Blast furnace cooling wall damage probability prediction method

Abstract

The invention discloses a blast furnace cooling wall damage probability prediction method, which relates to the field of industrial equipment maintenance, and comprises the following steps: arranging a temperature sensor array, a vibration sensor and an acoustic emission sensor, and collecting temperature gradient data, mechanical vibration frequency and high-frequency stress wave signals in real time; calculating the slag crust thickness of the hot surface of the cooling wall based on an unsteady state heat transfer equation, introducing an in-furnace gas flow parameter correction model, and generating a dynamic slag crust thickness distribution cloud picture; establishing temperature and vibration reference threshold values under different working conditions according to historical operation data of the cooling wall, and detecting temperature over-limit accumulated duration and a vibration energy spectrum abnormal frequency band in real time; taking the temperature gradient range, the slag skin thickness variation coefficient and the vibration dominant frequency offset as input characteristics, training a multi-dimensional coupled damage probability prediction model through a random forest algorithm, and outputting a recent dynamic damage probability value of each section of cooling wall; and dividing risk grades according to the damage probability value, dynamically adjusting a threshold interval in combination with the real-time smelting strength of the blast furnace, and generating a differentiated maintenance suggestion set.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment maintenance, and in particular to a method for predicting the probability of damage to a blast furnace cooling wall. Background Art

[0002] Blast furnaces are crucial equipment in steel production, and their internal cooling staves play a crucial role in protecting the furnace shell and maintaining normal operation. With the development of larger blast furnaces and higher-intensity ironmaking processes, the cooling staves face increasingly harsh operating environments. Predicting the probability of damage to these staves is crucial for ensuring safe production and extending the life of the blast furnace.

[0003] In modern industrial production, preventive maintenance of equipment is gaining increasing importance. Accurately predicting equipment failure probability allows for rational maintenance planning, reduces maintenance costs, and improves equipment reliability and production efficiency. Predicting the failure probability of blast furnace cooling staves, a key component of blast furnaces, is crucial for intelligent maintenance. With the advancement of sensor technology, big data analytics, and machine learning, fault prediction and diagnosis technologies are becoming increasingly widespread in industry. Real-time monitoring and analysis of equipment operating data can proactively identify potential faults, providing a scientific basis for equipment maintenance and management.

[0004] Existing technologies typically rely on periodic manual inspections and simple temperature monitoring to monitor the status of blast furnace cooling staves. This approach suffers from delayed detection and low accuracy, making it impossible to effectively predict the risk of cooling stave damage, impacting the long-term stable operation and safe production of the blast furnace. The lack of a real-time, accurate monitoring method for blast furnace cooling stave status makes it difficult to provide early warning of cooling stave damage risks, threatening the stability and safety of blast furnace operations. Furthermore, the inability to accurately quantify the probability of cooling stave damage limits the optimization and adjustment of blast furnace operating strategies. Summary of the Invention

[0005] In order to solve the above technical problems, a method for predicting the probability of blast furnace cooling wall damage is provided. This technical solution solves the above-mentioned lack of real-time and accurate blast furnace cooling wall status monitoring methods, which makes it difficult to provide early warning of the risk of cooling wall damage, resulting in threats to the stability and safety of blast furnace operation. It is impossible to accurately quantify the probability of cooling wall damage, which limits the optimization and adjustment of blast furnace operation strategies.

[0006] In order to achieve the above objects, the technical solution adopted by the present invention is: A method for predicting the probability of damage of a blast furnace cooling stave, comprising: Temperature sensor arrays are evenly distributed on the surface of each cooling wall, and vibration sensors and acoustic emission sensors are installed inside the cooling wall to collect temperature gradient data, mechanical vibration frequency, and high-frequency stress wave signals in real time; The slag thickness on the hot surface of the cooling wall is calculated based on the unsteady heat transfer equation, and a correction model for the gas flow parameters in the furnace is introduced to generate a dynamic slag thickness distribution cloud map. Based on the historical operating data of the cooling wall, the temperature and vibration benchmark thresholds under different working conditions are established. A sliding window algorithm is used to detect the cumulative duration of temperature exceeding the limit and the abnormal frequency band of the vibration energy spectrum in real time, and the number of abnormal event triggering in a single day is counted. The temperature gradient range, slag thickness variation coefficient, and vibration main frequency offset are used as input features. A multi-dimensional coupled damage probability prediction model is trained using the random forest algorithm to output the recent dynamic damage probability value of each cooling wall segment. The risk level is divided according to the damage probability value, the threshold range is dynamically adjusted based on the real-time smelting intensity of the blast furnace, and a differentiated maintenance recommendation set is generated.

