Online diagnosis system and method for gas purification heat exchanger based on multi-source data fusion

Abstract

The invention relates to the technical field of metallurgical chemical equipment monitoring, and particularly discloses a gas purification heat exchanger online diagnosis system and method based on multi-source data fusion, an infrared thermal imaging data acquisition unit adopts an explosion-proof uncooled infrared focal plane detector and is used for continuously acquiring temperature field distribution infrared video streams on the outer surface of a heat exchanger; the process parameter sensing unit comprises a temperature sensor, a pressure sensor and a flow sensor, and all sensor signals are accessed to a DCS (Distributed Control System) through a PROFIBUS-DP bus; the data integration and processing unit is an industrial server deployed in a control room, acquires process parameter time sequence data from a DCS system in real time through an OPC UA protocol, and receives an infrared video stream through a private network; according to the method, accurate identification and positioning of early hidden faults are realized, the situation that small diseases drag large disasters is avoided, the diagnosis result visualization degree is high, and the operation and maintenance efficiency and the response speed are improved.

Description

Technical Field

[0001] This invention relates to the field of monitoring technology for metallurgical and chemical equipment, specifically to an online diagnostic system and method for gas purification heat exchangers based on multi-source data fusion. Background Technology

[0002] In coal gas purification production lines in the metallurgical and chemical industries, a large number of shell-and-tube and spiral plate heat exchangers are widely installed. Current operation and maintenance strategies mainly rely on discrete inlet and outlet temperature, flow rate, and differential pressure data to estimate the overall heat transfer coefficient (K-value) to assess energy efficiency; and to roughly diagnose faults by monitoring abnormal increases (indicating scaling and blockage) or abnormal decreases (indicating internal leakage) in the tube-side / shell-side differential pressure (ΔP). This method suffers from significant monitoring lag, data silos, and reliance on human experience, making it unable to accurately diagnose and warn of early, latent faults such as localized blockage of heat exchanger tubes, low-flow internal leaks, and scale distribution. This leads to a passive operation and maintenance strategy, generally adopting a high-cost "reactive maintenance" model. High heat exchanger maintenance costs, coupled with unplanned shutdowns due to sudden leaks, severely restrict production continuity and economic efficiency. Therefore, exploring an online diagnostic system and method for coal gas purification heat exchangers based on multi-source data fusion has become a pressing technical problem for those skilled in the art. Summary of the Invention

[0003] To address the problems existing in the prior art, the present invention aims to provide an online diagnostic system and method for gas purification heat exchangers based on multi-source data fusion. The system collects data through the fusion of multiple source sensing units such as infrared thermal imaging and PROFIBUS-DP bus, and relies on mechanistic models and AI multimodal diagnostic algorithms to achieve online intelligent diagnosis of heat exchanger operating status, accurate fault location, and predictive maintenance decision support.

[0004] The technical solution adopted by this invention to solve its technical problem is: an online diagnostic system for gas purification heat exchangers based on multi-source data fusion, including an infrared thermal imaging data acquisition unit, a process parameter sensing unit, and a data integration and processing unit. The infrared thermal imaging data acquisition unit uses an explosion-proof uncooled infrared focal plane detector to continuously acquire infrared video streams of the temperature field distribution on the outer surface of the heat exchanger. The process parameter sensing unit includes temperature sensors, pressure sensors, and flow sensors. All sensor signals are connected to the DCS system via a PROFIBUS-DP bus. The data integration and processing unit is an industrial server deployed in the control room. It acquires process parameter timing data from the DCS system in real time via the OPC UA protocol and receives infrared video streams via a dedicated network.

[0005] Specifically, the data integration and processing unit has a built-in high-precision time synchronization protocol to align each frame of infrared image with all process parameters under the same timestamp, thereby constructing a multi-source fusion time series database.

[0006] Specifically, the data integration and processing unit executes the following software methods: Feature extraction and modeling: preprocessing infrared video frames and extracting temperature field distribution features; simultaneously, based on heat transfer principles and real-time process data, dynamically calculating the actual heat transfer coefficient, theoretical heat transfer coefficient, fouling thermal resistance, and efficiency coefficient of the heat exchanger; AI intelligent diagnosis and judgment: inputting the extracted infrared image feature sequence and the real-time calculated efficiency parameters into a pre-trained deep neural network model for multimodal fusion analysis, and outputting diagnostic results including fault type, fault location, and confidence level from the model.

