, "multiple" means two or more. Furthermore, "and / or" in the specification and claims indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0027] Please refer to Figure 1 The intelligent detection system for leaking flue gas waste heat recovery heat exchangers described in this embodiment is applied to the flue gas waste heat recovery heat exchanger at the tail end of a 300MW thermal power boiler. The shell side of the heat exchanger is supplied with high-temperature flue gas from the tail end of the boiler, and the tube side is supplied with demineralized water to heat the boiler feedwater, thereby realizing the recovery of flue gas waste heat.

[0028] The specific embodiments of the present invention are clearly and completely described below. The working principle, connection method and effect of all components are clear and achievable. Those skilled in the art can repeat the technical solution of the present invention, solve its technical problems and achieve the expected technical effects by following the contents of this specification without creative effort.

[0029] The system adopts a modular distributed architecture. Each module communicates with the other via industrial Ethernet. Data transmission uses the standard Modbus-TCP protocol, which is compatible with mainstream DCS systems and industrial control equipment in industrial settings.

[0030] See Figure 1 The overall system workflow is as follows: After system startup, the multi-source data acquisition module 1 collects full-dimensional detection data in real time at a preset frequency and transmits it to the data preprocessing and adaptive correction module; the data preprocessing and adaptive correction module 2 completes data preprocessing and adaptive compensation for operating conditions, and outputs a standardized detection dataset; the standardized dataset is synchronously input to the leakage feature identification and classification judgment module 3 and the leakage location module 4, which respectively complete the leakage classification judgment and leakage location location; the early warning and linkage response module 5 executes graded early warning and linkage control based on the judgment and location results; the full life cycle health management module 6 completes corrosion rate calculation and remaining life prediction based on the full data; all data is synchronized to the data storage and visualization module 7, and remote operation and maintenance is realized through the remote interaction module 8.

[0031] The multi-source data acquisition module is the core of the system's data source, including a tube-side medium detection unit, a shell-side flue gas detection unit, and an operating condition parameter acquisition unit. All detection units are connected to the data preprocessing and adaptive correction module via RS485 bus or industrial Ethernet. The sampling frequency is set to 1 time / second to ensure the real-time performance of the data.

[0032] The tube-side medium detection unit includes pressure sensors, temperature sensors, and electromagnetic flow sensors installed on the horizontal straight pipe sections at the inlet and outlet of the heat exchanger tubes, respectively. It also includes online conductivity meters and online pH meters installed at the lowest points of the straight pipe sections at the inlet and outlet, meeting the requirement of full-pipe, bubble-free installation. The pressure sensors use diffused silicon pressure transmitters with a measurement accuracy of ±0.2%FS, used to collect the pressure of the medium at the inlet and outlet of the tubes; a leak will result in an abnormally large pressure difference between the inlet and outlet. The temperature sensors use platinum resistance PT100 temperature transmitters with a measurement accuracy of ±0.1℃, used to collect the temperature of the medium at the inlet and outlet, assisting in the judgment of abnormal heat exchange efficiency. The electromagnetic flow sensors have a measurement accuracy of ±0.5%FS, used to collect the flow rate of the medium at the inlet and outlet; a leak will result in an abnormally large flow rate difference between the inlet and outlet. The online conductivity and pH meters are used to collect the conductivity and pH value of the medium in the tubes. When corrosive flue gas from the shell side leaks into the tubes, the medium conductivity will increase significantly and the pH value will decrease significantly, which are the core characteristic parameters of leakage.

[0033] Shell-side flue gas detection unit: Includes temperature sensors, pressure sensors, Pitot tube flue gas flow sensors, and zirconia oxygen content analyzers installed at the shell-side inlet of the heat exchanger, and temperature sensors, pressure sensors, capacitive humidity sensors, and ultraviolet differential SO2 / NO2 detectors installed at the shell-side outlet.x Concentration analyzer, laser dust concentration detector, and zirconia oxygen content analyzer. Among them, the capacitive humidity sensor has a measurement range of 0-100%RH and an accuracy of ±2%RH. It is used to detect the humidity at the flue gas outlet. When the tube-side medium leaks to the shell side, the flue gas humidity will increase significantly, making it a core auxiliary characteristic parameter of the leak; SO2 / NO... x The concentration analyzer is used to detect the concentration of acidic gases in flue gas and to help determine the concentration changes caused by the dissolution of acidic gases in leaked water; the other sensors are used to collect basic operating parameters of the flue gas to provide data support for subsequent adaptive correction.

