Blast furnace cooling wall service life prediction method based on multi-source data fusion analysis

Abstract

The present application belongs to the technical field of steel metallurgical equipment state monitoring and life prediction, and relates to a blast furnace cooling wall service life prediction method based on multi-source data fusion analysis. The present application collects multi-source data such as heat flow density, wall temperature, air temperature and slag skin thickness, identifies thermal shock in the order of air temperature, heat flow and wall temperature change rate, identifies slag skin peeling in synchronization with slag skin thinning and heat flow and wall temperature rise, synthesizes damage indexes by weighting the two types of damage and calculates the deterioration rate, reversely inverts the boundary thickness attenuation and determines whether it is slag skin thinning or erosion depth, generates degradation indexes by standardizing and superimposing the two, inputs the long short-term memory network to predict the remaining life and state grade, solves the problems of single physical quantity event identification, insufficient boundary attenuation discrimination and low prediction integration, realizes comprehensive quantification and reversible and irreversible loss differentiation of the cooling wall degradation, improves the coincidence degree of the prediction and the actual replacement cycle, and provides a basis for graded maintenance.

Description

Technical Field

[0001] This invention belongs to the field of condition monitoring and life prediction technology of iron and steel metallurgical equipment, and relates to a method for predicting the life of blast furnace cooling walls based on multi-source data fusion analysis. Background Technology

[0002] The belly, waist, and lower part of the blast furnace are subjected to high heat loads and intense mechanical wear for extended periods. The actual service life of copper cooling walls in these areas is typically significantly shorter than the overall design life of the blast furnace. Premature cooling wall failure not only increases equipment replacement costs but also hinders the long-term stable operation of the blast furnace. Thermal shock disturbance and slag shedding are two typical operating conditions that cause cooling wall performance degradation. These two conditions often occur simultaneously, and their combined effect on cooling wall durability is complex and coupled. Isolated assessments of a single process are insufficient to accurately reflect the overall degree of cooling wall deterioration.

[0003] In current monitoring practices, although a variety of parameters are collected on-site, each parameter is usually used for independent alarms or univariate trend observation. There are few cases where information on the coordinated changes of different parameters over time is systematically incorporated into degradation analysis. For key events such as thermal shock and slag shedding, their identification often relies on the amplitude changes of a single physical quantity, while the quantitative expression of the damage contribution of each event and their coupling effects in life assessment is relatively lacking.

[0004] The boundary condition of the hot surface of a cooling wall has a significant impact on its lifespan assessment. Existing heat transfer back-calculation methods can deduce the change in overall thermal resistance, and thus obtain the change in boundary thickness. However, the current criteria for determining whether this change originates from the thinning of the adhesion layer or the erosion of the bulk material are insufficient, which introduces uncertainty into the assessment of the degree of cooling wall lifespan loss.

[0005] Current lifespan prediction methods primarily rely on univariate trend extrapolation or fixed empirical formulas. The factors considered in the prediction inputs, such as event frequency, spatial location differences, and boundary decay characteristics, are relatively limited. These factors are poorly integrated into existing models, often resulting in observable biases between the predicted results and actual replacement cycles. Summary of the Invention

[0006] In view of this, in order to solve the problems mentioned in the background technology, a method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: a method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis, comprising: collecting multi-source monitoring data by cooling wall unit and unifying the spatiotemporal reference, extracting time-series features from each monitoring data, and the multi-source monitoring data including heat flux density, wall temperature, cooling water flow rate and temperature difference, blast temperature and slag thickness.

[0008] Thermal shock disturbances are identified when the wind temperature change precedes the heat flux density change rate response and the wall temperature change rate lags behind, and toughness indices are generated from the recovery time and overshoot. After thermal shock, when the slag thickness decreases and the heat flux density and wall temperature rise synchronously, slag shedding is identified, generating damage increments.

[0009] The toughness index is multiplied by the thermal shock life weight, and the damage increment is multiplied by the detachment life weight. The results are then multiplied by the life weight of the cooling wall section to obtain the damage index and calculate its degradation rate. The boundary thickness attenuation is obtained by inverse heat transfer inversion from the heat flux density and wall temperature time series, and is determined as the slag thinning amount or erosion depth. The degradation rate of the damage index and the boundary thickness attenuation amount are standardized and then added together to obtain the degradation index.

[0010] Input the time series of degradation indicators into the lifetime prediction model, and output the predicted remaining lifetime time and lifetime status level.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention identifies thermal shock disturbances and generates toughness index by using the time-series response relationship between wind temperature, heat flux density and wall temperature, identifies slag shedding and calculates damage increment by using the synchronous change of slag shedding thickness and heat flux density and wall temperature increase, and calculates damage index by coupling the two types of events with lifetime weight. This solves the problem that the identification of the two types of events is based only on the amplitude of a single physical quantity and the damage contribution lacks quantitative expression, and realizes the comprehensive quantification of cooling wall degradation by the time-series coordinated change of multi-source parameters.