[0007] Specifically, the deep neural network is a hybrid model combining long short-term memory networks and convolutional neural networks, used to simultaneously learn the temporal variation of performance parameters and the spatial distribution characteristics of infrared images.

[0008] A working method for an online diagnostic system for coal gas purification heat exchangers based on multi-source data fusion includes the following steps: S1. Data Synchronization and Fusion: The system has a built-in time synchronization protocol PTP, which accurately aligns each frame of infrared image with all process parameters under the same timestamp, builds a unified multi-source fusion database, and synchronously acquires infrared thermal imaging time series data and point-based process parameter time series data of the heat exchanger, laying a data foundation for analysis. S2. Feature Extraction and Modeling: Extract features reflecting the spatial distribution of the temperature field from infrared thermal imaging data, and calculate parameters reflecting the heat exchanger efficiency based on process parameters and mechanistic models. S3, AI Intelligent Diagnosis and Judgment: Infrared image features and performance parameters are fused and input into the AI ​​multimodal fault diagnosis model to identify and locate early faults in the heat exchanger.

[0009] Specifically, in step S2, the infrared thermal imaging data is first processed by infrared image processing. The infrared video frames are preprocessed by filtering and enhancing using the OpenCV library, and key features are extracted. The key features include the variance of the overall and regional temperature distribution, the coordinate trajectory of the highest temperature point, and the spatial distribution information of the rate of change of the thermal gradient.

[0010] Specifically, the mechanism model in step S2 is constructed based on the principle of heat transfer. A real-time performance calculation model of the heat exchanger is established. The model is input with real-time temperature and flow data, and dynamically calculates the current actual heat transfer coefficient K_actual and the theoretical heat transfer coefficient K_theory under clean conditions, thereby obtaining the key performance indicators of fouling thermal resistance Rf and performance coefficient η.

[0011] Specifically, the AI ​​multimodal fault diagnosis model in step S3 is trained based on a fault sample library built from historical operation and maintenance data and expert experience, establishing a sample library covering multiple fault modes and clarifying the characterization of each fault in infrared images and performance parameters. Multimodal fusion diagnosis: The extracted infrared image feature sequence and the performance parameters calculated in real time are input into a trained deep neural network for joint analysis; the AI ​​multimodal fault diagnosis model finally outputs the diagnosis results, fault location information and diagnosis confidence. Real-time tracking and decision support: Diagnostic results and early warning information are displayed on the human-machine interface (HMI) in real time and pushed to the mobile terminals of maintenance personnel; the system can intelligently recommend the best time window for predictive maintenance of PdM based on the decay trend of performance parameters, guiding purging, cleaning or planned maintenance.

[0012] Specifically, the diagnostic results include health status, scaling warning, blockage alarm, and internal leakage alarm.

[0013] Specifically, the early faults in step S3 include partial blockage of the tube bundle, low-flow internal leakage, and abnormal scale distribution. Tube bundle blockage is diagnosed by analyzing low-temperature anomalies in specific areas of the casing, and internal leakage is diagnosed by analyzing temperature field distribution distortion.

[0014] The present invention has the following beneficial effects: The online diagnostic system and method for gas purification heat exchangers designed in this invention based on multi-source data fusion are forward-looking and accurate, transforming passive post-event maintenance into proactive predictive maintenance, realizing accurate identification and location of early hidden faults, and avoiding "minor problems becoming major disasters".

[0015] The online diagnostic system and method for gas purification heat exchangers designed in this invention based on multi-source data fusion is comprehensive and intuitive. Infrared imaging provides a global and continuous temperature field distribution, overcoming the limitations of point-based measurements that are like "blind men touching an elephant," and the diagnostic results are highly visualized.

[0016] The present invention presents an online diagnostic system and method for gas purification heat exchangers based on multi-source data fusion. This system is intelligent and automated, reduces over-reliance on expert experience, and improves operation and maintenance efficiency and response speed through automatic analysis and decision-making via AI algorithms. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be described clearly, completely, and in further detail below. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0018] An online diagnostic system and method for gas purification heat exchangers based on multi-source data fusion is proposed. This system collects data by fusing data from multiple source sensing units such as infrared thermal imaging and PROFIBUS-DP bus. Relying on mechanistic models and AI multimodal diagnostic algorithms, it achieves online intelligent diagnosis of heat exchanger operating status, accurate fault location, and predictive maintenance decision support.