[0007] Preferably, the method of evenly distributing a temperature sensor array on the surface of each cooling wall section, installing a vibration sensor and an acoustic emission sensor inside the cooling wall section, and collecting temperature gradient data, mechanical vibration frequency, and high-frequency stress wave signals in real time specifically includes: evenly distributing monitoring points on the surface of each cooling wall section, installing digital temperature sensors at the monitoring points, and eliminating measurement errors caused by thermal expansion through a temperature compensation circuit; Collect the temperature values of each sensor in real time, generate a temperature field distribution map through spatial interpolation algorithm, calculate the temperature difference between adjacent sensors; combine with sliding window algorithm to calculate the cumulative duration of temperature exceeding the limit; Piezoelectric accelerometers are installed at key locations inside the cooling wall to detect mechanical vibration frequency. A three-axis vibration sensor is used to monitor its vibration energy, and the main frequency offset is extracted through frequency domain analysis. The original vibration signal is subjected to anti-aliasing filtering and FFT transformation to extract energy spectrum density data. An abnormal frequency band identification algorithm is used to detect sudden energy increases in the high-frequency band and determine whether the internal structure of the cooling wall is loose or whether microcracks are expanding. A broadband piezoelectric ceramic array is used to capture microcrack expansion and macroscopic deformation signals. Acoustic emission sensors are embedded in the cooling wall, and coupling agent sealing technology is used to ensure signal transmission efficiency. Electromagnetic shielding shells are used to reduce electromagnetic interference from the blast furnace. Real-time recording of acoustic emission event count rate and waveform parameters, combined with pattern recognition algorithms, can distinguish normal thermal stress fluctuations from abnormal crack signals.

[0008] Preferably, the calculation of the slag skin thickness on the hot surface of the cooling stave based on the unsteady-state heat transfer equation specifically includes: The heat transfer model is established based on the one-dimensional unsteady heat conduction differential equation. The finite difference method is used to discretize space and time, and the temperature field is iteratively calculated using explicit and implicit schemes. The equivalent specific heat method is introduced to deal with the latent heat of slag skin solidification and melting, and the functional relationship between thermal conductivity and temperature diffusivity with temperature is updated. According to the temperature field distribution, the position of the solid-liquid interface of the slag skin is calculated through the heat balance equation, and the thickness change rate is derived. The actual thickness of the slag skin is measured periodically using a laser rangefinder, and a temperature-thickness feedback compensation mechanism is established.

[0009] Preferably, the introducing of the furnace gas flow parameter correction model to generate a dynamic slag skin thickness distribution cloud map specifically includes: The gas flow velocity, pressure, and temperature distribution parameters are introduced, and the convective heat transfer coefficient is corrected using the Reynolds number and Nusselt number. The local heat flux density changes caused by the gas flow are embedded in the heat transfer equation to correct the slag skin thickness calculation model. By dividing the cooling wall surface into grid nodes and using a bilinear interpolation algorithm, the discrete thickness data is converted into a continuous distribution field. A visualization module is developed based on MATLAB to generate a pseudo-color cloud map that updates over time and a dynamic slag skin thickness distribution cloud map. Compare the model-predicted thickness with the laser-measured data, calculate the root mean square error, and optimize the mesh density and boundary condition parameters.

[0010] Preferably, the introducing of the furnace gas flow parameter correction model to generate a dynamic slag skin thickness distribution cloud map specifically includes: Based on the one-dimensional unsteady heat conduction differential equation: Where, is the rate of change of temperature with time, a signature term for unsteady-state heat conduction; To describe the curvature of the temperature field in space and reflect the driving force of heat diffusion; α is the thermal conductivity of the slag skin, T is the temperature field, x is the spatial coordinate, and t is the time; consider the third type of boundary conditions: Where λ is the thermal conductivity of the material, is the temperature change rate inside the object along the normal x-axis direction, h is the convection heat transfer coefficient of the coal gas to the slag skin in the furnace, and Tf is the coal gas temperature.

[0011] Preferably, establishing the temperature and vibration reference thresholds under different working conditions based on the historical operation data of the cooling wall specifically includes: The operating conditions are categorized according to the blast furnace smelting intensity, and temperature and vibration benchmark thresholds are established. The threshold range is determined using the quantile method. The temperature threshold is the lower and upper limits of the temperature at each measuring point under normal operating conditions in historical data. Temperatures exceeding the range are considered abnormal. The vibration threshold is calculated by calculating the moving average of the energy spectrum density of each frequency band ±3 times the standard deviation, and the abnormal frequency band range is dynamically adjusted. Establish a working condition-threshold mapping table and modify the threshold range based on the parameters of gas flow and furnace top pressure.

[0012] Preferably, the sliding window algorithm is used to monitor the accumulated duration of temperature exceeding the limit and the abnormal frequency band of the vibration energy spectrum in real time, and the number of abnormal event triggering in a single day is counted, specifically including: Linear interpolation is performed on the temperature data to fill missing values, and the vibration signal is detrended to complete the window data preprocessing; Real-time calculation of the continuous temperature exceeding limit time within the window and introduction of dynamic attenuation factor; Perform FFT transformation on the vibration signal to calculate the energy proportion of each frequency band within the predetermined range; use wavelet packet decomposition algorithm to detect sudden energy increases in high-frequency bands, and trigger an abnormal flag when the energy of a certain frequency band exceeds the baseline threshold to a certain extent; Record the number of temperature over-limit events and vibration frequency band abnormal events triggered each day, and classify them by severity; based on the recent abnormal event rate, automatically adjust the threshold range for the next day.