[0012] (2) The present invention obtains the boundary thickness attenuation by inverse heat transfer inversion of heat flux density and wall temperature time series, and extracts the temperature fluctuation range and monotonically rising duration of the wall temperature time series. The boundary thickness attenuation is determined as the amount of slag thinning or erosion depth, which solves the problem that the basis for distinguishing whether the change in boundary thickness is due to the thinning of the attachment layer or the erosion of the body is insufficient. It provides a distinguishable boundary state classification basis for judging the degree of cooling wall life loss.

[0013] (3) This invention calculates the degradation index by standardizing the degradation rate of the damage index and the boundary thickness attenuation, and inputs the degradation index time series into the lifetime prediction model composed of a long short-term memory network, and outputs the remaining lifetime prediction time value and lifetime status level. This solves the problem that lifetime prediction relies only on single variable trend extrapolation or fixed empirical formulas and has a low degree of integration of event frequency, spatial location and boundary attenuation factors, and improves the degree of consistency between the remaining lifetime prediction results and the actual replacement cycle. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of the blast furnace cooling wall service life prediction method based on multi-source data fusion analysis in this invention;

[0016] Figure 2 This is a flowchart illustrating the classification and determination process for boundary thickness attenuation in this invention.

[0017] Figure 3 This is a time-series response and synchronous correlation analysis diagram of thermal shock and slag shedding in this invention;

[0018] Figure 4 This is a comparison chart of the remaining lifetime prediction results in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] The specific scheme of the blast furnace cooling wall service life prediction method based on multi-source data fusion analysis provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Please see Figure 1 As shown, the implementation of the present invention includes S1 to S4: thermal shock and slag shedding are two major damage causes of cooling walls. The former causes instantaneous thermal stress damage, while the latter causes the body to lose protection and be exposed to high temperatures. Moreover, the former easily induces the latter. Although both can be characterized by changes in heat flux density and wall temperature, existing prediction methods mostly rely on a single parameter threshold for state judgment. It is difficult to quantify and separate the damage weights of the two types of events under the same framework, and it is also impossible to define whether the boundary decay belongs to reversible slag thinning or irreversible body erosion, resulting in insufficient accuracy in remaining lifetime prediction.

[0023] To overcome the aforementioned shortcomings, this invention identifies thermal shock and generates a toughness index by analyzing the temporal response relationship between wind temperature, heat flux density, and wall temperature. It identifies slag shedding and calculates damage increment by analyzing the synchronous correlation between slag thinning and the increase in heat flux density and wall temperature. A damage index is generated by fusing event frequency and location-specific heat load weights. This is then combined with the boundary thickness attenuation obtained from inverse heat transfer to determine whether it is slag thinning or bulk erosion. The degradation rate of the damage index is superimposed with the boundary thickness attenuation to form a degradation index, which is input into a long short-term memory network to output the remaining lifetime and state level. This method simultaneously considers event-driven discrete damage accumulation and the continuous degradation trend of the material itself within a single degradation index. It can distinguish between reversible performance fluctuations caused by slag shedding and irreversible lifetime loss caused by bulk erosion, providing a quantitative basis for graded maintenance decisions for cooling walls.

[0024] S1. Collect multi-source monitoring data according to the cooling wall unit and unify the spatiotemporal reference. Extract time-series features from each monitoring data. The multi-source monitoring data includes heat flux density, wall temperature, cooling water flow rate and temperature difference, air temperature and slag thickness.

[0025] The blast furnace cooling walls are distributed longitudinally and circumferentially along the furnace body. Each unit experiences different heat loads, slag adhesion states, and cooling conditions, requiring monitoring based on multiple parameters such as heat flux density, wall temperature, cooling water flow rate and temperature difference, blast temperature, and slag thickness. These parameters belong to different acquisition channels with varying sampling periods and dimensions; without spatial and temporal alignment, heterogeneous data cannot establish effective physical correlations. Therefore, this step collects multi-source monitoring data by cooling wall unit and unifies the spatiotemporal reference. Based on this, temporal features are extracted from each monitoring data set.

[0026] In one specific embodiment, each cooling wall in the blast furnace belly, waist, and lower part of the furnace body is first defined as a cooling wall unit. All cooling wall units are uniformly numbered according to the segment number from bottom to top in the longitudinal direction of the furnace body and the block number arranged sequentially along the circumference. Multi-source monitoring data are classified according to the cooling wall unit number. The multi-source monitoring data includes heat flux density, wall temperature, cooling water flow rate and temperature difference, blast temperature, and slag thickness.