[0019] (a) System hardware composition.

[0020] 1. Infrared thermal imaging data acquisition unit: It adopts an explosion-proof, high-resolution uncooled infrared focal plane detector (such as FLIR Axx series or equivalent models, supporting RTSP protocol streaming media transmission), which is installed in the key parts of the heat exchanger shell to continuously collect infrared video streams of its outer surface temperature field distribution.

[0021] 2. Process Parameter Sensing Unit: Upgrade existing sensors to ensure remote transmission of key data: The temperature sensor uses a Pt100 armored RTD to measure inlet and outlet temperatures; the pressure and flow sensors are integrated using a Rosemount 3051S series multivariable transmitter to simultaneously measure medium pressure and flow; all sensor signals are connected to the existing DCS system via PROFIBUS-DP bus.

[0022] 3. Data Integration and Processing Unit: An industrial server is deployed in the control room, with a built-in high-speed data acquisition card (such as NIPCIe-6353) and gigabit network card. It acquires real-time time-series data of all sensors from the DCS system through the OPCUA protocol and receives video streams from infrared cameras through a dedicated network.

[0023] (II) Software Methodology and Implementation Process: 1. Data Synchronization and Fusion: The system has a built-in high-precision time synchronization protocol (PTP) to align each frame of infrared image with all process parameters (temperature, pressure, flow rate) at the same timestamp, thus building a multi-source fusion database.

[0024] 2. Feature extraction and modeling: 2.1 Infrared Image Processing: The OpenCV library is used to preprocess infrared video frames (filtering, enhancement) and extract key features, such as the variance of the overall / regional temperature distribution, the coordinate trajectory of the highest temperature point, and the rate of change of the thermal gradient.

[0025] 2.2 Mechanism Model Construction: Based on the principle of heat transfer, a real-time performance calculation model for the heat exchanger is established. Real-time temperature and flow data are input, and the actual heat transfer coefficient K_actual and the theoretical heat transfer coefficient K_theory under clean conditions are dynamically calculated to obtain the fouling thermal resistance Rf and the performance coefficient η.

[0026] 3. AI-powered intelligent diagnosis and judgment: 3.1 Fault Knowledge Base Construction: Collect historical operation and maintenance data and expert experience to establish a fault sample library (e.g., tube blockage - corresponding to low temperature abnormality in a specific area of ​​the casing; internal leakage - corresponding to temperature field distribution distortion; scaling - linear decrease in efficiency η, etc.).

[0027] 3.2 Multimodal Fusion Diagnosis: The extracted infrared image feature sequences and the real-time calculated performance parameters (K value, η value, ΔP value) are input together into a trained deep neural network (such as a hybrid model of LSTM+CNN) for joint analysis. The AI ​​model outputs diagnostic results (health, scaling warning, blockage alarm, internal leakage alarm), fault location, and confidence level.

[0028] 3.3 Real-time Tracking and Decision Support: Diagnostic results and early warning information are displayed in real time on the Human-Machine Interface (HMI) and pushed to the mobile terminals of maintenance personnel. Based on the performance degradation trend, the system can recommend predictive maintenance (PdM) windows to guide purging, cleaning, or planned maintenance. Example 1

[0029] 1. Scenario Overview This embodiment focuses on a lean and rich liquid cooler in a coal gas purification anhydrous ammonia unit. After long-term operation, the tube side (crude oil side) of this heat exchanger is prone to scaling and blockage, leading to decreased efficiency, while the shell side (residue oil side) may also experience localized sedimentation due to the characteristics of the medium. Traditional DCS monitoring can only display macroscopic temperature and pressure, failing to detect early localized blockages or micro-leaks within the tube bundle in a timely manner, often resulting in unplanned shutdowns.

[0030] 2. System Hardware Deployment and Integration Infrared thermal imaging data acquisition unit: Equipment selection: One explosion-proof high-resolution infrared thermal imager was selected. This equipment meets the explosion-proof requirements on site, and its high resolution is sufficient to capture minute temperature differences at the tube bundle level.

