Converter gas cooler predictive maintenance system and method based on multi-source information fusion

Abstract

The invention belongs to the technical field of industrial equipment maintenance, and particularly relates to a converter gas cooler predictive maintenance system and method based on multi-source information fusion. The system comprises a sensing and data acquisition layer, a data aggregation and edge processing layer, a cloud platform or central server analysis layer and an application and display layer, and the sensing and data acquisition layer comprises a process parameter monitoring module and a physical state monitoring module. The process parameter monitoring module comprises a pressure difference transmitter, a temperature sensor, a pressure sensor and a flow sensor, and the physical state monitoring module comprises a plurality of fixed ultrasonic thickness measuring probes, a vibration monitoring sensor and an acoustic monitoring sensor. By integrating the fixed ultrasonic thickness measuring probe and the pressure difference transmitter, key physical quantities representing corrosion and blockage states can be directly and continuously obtained, quantitative evaluation and trend prediction are performed through a special analysis model, and synchronous and online quantitative prediction of the blockage and corrosion states of the converter gas cooler is realized.

Description

Technical Field

[0001] This invention belongs to the field of industrial equipment maintenance technology, and specifically relates to a predictive maintenance system and method for converter gas coolers based on multi-source information fusion. Background Technology

[0002] Converter gas coolers are critical equipment in the iron and steel metallurgical process. Operating under harsh conditions of high temperature, high dust, and strong corrosion, they are highly susceptible to blockages caused by accumulated gas dust and corrosion from condensation of acidic gases. In severe cases, this can lead to pipeline leaks, impacting production efficiency and posing significant safety hazards. Traditional equipment maintenance relies primarily on scheduled, planned inspections or reactive maintenance after a failure. This approach cannot accurately predict the actual health status of the equipment, often resulting in under- or over-maintenance, accompanied by high operating costs and the risk of unplanned downtime.

[0003] In recent years, with the development of Industrial Internet of Things (IIoT) and big data technologies, predictive maintenance (PdM) has become an important direction for improving the reliability of industrial equipment. Several general predictive maintenance system frameworks have emerged in existing technologies. For example, PCT patent WO2019216975A1 discloses a general predictive maintenance system for data acquisition, learning, and analysis in industrial environments. It integrates a data processing architecture from the edge to the cloud, utilizing sensor data such as vibration sensors for fault detection and maintenance recommendations. Another PCT patent, WO2019028269A2, also describes the use of multiple sensors, including vibration sensors, for data acquisition and fusion in an IIoT environment to achieve equipment fault diagnosis. These technologies provide general ideas and platform architectures for predictive maintenance. Meanwhile, academic research has also emphasized the core role of data fusion and digital twins in predictive maintenance. For example, the paper "The role of data fusion in predictive maintenance using digital twin" published at the AIP Conference Proceedings points out that data fusion technology can provide more accurate system state estimates for digital twin models, thereby improving the prediction accuracy of remaining useful life (RUL). In the specific field of heat exchange equipment, existing research primarily focuses on predicting fouling (or scaling) problems. For example, Tutpol Ardsomang et al., in their paper "Heat Exchanger Fouling and Estimation of Remaining Useful Life" published at the PHM Society Conference, proposed a data-driven prediction method for heat exchanger scaling, which characterizes equipment degradation and estimates remaining life by modeling process data. However, this method mainly relies on inferences from indirect process parameters and does not address online monitoring of corrosion problems. Furthermore, monitoring techniques for single failure modes also exist. For example, US Patent 7940189B2 describes a system for detecting leaks in industrial process valves using acoustic or pressure sensors. Regarding corrosion monitoring, US Patent 9732369B2 discloses a method for monitoring and assessing the risk of microbial influence corrosion (MIC) in oilfield equipment. While these techniques are effective in specific scenarios, they are isolated solutions for specific components (such as valves) or specific corrosion mechanisms (such as microbial corrosion).

[0004] In summary, existing technologies have the following shortcomings: 1. While general-purpose industrial IoT predictive maintenance platforms (such as those disclosed in WO2019216975A1) have advanced architectures, they lack dedicated sensing methods and analysis models for the complex problems of blockage and acid dew point corrosion faced by specialized equipment like converter gas coolers. 2. Predictive research on heat exchanger blockage (such as that described in "Heat Exchanger Fouling and Estimation of Remaining Useful Life") proposes methods for estimating RUL, but fails to address the equally critical corrosion issue. It lacks direct and continuous measurement of key physical states of the equipment (such as wall thickness), resulting in limited prediction accuracy and reliability. 3. Monitoring technologies targeting single issues such as leakage and corrosion (such as US7940189B2 and US9732369B2) have significantly different application scenarios and monitoring objects compared to converter gas coolers, and their solutions are isolated, failing to provide a comprehensive assessment of the overall health status of the equipment. Therefore, there is an urgent need for a predictive maintenance system specifically designed for converter gas coolers that can integrate multi-source information, especially one that can directly, online, and quantitatively monitor key degradation processes of the equipment (such as corrosion thinning and channel blockage), in order to achieve simultaneous prediction and proactive early warning of various potential faults such as blockage, corrosion, and leakage, thereby fundamentally solving the drawbacks of traditional maintenance models. Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this invention is to provide a predictive maintenance system and method for converter gas coolers based on multi-source information fusion. This invention integrates a dedicated fixed ultrasonic thickness probe and a differential pressure transmitter, enabling for the first time to directly and continuously acquire key physical quantities characterizing corrosion and blockage states. Through a proprietary analysis model, quantitative assessment and trend prediction are performed, achieving synchronous and online quantitative prediction of the blockage and corrosion states of converter gas coolers, thus completing proactive predictive maintenance of converter gas coolers.