[0027] Among the monitoring data, the wall temperature is measured by thermocouples installed in mounting holes near the hot surface of the cooling wall body; the cooling water flow rate is measured by electromagnetic flowmeters installed on the cooling water branch pipes of each cooling wall unit; the cooling water temperature difference is obtained by subtracting the inlet and return water temperatures measured by two platinum resistance thermometers installed at the inlet and return ends of the branch pipes respectively; the heat flux density is calculated according to the cooling water side heat balance, that is, the cooling water volume flow rate is converted into mass flow rate according to the density of water at the inlet temperature, multiplied by the specific heat capacity obtained from the water thermal property table according to the inlet temperature and the cooling water temperature difference to obtain the heat carried away by the cooling water of the unit, and then divided by the heated surface area of ​​the unit calculated from the hot surface dimensions of the cooling wall product drawing; the air temperature is measured by thermocouples installed on the hot blast main pipe of the blast furnace and is assigned to each cooling wall unit as a unified parameter for the whole furnace.

[0028] The slag skin thickness is calculated by inversely using the one-dimensional steady-state Fourier thermal conductivity law. Specifically, the difference between the design value of the furnace gas temperature and the wall temperature in the height range of the unit is taken as the temperature difference on both sides of the slag skin. This difference is multiplied by the thermal conductivity of the slag skin obtained from the slag sample taken during blast furnace maintenance and then divided by the heat flux density. The resulting slag skin thickness is the equivalent thickness of the hot surface boundary adhesion layer calculated based on the thermal conductivity of the slag skin.

[0029] In terms of time, each monitoring data is recorded by its own acquisition instrument at its own fixed period to form raw data. This fixed period is the original sampling period of the monitoring data. The original sampling period of wall temperature and air temperature is shorter than that of cooling water flow and cooling water temperature difference. The longest original sampling period among all monitoring data is taken as the unified sampling period. The arithmetic mean of all the sampled values ​​of the monitoring data whose original sampling period is shorter than the unified sampling period in each unified sampling period is taken and a timestamp of the start time of the unified sampling period is assigned. After being aligned point by point with the monitoring data whose original sampling period is equal to the unified sampling period, they are merged into an equally spaced time series of each monitoring data according to the cooling wall unit.

[0030] For sampling points that are interrupted or exceed the range, linear interpolation of the two adjacent valid sampling points is used to replace them. The unified sampling period determined in this way is shorter than the time-series transmission time of thermal shock disturbance from wind temperature change through heat flux density response to wall temperature rise, which can retain the order information of each rate of change inflection point required for subsequent event identification.

[0031] Based on this, time-series features are extracted from the time series of each monitoring data. The time-series features are divided into two categories: window statistical features and sampling point differential features. The window statistical features are calculated window by window using the feature window length and a sliding step size smaller than the feature window length. The feature window length is the maximum value of the entire process duration from the occurrence of each thermal shock disturbance to the stabilization of the wall temperature within the historical monitoring period, so as to fully cover the entire process of the event. The value of the sliding step size ensures that there is overlap between adjacent feature windows. The sampling point differential features are calculated by dividing the difference between the values ​​of two adjacent sampling points by a uniform sampling period, and the same timestamp is maintained as the original time series.

[0032] Specifically: the heat flux characteristics include the mean heat flux density, which is the arithmetic mean of the heat flux density time series within the feature window; the wall temperature characteristics include the peak wall temperature and the wall temperature change rate, where the peak wall temperature is the maximum value of the wall temperature time series within the feature window, and the wall temperature heating rate is the difference time series of the wall temperature time series at each sampling point.

[0033] The wind temperature characteristics include the wind temperature change rate, which is the difference time series of the wind temperature time series at each sampling point; the slag skin characteristics include the slag skin thickness reduction rate, which is the inverse sequence of the difference time series of the slag skin thickness time series at each sampling point; among them, the cooling water flow rate and cooling water temperature difference are used as process quantities for calculating heat flux density, and their time series characteristics are not extracted separately.

[0034] Furthermore, the heat flux density change rate time series is calculated using the same sampling point differential method. The heat flux density change rate time series, the air temperature change rate time series, and the wall temperature rise rate time series are used together to determine the order of inflection point timestamps in subsequent steps. The wall temperature peak value is used to calculate the overshoot in subsequent steps. The slag thickness reduction rate time series is used to calculate the synchronization correlation in subsequent steps. The average heat flux density of each cooling wall unit is averaged according to its longitudinal segment number and circumferential block number, and then used as the input for calculating the part heat load weight in subsequent steps.

[0035] S2. When the change in wind temperature precedes the change in heat flux density and the change in wall temperature follow behind, thermal shock disturbances are identified, and toughness indicators are generated from the recovery time and overshoot. After thermal shock, when the slag thickness decreases and the heat flux density and wall temperature rise synchronously, slag shedding is identified, generating damage increments.