[0031] Installation location: Install on a fixed platform 3 meters directly above the heat exchanger shell, ensuring that its field of view completely covers the entire surface of the heat exchanger shell. During installation, pay special attention to avoiding obstructions such as steam and water mist that may interfere with infrared measurements, and ensure that the pan-tilt unit remains stable under fan vibration.

[0032] Data Stream: The thermal imager continuously acquires the temperature field distribution on the surface of the enclosure at a frame rate of 10Hz, generating temperature matrix data (non-visible light video). This data stream is then transmitted in real-time to the data integration and processing unit in the control room via a gigabit industrial Ethernet network using the RTSP protocol. To cope with the complex electromagnetic environment on site, all network cables are armored shielded twisted-pair cables.

[0033] Process parameter sensing unit: Temperature measurement: The original Pt100 RTDs on the inlet and outlet pipes of the heat exchanger tube side were replaced with Class A armored Pt100 RTDs to improve the measurement accuracy to ±0.15°C.

[0034] Pressure and flow measurement: Multivariable transmitters are installed at the tube inlet, outlet, and shell inlet to accurately measure the pressure and pressure drop on the lean side (0.65 MPa -> 0.60 MPa, ΔP=0.05 MPa). The same series of transmitters are also installed at the shell inlet (cooling water side) to monitor the cooling water condition.

[0035] System integration: All sensor signals are connected to the factory’s existing DCS system (such as Siemens PCS 7) via PROFIBUS-DP bus.

[0036] Data Integration and Processing Unit: Hardware: A high-performance server equipped with an Intel Xeon Silver 4310 processor, 64GB of RAM, and an MCC DAQ high-speed data acquisition card and an Intel X550-T2 gigabit network card is deployed in the central control room. This configuration ensures high-speed simultaneous processing of infrared video streams and large amounts of process timing data.

[0037] Software and Communication: The core software of this system is deployed on the server. This software acts as an OPC UA client, reading all process parameters of the lean liquid heat exchanger in real time from the OPC UA server of the DCS system via the factory LAN (lean liquid inlet / outlet temperature (75 / 40°C), flow rate (50 m³ / h), pressure (0.65 / 0.60 MPa)). Simultaneously, it receives infrared temperature data streams from the FLIR A8580 via a dedicated network port.

[0038] 3. Software Method Implementation and Calculation Process Step 1: Data Synchronization and Integration Synchronization Mechanism: The system deploys the IEEE 1588 Precision Time Protocol (PTP) on the server and all intelligent terminals (infrared thermal imagers, multivariable transmitters) to control the time synchronization accuracy of the entire system to within 1 millisecond. This ensures that the infrared images and process parameters are strictly correlated in time during subsequent fusion analysis.

[0039] Fusion Database Construction: The system assigns a precise timestamp to each frame of infrared temperature matrix data (e.g., 2025-05-20 10:30:00) and extracts all process parameters under the same timestamp from the DCS, forming a fused data record, which is then stored in a time-series database (such as InfluxDB). This record contains: timestamp IR_frame: An m×n temperature matrix.

[0040] T_in, T_out: Inlet and outlet temperatures of the tube side (lean solution).

[0041] F: Pipeline flow rate (lean solution).

[0042] P_in, P_out: Inlet and outlet pressures of the tube (lean solution).

[0043] Step 2: Feature Extraction and Mechanism Modeling Infrared image feature extraction: 1) Preprocessing: Median filtering is used to remove noise. The temperature gradient of the shell of the lean liquid heat exchanger is usually not large, so it is necessary to effectively remove environmental reflected noise.

[0044] 2) Overall temperature distribution variance - reflects the overall thermal uniformity; scale buildup in lean solution will lead to uneven distribution.

[0045] 3) Divide the tube bundle into 9 regions (3x3 grid) and calculate the average temperature of each region to locate areas with severe blockage or scaling.

[0046] 4) The coordinate trajectory of the highest temperature point - tracking the movement of hotspots, extracting the feature value mechanism model construction and real-time calculation: calculating the actual total heat transfer coefficient K_actual; based on the basic heat transfer equation: Q = K*A*ΔTm.