[0013] Preferably, the temperature gradient extreme difference, slag skin thickness variation coefficient, and vibration main frequency offset are used as input features, and a multi-dimensional coupled damage probability prediction model is trained by a random forest algorithm to output the recent dynamic damage probability value of each cooling wall segment. Specifically, the following are included: The maximum temperature difference between adjacent sensors is calculated using the array data of the cooling wall surface temperature sensors. The ratio of the standard deviation to the mean of the thickness of a single cooling wall section is calculated based on the dynamic slag thickness distribution cloud map. The relative offset between the current main frequency and the historical benchmark main frequency is calculated by analyzing the vibration signal through FFT. Random forest model training uses cost-sensitive learning to address the class imbalance problem by assigning weight coefficients to cooling wall water pipe leakage samples; Multi-dimensional coupled prediction puts the above three input features into the trained random forest model, and finally generates a comprehensive probability through a majority voting mechanism; dynamic correction is made using time series.

[0014] The method for predicting the probability of damage of a blast furnace cooling stave according to claim 8 is characterized in that the multi-dimensional coupled prediction specifically includes: Generate the combined probability through majority voting mechanism: Where, P 破损 is the dynamic damage probability value of the final output, which is the average value of the predicted probability of all decision trees; N is the total number of decision trees; i is the number of the i-th decision tree; P i (y) is the predicted probability of the i-th decision tree for the input feature y, namely, the temperature gradient range, the coefficient of variation of the slag thickness, and the vibration main frequency offset, and the output value range is [0,1].

[0015] Preferably, the risk level is divided according to the damage probability value, the threshold interval is dynamically adjusted in combination with the real-time smelting intensity of the blast furnace, and a differentiated maintenance suggestion set is generated, specifically including: Automatically adjust the risk range based on real-time smelting intensity. During high-intensity smelting, the high-risk threshold is lowered to improve warning sensitivity; during low-intensity smelting, the high-risk threshold is raised to reduce false alarms. Low-risk maintenance plan, cooling water pH control, hardness control, to prevent scaling; daily monitoring of cooling wall temperature fluctuations, recording of vibration frequency baseline; Medium-risk intervention measures include increasing the purge frequency, controlling the pressure, and removing slag skin attachments; adjusting the purge angle with respect to the cooling wall surface to enhance coverage; and performing laser ranging and temperature calibration on sensors in abnormal areas to correct errors. High-risk emergency response: timely adjustments based on the spare parts replacement priority list and operating parameter adjustments; emergency risk response strategy, system interlock control, automatic triggering of blast furnace air reduction, and startup of the emergency water cooling system; calling the historical case library to match similar failure modes and push disposal plans.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes to evenly arrange a temperature sensor array on the surface of the cooling wall, and install vibration sensors and acoustic emission sensors inside, which can more comprehensively capture the operating status of the cooling wall under different working conditions and provide a rich data basis for subsequent analysis; calculate the slag skin thickness on the hot surface of the cooling wall based on the non-steady-state heat transfer equation, which can more accurately describe the dynamic heat transfer process of the cooling wall in actual operation and improve the accuracy of the slag skin thickness calculation; introduce a correction model for the gas flow parameters in the furnace to generate a dynamic slag skin thickness distribution cloud map, which can more realistically reflect the heat flow distribution in the furnace and further improve the prediction accuracy of the slag skin thickness distribution; through random The forest algorithm trains a multi-dimensional coupled damage probability prediction model. The random forest algorithm has good generalization ability and anti-overfitting performance, can effectively process high-dimensional data, and improve the accuracy and stability of the prediction model. The risk level is divided according to the damage probability value, which can intuitively reflect the damage risk level of each section of the cooling wall. Combined with the real-time smelting intensity of the blast furnace, the threshold interval is dynamically adjusted to make maintenance recommendations more in line with actual production conditions, and improve the scientificity and rationality of maintenance decisions. The generation of differentiated maintenance recommendation sets can provide targeted maintenance measures for cooling walls of different risk levels, effectively extending the service life of the blast furnace and reducing maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a method for predicting the probability of blast furnace cooling wall damage; Figure 2 This is a diagram of the sensor function. DETAILED DESCRIPTION

[0018] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0019] Reference Figure 1 As shown, a method for predicting the probability of damage of a blast furnace cooling stave comprises: Temperature sensor arrays are evenly distributed on the surface of each cooling wall, and vibration sensors and acoustic emission sensors are installed inside the cooling wall to collect temperature gradient data, mechanical vibration frequency, and high-frequency stress wave signals in real time; The slag thickness on the hot surface of the cooling wall is calculated based on the unsteady heat transfer equation, and a correction model for the gas flow parameters in the furnace is introduced to generate a dynamic slag thickness distribution cloud map. Based on the historical operating data of the cooling wall, the temperature and vibration benchmark thresholds under different working conditions are established. A sliding window algorithm is used to detect the cumulative duration of temperature exceeding the limit and the abnormal frequency band of the vibration energy spectrum in real time, and the number of abnormal event triggering in a single day is counted. The temperature gradient range, slag thickness variation coefficient, and vibration main frequency offset are used as input features. A multi-dimensional coupled damage probability prediction model is trained using the random forest algorithm to output the recent dynamic damage probability value of each cooling wall segment. The risk level is divided according to the damage probability value, the threshold range is dynamically adjusted based on the real-time smelting intensity of the blast furnace, and a differentiated maintenance recommendation set is generated.