[0036] Both thermal shock and slag shedding manifest as increases in heat flux density and wall temperature, and relying solely on a single parameter threshold cannot distinguish between the two. From a heat transfer perspective, thermal shock begins with a change in wind temperature, proceeds through heat flux density response to wall temperature increase, exhibiting a clear temporal progression. Slag shedding, on the other hand, occurs simultaneously with slag shedding thickness reduction and increases in heat flux density and wall temperature. These two events differ fundamentally in their temporal characteristics. Therefore, this step identifies thermal shock disturbances based on the aforementioned temporal sequence, generates toughness indices from recovery time and overshoot, and identifies slag shedding following thermal shock based on synchronous changes, calculating the damage increment.

[0037] In one specific embodiment, the time series of wind temperature change rate, heat flux density change rate, and wall temperature rise rate are first scanned point by point. The first time stamp in each time series where the value changes from non-positive to positive is determined as the rising inflection point time stamp of that time series. When the rising inflection point time stamp of the wind temperature change rate is earlier than the rising inflection point time stamp of the heat flux density change rate, and the rising inflection point time stamp of the wall temperature rise rate is later than the rising inflection point time stamp of the heat flux density change rate, it indicates that the disturbance starts from the change in supply air temperature, passes through the heat flux density response, and is completed by the wall heat storage and temperature rise. This conforms to the time sequence characteristics of thermal shock disturbance being transmitted step by step along the heat transfer path. A thermal shock disturbance event is determined to have occurred, and the starting time stamp of the thermal shock disturbance event is taken as the rising inflection point time stamp of the wind temperature change rate.

[0038] Subsequently, the time when the wall temperature first falls back to within the steady-state baseline range from the start time of the thermal shock disturbance event is determined as the end time of the thermal shock disturbance event. The maximum value of the wall temperature time series between the start and end times is taken as the wall temperature peak value. The difference between the wall temperature peak value and the steady-state baseline value at the peak time is taken as the overshoot, and the time span between the end and peak times is taken as the recovery time. The overshoot is divided by the steady-state baseline value at the peak time to obtain the dimensionless overshoot, and the recovery time is divided by the feature window length to obtain the dimensionless recovery time. The two are multiplied to generate a toughness index. The larger the value of the toughness index, the more severe the deviation of the wall temperature caused by the thermal shock and the degree of recovery lag, and the more severe the transient thermal stress damage suffered by the cooling wall body in this event.

[0039] The steady-state baseline is the moving average sequence of the heat flux density time series and the wall temperature time series within a historical time window in which no thermal shock disturbance event or slag detachment event has occurred, calculated with the characteristic window length as the sliding window. This historical time window is determined from the event-free period between the monitoring start time and the most recent event end time stamp. When no event has occurred, it is taken from the monitoring start time to the current monitoring period. The steady-state baseline interval is the range of values ​​with the moving average sequence as the center, floating above and below the corresponding time series within the historical time window by one standard deviation. The steady-state baseline value within the event period is the steady-state baseline value at the last time stamp before the start of that period.

[0040] Furthermore, after the thermal shock disturbance ends, the slag skin attached to the hot surface of the cooling wall may detach due to thermal shock, causing the body to lose its protective layer and be directly exposed to high-temperature gas. Therefore, a slag skin detachment discrimination window is constructed with the end time stamp of the thermal shock disturbance event as the starting point and the characteristic window length as the duration. Within this window, the Pearson correlation coefficients of the slag skin thickness reduction rate time series and the heat flux density change rate time series, the slag skin thickness reduction rate time series and the wall temperature rise rate time series, and the heat flux density change rate time series and the wall temperature rise rate time series are calculated.

[0041] When all three Pearson correlation coefficients are positive, it indicates that the thinning of the slag skin occurs simultaneously with the increase in heat flux density and the increase in wall temperature. This is consistent with the heat transfer characteristics of increased heat load on the hot surface after the slag skin falls off. Therefore, a slag skin falling off event is determined to have occurred. The start time stamp of the slag skin falling off event is the time stamp when the slag skin thickness reduction rate first turns positive within the discrimination window, and the end time stamp is the time stamp when the slag skin thickness reduction rate falls back to a non-positive value.

[0042] Finally, the difference between the heat flux density at the end time stamp and the heat flux density at the start time stamp of the slag shedding event is calculated as the heat flux density difference, and the difference between the wall temperature at the end time stamp and the wall temperature at the start time stamp is calculated as the wall temperature difference. The heat flux density difference and the wall temperature difference are divided by their respective steady-state baseline values ​​at the start time stamp to achieve dimensionless processing. The two dimensionless differences are added together to generate the damage increment. The larger the damage increment, the more significant the increase in boundary heat load caused by slag shedding, and the more severe the high-temperature exposure damage suffered by the cooling wall body in this event.