[0006] The technical solution of this invention is: a predictive maintenance system for converter gas coolers based on multi-source information fusion, comprising a sensing and data acquisition layer, a data aggregation and edge processing layer, a cloud platform or central server analysis layer, and an application and display layer, wherein: The sensing and data acquisition layer includes a process parameter monitoring module and a physical state monitoring module. The process parameter monitoring module includes a differential pressure transmitter for monitoring the pressure loss on the gas side of the cooler, temperature sensors, pressure sensors, and flow sensors for measuring the inlet and outlet of the gas and cooling water. The physical state monitoring module includes multiple fixed ultrasonic thickness probes, a vibration monitoring sensor for acquiring structural vibration spectrum data, and an acoustic monitoring sensor for listening for abnormal internal acoustic signals. The fixed ultrasonic thickness probes are installed in easily corroded parts of the cooler and are used to continuously acquire time-series data of metal wall thickness. The data aggregation and edge processing layer is communicatively connected to the sensing and data acquisition layer, and is used to preprocess and perform edge computing on the acquired data. The cloud platform or central server analysis layer is communicatively connected to the data aggregation and edge processing layer. The cloud platform or central server analysis layer includes a blockage analysis module, a corrosion analysis module, a leakage analysis module, a digital twin module, and a health status assessment and prediction module. The blockage analysis module receives pressure loss data from the differential pressure transmitter to predict the blockage development trend of the cooler. The corrosion analysis module receives time-series data on metal wall thickness from the fixed ultrasonic thickness gauge and, combined with the cooler's process parameters, calculates the real-time corrosion rate and remaining life using a physical-statistical fusion model. The health status assessment and prediction module integrates the outputs of the blockage analysis module and the corrosion analysis module to assess the overall health status of the cooler and generate predictive maintenance decision recommendations. The leakage analysis module fuses signals from the acoustic monitoring sensor and cooling water system data to identify early leaks. The digital twin module stores the cooler's three-dimensional model, material properties, and design operating parameters as a benchmark model for performance analysis. The application and presentation layer is used to dynamically display the health status of the cooler through a visual human-machine interface, and automatically generate a maintenance work order when an alarm is triggered.

[0007] The leakage analysis module performs time-frequency analysis on the acoustic signals of the acoustic monitoring sensor and calculates the signal energy within a specific frequency band. It compares the moving average of the acoustic signal energy with the historical baseline. When the energy continuously exceeds the preset standard deviation threshold and the cooling water replenishment volume simultaneously exceeds the preset percentage of normal loss, it is determined to be an early leakage event.

[0008] The physical-statistical fusion model used in the corrosion analysis module is a fusion model based on Kalman filtering. This model uses the acid dew point corrosion theory, which calculates the theoretical corrosion rate based on gas composition and temperature, as the equation of state, and uses the time series data of metal wall thickness collected by the fixed ultrasonic thickness measuring probe as the observation equation.

[0009] The blockage analysis module uses a regression model based on historical data to extrapolate the trend of pressure loss time series data. The regression model is either a linear regression model or a quadratic polynomial regression model.

[0010] The application and display layer dynamically displays the device health status in the form of a red, yellow, and green indicator light dashboard. When the health status assessment and prediction module triggers an early warning condition, it automatically generates a maintenance work order containing the fault type, risk level, recommended measures, and maintenance window, and pushes it to the maintenance management system.

[0011] A predictive maintenance system method for converter gas coolers based on multi-source information fusion, using the predictive maintenance system for converter gas coolers based on multi-source information fusion as described above, includes the following steps: S1: The process parameter monitoring module monitors the pressure loss data on the gas side of the cooler and the temperature, pressure and flow data of the gas and cooling water inlet and outlet of the cooler in real time. The physical state monitoring module continuously collects the time series data of the metal wall thickness of the easily corroded parts of the cooler. S2: The data aggregation and edge processing layer preprocesses the data collected by the sensing and data acquisition layer in step S; S3: Input the pressure loss data after data aggregation and edge processing to the blockage analysis module of the cloud platform or central server analysis layer to predict the blockage development trend; input the metal wall thickness time series data and related cooler process parameters to the corrosion analysis module, and calculate the real-time corrosion rate and remaining life through the physical-statistical fusion model; S4: Combining the blockage development trend and real-time corrosion rate obtained in step S3, the overall health status of the cooler is assessed and predictive maintenance decision recommendations are generated through the health status assessment and prediction module. S5: Based on the evaluation results in step S4, the application and display layer dynamically displays the health status, predictive alarm information, performance trend chart and remaining life prediction of the converter gas cooler through a visual human-machine interface.