[0020] It should be noted that the unsteady-state heat transfer equation is a mathematical equation that describes the heat transfer process of an object under unsteady conditions (i.e., temperature changes with time). Compared with the steady-state heat transfer equation, it can more accurately reflect the dynamic heat transfer of the cooling stave in actual operation, because the hot surface temperature and heat flux density of the blast furnace cooling stave will continuously change with the changes in the blast furnace operating conditions. By solving the unsteady-state heat transfer equation, the temperature distribution of the cooling stave hot surface can be calculated, and the thickness of the slag skin can be deduced. This is because the thickness of the slag skin affects the temperature distribution of the hot surface. Based on the principle of heat transfer, a relationship model between temperature and slag skin thickness can be established.

[0021] Introducing furnace gas flow parameter correction model: The gas flow parameters in the furnace have a significant impact on the heat transfer process of the cooling stave. Factors such as the gas flow velocity, temperature, and composition will change the heat flux density and temperature distribution on the hot surface of the cooling stave. Therefore, introducing a correction model for the gas flow parameters in the furnace can more accurately calculate the temperature of the hot surface of the cooling stave, thereby improving the accuracy of the slag skin thickness calculation. The modified model may be based on computational fluid dynamics (CFD) simulation. By considering the actual situation of gas flow, the calculation results of the unsteady-state heat transfer equation are modified to generate a dynamic slag skin thickness distribution cloud map that is more in line with reality.

[0022] Benchmark threshold establishment and anomaly detection based on historical operating data: Establishing benchmark thresholds: Based on the historical operating data of the cooling wall, temperature and vibration benchmark thresholds under different operating conditions are established to provide a reference standard for subsequent abnormality detection. The normal temperature and vibration range of the cooling wall under different operating conditions may be different, so it is necessary to establish corresponding benchmark thresholds for each operating condition. These benchmark thresholds can be obtained by statistical analysis of historical operating data, such as calculating the mean, standard deviation and other statistical quantities of temperature and vibration data under different operating conditions, and then determining the threshold range based on a certain confidence level.

[0023] Sliding window algorithm: This is a method for real-time data processing and analysis. In this invention, the sliding window algorithm is used to detect the cumulative duration of temperature exceeding the limit and the abnormal frequency band of the vibration energy spectrum in real time in order to promptly detect abnormal conditions in the operation of the cooling wall; The basic idea of the sliding window algorithm is to maintain a fixed-size time window. As time passes, the window continuously slides forward, processing only the data within the window at a time. For temperature data, the cumulative duration of temperature violations within the window can be calculated. If the cumulative duration exceeds a certain threshold, it is considered a temperature anomaly. For vibration data, the characteristics of the vibration energy spectrum within the window, such as the frequency band energy distribution, can be calculated. If an abnormal frequency band is found, it is considered a vibration anomaly.

[0024] Statistics on the number of abnormal event triggering: Statistics on the number of abnormal event triggering in a single day are used to quantify the abnormal conditions of the cooling wall within a day and provide data support for the subsequent prediction of the probability of damage. The more abnormal event triggering times, the more unstable the operating status of the cooling wall on that day, and the higher the risk of damage.

[0025] Damage probability prediction model based on random forest algorithm: Input feature selection: (1) Temperature gradient extremes: The temperature gradient extremes reflect the severity of the temperature change on the cooling wall surface. The larger the temperature gradient, the greater the thermal stress on the cooling wall and the higher the risk of damage. (2) Coefficient of variation of slag skin thickness: The coefficient of variation of slag skin thickness is the ratio of the standard deviation of slag skin thickness to the mean, which reflects the fluctuation of slag skin thickness. The larger the coefficient of variation of slag skin thickness, the worse the uniformity of slag skin thickness, which may lead to uneven heating of cooling wall and increase the probability of damage. (3) Vibration main frequency offset: The vibration main frequency offset refers to the deviation between the main frequency of the cooling wall vibration and the main frequency under normal working conditions. If the vibration main frequency is offset, it may mean that the mechanical structure of the cooling wall has changed, such as cracks, looseness, etc., thereby increasing the risk of damage.

[0026] Random Forest Algorithm: This is a supervised learning algorithm based on ensemble learning. It improves the accuracy and stability of the model by constructing multiple decision trees and combining their prediction results. In the present invention, the multi-dimensional coupled damage probability prediction model is trained by the random forest algorithm, which can effectively handle the complex relationship between input features and improve the accuracy of the prediction. It has good generalization ability, can process high-dimensional data, and has a certain degree of robustness to noise and outliers in the data. In addition, it can also provide feature importance assessment to help analyze which features have a greater impact on damage probability prediction.

[0027] Risk level classification and differentiated maintenance suggestions: (1) Risk level classification: The purpose of classifying risk levels according to the probability of damage is to quantify and classify the damage risk of the cooling wall, which is convenient for subsequent maintenance management. Risk levels can usually be divided into several levels, such as low risk, medium risk, and high risk. Different risk levels correspond to different maintenance strategies and measures. The risk level classification can be determined based on actual production experience and safety requirements. For example, one or more thresholds can be set to classify cooling wall sections with damage probability values in different ranges into different risk levels.