[0034] Operating Parameter Acquisition Unit: Establishes a two-way communication connection with the boiler DCS system via industrial Ethernet. The communication protocol is compatible with the OPC UA standard. It is used to acquire core operating parameters of the boiler in real time, such as real-time load, fuel type, coal feed rate, induced draft fan frequency, and flue gas damper opening. It provides reference data for adaptive correction of operating conditions and eliminates the interference of operating condition fluctuations on the detection parameters.

[0035] The data preprocessing and adaptive correction module is the core data processing unit of the system. It communicates bidirectionally with the multi-source data acquisition module and includes a data filtering unit, an outlier removal unit, and an adaptive correction unit. The module is implemented using an industrial-grade edge computing gateway with a computing power of no less than 4 TOPS, enabling real-time data processing.

[0036] Data filtering unit: The Kalman filter algorithm is used to filter and denoise the real-time data collected, eliminating high-frequency noise data caused by electromagnetic interference and sensor mechanical vibration in the industrial field, and outputting smooth and stable detection data to avoid noise data interfering with subsequent leak identification.

[0037] Outlier Removal Unit: Based on the 3σ criterion, outlier identification and removal are performed on the filtered data. Abrupt outliers exceeding three times the normal fluctuation range (such as jump data caused by sensor failure or line interference) are removed. At the same time, the mean of adjacent data is used to fill in the data to complete the data and avoid misjudgment caused by abnormal data.

[0038] Operating condition adaptive correction unit: It has a built-in preset operating condition-parameter mapping model. This model is constructed based on the historical operating data of the heat exchanger for the past 3 years through a multiple linear regression algorithm. The independent variables of the model are the real-time load of the boiler, flue gas flow rate, ambient temperature and humidity, and fuel calorific value. The dependent variables are the theoretical benchmark values ​​of flue gas humidity, tube pressure difference, tube medium conductivity, and inlet and outlet temperature difference under the current operating conditions. The model's multiple linear regression equation is: Y = β0 + β1X1 + β2X2 + β3X3 + β4X4 + ε, where Y is the theoretical baseline value of the detection parameters, β0 is the constant term, β1-β4 are the regression coefficients of each independent variable, X1-X4 are the boiler real-time load, flue gas flow rate, ambient temperature and humidity, and fuel calorific value, respectively, and ε is the random error term. The model solves for the regression coefficients using the least squares method, and the goodness of fit R0 is... 2 ≥0.95.

[0039] During operation, the calibration unit calculates the theoretical baseline values ​​of each detection parameter under the current operating condition based on the real-time collected operating parameters. It then dynamically compensates and corrects the pre-processed detection data to eliminate non-leakage parameter deviations caused by operating condition fluctuations. For example, when the boiler load increases from 50% to 100% of the rated load, the flue gas flow rate and initial moisture content will increase simultaneously. At this time, the calibration unit will dynamically compensate the flue gas humidity detection value according to the load change, eliminating the impact of increased humidity caused by the increased load and avoiding misjudging normal operating condition fluctuations as leaks. Finally, the calibration unit outputs a standardized detection dataset, which is transmitted to the leak feature identification and grading judgment module and the leak location module, respectively.

[0040] The leakage feature identification and classification judgment module is the core of the system's leakage identification. It is implemented using an industrial-grade AI edge computing host, with a built-in pre-trained CNN-LSTM hybrid deep learning model. It communicates bidirectionally with the data preprocessing and adaptive correction module, and the output ends are connected to the leakage location module, the early warning and linkage response module, and the full life cycle health management module, respectively.

[0041] The CNN-LSTM hybrid deep learning model built into this module has the following structure: The input layer has 12-dimensional feature inputs, corresponding to the 12 core detection parameters of the standardized detection dataset; the CNN layer contains two one-dimensional convolutional layers with kernel sizes of 3 and 5 respectively, a stride of 1, and uses the ReLU activation function to extract spatial correlation features between multi-source detection data; the pooling layer uses max pooling to reduce the dimensionality of the features output by the convolutional layers; the LSTM layer contains two LSTM network layers with 64 hidden neurons to extract the temporal variation features of the detection data and capture the temporal variation patterns of parameters caused by leakage; the output layer uses the Softmax activation function to output a leakage probability value between 0 and 1.

