Carburizing heat treatment furnace carbon potential monitoring system and method based on big data analysis

Abstract

The present application relates to the technical field of heat treatment process monitoring, and discloses a carburizing heat treatment furnace carbon potential monitoring system and method based on big data analysis, wherein the system comprises a transition data acquisition module, a reference trajectory generation module, a dynamics prediction module, a deviation analysis module, a parameter identification module, a correction monitoring module and a data feedback module. The reference transition trajectory template is generated by performing normalized time alignment and weighted statistical processing on the historical transition process data of multiple furnaces in the workshop, the online identification of the rate constant scaling factor is triggered by the deviation trend index, and the corrected parameters are used for Kalman filter prediction step update, and the carbon potential transient monitoring value is output.

Description

Technical Field

[0001] This invention relates to the field of heat treatment process monitoring technology, and more specifically, to a carbon potential monitoring system and method for carburizing heat treatment furnaces based on big data analysis. Background Technology

[0002] Carburizing heat treatment is an important process for improving the surface hardness and wear resistance of metal parts, and accurate monitoring of the carbon potential of the furnace gas is the core link to ensure the quality of carburizing. During the transition from the strong carburizing to the diffusion stage, the composition of the furnace gas undergoes transient changes, and the carbon potential monitoring system needs to maintain reliable prediction and tracking capabilities during this transition period.

[0003] Existing carbon potential monitoring methods typically employ a framework combining reaction kinetic rate equations with Kalman filtering. The rate constants upon which the rate equations depend are derived from literature standards or offline calibration values, and each furnace independently performs Runge-Kutta integration for carbon potential prediction. However, under actual furnace conditions, the rate constants deviate from the standard values ​​due to factors such as the catalytic effect of carbon deposition on the furnace wall, furnace airtightness, and fan stirring efficiency. These factors change slowly with furnace age and maintenance status. When the rate constant is inaccurate, the carbon potential prediction trajectory generated by Runge-Kutta integration exhibits cumulative deviations, causing the Kalman filter to continuously rely on observational corrections during the transition period, thus losing its predictive ability. Furthermore, the lack of external reference benchmarks in single-furnace systems makes it impossible to distinguish whether the prediction residuals originate from rate constant deviations or oxygen probe measurement errors. The large amount of completed transition process data accumulated at the workshop level is also not effectively utilized for parameter correction, making it difficult to improve monitoring accuracy with the accumulation of production data. Summary of the Invention

[0004] This invention provides a carbon potential monitoring system and method for carburizing heat treatment furnaces based on big data analysis, which solves the technical problems in related technologies such as insufficient accuracy of dynamic prediction of carbon potential during the transition period of stage switching of carburizing heat treatment furnaces, inability of kinetic parameters to adaptively correct with changes in furnace conditions, and low monitoring reliability due to the lack of statistical reference benchmarks based on historical big data.

[0005] This invention discloses a carbon potential monitoring system for a carburizing heat treatment furnace, comprising: The transition data acquisition module acquires historical transition data of multiple furnaces that have completed stage switching, forming a large dataset of the transition process. The reference trajectory generation module performs normalized time alignment and weighted statistics on the carbon potential curves of each furnace to generate a reference transition trajectory template and statistical confidence band. The kinetic prediction module substitutes the rate constant parameter into the reaction kinetic rate equations and performs integration to generate the target furnace carbon potential prediction trajectory. The deviation analysis module compares the carbon potential prediction trajectory with the reference transition trajectory template point by point and calculates trend indicators. The parameter identification module uses a reference transition trajectory template as a constraint target identification rate constant scaling factor when the trend indicator exceeds the threshold. The calibration monitoring module updates the kinetic parameters with the identified rate constant scaling factor, performs calibration integration and filtering update, and outputs the carbon potential monitoring value. The data feedback module writes the data into the big data set and updates the reference transition trajectory template after the transition is completed.

[0006] Furthermore, the reference trajectory generation module includes a time alignment submodule, which maps the actual transition duration of the carbon potential time series curves during the transition period of each furnace to a normalized time axis. ,in The corresponding step jump time, For each furnace, the determination time of the return to steady state is used, and an equally spaced carbon potential sequence is obtained on the normalized time axis using a linear interpolation algorithm. The anomaly removal submodule calculates the integral deviation between the carbon potential curve and the population mean curve at all normalized time points for each furnace, and uses the ratio of this integral deviation to the population standard deviation as the deviation index. Furnace data with deviations exceeding the population statistical threshold are removed entirely. The statistical calculation submodule calculates the weighted mean and weighted standard deviation of the carbon potential at each normalized time point for the valid furnace data after anomaly removal. The weighted mean at each normalized time point constitutes the reference transition trajectory template, and the upper and lower bounds corresponding to the weighted standard deviation constitute the statistical confidence band.

[0007] Furthermore, in the statistical calculation submodule, the weight of each furnace is assigned in a time decay manner according to the order of its completion time, with the furnace closer to the current time having a greater weight.