[0028] (2) Dynamic adjustment of threshold intervals: The threshold intervals are dynamically adjusted in combination with the real-time smelting intensity of the blast furnace in order to make the risk level classification more consistent with the actual production situation. The smelting intensity of the blast furnace is different, and the operating conditions and heating degree of the cooling wall are also different. Therefore, it is necessary to appropriately adjust the threshold intervals of the risk level classification according to the changes in the smelting intensity. For example, under high smelting intensity, the operating temperature and stress of the cooling wall may be higher. At this time, the threshold of the risk level classification can be appropriately increased to more strictly monitor the status of the cooling wall; under low smelting intensity, the threshold can be appropriately lowered to avoid excessive maintenance.

[0029] (3) Differentiated maintenance recommendation set: The purpose of generating a differentiated maintenance recommendation set is to provide targeted maintenance measures for cooling wall sections with different risk levels. For cooling wall sections with low risk levels, conventional maintenance measures can be taken, such as regular inspections and cleaning; for cooling wall sections with medium risk levels, the inspection frequency can be increased, and local inspections and repairs can be carried out; for cooling wall sections with high risk levels, emergency maintenance measures need to be taken immediately, such as shutting down the furnace for maintenance and replacing the cooling wall.

[0030] Reference Figure 2As shown, the temperature sensor array is evenly arranged on the surface of each cooling wall, and vibration sensors and acoustic emission sensors are installed inside the cooling wall to collect temperature gradient data, mechanical vibration frequency and high-frequency stress wave signals in real time. Specifically, the following are included: Monitoring points are evenly distributed on the surface of each cooling wall section, digital temperature sensors are installed at the monitoring points, and temperature compensation circuits are used to eliminate measurement errors caused by thermal expansion; Collect the temperature values of each sensor in real time, generate a temperature field distribution map through spatial interpolation algorithm, calculate the temperature difference between adjacent sensors; combine with sliding window algorithm to calculate the cumulative duration of temperature exceeding the limit; Piezoelectric accelerometers are installed at key locations inside the cooling wall to detect mechanical vibration frequency. A three-axis vibration sensor is used to monitor its vibration energy, and the main frequency offset is extracted through frequency domain analysis. The original vibration signal is subjected to anti-aliasing filtering and FFT transformation to extract energy spectrum density data. An abnormal frequency band identification algorithm is used to detect sudden energy increases in the high-frequency band and determine whether the internal structure of the cooling wall is loose or whether microcracks are expanding. A broadband piezoelectric ceramic array is used to capture microcrack expansion and macroscopic deformation signals. Acoustic emission sensors are embedded in the cooling wall, and coupling agent sealing technology is used to ensure signal transmission efficiency. Electromagnetic shielding shells are used to reduce electromagnetic interference from the blast furnace. Real-time recording of acoustic emission event count rate and waveform parameters, combined with pattern recognition algorithms, can distinguish normal thermal stress fluctuations from abnormal crack signals.

[0031] It should be noted that the sensor layout: Temperature sensor array: The temperature sensor array is evenly distributed on the surface of each cooling wall to comprehensively monitor the temperature distribution on the cooling wall. This uniform distribution ensures that temperature data from all areas of the cooling wall surface can be obtained, avoiding problems that cannot be discovered in time due to local temperature anomalies. Vibration sensor and acoustic emission sensor: Vibration sensor and acoustic emission sensor are installed inside the cooling wall to monitor the mechanical vibration and stress wave conditions inside the cooling wall. The vibration sensor is mainly used to monitor the mechanical vibration frequency of the cooling wall, while the acoustic emission sensor is used to capture the high-frequency stress wave signal inside the cooling wall. The combined use of these two sensors can more comprehensively reflect the mechanical state of the cooling wall.

[0032] Data collection: Temperature gradient data: The temperature data of the cooling wall surface is collected in real time through the temperature sensor array, and the temperature gradient is calculated, that is, the rate of change of temperature in space, which can reflect the heat transfer and thermal stress distribution of the cooling wall.

[0033] Mechanical vibration frequency and high-frequency stress wave signals: The mechanical vibration frequency and high-frequency stress wave signals collected in real time by vibration sensors and acoustic emission sensors can be used to analyze the mechanical vibration characteristics and internal stress changes of the cooling wall, providing a basis for subsequent anomaly detection and damage probability prediction.