[0043] Please see Figure 3 As shown, Figure 3Figure (a) shows the time sequence relationship of the rising inflection points of the wind temperature change rate, heat flux density change rate, and wall temperature change rate, i.e., the wall temperature rise rate, during the period of thermal shock disturbance. The inflection point timestamps of the three curves are marked as t1, t2, and t3, respectively. They satisfy that t1 is earlier than t2 and t3 is later than t2, that is, wind temperature takes precedence, heat flux density responds, and wall temperature follows with lag. Based on this sequence, thermal shock disturbance events are determined and toughness indices are generated.

[0044] Figure 3 Figure (b) shows the synchronous positive correlation trend of the slag thickness reduction rate, heat flux density change rate and wall temperature change rate within the slag detachment discrimination window after the thermal shock disturbance. The three curves rise synchronously and the pairwise Pearson correlation coefficients are all positive. Based on this synchronous change, the slag detachment event is judged and the damage increment is calculated. This figure intuitively reflects the differentiation and quantification of the two types of events, thermal shock disturbance and slag detachment, by the temporal coordinated change of multi-source parameters.

[0045] S3. Multiply the toughness index by the thermal shock life weight, multiply the damage increment by the detachment life weight, add them together, and then multiply by the life weight of the cooling wall part to obtain the damage index and calculate its degradation rate; obtain the boundary thickness attenuation by inverse heat transfer inversion from the heat flux density and wall temperature time series, and determine it as slag thinning or erosion depth; standardize the degradation rate of the damage index and the boundary thickness attenuation respectively, and add them together to obtain the degradation index.

[0046] The lifespan of a cooling wall is not solely determined by the severity of a single event: the contribution of thermal shock and slag shedding to wall damage depends on their respective frequencies, and different locations are affected differently due to variations in heat load. Using toughness indices or damage increments in isolation is insufficient to represent the overall degree of degradation. Furthermore, damage indices only reflect transient response characteristics; the slow decay of wall boundary thickness, as direct evidence of irreversible loss, needs to be obtained separately from a heat transfer perspective. Therefore, this step involves weighted synthesis of toughness indices and damage increments, multiplied by location weights to obtain the damage index and its degradation rate. Then, the boundary thickness decay is obtained through inverse heat transfer inversion and standardized and added to the degradation rate to calculate the degradation index.

[0047] In one specific embodiment, the frequency of thermal shock disturbance events and the frequency of slag peeling events within the historical time window from the start of monitoring to the current calculation time are first counted to obtain the sum of the two frequencies. The thermal shock lifetime weight is obtained by dividing the frequency of thermal shock disturbance events by the sum of frequencies, and the slag peeling event weight is obtained by dividing the frequency of slag peeling events by the sum of frequencies. The counting of the two types of events is performed according to the event start timestamp and end timestamp determined in the aforementioned steps. When the sum of frequencies is zero, both the thermal shock lifetime weight and the slag peeling lifetime weight are set to zero.

[0048] Subsequently, the average heat flux density of each cooling wall unit is averaged according to its longitudinal segment number and circumferential block number to obtain the statistical average heat load of the cooling wall in the longitudinal and circumferential regions of the furnace body. The statistical average heat load of the region is divided by the sum of the statistical average heat load of all regions of the entire furnace to obtain the life weight of the part.

[0049] Based on this, the toughness index is multiplied by the thermal shock life weight to obtain the first product, and the damage increment is multiplied by the detachment life weight to obtain the second product. The first product and the second product are added together and then multiplied by the location life weight to obtain the damage index. This damage index represents the comprehensive damage degree of the cooling wall unit under the thermal load conditions of its location, which is a weighted synthesis of two types of events according to their historical occurrence frequency. The larger the value, the worse the operating condition of the cooling wall.

[0050] Damage indices are calculated for each feature window by sliding step and arranged in chronological order to form a damage index time series. The first-order difference of the damage indices of adjacent feature windows is performed to obtain the degradation rate of the damage indices. Since the damage index increases with the accumulation of damage, this degradation rate characterizes the acceleration rate of the degree of damage to the cooling wall over time. The larger the value, the faster the damage accumulates.

[0051] Furthermore, the damage index and its degradation rate reflect the cumulative damage driven by two types of events. The slow decay of the boundary thickness is an irreversible loss of life and needs to be quantified by heat transfer methods and combined with the degradation rate of the damage index. Therefore, the degradation index is calculated as follows: the heat flux density time series and the wall temperature time series are used as known boundary conditions. The heat flux density time series is used to calculate the overall boundary thermal resistance, and the wall temperature time series is used to classify and determine the amount of boundary thickness decay. Combining the thermal conductivity properties of the cooling wall material and the convective heat transfer properties of the cooling water, the overall boundary thermal resistance time series of the cooling wall is calculated according to the following formula. : .