[0012] In step S3, the data collected and processed in real time is compared and analyzed with the benchmark parameters stored in the digital twin module.

[0013] The prediction of the blockage development trend in step S3 specifically includes: using a quadratic polynomial regression model to fit and extrapolate the differential pressure time series data, retraining the model daily using data from the past preset number of days, and predicting the time when the cleaning decision threshold is reached.

[0014] The technical advantages of this invention are as follows: 1. By integrating a dedicated fixed ultrasonic thickness probe and a differential pressure transmitter, this invention, for the first time, can directly and continuously acquire key physical quantities characterizing corrosion and blockage states. Through a proprietary analytical model, it performs quantitative evaluation and trend prediction, achieving simultaneous, online, and quantitative prediction of the two core faults of converter gas coolers: blockage and corrosion. This fills the technical gap in existing technologies that cannot simultaneously monitor these two issues online. 2. Through precise calculation of corrosion rate and remaining life, and real-time tracking of blockage trends, this system can provide accurate early warnings before catastrophic failures such as wall thinning, perforation, or severe blockage occur, effectively preventing major safety hazards such as gas leaks. 1. This invention ensures production safety and significantly improves the safety and reliability of equipment operation; 2. This invention can generate accurate maintenance decision suggestions based on the actual health status of the equipment, avoiding over-maintenance or under-maintenance in the traditional periodic maintenance mode, reducing unplanned downtime, improving the overall utilization rate of equipment, and reducing spare parts inventory and maintenance costs; 3. This invention, by introducing digital twins, edge computing, and a visualization application layer, constructs a complete technical architecture from data acquisition, edge processing, cloud analysis to decision execution, transforming complex analysis results into intuitive health status assessments and automated maintenance instructions, realizing the transformation from passive maintenance to predictive maintenance, and significantly reducing operation and maintenance costs.

[0015] The following will provide further explanation in conjunction with the accompanying drawings. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of a predictive maintenance system for converter gas coolers based on multi-source information fusion, according to the present invention.

[0017] Figure 2 This is a flowchart of a predictive maintenance method for converter gas coolers based on multi-source information fusion, according to the present invention. Detailed Implementation Example 1

[0018] like Figure 1 As shown, a predictive maintenance system for converter gas coolers based on multi-source information fusion includes a sensing and data acquisition layer 100, a data aggregation and edge processing layer 200, a cloud platform or central server analysis layer 300, and an application and display layer 400, wherein: The sensing and data acquisition layer 100 includes a process parameter monitoring module 110 and a physical state monitoring module 120. The process parameter monitoring module 110 includes a differential pressure transmitter for monitoring the pressure loss on the gas side of the cooler, a temperature sensor for measuring the gas and cooling water inlet and outlet, a pressure sensor, and a flow sensor. The physical state monitoring module 120 includes multiple fixed ultrasonic thickness probes, a vibration monitoring sensor for acquiring structural vibration spectrum data, and an acoustic monitoring sensor for listening to internal abnormal acoustic signals. The fixed ultrasonic thickness probes are installed in easily corroded parts of the cooler and are used to continuously acquire time-series data of metal wall thickness. The data aggregation and edge processing layer 200 is communicatively connected to the sensing and data acquisition layer 100 and is used to preprocess and perform edge computing on the acquired data. The cloud platform or central server analysis layer 300 is communicatively connected to the data aggregation and edge processing layer 200. The cloud platform or central server analysis layer 300 includes a blockage analysis module 310, a corrosion analysis module 320, a leakage analysis module 330, a digital twin module 340, and a health status assessment and prediction module 350. The blockage analysis module 310 receives pressure loss data from the differential pressure transmitter to predict the blockage development trend of the cooler. The corrosion analysis module 320 receives time-series data on metal wall thickness from the fixed ultrasonic thickness probe and combines it with the... The process parameters of the cooler are used to calculate the real-time corrosion rate and remaining life through a physical-statistical fusion model. The health status assessment and prediction module 350 integrates the outputs of the blockage analysis module 310 and the corrosion analysis module 320 to assess the overall health status of the cooler and generate predictive maintenance decision recommendations. The leakage analysis module 330 is used to fuse the signals from the acoustic monitoring sensor and the cooling water system data to identify early leaks. The digital twin module 340 is used to store the three-dimensional model, material properties, and design operating parameters of the cooler as a benchmark model for performance analysis. The application and presentation layer 400 is used to dynamically display the health status of the cooler through a visual human-machine interface, and automatically generate a maintenance work order when an alarm is triggered.