[0042] The model uses cross-entropy loss function and Adam optimizer with a learning rate of 0.001. The training set contains nearly 5 years of full-condition data for the heat exchanger, including 100,000 sets of normal operating conditions, 20,000 sets of leakage conditions with different leakage amounts, and 50,000 sets of disturbance conditions such as boiler load changes, fuel type changes, ambient temperature and humidity changes, and boiler start-up and shutdown. The ratio of training set to test set is 8:2. The leak identification accuracy of the trained model on the test set is no less than 98%, which can effectively distinguish between leakage conditions and various non-leakage disturbance conditions, avoiding false alarms.

[0043] The module sets three levels of judgment thresholds based on historical leakage condition data: a micro-leakage early warning threshold corresponding to a leakage probability of 30%-60%, a medium-leakage alarm threshold corresponding to a leakage probability of 60%-90%, and a large-leakage emergency shutdown threshold corresponding to a leakage probability of ≥90%.

[0044] During operation, the module inputs the received standardized detection dataset into a pre-trained leak detection model. The model outputs the leak probability value under the current operating conditions. Based on the threshold range corresponding to the leak probability value, it completes the classification and determination of leak risk, and outputs the corresponding leak feature data. For example, when the leak probability is 45%, it is determined to be a minor leak warning state; when the leak probability is 75%, it is determined to be a medium leak alarm state; and when the leak probability is 95%, it is determined to be a major leak emergency shutdown state.

[0045] The leak location module is used to accurately locate the leak location. It communicates with the data preprocessing and adaptive correction module and the leak feature identification and classification judgment module. The output end is connected to the early warning and linkage response module and the data storage and visualization module, including the heat exchange tube zoning management unit and the spatiotemporal feature matching unit.

[0046] Heat exchanger tube zone management unit: The heat exchanger tubes are pre-divided into 8 independent detection zones according to tube rows and flow, each with its own independent parameter monitoring point. Each zone is equipped with 4 wall temperature sensors and has its own independent inlet and outlet parameter monitoring points, thus realizing zoned parameter monitoring.

[0047] Spatiotemporal Feature Matching Unit: This unit incorporates a built-in leak feature library containing the spatiotemporal variation characteristics of various detection parameters when leaks occur in different zones. This includes wall temperature abrupt changes corresponding to the leaking zone, time difference characteristics of flue gas humidity changes, and zone distribution characteristics of tube-side flow rate changes. During operation, the module extracts the spatiotemporal variation characteristics of detection parameters for each zone based on the received standardized detection dataset and leak feature data. It then performs similarity matching with the standard feature data in the leak feature library. The zone with the highest matching degree is the zone where the leak occurred. Furthermore, based on the wall temperature change data of individual heat exchange tubes within the zone, it can further locate the leaking individual heat exchange tube with a positioning accuracy down to the individual heat exchange tube level.

[0048] The early warning and coordinated response module is the core of the system. It is connected to the leakage feature identification and classification judgment module and the leakage location module. It includes a graded early warning unit and a coordinated control unit, which can realize graded handling of leakage faults and prevent the fault from escalating.

[0049] Tiered Early Warning Unit: Equipped with an explosion-proof audible and visual alarm and 4G wireless communication components, it outputs different levels of early warning signals based on the received leakage classification results: When a micro-leakage early warning state is detected, an SMS message and platform early warning information are sent to maintenance personnel via the remote maintenance platform, prompting them to pay attention to the equipment status and arrange planned maintenance; when a medium-leakage alarm state is detected, the on-site audible and visual alarm is activated, and an emergency alarm message and leakage location results are sent to the maintenance supervisor, prompting immediate action; when a large-leakage emergency shutdown state is detected, the highest-level audible and visual alarm is triggered, and an emergency shutdown signal is simultaneously sent to the boiler DCS system to prevent further escalation of the fault.

