Air-fuel ratio optimization control method based on machine vision recognition

Abstract

The invention discloses an air-fuel ratio optimization control method based on machine vision recognition, which comprises the following steps of: establishing a temperature prediction model through historical data, and recognizing the flame temperature in a furnace and the state of a sintering material surface in real time by utilizing a CCD (Charge Coupled Device) camera; the core of the system is that dual-mode control is adopted: in a basic mode, a valve is adjusted according to a deviation ratio between a target temperature and a predicted temperature; and when the system is stable, an air-fuel ratio optimization program is started, and the optimal air-fuel ratio is dynamically optimized and locked by finely adjusting an air valve and comprehensively evaluating combustion efficiency and charge level image features. Closed-loop precise control and online self-optimization of the combustion process are achieved, and the combustion efficiency, the system stability and the automation level are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent combustion control optimization, and more specifically to an air-fuel ratio optimization control method based on machine vision recognition. Background Technology

[0002] As the core equipment of the sintering production line, the sintering ignition furnace in the sintering raw material plant has a decisive impact on the quality of sinter and the technical and economic indicators of the entire sintering process. However, in actual operation, it mainly relies on manual adjustment of the opening of electric regulating valves on the blast furnace gas, coke oven gas, and combustion air pipelines based on the temperature signal detected by the ignition furnace thermocouple and the estimated thickness of the material layer to achieve mixed combustion. This traditional manual control mode has significant inherent defects, mainly reflected in the lag of adjustment and the difficulty in continuously maintaining the optimal air-fuel ratio, resulting in large fluctuations in ignition temperature, increased gas consumption, and ultimately affecting the yield and overall energy efficiency of sinter, leading to "over-burning" or "under-burning". When over-burning occurs, the surface sinter is over-melted, the FeO content increases, the reducibility deteriorates, and the permeability of the material layer decreases; when under-burning occurs, the surface sinter cake has poor strength, or even fails to form a sinter cake, leading to an increase in the amount of returned ore. This not only seriously affects the drum strength and yield of sinter, but also causes significant gas waste. Maintaining the optimal air-fuel ratio (the volume ratio of air to gas) is key to ensuring complete combustion of fuel and improving thermal efficiency.

[0003] Therefore, there is an urgent need for a new method that can precisely and automatically control the sintering system and maintain the optimal fuel-air ratio, thereby reducing energy waste and further improving the quality of sintered ore. Summary of the Invention

[0004] Purpose of the invention: To address the problems mentioned in the background art, this invention discloses an air-fuel ratio optimization control method based on machine vision recognition. By acquiring data through temperature prediction and measurement models, the method assists the sintering system in automatically adjusting the valve opening. The method adaptively optimizes and locks the air-fuel ratio using a progressive air-fuel ratio search algorithm, achieving a fine-tuning effect superior to manual calculation. This improves sintering quality and reduces energy costs while automating the adjustment.

[0005] Technical solution:

[0006] An air-fuel ratio optimization control method based on machine vision recognition, the method comprising the following steps:

[0007] S1: Based on historical data of the sintering system, the influence coefficients are estimated using the least squares method. An initial temperature prediction model is constructed using non-intercept multiple linear regression to obtain the predicted temperature. ;

[0008] S2: Capture RGB frame images from video of the flame inside the sintering furnace and preprocess them. Pair these images with real-time thermocouple temperatures to form a calibration dataset. Use this dataset to establish a linear relationship between RGB values ​​and radiation intensity, construct a colorimetric temperature measurement model, and obtain the target temperature. ;

[0009] S3: Convert RGB frame images to HSV color space, set H, S, V threshold ranges, and determine the over- or over-melted state of the material surface in real time.

[0010] S4: Calculate the target temperature With predicted temperature The ratio is used as the temperature deviation ratio (Ratio), and the valve opening of the actuator is adjusted sequentially according to the ratio and the material level.

[0011] S5: Based on the progressive air-fuel ratio search algorithm, the air-fuel ratio is adaptively optimized, the current comprehensive combustion efficiency index η is calculated and optimized to the maximum, and the current optimal air-fuel ratio is locked.

[0012] Furthermore, the historical data mentioned in S1 includes gas flow rate, air flow rate, opening degree, and pressure. The initial temperature prediction model formula is as follows:

[0013]

[0014] in, G is the predicted temperature; A is the gas flow rate; α is the gas flow rate; β is the air flow rate; α is the gas flow rate coefficient; β is the air flow rate coefficient; the coefficients α and β are estimated using the least squares method based on historical data.