[0019] The leakage analysis module 330 performs time-frequency analysis on the acoustic signal of the acoustic monitoring sensor and calculates the signal energy in a specific frequency band. It compares the moving average of the acoustic signal energy with the historical baseline. When the energy continuously exceeds the preset standard deviation threshold and the cooling water replenishment volume also exceeds the preset percentage of normal loss, it is determined to be an early leakage event.

[0020] The physical-statistical fusion model used in the corrosion analysis module 320 is a fusion model based on Kalman filtering. This model uses the acid dew point corrosion theory, which calculates the theoretical corrosion rate based on the gas composition and temperature, as the equation of state, and uses the time series data of metal wall thickness collected by the fixed ultrasonic thickness measuring probe as the observation equation.

[0021] The blockage analysis module 310 uses a regression model based on historical data to extrapolate the trend of pressure loss time series data. The regression model is either a linear regression model or a quadratic polynomial regression model.

[0022] The application and display layer 400 dynamically displays the device health status in the form of a red, yellow, and green three-color light dashboard; when the health status assessment and prediction module 330 triggers the early warning condition, it automatically generates a maintenance work order containing the fault type, risk level, suggested measures, and maintenance window, and pushes it to the maintenance management system.

[0023] The sensing and data acquisition layer of this invention is responsible for acquiring raw data on the equipment status. This layer specifically includes a process parameter monitoring module and a physical condition monitoring module. The process parameter monitoring module includes a differential pressure transmitter for real-time monitoring of the pressure loss on the gas side of the cooler, as well as sensors for measuring the inlet and outlet temperatures, pressures, and flow rates of the gas and cooling water. The physical condition monitoring module includes multiple fixed ultrasonic thickness gauges permanently installed on easily corroded parts of the cooler (such as the shell, heat exchanger tube bends, etc.) for continuously acquiring time-series data on metal wall thickness. Furthermore, this module can also integrate a vibration monitoring sensor for acquiring structural vibration spectrum data and an acoustic monitoring sensor for detecting internal abnormal acoustic signals.

[0024] The data aggregation and edge processing layer, located between the sensing and analysis layers, is used to clean, format, and preprocess the collected multi-source data. Through an edge computing gateway, this layer can perform real-time calculations on high-frequency data, such as calculating real-time heat transfer coefficients or fouling thermal resistance based on collected temperature and flow data. This reduces the computational burden on cloud servers and improves system response speed.

[0025] The cloud platform / central server analysis layer is the core of the system, communicating with the sensing and data acquisition layer to perform in-depth data analysis and prediction. This analysis layer includes a digital twin module, which pre-stores the 3D model, material properties, and design parameters of the gas cooler as a benchmark model for performance analysis. The analysis layer also includes several other analysis modules: a blockage analysis module, configured to receive pressure loss data from a differential pressure transmitter and, combined with the digital twin benchmark, predict the blockage development trend of the cooler using a trend extrapolation algorithm; a corrosion analysis module, configured to receive time-series data on metal wall thickness from a fixed ultrasonic thickness gauge and, combined with the cooler's process parameters (such as gas temperature and composition), calculate the real-time corrosion rate and remaining life using a physical-statistical fusion model; and a leakage analysis module, used to fuse signals from the acoustic monitoring unit and cooling water system water balance data, using an anomaly detection algorithm to identify early leaks. Finally, a health status assessment and prediction module is configured to integrate the outputs of all the above analysis modules to assess the overall health status of the cooler and generate predictive maintenance decision recommendations.

[0026] As a preferred implementation, the physical-statistical fusion model used in the corrosion analysis module is a fusion model based on Kalman filtering. This model innovatively uses the physical acid dew point corrosion theory (e.g., the theoretical corrosion rate calculated based on gas composition and temperature) as the equation of state, and the time-series data of metal wall thickness obtained in the acquisition step as the observation equation. Through filtering iteration, it achieves accurate and dynamic estimation of the actual corrosion rate and remaining life.

[0027] The application and presentation layer serves as the interface between the system and the user. This layer uses a visual human-machine interface (such as a web interface or a large mobile screen) to dynamically display the cooler's health status, predictive alarm information, performance trend graphs, and remaining life predictions in an intuitive manner, using red, yellow, and green indicator lights. When the health status assessment and prediction module triggers an alert, the system can automatically generate a maintenance work order containing the fault type, risk level, recommended measures, and maintenance window, and push it to relevant maintenance personnel, achieving closed-loop management from monitoring and analysis to decision-making and execution. Example 2