[0050] The linkage control unit establishes a two-way communication connection with the boiler DCS system, the electric control valves for the tubular medium, and the electric actuators for the flue gas dampers via industrial Ethernet. Based on the leakage classification results, it can trigger corresponding control actions: In the case of a minor leak, it fine-tunes the tubular medium pressure to reduce the leakage rate and simultaneously adjusts the flue gas damper opening to optimize flue gas flow; in the case of a medium leak, it reduces the boiler operating load, closes the flue gas damper opening, and simultaneously shuts off the tubular bypass valves to reduce equipment operating risks; in the case of a major leak, it urgently shuts off the tubular medium inlet and outlet valves to cut off the medium supply and triggers the boiler emergency shutdown procedure to ensure equipment and personnel safety. All linkage actions are synchronously recorded to the data storage and visualization module.

[0051] The full lifecycle health management module is used to implement preventive maintenance of equipment. It communicates with the leak feature identification and classification module and its output is connected to the data storage and visualization module and the remote interaction module, including a corrosion rate calculation unit, a remaining life prediction unit, and a maintenance suggestion generation unit.

[0052] Corrosion rate calculation unit: Based on historical operating data and real-time monitoring data of the equipment, combined with the pH value and conductivity changes of the tube-side medium, the corrosive parameters of the flue gas, and the cumulative operating time of the equipment, the unit calculates the real-time corrosion rate and cumulative corrosion amount of the heat exchange tubes using Faraday's law and the corrosion weight loss method.

[0053] Remaining life prediction unit: It has a built-in corrosion aging prediction model based on LSTM algorithm. It can accurately predict the remaining life of heat exchange tubes based on real-time corrosion rate and historical corrosion trend, combined with the design wall thickness and minimum allowable wall thickness of heat exchange tubes. When the remaining life is lower than the preset 1-year threshold, it will output a replacement warning in advance.

[0054] Maintenance suggestion generation unit: Based on the remaining life prediction results, leak detection results, and equipment operating status, it generates targeted preventive maintenance suggestions, including heat exchange tube acid pickling and passivation, zoned non-destructive testing, heat exchange tube replacement, and operating condition optimization, to guide maintenance personnel in managing the equipment throughout its entire life cycle and achieve preventive maintenance.

[0055] The data storage and visualization module communicates with all other core modules of the system, and integrates an industrial-grade storage server and a visualization industrial control computer. The storage server adopts a RAID5 redundant architecture with a storage capacity of no less than 4TB, used to store all system operation data, detection data, judgment results, handling records, maintenance data, etc., with a data storage cycle of no less than 3 years, and historical data can be retrieved at any time for retrospective analysis. The visualization industrial control computer is equipped with a 21.5-inch industrial touch screen with a built-in visualization configuration interface, which can display the heat exchanger's operating status, real-time curves of various detection parameters, leakage risk level, leakage location results, remaining service life prediction data, maintenance suggestions, etc., in real time, facilitating intuitive viewing and operation by on-site maintenance personnel.

[0056] The remote interaction module communicates with the data storage and visualization module, and has built-in 5G wireless communication components and gigabit Ethernet communication components. It supports VPN encrypted transmission and can synchronize system detection data, alarm information, leak location results, maintenance suggestions and other data to the remote operation and maintenance cloud platform. At the same time, it can receive control commands, model update parameters and threshold adjustment parameters issued by the remote operation and maintenance platform, realize remote monitoring, remote operation and maintenance of equipment, and remote iterative updates of models, and complete system optimization and upgrades without on-site personnel operation.

[0057] The working principle of this invention is: Step 1: The principle of real-time acquisition of multi-source data in all dimensions; After the system starts up, the multi-source data acquisition module, with a high sampling frequency of 1 time per second, completes real-time synchronous acquisition of core data across all dimensions through three main units: the tube-side medium detection unit, the shell-side flue gas detection unit, and the operating condition parameter acquisition unit. This provides a complete data source for subsequent leak identification. The core acquisition logic and principle are as follows: Tube-side medium detection unit: This unit uses sensors installed at the inlet and outlet of the tube side to collect real-time data on key parameters of the tube-side medium, including pressure, flow rate, inlet and outlet temperatures, conductivity, and pH value. Its underlying principle is: when shell-side flue gas leaks into the tube-side medium, SO2 and NO in the flue gas... x Acidic gases can dissolve in the tube-side medium, leading to a significant increase in the medium's conductivity and a significant decrease in its pH value. When the tube-side medium leaks into the shell-side flue gas, it causes an abnormal increase in the pressure difference between the tube-side inlet and outlet and an abnormal increase in the inlet and outlet flow rate. These parameters are the core direct characteristics of the leakage.