[0047] in: Q = heat load = m*Cp*(T_ou -T_in) =ρ*F*Cp*(T_in-T_out). Here, the lean liquid density ρ and specific heat capacity Cp are obtained in real-time from a pre-set property database based on its concentration and temperature, rather than fixed values, to improve calculation accuracy.

[0048] A = Design heat transfer area of ​​the heat exchanger (a known constant).

[0049] ΔTm = logarithmic mean temperature difference, calculated based on the inlet and outlet temperatures of lean solution and cooling water.

[0050] Therefore, the real-time K value is: K_actual=(F*ρ*Cp*(T_out-T_in)) / (A*ΔTm).

[0051] Calculate the theoretical / clean heat transfer coefficient K_theory: K_theory is the baseline value for the heat exchanger under clean, scale-free conditions.

[0052] In this embodiment, the average value of K_actual during stable operation under conditions of a lean liquid flow rate of 50 m³ / h and an inlet temperature of around 75°C after new commissioning or thorough cleaning is taken as the baseline value of K_theory.

[0053] Calculate the scaling thermal resistance Rf and efficiency coefficient η: Rf = (1 / K_actual-1 / K_theory) / / Impurities, degradation products, or ammonium salt crystals in lean solution can cause scaling, which increases Rf.

[0054] η = K_actual / K_theory / / Efficiency coefficient, which should be close to 1 in a healthy state and decreases with scaling. In this system, η is initially set to 1.

[0055] Calculate the tube-side pressure drop: ΔP = P_in - P_out. The normal pressure drop on the lean side is 0.05 MPa. If tube-side blockage occurs (such as foreign matter or severe scaling), ΔP will increase significantly, which is a critical alarm parameter.

[0056] Step 3: AI Intelligent Diagnosis and Judgment Failure Mode Infrared image features Process parameter characteristics healthy The temperature distribution is uniform with small variance and no stable hot spots. K_actual ≈ K_theory, η > 0.95, ΔP is stable at ~0.05 MPa. Tube scaling The overall temperature distribution began to become uneven. The temperature drop in the lean liquid outlet area (low temperature side) was less than expected, and the overall shell temperature increased, resulting in a larger variance. K_actual decreases slowly, η decreases with a trend, Rf increases slowly, and ΔP may increase slightly due to the narrowing of the flow channel. Single tube / partial blockage Obvious low-temperature bands or patches appear (corresponding to blocked tube bundles, reduced lean solution flow leading to abnormal heat transfer), and their locations are fixed. The standard deviation of regional temperature increases significantly. As K_actual decreases, η decreases. A significant increase in ΔP (e.g., from 0.05 MPa to 0.08 MPa) is important supporting evidence. Internal leakage (leakage from the tube side to the shell side) An abnormally high localized high temperature area (hot spot) is caused by the leakage of high-pressure lean liquid into the low-pressure shell side, resulting in instantaneous vaporization or heating of the local shell. The T_out (lean liquid outlet temperature) is abnormally high (due to lean liquid short circuit) instead of decreasing; the calculated K_actual value is abnormal, and the flow rate may be unbalanced. Multimodal fusion diagnostics (AI analysis): Model architecture: A hybrid deep learning model is used. The input layer is divided into two branches: Timing Branching (LSTM): Input timing process features within a time window (e.g., the past 30 minutes): [K_actual, η, Rf, ΔP, F].

[0057] Spatial branch (CNN): Input is an aligned sequence of infrared image features within the same time window. [temp_variance, regional_std, ...] and the coordinates of the highest temperature point.

[0058] Model Training: The model was trained using over one year of historical normal operation data and data from several failure shutdowns before and after the initial operation. Supervised learning enabled the model to map multimodal inputs to corresponding fault categories, locations, and severity. Sample reinforcement was specifically implemented for ammonium salt crystallization blockage patterns commonly found in lean fluid systems.

[0059] Real-time inference: The system performs inference once per minute. The latest 30 sets of fused features are input into the trained AI model.