[0008] Furthermore, the kinetic prediction module includes: a real-time data acquisition submodule, which acquires the measured values ​​of the oxygen probe electromotive force, furnace temperature, and enriched gas flow step change markers of the target furnace, and reads the output value of the Kalman filter in the steady-state stage before the step change as the initial component state; if the target furnace has furnace condition marker values ​​written in previous furnaces, the current rate constant parameter is initialized with the rate constant scaling factor recorded by the furnace condition marker value; otherwise, the standard rate constant in the literature is used as the initial parameter; a kinetic integration submodule, which, with the initial component state as the starting condition, substitutes the current rate constant parameter into the reaction kinetic rate equation system containing the forward and reverse rate terms of the Boudouard reaction and the water-gas reaction, performs time integration using the fourth-order Runge-Kutta method, and calculates the kinetic prediction values ​​of the gas component concentrations at each time step; and a carbon potential conversion submodule, which converts the kinetic prediction values ​​of the gas component concentrations at each time step into the carbon potential prediction trajectory according to the thermodynamic equilibrium conversion relationship between carbon activity and carbon potential, and performs time mapping according to the same normalized time axis as the reference transition trajectory template.

[0009] Furthermore, the trend indicators calculated by the deviation analysis module include monotonic growth indicators. and cumulative deviation The deviation analysis module generates a deviation sequence. ,in For the first The difference between the predicted carbon potential value of the carbon potential trajectory at each normalized time point and the carbon potential value of the reference transition trajectory template. The current number of normalized time points; the monotonic growth indicator The calculation method is as follows: calculate the difference between adjacent values. Count the number of positive values. and the number of negative values ,according to Calculate; the cumulative deviation according to calculate.

[0010] Furthermore, the parameter identification module in the monotonically increasing index and the cumulative deviation Execution is triggered when all parameters exceed their respective confidence thresholds; the parameter identification module scales the rate constant of the Boudouard reaction by a factor. Scaling factor for the rate constant of the reaction with water and gas Let the parameter be set as the one to be identified, and the rate constant scaling factor be defined as the ratio of the actual effective rate constant to the standard rate constant in the literature; the objective function is minimized using a nonlinear least squares optimization method: ; in For the weighted sum of squared residuals, For and For the parameter in the first Normalization time point The predicted carbon potential at that location, In order to refer to the transition trajectory template in The carbon potential at that point, Take the statistical confidence band at The reciprocal of the width.

[0011] Furthermore, the correction monitoring module includes: a parameter update submodule, which writes the rate constant scaling factor into the reaction kinetic rate equations of the target furnace and updates the forward and reverse rate constants of the Boudouard reaction and the water-gas reaction; a correction integration submodule, which uses the concentration values ​​of each component in the current Kalman filter state estimate as starting conditions, performs a fourth-order Runge-Kutta integral with the updated rate constant parameters, calculates the predicted values ​​of each component concentration in the next time step, and converts them into carbon potential correction prediction values; and a filter update submodule, which uses the carbon potential correction prediction values ​​as the output of the Kalman filter prediction step, combines the measured value of the oxygen probe electromotive force with the carbon potential observation value obtained by the Nernst equation transformation, performs a Kalman filter update step, and outputs the fused carbon potential monitoring value.

[0012] Furthermore, the data feedback module includes: a data writing submodule, which writes the complete transition trajectory data and rate constant scaling factor of the target furnace into the transition process big data set as a new furnace record; a template updating submodule, which uses the transition process big data set after the new record is written as the data source, and re-executes normalized time alignment, anomaly removal and weighted statistical calculation according to time decay weight to generate an updated reference transition trajectory template and statistical confidence band; and a furnace condition marking submodule, which stores the rate constant scaling factor obtained in this identification as the current furnace condition marking value of the target furnace, for the kinetic prediction module of the subsequent furnaces of this furnace to read for the initialization of the rate constant.

[0013] Furthermore, the initial component state includes the initial values ​​of carbon monoxide concentration, carbon dioxide concentration, water vapor partial pressure, hydrogen concentration, and methane concentration; the data collected by the transition data acquisition module for each furnace includes the carbon potential tracking time series curve, furnace temperature time series, enriched gas flow step parameter, and final steady-state convergent carbon potential value output by Kalman filtering during the transition period.

[0014] This invention discloses a method for monitoring the carbon potential of a carburizing heat treatment furnace based on big data analysis, used to execute the aforementioned carbon potential monitoring system for a carburizing heat treatment furnace based on big data analysis, comprising: Step 1: Obtain historical transition data of multiple furnaces that have completed stage switching through the transition data acquisition module to form a large dataset of the transition process; Step 2: Normalize the time alignment and weighted statistics of the carbon potential curves of each furnace by using the reference trajectory generation module to generate a reference transition trajectory template and statistical confidence band; Step 3: Substitute the rate constant parameter into the reaction kinetic rate equations through the kinetic prediction module to perform integration and generate the target furnace carbon potential prediction trajectory. Step 4: The carbon potential prediction trajectory is compared point by point with the reference transition trajectory template through the deviation analysis module to calculate the trend index. Step 5: When the trend indicator exceeds the threshold, the parameter identification module identifies the rate constant scaling factor by using the reference transition trajectory template as the constraint target. Step 6: Update the kinetic parameters with the identified rate constant scaling factor through the calibration monitoring module, perform calibration integration and filtering update, and output the carbon potential monitoring value; Step 7: After the transition is completed, the data is written into the big dataset and the reference transition trajectory template is updated through the data feedback module.