[0058] Shell-side flue gas monitoring unit: This unit uses monitoring equipment installed at the shell-side inlet and outlet to collect real-time data on the temperature, pressure, inlet and outlet flow rates, humidity, and SO2 / NO2 ratio of the shell-side flue gas. x Concentration, oxygen content, and dust concentration parameters. The underlying principle is: when the tube-side medium leaks into the shell-side flue gas, the vaporization of the medium leads to a significant increase in the humidity at the flue gas outlet. Simultaneously, acidic gases dissolve in the leaked medium, causing an increase in the SO2 / NO2 ratio in the flue gas. x Abnormal changes in concentration; the above parameters are key auxiliary characteristics of a leak.

[0059] Operating Parameter Acquisition Unit: Through bidirectional communication with the boiler DCS system via the OPC UA standard protocol, it acquires real-time operating parameters such as boiler load, fuel type, coal feed rate, induced draft fan frequency, and flue gas damper opening, providing reference data for subsequent elimination of non-leakage operating interference.

[0060] All acquired data is transmitted in real time to the data preprocessing and adaptive correction module via RS485 bus or industrial Ethernet, completing the closed loop of the data acquisition process.

[0061] Step 2: Data preprocessing and operating condition adaptive correction principle; After receiving the collected data, the data preprocessing and adaptive correction module completes data cleaning and operational interference elimination through a three-level processing flow, outputting a standardized detection dataset. This addresses the false alarm problem caused by operational fluctuations at its source. The core principle is as follows: Data filtering and noise reduction: The raw data is filtered using the Kalman filter algorithm to eliminate high-frequency noise caused by electromagnetic interference and sensor mechanical vibration in the industrial field, and output smooth and stable detection data to avoid noise data interfering with subsequent leak identification.

[0062] Outlier Removal and Completion: Based on the 3σ criterion, outlier identification is performed on the filtered data. Abrupt outliers that exceed three times the normal fluctuation range (such as jump data caused by sensor failure or line interference) are removed. At the same time, the mean of adjacent data is used to fill in the data to complete the data and avoid misjudgment caused by outlier data.

[0063] Adaptive Dynamic Correction Based on Operating Conditions: The module incorporates an operating condition-parameter mapping model built using a multiple linear regression algorithm based on three years of historical operating data from the equipment. The model uses real-time boiler load, flue gas flow rate, ambient temperature and humidity, and fuel calorific value as independent variables, and theoretical baseline values ​​of core detection parameters such as flue gas humidity, tube-side pressure difference, and conductivity under current operating conditions as dependent variables. The regression coefficients are solved using the least squares method, and the goodness-of-fit R-value is calculated. 2 ≥0.95. During operation, the module calculates the theoretical baseline values ​​of each detection parameter under the current operating condition based on the real-time collected operating parameters, and performs dynamic compensation and correction on the preprocessed detection data to completely eliminate parameter deviations caused by non-leakage operating condition fluctuations such as boiler load fluctuations, fuel changes, and ambient temperature and humidity changes. Finally, it outputs a standardized detection dataset without interference, which is synchronously transmitted to the leakage feature identification and classification judgment module and the leakage location module.

[0064] Step 3: The principle of intelligent extraction and hierarchical determination of leakage characteristics; After receiving the standardized detection dataset, the leak feature identification and risk classification module uses a built-in pre-trained CNN-LSTM hybrid deep learning model to complete leak feature extraction, leak condition identification, and risk classification, solving the core pain points of difficulty in identifying micro-leaks and difficulty in distinguishing leaks from fluctuations in operating conditions. The core principle is as follows: The hybrid model employs a dual-dimensional feature extraction approach: The input layer receives 12 core detection parameters. First, two one-dimensional CNN convolutional layers extract spatial correlation features between multi-source detection data, capturing the interconnected changes between different parameters. Then, a two-layer LSTM network extracts temporal variation features of the detection data, capturing the temporal evolution of parameters caused by leakage, thus achieving full-dimensional extraction of the spatiotemporal features of the leak. This model is pre-trained based on five years of full-condition data from the equipment. The training set includes 100,000 sets of normal operating condition data, 20,000 sets of leakage condition data with different leakage amounts, and 50,000 sets of various interference condition data. After training, the leak identification accuracy on the test set is no less than 98%, accurately distinguishing leakage conditions from non-leakage interference conditions such as normal operating condition fluctuations, fuel changes, and load adjustments, fundamentally reducing the false alarm rate.