[0060] Model output: A JSON object containing the following information: json { "status": "Congestion alarm", "confidence": 0.94, "location": "Area 5 (Lower Middle Section)", "severity": "moderate" "recommendation": "The pressure differential ΔP on the lean fluid side has increased from 0.05 MPa to 0.07 MPa. Based on the infrared signature, this indicates a localized blockage. It is recommended that this area of ​​the tubing be inspected and cleaned during the next scheduled maintenance (expected in 14 days)." "trend": "The η value decreased from 0.92 to 0.85 over the past 7 days, showing a clear downward trend." Step 4: Real-time tracking and decision support HMI Interface: On the HMI screen in the central control room, a virtual 3D model of the lean liquid heat exchanger is displayed. The model surface is mapped with a pseudo-color infrared temperature field in real time. The sidebar clearly displays the AI ​​diagnostic results, confidence level, and fault location (highlighted as "Area 5" on the 3D model). It also prominently displays the key performance indicators of the lean liquid side: K_actual, η, ΔP, and their historical trend curves.

[0061] Mobile push: When the diagnostic confidence exceeds 90% or the efficiency coefficient η is below the threshold of 0.88 for 3 consecutive hours, or the pressure difference ΔP on the lean side exceeds the safety threshold (e.g., 0.065 MPa) for 1 consecutive hour, the system will automatically generate alarm information.

[0062] Predictive maintenance decision-making: The system continuously analyzes the long-term trends of η and Rf in the background. It predicts the time when the η value will drop to an unacceptable level (e.g., 0.7) through exponential smoothing or linear regression. When the predicted remaining time enters the next two-week window, or when the upward trend of ΔP indicates that the operating limit is about to be reached, the system automatically generates a "predictive cleaning work order" in the maintenance management system and recommends the optimal time window, thus realizing the shift from "post-failure maintenance" to "predictive maintenance." For lean liquid systems, this effectively prevents emergency shutdowns caused by the deterioration of ammonium salt crystallization.

[0063] Through the implementation of this system, the factory has achieved transparent and intelligent management of the status of the anhydrous ammonia lean and rich liquid cooler in this embodiment: Early and accurate diagnosis: It successfully issued an alarm and accurately located the blockage when a pipe was 30% blocked, avoiding the serious decline in the efficiency of the entire heat exchanger caused by late detection in the past.

[0064] Avoid unplanned downtime: AI-based predictive maintenance recommendations enabled maintenance activities to be integrated into planned overhauls across the plant, preventing two unexpected unit download operations caused by heat exchanger issues.

[0065] Reduced maintenance costs: The shift from periodic, indiscriminate overall cleaning to on-demand, targeted precision maintenance saves on cleaning agents, labor costs, and production losses due to downtime. It is estimated that this saves approximately RMB 350,000 annually in maintenance costs and avoids downtime losses for this single unit.

[0066] This invention proposes a method for high-precision synchronization and fusion of spatiotemporal data of infrared thermal imaging temperature fields with point-based process parameters in a DCS system. An AI multimodal model is created that combines spatial features of infrared images with performance time-series parameters and utilizes deep learning algorithms for joint diagnosis. A specific system integration architecture is designed to achieve efficient and safe linkage between infrared equipment and industrial control networks (DCS / SCADA). A novel, quantitative, and data-driven fault criterion is established for early-stage latent faults (such as micro-leaks in single pipes).

[0067] Calculations show that the system of this invention can extend the life of heat exchangers by more than 20% by optimizing the maintenance cycle, and save an average of 15% in maintenance costs per unit per year, resulting in significant economic benefits.

[0068] This invention is not limited to the above-described embodiments. Anyone should know that any structural changes made under the guidance of this invention, and any technical solutions that are the same as or similar to this invention, fall within the protection scope of this invention.

[0069] The technologies, shapes, and structures not described in detail in this invention are all known technologies.

Claims

1. An online diagnostic system for gas purification heat exchangers based on multi-source data fusion, characterized in that, It includes an infrared thermal imaging data acquisition unit, a process parameter sensing unit, and a data integration and processing unit. The infrared thermal imaging data acquisition unit uses an explosion-proof uncooled infrared focal plane detector to continuously acquire infrared video streams of the temperature field distribution on the outer surface of the heat exchanger. The process parameter sensing unit includes temperature sensors, pressure sensors, and flow sensors. All sensor signals are connected to the DCS system via a PROFIBUS-DP bus. The data integration and processing unit is an industrial server deployed in the control room. It acquires process parameter timing data from the DCS system in real time via the OPC UA protocol and receives infrared video streams via a dedicated network.