[0015] The present invention has the following beneficial effects: By performing normalized time alignment and time decay weighted statistical processing on historical data from multiple carburizing heat treatment furnaces in the workshop that have completed the transition process, a reference transition trajectory template and statistical confidence band are generated. The random measurement errors of a single oxygen probe cancel each other out during the weighted statistical process across multiple furnaces, and the preserved carbon potential change trend represents the typical transition behavior of the furnace group under normal rate constant conditions. Using this as an external anchoring benchmark, when a systematic unidirectional drift occurs between the target furnace's carbon potential kinetic prediction trajectory and the reference transition trajectory template, this drift can be attributed to rate constant deviation rather than oxygen probe measurement error, thus solving the problem of not being able to distinguish between the two types of error sources in a single furnace system.

[0016] The identification rate constant scaling factor is determined by performing nonlinear least squares optimization with a reference transition trajectory template as the constraint objective and the inverse of the statistical confidence band width as the weight. and The parameters of the reaction kinetic rate equations can be adapted to the combined effects of factors such as the catalytic effect of carbon deposition on the furnace wall, the airtightness of the furnace, and the efficiency of the fan stirring, which are present in the current actual furnace conditions of the target furnace. The deviation between the predicted carbon potential generated by performing Runge-Kutta integration with the corrected rate constant and the actual transition trajectory is reduced. The predicted step output of the Kalman filter no longer continuously deviates from the observed value, and the filter can restore the normal cooperative relationship between prediction and observation, overcoming the problem that the filter loses its predictive ability during the transition period due to a fixed standard rate constant.

[0017] After each transition, the data feedback module injects new data into the multi-furnace transition process dataset and updates the reference transition trajectory template with time decay weights. The reference transition trajectory template can track the slow changing trend of the overall furnace condition of the workshop furnace group. At the same time, the rate constant scaling factor is used as the furnace condition label value for parameter initialization of subsequent furnace cycles. This allows the kinetic integral of subsequent transition periods to start from a parameter starting point closer to the actual furnace condition, reducing the initial accumulation of deviations and achieving a gradual improvement in monitoring accuracy with data accumulation. Attached Figure Description

[0018] Figure 1This is a flowchart of a method for monitoring the carbon potential of a carburizing heat treatment furnace based on big data analysis, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the overall data flow of the carburizing transition monitoring system provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the reference transition trajectory template and confidence band provided in an embodiment of the present invention; Figure 4 This is a schematic diagram comparing the predicted carbon potential of furnace 3 based on the kinetics with a reference template, provided in an embodiment of the present invention. Figure 5 This is a schematic diagram of the deviation sequence distribution (representative normalized time points) provided in an embodiment of the present invention; Figure 6 This is a schematic diagram comparing the predicted and monitored carbon potential values ​​before and after correction, provided in an embodiment of the present invention. Figure 7 This is a schematic diagram showing the distribution of actual transition times for each historical furnace batch provided in the embodiments of the present invention; Figure 8 This is a schematic diagram of the internal flowchart of the parameter identification module provided in an embodiment of the present invention; Figure 9 This is a schematic diagram illustrating the relationship between the confidence band width and the normalized time point provided in an embodiment of the present invention. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0020] At least one embodiment of the present invention discloses a carbon potential monitoring system for a carburizing heat treatment furnace based on big data analysis, wherein the method performed by the system is as follows: Figure 1 As shown, the system includes: The transition data acquisition module is configured to acquire historical transition process data of multiple furnaces that have completed stage switching, and aggregate them to form a large dataset of multi-furnace transition processes. The transition data acquisition module retrieves transition process data from the historical records of multiple carburizing heat treatment furnaces within the workshop, showing furnaces that have completed the same type of stage switching and returned to steady state. For each historical furnace, the acquired data includes: the carbon potential tracking time-series curve output by Kalman filtering during the transition period, the corresponding furnace temperature time-series sequence, the enrichment gas flow step parameter, and the final steady-state converged carbon potential value. The data from each furnace are organized according to a unified data format and timestamp standard, and aggregated to form a large dataset of multi-furnace transition processes.

[0021] The reference trajectory generation module is configured to perform time alignment and statistical processing on historical transition data from multiple furnaces, generating a reference transition trajectory template and statistical confidence band.

[0022] The reference trajectory generation module performs the following processing on the carbon potential time-series curves of each furnace in the multi-furnace transition process big data set: The time alignment submodule aligns the carbon potential time-series curves of each furnace's transition period according to the normalized transition time. It maps the actual transition duration of each curve to a unified normalized time axis. ,in The corresponding step jump time, The determination time for the return to steady state of each furnace is corresponding to the time of the furnace, thereby eliminating the inconsistency in transition time caused by differences in furnace heat capacity and gas path response among furnaces. The sampling points of the carbon potential curve of each furnace on the normalized time axis are obtained by linear interpolation algorithm to obtain an equally spaced normalized time point sequence. The input is the original time series carbon potential curve and its actual timestamp, and the output is the equally spaced carbon potential sequence of each furnace on the unified normalized time axis.

[0023] The anomaly removal submodule calculates population statistics for the aligned carbon potential curves of each furnace at each normalized time point. For each furnace, it calculates the integral deviation between its carbon potential curve and the current population mean curve at all normalized time points, and uses the ratio of this integral deviation to the population standard deviation as the deviation index for that furnace. Furnaces with deviations exceeding the population statistical threshold are removed entirely.

[0024] The statistical calculation submodule calculates the weighted mean of carbon potential and confidence intervals at each normalized time point for valid furnace data after anomaly removal. In the weighted calculation, the weight of each furnace is assigned in a time-decay manner according to its completion time, with furnaces closer to the current time having a larger weight, so that the reference transition trajectory template reflects recent furnace conditions more comprehensively. The weighted mean at each normalized time point constitutes the reference transition trajectory template, and the upper and lower bounds corresponding to the weighted standard deviation constitute the statistical confidence band.