[0052] in, This refers to the design value of the furnace gas temperature within the height range of the cooling wall unit. for Cooling water temperature at any given time for Heat flux density at any given time For the thickness of the cooling wall body, The thermal conductivity of the cooling wall material is given. The convective heat transfer coefficient between the cooling water and the inner wall of the branch pipe is given. The cooling water temperature is taken as the arithmetic mean of the platinum resistance measurements at the inlet and outlet of the branch pipe. The convective heat transfer coefficient is calculated by the cooling water flow velocity and the diameter of the cooling water branch pipe according to the correlation formula of forced convective heat transfer in the pipe. The cooling water flow velocity is obtained by dividing the cooling water flow rate by the cross-sectional area of ​​the branch pipe.

[0053] The total thermal resistance from the furnace to the cooling water. The thermal resistance of the cooling wall body is the thermal resistance of the cooling wall body. The convective heat transfer thermal resistance on the cooling water side is represented by this formula. The formula as a whole represents the total thermal resistance of the remaining hot surface boundary layer after deducting the thermal resistance of the body and the convective heat transfer thermal resistance on the water side from the total thermal resistance.

[0054] The difference between the maximum value of the boundary comprehensive thermal resistance time series from the start of monitoring to the current calculation time and the current value is extracted as the decrease difference. The decrease difference is used as the thermal resistance attenuation. According to the one-dimensional steady-state thermal conduction law, the thermal resistance attenuation is multiplied by the thermal conductivity corresponding to the judgment result to obtain the boundary thickness attenuation.

[0055] Since the decrease in the overall boundary thermal resistance can be due to either the thinning of the slag coating or the erosion of the cooling wall material, the former is a reversible change that can be recovered with the re-attachment of slag, while the latter is an irreversible loss of life. The two have different meanings, so it is necessary to determine whether the boundary thickness reduction is due to the thinning of the slag coating or the depth of erosion.

[0056] Please see Figure 2 As shown, the steps for determining the boundary thickness attenuation as slag thinning or erosion depth include: S301, extracting the temperature fluctuation range and monotonically increasing duration of the wall temperature time series within a time window, with the time window length being the feature window length; the extraction of the temperature fluctuation range is as follows: traverse all sampled values ​​of the wall temperature within the time window, take the maximum and minimum values, and subtract the minimum value from the maximum value to obtain the temperature fluctuation range; the extraction of the monotonically increasing duration is as follows: calculate the difference in wall temperature for each adjacent sample point within the time window, connect adjacent sample points with positive differences to form an increasing segment, and terminate the increasing segment when encountering an adjacent sample point with a non-positive difference, multiply the number of sample points contained in each increasing segment by a uniform sampling period to obtain the duration of the increasing segment, and take the maximum value among the durations of each increasing segment to obtain the monotonically increasing duration.

[0057] S302. When the temperature fluctuation range is not less than the historical average fluctuation range and the monotonous rise lasts for no more than half a time window, it indicates that the wall temperature fluctuates violently in a short period of time and then falls back. This is consistent with the characteristics of rapid changes in the state of the attached layer caused by the alternation of slag skin shedding and re-attachment. It is determined that the state of the attached layer on the outside of the cooling wall has changed. The difference in temperature drop is multiplied by the thermal conductivity of the slag skin and converted into the boundary thickness reduction amount, which is marked as the slag skin thinning amount.

[0058] S303. When the temperature fluctuation range is less than the historical average fluctuation range and the monotonous rise lasts for more than half a time window, it indicates that the wall temperature is rising slowly with slight fluctuations. This is consistent with the heat transfer characteristics of the continuous decrease in boundary thermal resistance caused by the erosion and thinning of the body material. It is determined that the cooling wall body material has suffered irreversible loss. The difference in temperature is multiplied by the thermal conductivity of the cooling wall body material and converted into the boundary thickness attenuation, which is marked as the erosion depth.

[0059] When the judgment directions of the two criteria, temperature fluctuation range and duration of monotonic increase, are inconsistent, the judgment direction of duration of monotonic increase shall be used for conversion and marking. Among them, the historical average fluctuation range is the arithmetic mean of the temperature fluctuation range calculated window by window with characteristic window length within the historical time window in which no thermal shock disturbance event or slag skin detachment event occurred.

[0060] Finally, the degradation rate and boundary thickness attenuation of the damage index are respectively subtracted from their respective historical time windows from the start of self-monitoring to the current calculation time, and then divided by the standard deviation within that window to complete the standardization process. When the standard deviation is zero, the corresponding standardized value is set to zero. The two standardized values ​​are added together to calculate the degradation index. This degradation index represents a comprehensive quantification of the event-driven damage accumulation rate and the degree of irreversible boundary loss. The larger the value, the deeper the overall degradation of the cooling wall. The degradation index is calculated step by step according to the feature window and arranged in chronological order to form a degradation index time series.