[0034] The use process of the present invention is: Step 1: Evenly distribute monitoring points on the surface of each cooling wall section, install digital temperature sensors at the monitoring points, and use temperature compensation circuits to eliminate measurement errors caused by thermal expansion; Step 2: Collect the temperature values of each sensor in real time, generate a temperature field distribution map using a spatial interpolation algorithm, and calculate the temperature difference between adjacent sensors. Combined with a sliding window algorithm, calculate the cumulative duration of temperature exceeding the limit. Step 3: Install piezoelectric accelerometers at key locations inside the cooling wall to detect mechanical vibration frequency. Use a three-axis vibration sensor to monitor its vibration energy and extract the main frequency offset through frequency domain analysis. Step 4: Perform anti-aliasing filtering and FFT transformation on the original vibration signal to extract energy spectral density data. Use an abnormal frequency band identification algorithm to detect sudden energy increases in the high-frequency band and determine whether the internal structure of the cooling wall is loose or microcracks are growing. Use a broadband piezoelectric ceramic array to capture microcrack growth and macroscopic deformation signals. Step 5: Embed the acoustic emission sensor inside the cooling wall, use coupling agent sealing technology to ensure signal transmission efficiency, and use electromagnetic shielding shell to reduce electromagnetic interference of the blast furnace; Step 6: Real-time recording of acoustic emission event count rate and waveform parameters, combined with pattern recognition algorithms, to distinguish normal thermal stress fluctuations from abnormal crack signals; Step 7: The heat transfer model is established based on the one-dimensional unsteady heat conduction differential equation. The finite difference method is used to discretize space and time, and the temperature field is iteratively calculated using explicit and implicit schemes. Step 8: Introduce the equivalent specific heat method to deal with the latent heat of slag skin solidification and melting, and update the functional relationship between thermal conductivity and temperature diffusivity with temperature; Step 9: Based on the temperature field distribution, calculate the position of the solid-liquid interface of the slag skin through the heat balance equation and deduce the thickness change rate; Step 10: Periodically use a laser rangefinder to measure the actual thickness of the slag skin and establish a temperature-thickness feedback compensation mechanism; Step 11: Introduce the gas flow velocity, pressure, and temperature distribution parameters, and modify the convective heat transfer coefficient using the Reynolds number and Nusselt number. Embed the local heat flux density changes caused by the gas flow into the heat transfer equation and modify the slag skin thickness calculation model. Step 12: By dividing the cooling wall surface into grid nodes, the discrete thickness data is converted into a continuous distribution field using a bilinear interpolation algorithm. A visualization module is developed based on MATLAB to generate a pseudo-color cloud map that updates over time and a dynamic slag skin thickness distribution cloud map. Step 13: Compare the model-predicted thickness with the laser-measured data, calculate the root mean square error, and optimize the mesh density and boundary condition parameters; Step 14: Classify the working conditions according to the blast furnace smelting intensity, establish the temperature and vibration reference thresholds respectively, and use the quantile method to determine the threshold range; Step 15: The temperature threshold is the lower and upper limits of the temperature at each measuring point under normal operating conditions in historical data. Any temperature outside the range is considered abnormal. The vibration threshold is calculated by calculating the moving average of the energy spectrum density of each frequency band ±3 times the standard deviation, and the abnormal frequency band range is dynamically adjusted. Step 16: Create a working condition-threshold mapping table and modify the threshold interval based on the parameters of gas flow and furnace top pressure; Step 17: Linear interpolation is performed on the temperature data to fill in missing values, and the vibration signal is detrended to complete the window data preprocessing; Step 18: Calculate the continuous temperature exceeding limit time in the window in real time and introduce a dynamic attenuation factor; Step 19: Perform an FFT transform on the vibration signal to calculate the energy proportion of each frequency band within the predetermined range; use a wavelet packet decomposition algorithm to detect sudden energy increases in high-frequency bands, and trigger an abnormal flag when the energy of a certain frequency band exceeds a certain threshold; Step 20: Record the number of temperature over-limit events and vibration frequency band abnormal events triggered each day, and classify them by severity; based on the recent abnormal event rate, automatically adjust the threshold range for the next day; Step 21: Calculate the maximum temperature difference between adjacent sensors using the array data of the cooling wall surface temperature sensors; calculate the ratio of the standard deviation to the mean of the thickness of the single-section cooling wall based on the dynamic slag thickness distribution cloud map; and calculate the relative offset between the current main frequency and the historical benchmark main frequency by analyzing the vibration signal through FFT. Step 22: Random forest model training, by assigning weight coefficients to cooling wall water pipe leakage samples, using cost-sensitive learning to solve the class imbalance problem; Step 23: Multi-dimensional coupled prediction is performed by feeding the above three input features into the trained random forest model and finally generating a comprehensive probability through a majority voting mechanism; dynamic correction is applied using time series; Step 24: Automatically adjust the risk range based on the real-time smelting intensity. When the smelting intensity is high, the high-risk threshold is lowered to improve the warning sensitivity; when the smelting intensity is low, the high-risk threshold is raised to reduce false alarms. Step 25: Low-risk maintenance plan: adjust the pH value and hardness of cooling water to prevent scaling; monitor the temperature fluctuation of the cooling wall daily and record the vibration frequency baseline; Step 26: Medium-risk intervention measures: Increase the purge frequency, control the pressure, and remove slag skin attachments; adjust the purge angle with respect to the cooling wall surface to enhance coverage; perform laser ranging and temperature calibration on sensors in abnormal areas and perform error correction; Step 27: High-risk emergency handling: timely adjustment based on spare parts replacement priority list and operating parameter adjustment; emergency risk response strategy, system interlock control, automatic triggering of blast furnace air reduction, start-up of emergency water cooling system; call the historical case library to match similar failure modes and push the disposal plan.

[0035] In summary, the advantages of the present invention are: by arranging multiple sensors on the surface and inside of the cooling wall, multi-dimensional data such as temperature, vibration and acoustic emission are collected in real time, providing a comprehensive information basis for cooling wall status monitoring; based on the non-steady-state heat transfer equation and combined with the correction of gas flow parameters, the slag skin thickness distribution is accurately calculated and dynamically generated to accurately reflect the thermal status of the cooling wall; historical data is used to establish a benchmark threshold, and the sliding window algorithm is combined to detect temperature and vibration anomalies in real time to quickly identify potential faults; with multi-dimensional features as input, the random forest algorithm is used to train a damage probability prediction model to provide accurate dynamic damage probability prediction; risk levels are divided according to the damage probability, and the threshold is adjusted in combination with real-time smelting intensity to generate differentiated maintenance suggestions, effectively extending the service life of the blast furnace.