[0028] like Figure 2 As shown, a predictive maintenance system method for converter gas coolers based on multi-source information fusion is described above. The method includes the following steps: S1: The process parameter monitoring module 110 monitors the pressure loss data on the gas side of the cooler and the temperature, pressure and flow data of the gas and cooling water inlet and outlet of the cooler in real time. The physical state monitoring module 120 continuously collects the time series data of the metal wall thickness of the easily corroded parts of the cooler. S2: The data aggregation and edge processing layer 200 preprocesses the data collected by the sensing and data acquisition layer 100 in step S1; S3: Input the pressure loss data after data aggregation and edge processing layer 200 into the blockage analysis module 310 of the cloud platform or central server analysis layer 300 to predict the blockage development trend; input the metal wall thickness time series data and related cooler process parameters into the corrosion analysis module 320, and calculate the real-time corrosion rate and remaining life through the physical-statistical fusion model; S4: Combining the blockage development trend and real-time corrosion rate obtained in step S3, the overall health status of the cooler is assessed and predictive maintenance decision recommendations are generated through the health status assessment and prediction module 350. S5: Based on the evaluation results in step S4, the application and display layer 400 dynamically displays the health status, predictive alarm information, performance trend chart and remaining life prediction of the converter gas cooler through a visual human-machine interface.

[0029] In step S3, the data collected and processed in real time is compared and analyzed with the benchmark parameters stored in the digital twin module 340.

[0030] The prediction of the blockage development trend in step S3 specifically includes: using a quadratic polynomial regression model to fit and extrapolate the differential pressure time series data, retraining the model daily using data from the past preset number of days, and predicting the time when the cleaning decision threshold is reached. Example 3

[0031] A predictive maintenance system method for converter gas coolers based on multi-source information fusion is proposed. Using the predictive maintenance system for converter gas coolers based on multi-source information fusion as described in Example 1, simultaneous prediction of corrosion and blockage is performed based on a Kalman filter fusion model. The process is as follows: This embodiment focuses on a saturated steam-gas cooler for a 180-ton converter in a steel plant. The heat exchange tube material is 20g, and the designed wall thickness is 12mm. The system deployment of this invention is as follows: Principle and Mechanism: The core of this embodiment lies in the physical-statistical fusion model based on Kalman filtering employed in the corrosion analysis module. In the physical model, based on the typical components of converter gas (CO, CO2, H2S, H2O, etc.) and real-time temperature, a basic physical equation for the corrosion rate, vphy = f(T, Cgas), is established using acid dew point corrosion theory. This equation serves as the state equation for the Kalman filter, used to predict the wall thickness at the next moment. In the statistical model, the wall thickness time-series data dmeasure collected by a fixed ultrasonic thickness gauge is used as the observation equation to correct the predicted values ​​of the physical model. The Kalman filter, through recursive calculation, optimally fuses the prediction uncertainty of the physical model with the noise uncertainty of the sensor measurements, outputting a wall thickness estimate and real-time corrosion rate that more closely approximates the true value.

[0032] System deployment and parameter settings: 1. Sensing and Data Acquisition Layer: Permanently install 5MHz fixed ultrasonic thickness gauges at five key corrosion points, including the gas inlet section of the cooler and the bends in the heat exchanger tube bundle, with a sampling frequency of once per hour. Install differential pressure transmitters with a range of 0-15kPa on the main gas inlet and outlet pipes, with a sampling frequency of once per minute. Temperature, pressure, and flow sensors are configured as usual.

[0033] 2. Data aggregation and edge processing layer: The edge computing gateway calculates the fouling thermal resistance Rf in real time. The calculation formula is Rf(t) = (1 / U(t)) - (1 / Uclean), where U(t) is the instantaneous total heat transfer coefficient calculated based on real-time temperature and flow data, and Uclean is the total heat transfer coefficient stored in the digital twin model under the clean state.

[0034] 3. Cloud Platform / Central Server Analysis Layer: The digital twin module stores baseline parameters: initial wall thickness 12mm, clean state pressure difference 2.5kPa. The blockage analysis module performs linear regression on the Rf and pressure difference time series data to predict when the pressure difference will reach the preset cleaning threshold of 6.0kPa. The corrosion analysis module runs the above Kalman filter model and triggers a high-level warning when the predicted remaining lifetime (RUL) is less than 6 months.

[0035] 4. Application and Presentation Layer: The visual interface displays the health status (out of 100 points) in a dashboard format, with a red light indicating a score below 60. When the RUL (Round Usage Limit) is less than 6 months or the predicted blockage cleaning time is less than 15 days, an A-level maintenance work order is automatically generated.