2. The online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to claim 1, characterized in that, The data integration and processing unit has a built-in high-precision time synchronization protocol, which is used to align each frame of infrared image with all process parameters under the same timestamp to build a multi-source fusion time series database.

3. The online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to claim 2, characterized in that, The data integration and processing unit executes the following software methods: Feature extraction and modeling: Preprocessing infrared video frames and extracting temperature field distribution features; simultaneously, based on heat transfer principles and real-time process data, dynamically calculating the actual heat transfer coefficient, theoretical heat transfer coefficient, fouling thermal resistance, and efficiency coefficient of the heat exchanger; AI intelligent diagnosis and judgment: Inputting the extracted infrared image feature sequence and the real-time calculated efficiency parameters into a pre-trained deep neural network model for multimodal fusion analysis, and outputting diagnostic results including fault type, fault location, and confidence level from the model.

4. The online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to claim 3, characterized in that, The deep neural network is a hybrid model combining long short-term memory networks and convolutional neural networks, used to simultaneously learn the temporal variation of performance parameters and the spatial distribution characteristics of infrared images.

5. The working method of the online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to any one of claims 1-4, characterized in that, Includes the following steps: S1. Data Synchronization and Fusion: The system has a built-in time synchronization protocol PTP, which accurately aligns each frame of infrared image with all process parameters under the same timestamp, builds a unified multi-source fusion database, and synchronously acquires infrared thermal imaging time series data and point-based process parameter time series data of the heat exchanger, laying a data foundation for analysis. S2. Feature Extraction and Modeling: Extract features reflecting the spatial distribution of the temperature field from infrared thermal imaging data, and calculate parameters reflecting the heat exchanger efficiency based on process parameters and mechanistic models. S3, AI Intelligent Diagnosis and Judgment: Infrared image features and performance parameters are fused and input into the AI ​​multimodal fault diagnosis model to identify and locate early faults in the heat exchanger.

6. The working method of the online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to claim 5, characterized in that, In step S2, the infrared thermal imaging data is first processed by infrared image processing. The infrared video frames are preprocessed by filtering and enhancing using the OpenCV library, and key features are extracted. Key features include the variance of overall and regional temperature distribution, the coordinate trajectory of the highest temperature point, and the spatial distribution information of the rate of change of thermal gradient.

7. The working method of the online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to claim 5, characterized in that, The mechanism model in step S2 is constructed based on the principle of heat transfer. A real-time performance calculation model of the heat exchanger is established. The model is input with real-time temperature and flow data, and dynamically calculates the current actual heat transfer coefficient K_actual and the theoretical heat transfer coefficient K_theory under clean conditions. Then, the key performance indicators of fouling thermal resistance Rf and performance coefficient η are obtained.

8. The working method of the online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to claim 5, characterized in that, The AI ​​multimodal fault diagnosis model in step S3 is trained based on a fault sample library built from historical operation and maintenance data and expert experience. A sample library covering multiple fault modes is established to clarify the characterization of each fault in infrared images and performance parameters. Multimodal fusion diagnosis: The extracted infrared image feature sequence and the performance parameters calculated in real time are input into a trained deep neural network for joint analysis; the AI ​​multimodal fault diagnosis model finally outputs the diagnosis results, fault location information and diagnosis confidence. Real-time tracking and decision support: Diagnostic results and early warning information are displayed on the human-machine interface (HMI) in real time and pushed to the mobile terminals of maintenance personnel; the system can intelligently recommend the best time window for predictive maintenance of PdM based on the decay trend of performance parameters, guiding purging, cleaning or planned maintenance.

9. The working method of the online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to claim 8, characterized in that, The diagnostic results include health status, scaling warning, blockage alarm, and internal leakage alarm.

10. The working method of the online diagnostic system for gas purification heat exchangers based on multi-source data fusion according to claim 5, characterized in that, The early faults in step S3 include local blockage of the tube bundle, low-flow internal leakage, and abnormal scale distribution. Tube bundle blockage is diagnosed by analyzing low-temperature anomalies in specific areas of the casing, and internal leakage is diagnosed by analyzing temperature field distribution distortion.