[0036] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the invention as claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the probability of damage to a blast furnace cooling stave, characterized in that: include: Temperature sensor arrays are evenly distributed on the surface of each cooling wall, and vibration sensors and acoustic emission sensors are installed inside the cooling wall to collect temperature gradient data, mechanical vibration frequency, and high-frequency stress wave signals in real time; The slag thickness on the hot surface of the cooling wall is calculated based on the unsteady heat transfer equation, and a correction model for the gas flow parameters in the furnace is introduced to generate a dynamic slag thickness distribution cloud map. Based on the historical operating data of the cooling wall, the temperature and vibration benchmark thresholds under different working conditions are established. A sliding window algorithm is used to detect the cumulative duration of temperature exceeding the limit and the abnormal frequency band of the vibration energy spectrum in real time, and the number of abnormal event triggering in a single day is counted. The temperature gradient range, slag thickness variation coefficient, and vibration main frequency offset are used as input features. A multi-dimensional coupled damage probability prediction model is trained using the random forest algorithm to output the recent dynamic damage probability value of each cooling wall segment. The risk level is divided according to the damage probability value, the threshold range is dynamically adjusted based on the real-time smelting intensity of the blast furnace, and a differentiated maintenance recommendation set is generated.

2. The method for predicting the probability of damage of a blast furnace cooling stave according to claim 1, characterized in that: The temperature sensor array is evenly arranged on the surface of each cooling wall, and vibration sensors and acoustic emission sensors are installed inside the cooling wall to collect temperature gradient data, mechanical vibration frequency and high-frequency stress wave signals in real time. Specifically, the method includes: Monitoring points are evenly distributed on the surface of each cooling wall section, digital temperature sensors are installed at the monitoring points, and temperature compensation circuits are used to eliminate measurement errors caused by thermal expansion; Collect the temperature values of each sensor in real time, generate a temperature field distribution map through spatial interpolation algorithm, calculate the temperature difference between adjacent sensors; combine with sliding window algorithm to calculate the cumulative duration of temperature exceeding the limit; Piezoelectric accelerometers are installed at key locations inside the cooling wall to detect mechanical vibration frequency. A three-axis vibration sensor is used to monitor its vibration energy, and the main frequency offset is extracted through frequency domain analysis. The original vibration signal is subjected to anti-aliasing filtering and FFT transformation to extract energy spectrum density data. An abnormal frequency band identification algorithm is used to detect sudden energy increases in the high-frequency band and determine whether the internal structure of the cooling wall is loose or whether microcracks are expanding. A broadband piezoelectric ceramic array is used to capture microcrack expansion and macroscopic deformation signals. Acoustic emission sensors are embedded in the cooling wall, and coupling agent sealing technology is used to ensure signal transmission efficiency. Electromagnetic shielding shells are used to reduce electromagnetic interference from the blast furnace. Real-time recording of acoustic emission event count rate and waveform parameters, combined with pattern recognition algorithms, can distinguish normal thermal stress fluctuations from abnormal crack signals.

3. The method for predicting the probability of damage of a blast furnace cooling stave according to claim 2, characterized in that: The calculation of the slag skin thickness on the hot surface of the cooling stave based on the unsteady-state heat transfer equation specifically includes: The heat transfer model is established based on the one-dimensional unsteady heat conduction differential equation. The finite difference method is used to discretize space and time, and the temperature field is iteratively calculated using explicit and implicit schemes. The equivalent specific heat method is introduced to deal with the latent heat of slag skin solidification and melting, and the functional relationship between thermal conductivity and temperature diffusivity with temperature is updated. According to the temperature field distribution, the position of the solid-liquid interface of the slag skin is calculated through the heat balance equation, and the thickness change rate is derived. The actual thickness of the slag skin is measured periodically using a laser rangefinder, and a temperature-thickness feedback compensation mechanism is established.

4. A method for predicting the probability of damage to a blast furnace cooling stave according to claim 3, characterized in that: The introduction of the furnace gas flow parameter correction model to generate a dynamic slag skin thickness distribution cloud map specifically includes: The gas flow velocity, pressure, and temperature distribution parameters are introduced, and the convective heat transfer coefficient is corrected using the Reynolds number and Nusselt number. The local heat flux density changes caused by the gas flow are embedded in the heat transfer equation to correct the slag skin thickness calculation model. By dividing the cooling wall surface into grid nodes and using a bilinear interpolation algorithm, the discrete thickness data is converted into a continuous distribution field. A visualization module is developed based on MATLAB to generate a pseudo-color cloud map that updates over time and a dynamic slag skin thickness distribution cloud map. Compare the model-predicted thickness with the laser-measured data, calculate the root mean square error, and optimize the mesh density and boundary condition parameters.