[0025] The kinetic prediction module is configured to acquire real-time data of the current transient process of the target furnace and perform reaction kinetic integration to generate a carbon potential kinetic prediction trajectory.

[0026] The kinetic prediction module performs the following processing on the target furnace currently undergoing a phase switch: The real-time data acquisition submodule acquires the measured values ​​of the oxygen probe electromotive force, furnace temperature, and enrichment gas flow step change markers for the target furnace. Simultaneously, it reads the Kalman filter output value from the steady-state phase before the step change as the initial component state, including initial values ​​for carbon monoxide concentration, carbon dioxide concentration, water vapor partial pressure, hydrogen concentration, and methane concentration. If the target furnace has furnace condition marker values ​​written in previous furnace cycles, the current rate constant parameter is initialized using the rate constant scaling factor recorded in the furnace condition marker value; otherwise, the standard rate constant from the literature is used as the initial parameter.

[0027] The kinetic integration submodule starts with the initial component states as the starting condition, substituting the current rate constant parameters into the reaction kinetic rate equations, which include forward and reverse rate terms for the Boudouard reaction and the water-gas reaction. The rate equations are integrated over time using the fourth-order Runge-Kutta method. The measured furnace temperature and enriched gas flow rate at each integration time step are used as external input conditions to calculate the kinetic predictions of carbon monoxide, carbon dioxide, water vapor, hydrogen, and methane concentrations at each time step.

[0028] The carbon potential conversion submodule converts the predicted gas component concentrations at each time step into a carbon potential kinetic prediction trajectory based on the thermodynamic equilibrium relationship between carbon activity and carbon potential. This carbon potential kinetic prediction trajectory is time-mapped using the same normalized time axis as the reference transition trajectory template, ensuring subsequent comparisons are performed under a unified time coordinate system.

[0029] The deviation analysis module is configured to compare the carbon potential dynamics prediction trajectory with the reference transition trajectory template point by point, generate a deviation sequence, and calculate trend indicators.

[0030] The deviation analysis module performs the following processing: Point-by-point comparison submodule: This module calculates the difference between the carbon potential value at each normalized transition time point of the predicted carbon potential trajectory and the corresponding carbon potential value of the reference transition trajectory template, generating a deviation sequence. ,in For the first The difference between the kinetic predicted carbon potential at each normalized time point and the carbon potential of the reference transition trajectory template. This represents the number of normalized time points that have already occurred. This represents the sequence number of the normalized time point. Both the carbon potential kinetic prediction trajectory and the reference transition trajectory template are expressed in terms of carbon potential (unit: mass fraction, dimensionless). The difference operation is performed point by point with the same dimension, requiring no additional preprocessing.

[0031] Trend Analysis Submodule: For Deviation Sequences Perform trend analysis and calculate the following two indicators: Monotonic growth indicators Calculate the difference between adjacent values ​​in the deviation sequence. ( ),in For the first Count the number of positive values ​​among the differences between adjacent deviations. and the number of negative values ,in for Take the number of positive values. for Count the number of negative values ​​and calculate. ; in As a monotonic growth indicator, This represents the total number of adjacent differences. The closer the value is This indicates that the more consistent the direction of change of the deviation sequence, the more likely the deviation exhibits a unidirectional continuous drift characteristic rather than random fluctuation.

[0032] Cumulative deviation : Calculate the cumulative sum of the absolute values ​​of each point in the deviation sequence ; in The cumulative deviation reflects the overall degree of deviation between the carbon potential dynamics predicted trajectory and the reference transition trajectory template. It is a dimensionless ratio. The dimensions are consistent with the carbon potential, and both are independently compared with their respective parameter confidence thresholds, without any cross-dimensional addition or subtraction operations.

[0033] The parameter identification module is configured to identify the rate constant scaling factor by using the reference transition trajectory template as the constraint target when the deviation trend index exceeds the threshold.

[0034] The parameter identification module in monotonically increasing indicators and cumulative deviation Execution is triggered when all parameters exceed their respective confidence thresholds. The process is as follows: Identify variable definition submodule: scale the rate constant of the Boudouard reaction. Scaling factor for the rate constant of the reaction with water and gas Set as the parameter to be identified. The rate constant scaling factor is defined as the ratio of the actual effective rate constant to the literature standard rate constant, that is, the actual effective rate constant of the Boudouard reaction is equal to... The actual effective rate constant of the water-gas reaction is equal to its standard rate constant. The product of its standard rate constant.

[0035] Optimization solution submodule: using the carbon potential values ​​at each normalized time point of the reference transition trajectory template. As the constraint target value, For reference, the transition trajectory template is in the first Normalization time point The carbon potential value at that location. and Substituting the reaction kinetic rate equations and re-executing the fourth-order Runge-Kutta integral from the initial component states, a parameterized carbon potential prediction trajectory is generated. ,in For and For the parameter in the first Normalization time point The predicted carbon potential at that location, , Let be the normalized total number of time points. Minimize the objective function using a nonlinear least squares optimization method: ; in For the weighted sum of squared residuals, For the first The weights of each normalized time point. and All are expressed as carbon potential mass fractions, with consistent dimensions, and the residual squared terms have the same dimensions. Take the statistical confidence band at The reciprocal of the width of the statistical confidence band, which has the same dimensions as the carbon potential, is used as a weight to weight the squared residuals. The units are the same and do not affect the comparison operation of the optimization objective function. The narrower the statistical confidence band, the higher the statistical consistency of the multiple furnaces at that time point, and the stronger the reliability of the corresponding reference value, so it is given a larger weight. The input of the nonlinear least squares optimization is the point-by-point residual and weight of the parameterized carbon potential prediction trajectory and the reference transition trajectory template carbon potential value, and the output is to make the... Take the minimum value and .