[0036] Key parameters: Ultrasonic probe sampling frequency: 1 time / hour; Differential pressure transmitter sampling frequency: 1 time / minute; Blockage warning threshold (differential pressure): 6.0 kPa; Corrosion warning threshold (remaining life): <6 months; Implementation steps: Step 1: System Deployment and Initialization. Install all sensors on the cooler and build a digital twin model on the cloud platform, inputting design parameters as a baseline; Step 2: Continuous data acquisition and edge preprocessing; After the system starts, each sensor continuously acquires data, the edge gateway calculates the fouling thermal resistance and pressure difference in real time, and uploads the processed data to the cloud platform; Step 3: Cloud-based integrated analysis; The blockage analysis module performs trend extrapolation prediction on differential pressure and fouling thermal resistance data every hour; The corrosion analysis module runs the Kalman filter model every 24 hours to update the real-time corrosion rate and RUL of all measuring points; Step 4: Health assessment and decision generation; The health status assessment module integrates the congestion prediction time and the minimum RUL, and updates the total health score of the equipment every 24 hours; If any indicator reaches the warning threshold, a corresponding level of maintenance recommendation is generated. Step 5: Results Display and Work Order Issuance; Users can view health status, data trend charts, and RUL at any time through the web interface; after an alert is triggered, the system automatically sends a work order containing the fault location, type, and recommended measures to the maintenance department's MES system.

[0037] Technical Results: After three months of system operation, the system successfully monitored that at monitoring point 2, due to localized flue gas erosion and low temperature, the corrosion rate reached 0.8 mm / year, significantly higher than other areas. The system predicted a RUL (Range Ultra-Likely Usage Time) of 15 months and issued a yellow alert. Simultaneously, differential pressure data showed a steady upward trend, and the system predicted that the cleaning threshold would be reached within the next 40 days. This information provided precise data support for the maintenance department to plan key inspection areas and cleaning operations for the next overhaul. Example 4

[0038] A predictive maintenance system method for converter gas coolers based on multi-source information fusion is presented. Using the predictive maintenance system for converter gas coolers based on multi-source information fusion as described in Example 1, early leak detection is performed by integrating acoustic signals. This embodiment, based on the system in Example 3, further integrates leak detection functionality for the same gas cooler. The physical condition monitoring module adds two wideband (20kHz-100kHz) ultrasonic acoustic sensors, installed on the outer wall of the cooler's inlet and outlet water pipes, to listen for ultrasonic signals generated by early leaks on the water or gas side.

[0039] The leak analysis module employs an energy anomaly-based detection algorithm. This algorithm first performs time-frequency analysis (e.g., short-time Fourier transform) on the acquired acoustic signals to calculate the signal energy within the 25kHz-60kHz frequency band. Simultaneously, it obtains the cooling water replenishment data from the factory's DCS system. The algorithm compares the moving average of the acoustic signal energy with a historical baseline. If a continuous energy spike exceeding three standard deviations occurs, accompanied by an unexplained small increase in cooling water replenishment (e.g., greater than 0.5% of normal evaporation loss), it is identified as an early leak event.

[0040] Key parameters: Acoustic sensor frequency range: 20kHz - 100kHz; Leak detection frequency band: 25kHz - 60kHz; Acoustic energy anomaly threshold: >3 standard deviations; Water balance anomaly threshold: >0.5% of normal loss; Parameter range: The turbulence and bubble bursts generated by a leak produce high-frequency acoustic emission signals. 25kHz-60kHz is a typical leak signal frequency band, effectively avoiding low-frequency mechanical noise in the factory environment. Fusion of the acoustic signals with water balance data can significantly improve the accuracy of leak detection and avoid interference caused by false alarms from a single sensor.

[0041] Implementation steps: Step 1: Based on the hardware of Example 1, add an acoustic sensor and configure a leakage analysis module; Step 2: The system continuously monitors acoustic signals and water replenishment data, and establishes a data baseline under normal operating conditions; Step 3: The leakage analysis module performs real-time calculations for the anomaly detection algorithm; Step 4: When an abnormality in acoustic energy and water balance is detected simultaneously, the system immediately triggers the highest level of leak alarm and automatically generates an emergency inspection work order.

[0042] Technical effects: In a simulation test, a minor leak (approximately 0.1% of the total circulating water volume) was simulated by slightly opening a bypass valve. Within 3 minutes, the system successfully detected a continuous anomaly in the acoustic signal and, combined with subtle fluctuations in water volume data, accurately issued an early leak warning. Meanwhile, traditional pressure gauges and level gauges showed no significant changes, demonstrating the system's superior performance in identifying early-stage minor leaks. Example 5

[0043] A predictive maintenance system method for converter gas coolers based on multi-source information fusion is disclosed. This method utilizes a predictive maintenance system for converter gas coolers based on multi-source information fusion as described in Example 1 to predict blockage trends and optimize cleaning under high-dust conditions. This example is applied to a gas cooler that processes primary converter flue gas (without deep dust removal), characterized by high dust content and rapid blockage development. In this scenario, the system of this invention focuses on leveraging the blockage analysis module.

[0044] The blockage analysis module uses a quadratic polynomial regression model to fit and extrapolate the differential pressure time series data. The model is... ,in Here, t represents pressure difference, and t represents time. The system retrains the model parameters a, b, and c daily using data from the past week, and then solves the equations. This allows for the prediction of the congestion time. This nonlinear model is better able to capture the trend of accelerated development in the later stages of congestion.