5. The method for predicting the probability of damage of a blast furnace cooling stave according to claim 4, characterized in that: The introduction of the furnace gas flow parameter correction model to generate a dynamic slag skin thickness distribution cloud map specifically includes: Based on the one-dimensional unsteady heat conduction differential equation: Where, is the rate of change of temperature with time, a signature term for unsteady-state heat conduction; To describe the curvature of the temperature field in space and reflect the driving force of heat diffusion; α is the thermal conductivity of the slag skin, T is the temperature field, x is the spatial coordinate, and t is the time; consider the third type of boundary conditions: Where λ is the thermal conductivity of the material, is the temperature change rate inside the object along the normal x-axis direction, h is the convection heat transfer coefficient of the furnace gas to the slag skin, T f is the gas temperature.

6. A method for predicting the probability of damage to a blast furnace cooling stave according to claim 5, characterized in that: The establishment of temperature and vibration reference thresholds under different working conditions based on historical operating data of the cooling wall specifically includes: The operating conditions are categorized according to the blast furnace smelting intensity, and temperature and vibration benchmark thresholds are established. The threshold range is determined using the quantile method. The temperature threshold is the lower and upper limits of the temperature at each measuring point under normal operating conditions in historical data. Temperatures exceeding the range are considered abnormal. The vibration threshold is calculated by calculating the moving average of the energy spectrum density of each frequency band ±3 times the standard deviation, and the abnormal frequency band range is dynamically adjusted. Establish a working condition-threshold mapping table and modify the threshold range based on the parameters of gas flow and furnace top pressure.

7. A method for predicting the probability of damage to a blast furnace cooling stave according to claim 6, characterized in that: The sliding window algorithm is used to detect the cumulative duration of temperature exceeding the limit and the abnormal frequency band of the vibration energy spectrum in real time, and the number of abnormal event triggering in a single day is counted. Specifically, the following are included: Linear interpolation is performed on temperature data to fill missing values, and vibration signals are detrended to complete window data preprocessing. Real-time calculation of the continuous temperature exceeding limit time within the window and introduction of dynamic attenuation factor; Perform FFT transformation on the vibration signal to calculate the energy proportion of each frequency band within the predetermined range; use wavelet packet decomposition algorithm to detect sudden energy increases in high-frequency bands, and trigger an abnormal flag when the energy of a certain frequency band exceeds the baseline threshold to a certain extent; Record the number of temperature over-limit events and vibration frequency band abnormal events triggered each day, and classify them by severity; based on the recent abnormal event rate, automatically adjust the threshold range for the next day.

8. A method for predicting the probability of damage to a blast furnace cooling stave according to claim 7, characterized in that: The temperature gradient extreme difference, slag thickness variation coefficient, and vibration main frequency offset are used as input features, and a multi-dimensional coupled damage probability prediction model is trained through a random forest algorithm to output the recent dynamic damage probability value of each cooling wall segment. Specifically, the following are included: The maximum temperature difference between adjacent sensors is calculated using the array data of the cooling wall surface temperature sensors. The ratio of the standard deviation to the mean of the thickness of a single cooling wall section is calculated based on the dynamic slag thickness distribution cloud map. The relative offset between the current main frequency and the historical benchmark main frequency is calculated by analyzing the vibration signal through FFT. Random forest model training uses cost-sensitive learning to address the class imbalance problem by assigning weight coefficients to cooling wall water pipe leakage samples; Multi-dimensional coupled prediction puts the above three input features into the trained random forest model, and finally generates a comprehensive probability through a majority voting mechanism; dynamic correction is made using time series.

9. A method for predicting the probability of damage to a blast furnace cooling stave according to claim 8, characterized in that: The multi-dimensional coupling prediction specifically includes: Generate the combined probability through majority voting mechanism: Where, P 破损 is the dynamic damage probability value of the final output, which is the average value of the predicted probability of all decision trees; N is the total number of decision trees; i is the number of the i-th decision tree; P i (y) is the predicted probability of the i-th decision tree for the input feature y, namely, the temperature gradient range, the coefficient of variation of the slag thickness, and the vibration main frequency offset, and the output value range is [0,1].

10. A method for predicting the probability of damage to a blast furnace cooling stave according to claim 9, characterized in that: The risk level is divided according to the damage probability value, the threshold range is dynamically adjusted in combination with the real-time smelting intensity of the blast furnace, and a differentiated maintenance recommendation set is generated, specifically including: Automatically adjust the risk range based on real-time smelting intensity. During high-intensity smelting, the high-risk threshold is lowered to improve warning sensitivity; during low-intensity smelting, the high-risk threshold is raised to reduce false alarms. Low-risk maintenance plan, cooling water pH control, hardness control, to prevent scaling; daily monitoring of cooling wall temperature fluctuations, recording of vibration frequency baseline; Medium-risk intervention measures include increasing the purge frequency, controlling the pressure, and removing slag skin attachments; adjusting the purge angle with respect to the cooling wall surface to enhance coverage; and performing laser ranging and temperature calibration on sensors in abnormal areas to correct errors. High-risk emergency response: timely adjustments based on the spare parts replacement priority list and operating parameter adjustments; emergency risk response strategy, system interlock control, automatic triggering of blast furnace air reduction, and startup of the emergency water cooling system; calling the historical case library to match similar failure modes and push disposal plans.

Images