[0036] The identification result output submodule outputs the scaling factor of the rate constant obtained from online identification. and .

[0037] The calibration monitoring module is configured to update the kinetic parameters using the identified rate constant scaling factor and perform calibration prediction and filtering updates, outputting the transient monitoring value of the carbon potential.

[0038] The calibration monitoring module performs the following processing: Parameter update submodule: scales the rate constant. and The reaction kinetic rate equations of the target furnace are written into the system to update the forward and reverse rate constants of the Boudouard reaction and the water-gas reaction, respectively.

[0039] The correction integral submodule starts with the concentration values ​​of each component in the current Kalman filter state estimate as the initial condition, performs a fourth-order Runge-Kutta integral with the updated rate constant parameter, calculates the predicted values ​​of each component concentration in the next time step, and converts them into carbon potential correction prediction values.

[0040] The filter update submodule uses the carbon potential correction prediction value as the output of the Kalman filter prediction step. It then combines this with the carbon potential observation value obtained by transforming the measured electromotive force of the oxygen probe using the Nernst equation to perform the Kalman filter update step, outputting the fused transient carbon potential monitoring value. Subsequent integration time steps continue to perform prediction and filter update cycles with the updated rate constant parameters until the current transition process ends.

[0041] The data feedback module is configured to write the transition data to the multi-furnace transition process big data set after the target furnace transition is completed, and update the reference transition trajectory template and furnace condition parameter markers.

[0042] After the target furnace completes its transition and returns to steady state, the data feedback module performs the following processing: The data writing submodule stores the complete transition trajectory data of the target furnace (including the carbon potential time series curve, furnace temperature sequence, enrichment gas flow step parameters, and steady-state convergent carbon potential value) and the rate constant scaling factor. and Write the data into the multi-furnace transition process big dataset as a record of newly added furnaces.

[0043] The template update submodule uses the large dataset of the multi-furnace transition process after the new record is written as the data source. It re-executes normalized time alignment, anomaly removal, and weighted statistical calculation according to time decay weights to generate an updated reference transition trajectory template and statistical confidence band. Since the newly written furnace record is the closest to the current time, it receives the maximum weight in the weighted calculation, causing the updated reference transition trajectory template to shift towards the latest furnace condition characteristics.

[0044] Furnace condition marking submodule: Scales the rate constant obtained from this identification using a scaling factor. and The current furnace condition flag value is stored as the target furnace. When the furnace undergoes a phase transition in subsequent furnace cycles, the kinetic prediction module reads the furnace condition flag value to initialize the rate constant, so that the kinetic integral starts from a parameter starting point close to the current actual furnace condition, reducing the initial accumulation of kinetic prediction deviations during the subsequent transition period.

[0045] Output the updated reference transition trajectory template and target furnace condition parameter markers.

[0046] This invention generates a reference transition trajectory template and its statistical confidence band by performing normalized time alignment and weighted statistical processing on large-scale historical data of multiple carburizing heat treatment furnaces in a workshop that have completed the transition process. The reference transition trajectory template aggregates the statistical characteristics of the group transition behavior of multiple furnaces under the same process formula. The random measurement errors of a single oxygen probe are mutually canceled out in the weighted statistical process of multiple furnaces, and the retained carbon potential change trend represents the typical transition behavior of the current workshop furnace group under normal rate constant conditions. Based on this, when a systematic unidirectional drift detected by the deviation analysis module occurs between the carbon potential kinetic prediction trajectory of the target furnace and the reference transition trajectory template, this drift can be attributed to the deviation of the reaction kinetic rate constant of the target furnace rather than the measurement error of the oxygen probe, thus solving the problem of not being able to distinguish between the two types of error sources in a single furnace system.

[0047] The identification rate constant scaling factor is determined by performing nonlinear least squares optimization with a reference transition trajectory template as the constraint objective and the inverse of the statistical confidence band width as the weight. and The parameters of the reaction kinetic rate equations can be adapted to the combined effects of factors such as the catalytic effect of carbon deposition on the furnace wall, the airtightness of the furnace, and the efficiency of the fan stirring, which are present in the current actual furnace conditions of the target furnace. The deviation between the predicted carbon potential generated by performing Runge-Kutta integration with the corrected rate constant and the actual transition trajectory is reduced. The predicted step output of the Kalman filter no longer continuously deviates from the observed value, and the filter restores the normal cooperative relationship between prediction and observation. This overcomes the problem of the filter losing its predictive ability during the transition period due to a fixed standard rate constant.

[0048] After each transition, the data feedback module injects new data into the multi-furnace transition process dataset and updates the reference transition trajectory template with time decay weights. The reference transition trajectory template can track the slow changing trend of the overall furnace condition of the workshop furnace group. At the same time, the rate constant scaling factor is used as the furnace condition label value for parameter initialization of subsequent furnace cycles. This allows the kinetic integral of subsequent transition periods to start from a parameter starting point closer to the actual furnace condition, reducing the initial accumulation of deviations and achieving a gradual improvement in monitoring accuracy with data accumulation.