[0065] Leakage risk classification in three levels: The model output layer uses the Softmax activation function to output a leakage probability value between 0 and 1. The module presets three-level judgment thresholds based on historical leakage condition data to complete the accurate classification of leakage risk. The probability of leakage is 30%-60%, which is considered a micro-leakage warning state. The probability of leakage is 60%-90%, which is considered a medium leakage alarm state. If the leakage probability is ≥90%, it is determined to be a major leakage emergency shutdown state. The module finally outputs the leakage classification result and leakage characteristic data, which are synchronously transmitted to the leakage location module, the early warning and linkage response module, and the full life cycle health management module.

[0066] Step 4: The principle of spatiotemporal feature matching and precise location of leakage; After synchronously receiving standardized detection datasets and leak characteristic data, the leak location module uses a dual logic of partitioned management and spatiotemporal feature matching to achieve precise location of the leak, solving the problems of traditional solutions being unable to locate the leak and having high maintenance costs. The core principle is as follows: Zoned management of heat exchange tubes: The heat exchange tubes of the heat exchanger are pre-divided into multiple independent detection zones according to tube rows and flow rates. Each zone corresponds to an independent parameter monitoring point and wall temperature sensor, realizing independent parameter monitoring in zones and providing basic spatial data for positioning.

[0067] Spatiotemporal Feature Matching and Localization: The module has a built-in leak feature library, which contains the spatiotemporal variation characteristics of various detection parameters when leaks occur in different zones (including the characteristics of sudden changes in wall temperature in the leaking zone, the time difference characteristics of flue gas humidity changes, and the zonal distribution characteristics of tube flow changes, etc.). During operation, the module extracts the spatiotemporal variation characteristics of the detection parameters in each zone and performs similarity matching with the standard feature data in the leak feature library. The zone with the highest matching degree is the zone where the leak occurred. At the same time, combined with the wall temperature change data of a single heat exchange tube within the zone, the module further locates the leaking single heat exchange tube, achieving precise single-tube-level localization. The module finally outputs the leak localization result and transmits it synchronously to the early warning and linkage response module and the data storage and visualization module.

[0068] Step 5: The principle of tiered early warning and coordinated closed-loop response; After receiving the leakage classification and location results, the early warning and coordinated response module executes classified early warning and coordinated control to achieve closed-loop handling of leakage faults and prevent the fault from escalating. The core principle is as follows: Tiered early warning trigger: Based on the leakage classification results, the module outputs early warning signals of corresponding levels, forming a tiered early warning mechanism. Micro-leakage status: Send SMS messages and platform alerts to maintenance personnel via the remote maintenance platform to prompt planned maintenance; Leakage status: Activate the on-site explosion-proof audible and visual alarm, send emergency alarm information and location results to the operation and maintenance manager, and prompt immediate action; Major leakage condition: Triggers the highest level audible and visual alarm, simultaneously sends an emergency shutdown signal to the boiler DCS system, and triggers safety protection.

[0069] Closed-loop linkage control: The module communicates bidirectionally with the boiler DCS system and field electric actuators via industrial Ethernet, and executes corresponding control actions based on the leakage classification results to achieve proactive fault intervention. Micro-leakage status: Fine-tune the medium pressure in the tube side to reduce the leakage rate, and adjust the flue gas damper opening to optimize the flue gas flow. Leakage status: The boiler operating load is reduced by linkage, the flue gas damper opening is reduced, and the tube bypass valve is closed to reduce the operating risk; In the event of a major leak: The inlet and outlet valves of the medium in the pipeline are immediately shut off to cut off the medium supply, simultaneously triggering the boiler emergency shutdown procedure to ensure the safety of equipment and personnel. All early warning information and coordinated response actions are simultaneously recorded to the data storage and visualization module, forming a complete traceability of the entire response process.