[0045] Key parameters: Congestion prediction model: quadratic polynomial regression; model training data window: past 7 days; cleaning decision threshold: predicted congestion time < 7 days.

[0046] Application scenarios: In this scenario, the cleaning cycle of the cooler is crucial. Too frequent cleaning leads to unnecessary downtime and resource waste, while delayed cleaning results in a sharp decline in heat exchange efficiency and even safety risks. The goal of this embodiment is to accurately predict the optimal cleaning time so that it can be incorporated into the plant's shortest planned downtime window (e.g., weekend shift maintenance).

[0047] Implementation steps: Step 1: Deploy a sensing and data acquisition system similar to that in Example 1; Step 2: The blockage analysis module automatically retrieves the latest differential pressure data daily and updates the quadratic polynomial regression model; Step 3: The system compares the predicted cleaning dates with the factory's production and maintenance schedule; Step 4: When the predicted cleaning date is earlier than the next planned maintenance date, the system will issue an early warning and suggest that the cleaning operation be given higher priority or that a temporary maintenance window be requested.

[0048] Technical effects: After implementing this system, the cleaning cycle of the cooler changed from a fixed "once every two months" to a dynamic cycle of "35-50 days" based on prediction. During the six-month operation period, the system successfully predicted three blockage events, issuing warnings an average of 10 days in advance, ensuring that all cleaning operations could be completed on schedule. This completely avoided unplanned production reductions or downtime caused by blockages, resulting in an overall improvement in operating efficiency of approximately 15%. Example 6

[0049] The converter gas cooler described in Example 3 and the system of the present invention were used for continuous monitoring for a period of 6 months.

[0050] Experimental Results: On day 95, the system detected an abnormal corrosion rate at measuring point 2, predicting a RUL of 15 months and issuing a yellow alert. On day 120, the system predicted that the equipment pressure differential would reach the cleaning threshold in 40 days and issued a cleaning preparation notice. Based on the alert information, maintenance personnel conducted a focused inspection and reinforcement of the area around measuring point 2 during planned maintenance on day 160, and completed the dust removal operation on time. There were no unplanned shutdowns throughout the entire cycle.

[0051] Data Analysis: Through online and continuous data acquisition and model analysis, the system transforms potential corrosion and clogging problems from "unknown" to "known" and "predictable," enabling a shift from reactive maintenance to proactive and predictive maintenance. Maintenance decisions are based on quantitative data, significantly improving accuracy and efficiency.

[0052] Comparative Example 1 The existing periodic maintenance method was adopted: Experimental conditions: On the same converter gas cooler as in Example 6, the traditional maintenance method was adopted, that is, only the conventional process parameters (main pipe temperature and pressure) were monitored, and the machine was shut down every 6 months (180 days) for opening the cover and manually measuring the thickness.

[0053] Experimental Results: During operation, operators did not detect any significant abnormalities in the routine process parameters. During the 180th day of shutdown for maintenance, manual ultrasonic thickness measurement revealed that the wall thickness in area 2 had been severely reduced, with the remaining thickness only 110% of the safe thickness. Due to the urgency of the situation, the originally planned 2-day maintenance was extended to 5 days for emergency repairs. Simultaneously, it was discovered that internal blockages were quite severe, resulting in a dust removal workload and time far exceeding expectations.

[0054] Comparative analysis: Compared to Embodiment 6 of this invention, existing maintenance methods have significant blind spots. They cannot monitor the accelerated development of localized corrosion online, nor can they quantify the progress of blockage. This often results in problems being discovered when they are nearing a critical state, leading to higher safety risks and longer unplanned downtime (an additional 3 days of downtime). The system of this invention provides early warning of corrosion risks 65 days in advance (day 95 vs. day 160) and early warning of blockages 40 days in advance, transforming emergency repairs into planned maintenance, avoiding at least 3 days of production loss, and significantly improving the safety of equipment operation.