[0049] See Figures 2-9 The following application examples are proposed: A workshop has six identical carburizing heat treatment furnaces (numbered Furnace 1 to Furnace 6), all performing carburizing treatment on gear parts according to the "strong carburizing → diffusion" process formula. The target carbon potential is maintained at 1.20% (mass fraction) during the strong carburizing stage, and drops to 0.85% after switching to the diffusion stage. The workshop has accumulated several months of historical transition process data. Furnace 3 is currently performing this batch's transition from the strong carburizing to the diffusion stage. It has a relatively long furnace life, and there is a certain degree of carbon deposition on the furnace wall. The initial rate constant is started using the literature standard value. This application example uses the current transition monitoring process of Furnace 3 as the main thread to demonstrate the complete data flow of each module.

[0050] Transitional data acquisition module: Data from furnaces 1 through 6 that completed the transition from strong percolation to diffusion was extracted from the workshop's historical records. Each record includes the carbon potential time-series curve output by the Kalman filter, the furnace temperature sequence, the enrichment gas flow step parameter, and the steady-state convergent carbon potential value. A total of 42 valid historical furnace records were extracted and aggregated to form a large dataset of multi-furnace transition processes. The table below shows the basic information of 6 representative records.

[0051] Table 1. Representative records of the multi-furnace historical transition process big data set

[0052] Reference trajectory generation module: The time alignment submodule maps the actual time axis of each of the 42 historical carbon potential curves to a normalized time axis. To eliminate the influence of differences in transition time between different furnace batches (ranging from 36.8 to 41.2 min), 20 equally spaced sampling points were taken on the normalized time axis using linear interpolation. ).

[0053] The anomaly removal submodule calculates the integral deviation of each alignment curve from the current group mean curve, and uses the ratio of the integral deviation to the group standard deviation as the deviation index. Of the 42 records, 3 had deviations exceeding the group statistical threshold (set to 3 times the group standard deviation) and were therefore removed, leaving 39 valid furnace batches.

[0054] The statistical calculation submodule assigns time decay weights to the 39 valid furnace cycles based on their completion time, with furnace cycles completed in year C having the highest weights and those completed in year A having the lowest. Weighted mean and weighted standard deviation are calculated at each normalized time point to form a reference transition trajectory template and statistical confidence band. The table below shows the statistical results for some normalized time points.

[0055] Table 2. Reference Transition Trajectory Template (Partially Normalized Time Points)

[0056] It can be seen that the closer the normalization time point is to the steady state ( The narrower the confidence band, the higher the statistical consistency of multiple furnaces near the steady state.

[0057] Dynamics prediction module: The real-time data acquisition submodule reads the initial state of furnace 3 during the current transition period: the initial concentrations of each component output by the Kalman filter before the step occurs are shown in the table below. Since furnace 3 has no furnace condition marker values ​​written in previous furnace cycles, the standard rate constant (from the literature) is used. , ) as initial parameters.

[0058] Table 3 Initial composition state of Furnace 3 (time of step occurrence)

[0059] The kinetics integrator module substitutes the above initial state into the reaction kinetic rate equations, using the measured furnace temperature (920). The external inputs are the stable concentration and the enriched gas flow rate step (12.0→6.5 L / min), and the time integration is performed using the fourth-order Runge-Kutta method. The carbon potential conversion submodule converts the predicted component concentrations at each time step into carbon potentials according to the thermodynamic equilibrium relationship and maps them to the normalized time axis. The table below shows a comparison between the carbon potential trajectory predicted by the furnace 3 kinetics and the reference transition trajectory template.

[0060] Table 4 Comparison of Furnace 3 Carbon Potential Kinetic Prediction Trajectory with Reference Template

[0061] Deviation Analysis Module: The point-by-point alignment submodule generates a complete deviation sequence at 20 normalized time points. Table 4 shows the deviation values ​​at five representative time points.

[0062] The trend analysis submodule calculates two indicators for the complete deviation series: calculating the adjacent differences. (Total 19), of which the number of positive values Number of negative values Monotonic growth indicator is The cumulative deviation is ; The system's preset parameter confidence threshold is , .current and Both indicators exceeded the threshold, triggering the parameter identification module. The deviation sequence showed a continuous positive drift in one direction, indicating that the actual reaction kinetic rate constant of furnace 3 was higher than the standard value in the literature, resulting in a continuously higher predicted carbon potential, rather than being caused by random measurement errors of the oxygen probe.

[0063] Parameter identification module: Optimize the solution submodule to reference the carbon potential value of the transition trajectory template. To constrain the objective, the inverse of the confidence band width is used as the weight (normalized time points). The narrowest width is 0.02%, corresponding to a maximum weight of 50; (Wide width is 0.10%, corresponding to a weight of 10) Parameters to be identified and Perform nonlinear least squares optimization, with the objective function being: ; Optimization iteration from the initial point Start, converge, and output the recognition result. , This indicates that the actual Boudouard reaction rate constant of Furnace 3 is approximately 1.18 times the standard value in the literature, and the water-gas reaction rate constant is approximately 1.09 times the standard value, which is consistent with the actual situation of Furnace 3 having a long furnace life and enhanced catalytic effect of carbon deposition on the furnace wall.