[0070] Step Six: The Principles of Full Life Cycle Health Management and Preventive Maintenance; After receiving real-time leak detection data and historical operating data, the full lifecycle health management module transforms equipment maintenance from reactive fault handling to proactive preventative maintenance through three main steps: corrosion rate calculation, remaining life prediction, and maintenance suggestion generation. The core principle is as follows: Real-time corrosion rate calculation: Based on historical and real-time monitoring of tube-side medium pH, conductivity, and flue gas corrosivity parameters, combined with the equipment's cumulative operating time, the real-time corrosion rate and cumulative corrosion amount of the heat exchange tubes are accurately calculated using Faraday's law and the corrosion weight loss method, thus quantifying the degree of corrosion aging of the heat exchange tubes.

[0071] Remaining service life prediction: The module has a built-in corrosion aging prediction model based on the LSTM algorithm. Based on the real-time corrosion rate and historical corrosion trend, combined with the design wall thickness and minimum allowable wall thickness of the heat exchange tube, it accurately predicts the remaining service life of the heat exchange tube. When the remaining service life is lower than the preset 1-year threshold, it outputs a replacement warning in advance.

[0072] Preventive maintenance recommendations are generated based on remaining life predictions, leak detection results, and equipment operating status. These recommendations include acid pickling and passivation of heat exchange tubes, non-destructive testing of different zones, replacement of heat exchange tubes, and optimization of operating conditions. This guides maintenance personnel to complete the full lifecycle management of the equipment and reduces the probability of leaks and failures from the source.

[0073] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A smart detection system for leaks in a flue gas waste heat recovery heat exchanger, characterized in that, It includes a multi-source data acquisition module, a data preprocessing and adaptive correction module, a leak feature identification and classification module, a leak location module, an early warning and coordinated response module, and a full life cycle health management module; The multi-source data acquisition module is used to collect in real time the pressure, flow rate, conductivity, and pH value of the heat exchanger tube-side medium, the humidity, SO2 concentration, temperature, and pressure parameters of the shell-side flue gas, as well as the equipment operating condition parameters such as the boiler real-time load, induced draft fan frequency, and flue gas damper opening. The data preprocessing and adaptive correction module is connected to the multi-source data acquisition module. It is used to filter and denoise the acquired multi-source data, remove outliers, and has a built-in working condition-parameter mapping model. It calculates the benchmark values ​​of each detection parameter under the current working condition based on the real-time working condition parameters, performs dynamic compensation and correction on the preprocessed detection data, eliminates parameter deviations caused by non-leakage working condition fluctuations such as boiler load fluctuations, fuel changes, and ambient temperature and humidity changes, and outputs a standardized detection dataset. The leakage feature identification and classification module is connected to the data preprocessing and adaptive correction module. It has a built-in CNN-LSTM hybrid deep learning model trained on multi-source data. It is used to extract the spatiotemporal features of leakage based on the standardized detection dataset, identify the leakage conditions and complete the three-level classification of leakage risk, and output the leakage classification results and leakage feature data. The leak location module is connected to the data preprocessing and adaptive correction module and the leak feature recognition and classification judgment module, respectively. It is used to match the spatiotemporal distribution characteristics of the leak based on the standardized detection dataset and the leak feature data to locate the specific zone where the leak occurs and the location of the heat exchange tube. The early warning and linkage response module is communicatively connected to the leakage feature identification and classification judgment module and the leakage location module, respectively. It is used to output the corresponding level of early warning alarm signal according to the leakage classification result and the leakage location result, and to link and execute the corresponding operating condition adjustment and safety protection actions. The full life cycle health management module is connected to the leak feature identification and classification module. It is used to build a corrosion and aging model of the heat exchanger tubes based on historical operating data and real-time detection data, predict the remaining service life, and output preventive maintenance suggestions.

2. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 1, characterized in that, The multi-source data acquisition module includes a tube-side medium detection unit, a shell-side flue gas detection unit, and an operating condition parameter acquisition unit. The tube-side medium detection unit is also used to acquire the inlet and outlet temperature parameters of the heat exchanger tube-side medium. The shell-side flue gas detection unit is also used to acquire the inlet and outlet flow rates, oxygen content, and NO content of the heat exchanger shell-side flue gas. x Concentration and dust concentration parameters; the operating condition parameter acquisition unit is also used to acquire fuel type parameters.

3. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 1, characterized in that, The data preprocessing and adaptive correction module includes a data filtering unit, an outlier removal unit, and an operating condition adaptive correction unit. The data filtering unit uses the Kalman filter algorithm to filter and denoise the collected data. The outlier removal unit completes outlier identification, removal, and data completion based on the 3σ criterion. The operating condition adaptive correction unit has a built-in operating condition-parameter mapping model, which is a parameter benchmark calculation model constructed based on historical operating data using a multiple linear regression algorithm.

4. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 1, characterized in that, The leakage feature identification and classification judgment module sets three judgment thresholds based on historical leakage operating data: a micro-leakage early warning threshold corresponding to a leakage probability of 30%-60%, a medium-leakage alarm threshold corresponding to a leakage probability of 60%-90%, and a large-leakage emergency shutdown threshold corresponding to a leakage probability ≥90%. The leakage risk can be classified and judged according to the threshold range in which the leakage probability value is located.

5. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 4, characterized in that, The leakage feature identification and classification module has a built-in CNN-LSTM hybrid deep learning model. The CNN network extracts the spatial correlation features of multi-source detection data, and the LSTM network extracts the time series change features of the detection data. The model is pre-trained based on historical leakage condition data, normal condition data, and various interference condition data. It can distinguish between leakage conditions and non-leakage interference conditions such as fluctuations in normal conditions, fuel changes, load adjustments, and changes in ambient temperature and humidity.

6. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 1, characterized in that, The leak location module includes a heat exchange tube partition management unit and a spatiotemporal feature matching unit. The heat exchange tube partition management unit is used to divide the heat exchange tubes of the heat exchanger into several independent detection partitions with corresponding independent parameter monitoring points according to the tube row and the process. The spatiotemporal feature matching unit has a built-in leak feature library, which can extract the spatiotemporal change features of the detection parameters of each partition, match them with the leak feature library, and locate the partition and single heat exchange tube where the leak is located.

7. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 1, characterized in that, The early warning and linkage response module includes a graded early warning unit and a linkage control unit. The graded early warning unit can output graded early warning signals such as on-site audible and visual early warning, remote platform early warning, and emergency shutdown alarm according to the leakage classification results. The linkage control unit can communicate with the boiler DCS system and on-site actuators, and adjust the boiler load, flue gas damper opening, and pipe-side medium valve opening according to the leakage classification results, and trigger the shutdown protection program in emergency conditions.

8. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 1, characterized in that, The full lifecycle health management module includes a corrosion rate calculation unit, a remaining life prediction unit, and a maintenance suggestion generation unit. The corrosion rate calculation unit calculates the real-time corrosion rate and cumulative corrosion amount of the heat exchange tubes based on historical and real-time monitored parameters such as the pH value, conductivity, and flue gas corrosivity of the tube-side medium, using Faraday's law and the corrosion weight loss method. The remaining life prediction unit incorporates an LSTM corrosion aging prediction model, which can predict the remaining service life of the heat exchange tubes based on the real-time corrosion rate and historical corrosion trends, combined with the design wall thickness and minimum allowable wall thickness. The maintenance suggestion generation unit can output corresponding preventive maintenance and replacement suggestions based on the prediction results.

9. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 1, characterized in that, It also includes a data storage and visualization module, which is connected to a multi-source data acquisition module, a data preprocessing and adaptive correction module, a leak feature identification and classification judgment module, a leak location module, an early warning and linkage response module, and a full life cycle health management module. It is used to store all operating data, detection data, judgment results, handling records and maintenance data, and at the same time, it displays the heat exchanger operating status, leak risk level, location results and remaining life prediction data in real time through a visualization interface.

10. The intelligent detection system for leaks in a flue gas waste heat recovery heat exchanger according to claim 9, characterized in that, It also includes a remote interaction module, which is communicatively connected to the data storage and visualization module. This module is used to synchronize detection data, alarm information, and maintenance suggestions to the remote operation and maintenance platform, and to receive remotely issued control commands and model update parameters, thereby enabling remote operation and maintenance and system iteration.

Images