[0055] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A predictive maintenance system for converter gas coolers based on multi-source information fusion, characterized in that: It includes a sensing and data acquisition layer (100), a data aggregation and edge processing layer (200), a cloud platform or central server analysis layer (300), and an application and presentation layer (400), wherein: The sensing and data acquisition layer (100) includes a process parameter monitoring module (110) and a physical state monitoring module (120). The process parameter monitoring module (110) includes a differential pressure transmitter for monitoring the pressure loss on the gas side of the cooler, a temperature sensor for measuring the gas and cooling water inlet and outlet, a pressure sensor, and a flow sensor. The physical state monitoring module (120) includes multiple fixed ultrasonic thickness probes, a vibration monitoring sensor for collecting structural vibration spectrum data, and an acoustic monitoring sensor for listening to internal abnormal acoustic signals. The fixed ultrasonic thickness probes are installed in easily corroded parts of the cooler and are used to continuously collect time-series data of metal wall thickness. The data aggregation and edge processing layer (200) is communicatively connected to the sensing and data acquisition layer (100) and is used to preprocess and perform edge computing on the acquired data; The cloud platform or central server analysis layer (300) is communicatively connected to the data aggregation and edge processing layer (200). The cloud platform or central server analysis layer (300) includes a blockage analysis module (310), a corrosion analysis module (320), a leakage analysis module (330), a digital twin module (340), and a health status assessment and prediction module (350). The blockage analysis module (310) receives pressure loss data from the differential pressure transmitter to predict the blockage development trend of the cooler. The corrosion analysis module (320) receives time-series data of metal wall thickness from the fixed ultrasonic thickness probe. In conjunction with the process parameters of the cooler, the real-time corrosion rate and remaining life are calculated through a physical-statistical fusion model; the health status assessment and prediction module (350) integrates the outputs of the blockage analysis module (310) and the corrosion analysis module (320) to assess the overall health status of the cooler and generate predictive maintenance decision recommendations; the leakage analysis module (330) is used to fuse the signals of the acoustic monitoring sensor and the cooling water system data to identify early leaks; the digital twin module (340) is used to store the three-dimensional model, material properties and design operating parameters of the cooler as a benchmark model for performance analysis; The application and presentation layer (400) is used to dynamically display the health status of the cooler through a visual human-machine interface, and automatically generate a maintenance work order when an alarm is triggered.

2. The predictive maintenance system for converter gas coolers based on multi-source information fusion according to claim 1, characterized in that: The leakage analysis module (330) performs time-frequency analysis on the acoustic signal of the acoustic monitoring sensor and calculates the signal energy in a specific frequency band. It compares the moving average of the acoustic signal energy with the historical baseline. When the energy continuously exceeds the preset standard deviation threshold and the cooling water replenishment volume also exceeds the preset percentage of normal loss, it is determined to be an early leakage event.

3. The predictive maintenance system for converter gas coolers based on multi-source information fusion according to claim 1, characterized in that: The physical-statistical fusion model used in the corrosion analysis module (320) is a fusion model based on Kalman filtering. This model uses the acid dew point corrosion theory, which calculates the theoretical corrosion rate based on the gas composition and temperature, as the equation of state, and uses the time series data of metal wall thickness collected by the fixed ultrasonic thickness measuring probe as the observation equation.

4. The predictive maintenance system for converter gas coolers based on multi-source information fusion according to claim 1, characterized in that: The blockage analysis module (310) uses a regression model based on historical data to extrapolate the trend of pressure loss time series data. The regression model is a linear regression model or a quadratic polynomial regression model.

5. The predictive maintenance system for converter gas coolers based on multi-source information fusion according to claim 1, characterized in that: The application and display layer (400) dynamically displays the health status of the equipment in the form of a red, yellow and green three-color light dashboard; when the health status assessment and prediction module (330) triggers the early warning condition, it automatically generates a maintenance work order containing the fault type, risk level, suggested measures and maintenance window, and pushes it to the maintenance management system.

6. A predictive maintenance system method for converter gas coolers based on multi-source information fusion, using the predictive maintenance system for converter gas coolers based on multi-source information fusion as described in claim 1, characterized in that: Includes the following steps: S1: The pressure loss data of the gas side of the cooler and the temperature, pressure and flow data of the gas and cooling water inlet and outlet of the cooler are monitored in real time through the process parameter monitoring module (110). The metal wall thickness time series data of the easily corroded parts of the cooler are continuously collected through the physical state monitoring module (120). S2: The data aggregation and edge processing layer (200) preprocesses the data collected by the sensing and data acquisition layer (100) in step S1; S3: Input the pressure loss data after data aggregation and processing by the edge processing layer (200) into the congestion analysis module (310) of the cloud platform or central server analysis layer (300) to predict the congestion development trend; The time series data of metal wall thickness and the process parameters of the relevant cooler are input into the corrosion analysis module (320), and the real-time corrosion rate and remaining lifetime are calculated by the physical-statistical fusion model. S4: Combining the blockage development trend and real-time corrosion rate obtained in step S3, the overall health status of the cooler is assessed and predictive maintenance decision recommendations are generated through the health status assessment and prediction module (350). S5: Based on the evaluation results in step S4, the application and display layer (400) dynamically displays the health status, predictive alarm information, performance trend chart and remaining life prediction of the converter gas cooler through a visual human-machine interface.

7. The predictive maintenance system method for converter gas coolers based on multi-source information fusion according to claim 6, characterized in that: In step S3, the data collected and processed in real time is compared and analyzed with the benchmark parameters stored in the digital twin module (340).

8. The predictive maintenance system method for converter gas coolers based on multi-source information fusion according to claim 6, characterized in that: The prediction of the blockage development trend in step S3 specifically includes: using a quadratic polynomial regression model to fit and extrapolate the differential pressure time series data, retraining the model daily using data from the past preset number of days, and predicting the time when the cleaning decision threshold is reached.

Images