[0064] Calibration monitoring module: The parameter update submodule will , The reaction kinetic rate equations for furnace 3 are written into the system, updating the forward and reverse rate constants. The correction integral submodule uses the component concentrations estimated by the current Kalman filter state as starting conditions and performs a fourth-order Runge-Kutta integral with the updated rate constants to generate a carbon potential correction prediction. The filter update submodule fuses the carbon potential correction prediction with the carbon potential observation value converted from the oxygen probe using the Nernst equation, outputting the transient carbon potential monitoring value. The table below shows a comparison between the carbon potential prediction value before and after correction and the Kalman filter output (taking the normalized time point around 0.75 as an example).

[0065] Table 5 Comparison of predicted and monitored carbon potential values ​​before and after correction

[0066] After correction, the residual between the predicted carbon potential and the carbon potential observed by the oxygen probe decreased from about 0.08% before correction to less than 0.01%. The Kalman filter no longer needs to rely on large corrections of the observed values, and the normal cooperative relationship between prediction and observation is restored.

[0067] Data feedback module: After Furnace 3 completes this transition and returns to steady state (steady-state convergence carbon potential 0.85%), the data writing submodule will store the complete transition data and identification results. , The new furnace record R043 is written into the multi-furnace transition process big data set, updating the total number of valid furnace records to 40. The template update submodule re-executes normalized time alignment, anomaly removal, and weighted statistics on the 40 records according to time decay weights, generating an updated reference transition trajectory template. R043, as the latest furnace record, receives the maximum weight, causing the template to slightly shift towards the actual transition behavior of furnace 3. The furnace condition marking submodule will... , Stored as the current furnace condition flag value for furnace 3, for use in the initialization of kinetic integration in subsequent furnace cycles of furnace 3.

[0068] Throughout the process, data is generated from historical records of multiple furnaces, aggregated into a large dataset by the transition data acquisition module, statistically refined into a reference transition trajectory template by the reference trajectory generation module (Table 2), and then the carbon potential prediction trajectory of furnace 3 is generated by the kinetic prediction module (Table 4). The deviation analysis module extracts trend indicators from the point-by-point difference between the predicted trajectory and the template. , The parameter identification module uses the template as a constraint target to inversely derive the rate constant scaling factor. , The calibration and monitoring module drives the Kalman filter to output accurate carbon potential monitoring values ​​using the calibrated parameters (Table 5). Finally, the data feedback module writes the transition results back to the big data set and updates the template and furnace condition markers, completing one closed-loop iteration. The output data of each module has a clear receiving and usage relationship in subsequent modules, ensuring the continuity and logical consistency of data flow.

[0069] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A carbon potential monitoring system for a carburizing heat treatment furnace based on big data analysis, characterized in that, include: The transition data acquisition module acquires historical transition data of multiple furnaces that have completed stage switching, forming a large dataset of the transition process. The reference trajectory generation module performs normalized time alignment and weighted statistics on the carbon potential curves of each furnace to generate a reference transition trajectory template and statistical confidence band. The kinetic prediction module substitutes the rate constant parameter into the reaction kinetic rate equations and performs integration to generate the target furnace carbon potential prediction trajectory. The deviation analysis module compares the carbon potential prediction trajectory with the reference transition trajectory template point by point and calculates trend indicators. The parameter identification module uses a reference transition trajectory template as a constraint target identification rate constant scaling factor when the trend indicator exceeds the threshold. The calibration monitoring module updates the kinetic parameters with the identified rate constant scaling factor, performs calibration integration and filtering update, and outputs the carbon potential monitoring value. The data feedback module writes the data into the big data set and updates the reference transition trajectory template after the transition is completed.

2. The carbon potential monitoring system for carburizing heat treatment furnace based on big data analysis according to claim 1, characterized in that, The reference trajectory generation module includes: The time alignment submodule maps the actual transition duration of the carbon potential time series curves for each furnace transition period to the normalized time axis. ,in The corresponding step jump time, The determination time of the return to steady state of the furnace batch is corresponding to the time of the furnace batch, and the equally spaced carbon potential sequence is obtained on the normalized time axis by linear interpolation algorithm; The anomaly removal submodule calculates the integral deviation between the carbon potential curve and the population mean curve of each furnace at all normalized time points, uses the ratio of the integral deviation to the population standard deviation as the deviation index, and removes the data of furnaces whose deviation exceeds the population statistical threshold. The statistical calculation submodule calculates the carbon potential weighted mean and weighted standard deviation at each normalized time point for the valid furnace data after removing anomalies. The weighted mean at each normalized time point constitutes the reference transition trajectory template, and the upper and lower bounds corresponding to the weighted standard deviation constitute the statistical confidence band.

3. The carbon potential monitoring system for carburizing heat treatment furnace based on big data analysis according to claim 2, characterized in that, In the statistical calculation submodule, the weight of each furnace is assigned in a time decay manner according to the order of their completion time, with the furnace closer to the current time having a larger weight.

4. The carbon potential monitoring system for carburizing heat treatment furnace based on big data analysis according to claim 1, characterized in that, The dynamics prediction module includes: The real-time data acquisition submodule acquires the measured values ​​of the oxygen probe electromotive force, furnace temperature, and enriched gas flow step change markers of the target furnace. It reads the output value of the Kalman filter in the steady-state stage before the step change occurs as the initial component state. If the target furnace has furnace condition marker values ​​written in the previous furnace, the current rate constant parameter is initialized with the rate constant scaling factor recorded in the furnace condition marker value; otherwise, the standard rate constant in the literature is used as the initial parameter. The kinetic integration submodule takes the initial component state as the starting condition, substitutes the current rate constant parameter into the reaction kinetic rate equation system containing the forward and reverse rate terms of the Boudouard reaction and the water-gas reaction, performs time integration through the fourth-order Runge-Kutta method, and calculates the kinetic prediction value of the concentration of each gas component at each time step. The carbon potential conversion submodule converts the predicted gas component concentration kinetics values ​​at each time step into the predicted carbon potential trajectory according to the thermodynamic equilibrium conversion relationship between carbon activity and carbon potential, and performs time mapping according to the same normalized time axis as the reference transition trajectory template.