[0061] S4. Input the degradation index time series into the lifetime prediction model and output the remaining lifetime prediction time value and lifetime status level.

[0062] Although the degradation index integrates discrete damage and continuous degradation information, its single-moment value only reflects the current state and cannot characterize the acceleration or deceleration trend of the degradation rate. The deterioration process of the blast furnace cooling wall is nonlinear and time-varying, and threshold judgment or linear extrapolation is difficult to capture the dynamic evolution of the degradation trajectory. Moreover, on-site operation and maintenance require tiered early warning rather than a single value. Therefore, this step inputs the degradation index time series into the life prediction model, and the model outputs the remaining life prediction time value and the life status level.

[0063] In one specific embodiment, the time series of degradation indicators is input into the lifetime prediction model, and the remaining lifetime prediction time value is output. The lifetime prediction model is composed of an input layer, a long short-term memory network layer, a fully connected layer, and an output layer connected in series. The input layer receives the time series of degradation indicators arranged by sliding step size from the monitoring start time to the current calculation time. The long short-term memory network layer extracts the temporal features from the time series. The fully connected layer performs a linear mapping on the temporal features. The output layer is composed of a single neuron and converts the linear mapping result into a remaining lifetime prediction time value, which represents the time from the current calculation time to the predicted replacement time of the cooling wall.

[0064] The training process of the life prediction model is as follows: the degradation index time series of the same type of cooling wall during the historical monitoring period is used as the training input, the duration from each historical calculation time to the actual replacement time of the cooling wall is used as the training label, the actual replacement time of the cooling wall is taken as the replacement timestamp in the blast furnace maintenance record, the mean square error between the remaining life prediction time value and the training label is used as the loss function, and the Adam optimizer is used to iteratively update the network parameters until the loss function value no longer decreases.

[0065] Subsequently, the ratio of the remaining life prediction time to the cooling wall design life benchmark is calculated. The cooling wall design life benchmark is the design value of the cooling wall replacement cycle determined in the blast furnace design.

[0066] The total range of the ratio is divided into three consecutive and mutually exclusive ratio intervals according to the principle of equal division. When the ratio is not less than two-thirds, the ratio falls into the maximum value ratio interval, and the normal state level is output.

[0067] When the ratio is not less than one-third and less than two-thirds, the ratio falls into the middle value range, and the attention status level is output.

[0068] When the ratio is less than one-third, the ratio falls into the minimum numerical proportion range, and the degradation status level is output.

[0069] Please see Figure 4 As shown, Figure 4 The remaining life prediction time of the cooling wall output by the life prediction model is compared with the actual remaining life calculated from the replacement timestamp in the blast furnace maintenance record using two curves. The solid line is the actual remaining life curve, which decreases over time and drops to zero at the actual replacement time. The dashed line is the remaining life prediction time curve output by the life prediction model. The two curves show the same trend and have small numerical deviations, which intuitively demonstrates the degree of agreement between the prediction results of this invention and the actual service life, and verifies the effectiveness of the life prediction method.

[0070] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0071] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0072] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0073] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0074] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis, characterized in that, include: Multi-source monitoring data were collected for each cooling wall unit and a unified spatiotemporal reference was established. Temporal features were extracted from each monitoring data. The multi-source monitoring data included heat flux density, wall temperature, cooling water flow rate and temperature difference, air temperature and slag thickness. When the change in wind temperature precedes the change in heat flux density and the change in wall temperature follow behind, thermal shock disturbances are identified, and toughness indices are generated from the recovery time and overshoot. After thermal shock, when the slag thickness decreases and the heat flux density and wall temperature rise synchronously, slag shedding is identified, generating damage increments. The toughness index is multiplied by the thermal shock life weight, the damage increment is multiplied by the detachment life weight, and the results are added together. Then, the results are multiplied by the life weight of the cooling wall section to obtain the damage index and calculate its degradation rate. The boundary thickness attenuation is obtained by inverse heat transfer inversion from the heat flux density and wall temperature time series, and is determined as the slag thinning amount or erosion depth. The degradation rate of the damage index and the boundary thickness attenuation amount are standardized and added together to obtain the degradation index. Input the time series of degradation indicators into the lifetime prediction model, and output the predicted remaining lifetime time and lifetime status level.

2. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 1, characterized in that, The time-series characteristics are the statistical characteristics of each monitoring data, including at least one of mean, peak value, rate of change, or volatility.

3. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 1, characterized in that, The acquisition of the resilience index includes: A thermal shock disturbance event is determined to have occurred when the time stamp of the inflection point of the wind temperature change rate is earlier than the time stamp of the inflection point of the heat flux density change rate, and the time stamp of the inflection point of the wall temperature change rate is later than the time stamp of the inflection point of the heat flux density change rate. The difference between the peak wall temperature and the steady-state baseline during a thermal shock disturbance event is calculated as the overshoot, and the time span during which the wall temperature falls back to the steady-state baseline is calculated as the recovery time. The overshoot and recovery time are multiplied after being dimensionless to generate a toughness index.

4. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 3, characterized in that, The acquisition of the damage increment includes: Within the time window after the thermal shock disturbance event ends, calculate the Pearson correlation coefficients of the slag thickness reduction rate, the heat flux density change rate, and the wall temperature change rate. When all three correlation coefficients show a positive upward trend, it is determined that a slag detachment event has occurred. The difference between the heat flux density and the wall temperature before and after the slag peeling event is calculated. The two are then dimensionless and added together to generate the damage increment.

5. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 4, characterized in that, The steady-state baseline is the moving average sequence of the heat flux density time series and the wall temperature time series within a historical time window during which no thermal shock disturbances or slag peeling events have occurred; the steady-state baseline interval is the numerical range of the moving average sequence fluctuating by one standard deviation above and below.

6. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 1, characterized in that, The calculation of the degradation rate of the damage index includes: The frequency of thermal shock disturbance events and the frequency of slag peeling events within the historical time window are statistically analyzed. The sum of the two frequencies is calculated. The thermal shock lifetime weight is calculated by dividing the frequency of thermal shock disturbance events by the sum of the frequencies. The slag peeling lifetime weight is obtained by dividing the frequency of slag peeling events by the sum of the frequencies. Calculate the statistical mean of the heat load of the cooling wall in the longitudinal and circumferential regions of the furnace body, and divide the statistical mean of the heat load of the region by the sum of the statistical mean of the heat load of the cooling wall of the whole furnace to obtain the life weight of the part. The first product is calculated by multiplying the toughness index by the thermal shock life weight, and the second product is calculated by multiplying the damage increment by the detachment life weight. The first and second products are added together and then multiplied by the location life weight to obtain the damage index. The time series of damage indicators are extracted, and the first-order difference is calculated on the time series to obtain the degradation rate of the damage indicators.

7. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 1, characterized in that, The calculation of the degradation index includes: Using the heat flux density time series and the wall temperature time series as known boundary conditions, and combining the thermal conductivity of the cooling wall material and the convective heat transfer properties of the cooling water, the comprehensive thermal resistance time series of the cooling wall boundary is calculated. Extract the decreasing difference in the time series of the boundary comprehensive thermal resistance, substitute the decreasing difference into the heat transfer inverse problem solving algorithm, and calculate the boundary thickness attenuation; the boundary thickness attenuation is the amount of slag thinning or the erosion depth. The degradation index is calculated by standardizing the degradation rate of the damage index and the boundary thickness attenuation, and then adding them together.

8. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 7, characterized in that, The process of determining the boundary thickness attenuation as either slag thinning or erosion depth includes: Extract the temperature fluctuation range and the duration of monotonic rise within the time window of the wall temperature time series; When the temperature fluctuation range is not less than the historical average fluctuation range and the monotonous rise duration is not greater than half the time window length, it is determined that the state of the outer layer of the cooling wall has changed, and the boundary thickness attenuation is marked as the slag skin thinning amount. When the temperature fluctuation range is less than the historical average fluctuation range and the monotonous increase lasts for more than half a time window, it is determined that the cooling wall material has suffered irreversible loss, and the boundary thickness attenuation is marked as the erosion depth.

9. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 1, characterized in that, The lifetime prediction model consists of an input layer, a long short-term memory network layer, a fully connected layer, and an output layer connected in series. The time series of degradation indicators is input into the long short-term memory network layer through the input layer to extract temporal features. After being calculated by the fully connected layer, it is converted into the remaining lifetime prediction time value by the output layer.

10. The method for predicting the service life of blast furnace cooling walls based on multi-source data fusion analysis as described in claim 1, characterized in that, The output of the remaining lifetime prediction time and lifetime status level specifically includes: Input the time series of degradation indicators into the life prediction model and output the remaining life prediction time value. Calculate the ratio of the predicted remaining lifespan to the baseline design lifespan of the cooling wall; The life prediction model is divided into three consecutive and mutually exclusive ratio intervals based on the magnitude of the ratio. When the ratio falls into the interval with the largest value, the life prediction model outputs the normal state level. When the ratio falls within the middle range of numerical proportions, the life prediction model outputs the level of concern. When the ratio falls within the minimum numerical proportion range, the lifetime prediction model outputs the degradation state level.

Images