5. The carbon potential monitoring system for carburizing heat treatment furnace based on big data analysis according to claim 1, characterized in that, The trend indicators calculated by the deviation analysis module include monotonic growth indicators. and cumulative deviation ; The deviation analysis module generates a deviation sequence. ,in For the first The difference between the predicted carbon potential value of the carbon potential trajectory at each normalized time point and the carbon potential value of the reference transition trajectory template. This represents the number of normalized time points that have already been experienced. The monotonic growth indicator The calculation method is as follows: calculate the difference between adjacent values. Count the number of positive values. and the number of negative values ,according to calculate; The cumulative deviation according to calculate.

6. The carbon potential monitoring system for carburizing heat treatment furnace based on big data analysis according to claim 5, characterized in that, The parameter identification module is in the monotonically increasing index and the cumulative deviation Execution is triggered when all parameters exceed their respective confidence thresholds; the parameter identification module scales the rate constant of the Boudouard reaction by a factor. Scaling factor for the rate constant of the reaction with water and gas Let the parameter be set as the one to be identified, and the rate constant scaling factor be defined as the ratio of the actual effective rate constant to the standard rate constant in the literature; the objective function is minimized using a nonlinear least squares optimization method: ; in For the weighted sum of squared residuals, For and For the parameter in the first Normalization time point The predicted carbon potential at that location, In order to refer to the transition trajectory template in The carbon potential at that point, Take the statistical confidence band at The reciprocal of the width.

7. The carbon potential monitoring system for carburizing heat treatment furnace based on big data analysis according to claim 1, characterized in that, The calibration monitoring module includes: The parameter update submodule writes the rate constant scaling factor into the reaction kinetic rate equation set of the target furnace and updates the forward and reverse rate constants of the Boudouard reaction and the water-gas reaction. The calibration integral submodule starts with the concentration values ​​of each component in the current Kalman filter state estimate as the initial condition, performs a fourth-order Runge-Kutta integral with the updated rate constant parameter, calculates the predicted values ​​of each component concentration in the next time step, and converts them into carbon potential calibration prediction values. The filter update submodule takes the carbon potential correction prediction value as the output of the Kalman filter prediction step, combines it with the carbon potential observation value obtained by converting the measured value of the oxygen probe electromotive force through the Nernst equation, performs the Kalman filter update step, and outputs the fused carbon potential monitoring value.

8. The carbon potential monitoring system for carburizing heat treatment furnace based on big data analysis according to claim 1, characterized in that, The data feedback module includes: The data writing submodule writes the complete transition trajectory data of the target furnace and the rate constant scaling factor into the transition process big data set as a new furnace record. The template update submodule uses the large dataset of the transition process after the new record is written as the data source, and re-executes normalized time alignment, anomaly removal and weighted statistical calculation according to the time decay weight to generate the updated reference transition trajectory template and statistical confidence band. The furnace condition marking submodule stores the rate constant scaling factor obtained from this identification as the current furnace condition marking value of the target furnace, which is then read by the kinetic prediction module of subsequent furnace cycles for rate constant initialization.

9. The carbon potential monitoring system for carburizing heat treatment furnace based on big data analysis according to claim 4, characterized in that, The initial component states include the initial values ​​of carbon monoxide concentration, carbon dioxide concentration, water vapor partial pressure, hydrogen concentration, and methane concentration; the data collected by the transition data acquisition module for each furnace includes the carbon potential tracking time series curve, furnace temperature time series, enriched gas flow step parameter, and final steady-state convergent carbon potential value output by Kalman filtering during the transition period.

10. A method for monitoring the carbon potential of a carburizing heat treatment furnace based on big data analysis, used to execute the carbon potential monitoring system for a carburizing heat treatment furnace based on big data analysis as described in any one of claims 1 to 9, characterized in that, include: Step 1: Obtain historical transition data of multiple furnaces that have completed stage switching through the transition data acquisition module to form a large dataset of the transition process; Step 2: Normalize the time alignment and weighted statistics of the carbon potential curves of each furnace by using the reference trajectory generation module to generate a reference transition trajectory template and statistical confidence band; Step 3: Substitute the rate constant parameter into the reaction kinetic rate equations through the kinetic prediction module to perform integration and generate the target furnace carbon potential prediction trajectory. Step 4: The carbon potential prediction trajectory is compared point by point with the reference transition trajectory template through the deviation analysis module to calculate the trend index. Step 5: When the trend indicator exceeds the threshold, the parameter identification module identifies the rate constant scaling factor by using the reference transition trajectory template as the constraint target. Step 6: Update the kinetic parameters with the identified rate constant scaling factor through the calibration monitoring module, perform calibration integration and filtering update, and output the carbon potential monitoring value; Step 7: After the transition is completed, the data is written into the big dataset and the reference transition trajectory template is updated through the data feedback module.

Images