[0015] Furthermore, the RGB frame image preprocessing process described in S2 includes: extracting the RGB image of each frame, using Gaussian filtering to eliminate image noise, and normalizing the image pixel values ​​to the [0,1] interval.

[0016] Furthermore, the colorimetric temperature measurement model described in S2 is specifically constructed as follows:

[0017] In the preprocessed image, a pixel region corresponding to the thermocouple measurement point is selected, and the average values ​​of the R, G, and B components of all pixels within that region are calculated to obtain a set of RGB mean values ​​(R, G, B) representing that temperature point. avg G avg B avg ), and combine data collected at multiple temperature points (R) avg G avg B avg → This forms the calibration dataset;

[0018] Establish RGB values ​​and radiation intensity using a calibration dataset. , The linear relationship between radiation intensity and RGB value satisfies:

[0019]

[0020] in , , The proportionality coefficient is obtained through thermocouple calibration;

[0021] Based on the colorimetric method for temperature calculation, the temperature calculation formula is expressed as follows:

[0022]

[0023] A, B, and C are model parameters, obtained by fitting a calibration dataset.

[0024] Furthermore, the adjustment of the sintering furnace valve opening degree according to the ratio and material level conditions as described in S4 is specifically as follows:

[0025] If Ratio > 1, meaning the target temperature is higher than the predicted value, increase the gas and air valves proportionally; if Ratio < 1, meaning the target temperature is lower than the predicted value, decrease the valves proportionally.

[0026]

[0027] Where Kcurrent is the actual valve opening degree, and Knew is the calculated valve opening degree.

[0028] The sintering system sends commands to the actuators to perform valve regulation operations, changing the supply of gas and air, and observing and evaluating the ignition status of the material surface:

[0029] If the material surface is too raw, increase the gas and air valves proportionally; if the material surface is too molten, decrease the valves proportionally until the material surface is within acceptable limits.

[0030] Furthermore, the activation condition for the adaptively optimized air-fuel ratio described in S5 is:

[0031] The sintering system continuously monitored that the material surface was in good ignition condition and the flame coverage was uniform and stable; the thermocouple measured that the temperature fluctuation rate inside the furnace was lower than the temperature threshold.

[0032] Furthermore, the adaptive optimization of the air-fuel ratio based on the progressive air-fuel ratio search algorithm described in S5 is as follows:

[0033] The sintering system increases the opening parameter V_air of the air valve in fixed steps, waits for the preset process response time, and allows the combustion state to reach a new equilibrium.

[0034] After the response time ends, execute S3 to acquire the material surface image features and simultaneously read the thermocouple temperature value to calculate the current comprehensive combustion efficiency index η(k):

[0035]

[0036] Where F_image(I_k) is the combustion state function, T_thermocouple(k) is the current thermocouple temperature, and T_target is the target temperature. and These are weighting coefficients;

[0037] After the k-th adjustment, the calculated combustion efficiency satisfies η(k)>η(k-1), indicating an improvement in efficiency. The system will then automatically execute the next adjustment, i.e., k+1.

[0038] When the S3 image recognition result determines that the material surface is too raw or the calculation shows that η(k)≤η(k-1), the sintering system sets the air valve opening degree to the previous value, that is, V_air(optimal)=V_air(k-1), and locks the air-fuel ratio as the optimal parameter for the current production condition.

[0039] Furthermore, after the optimal air-fuel ratio is locked as described in S5, the corresponding operating parameters are recorded in the database. When the sintering system performs air-fuel ratio optimization again, if the operating conditions are the same, the corresponding air-fuel ratio can be directly adopted.

[0040] Beneficial effects:

[0041] 1. This invention constructs a non-intercept multiple linear regression model using historical data, which can quickly predict the impact of gas and air flow on temperature, achieving more accurate temperature prediction, providing a data basis for subsequent valve regulation, and further reducing the error of the optimal fuel-air ratio data.

[0042] 2. This invention utilizes the correspondence between thermocouples and images to form a calibration dataset. Based on the image, it measures the temperature of the sintering furnace, overcoming the shortcomings of conventional thermocouple detection, such as easy damage and strong contact requirements. This enables non-contact and wide-range temperature detection, reducing equipment costs while maintaining stable temperature data output, thus improving the robustness of the invention.

[0043] 3. This invention analyzes the material surface state through the HSV color space, which has strong resistance to light interference and can accurately identify the material surface state such as "over-raw" and "over-melted", providing a reliable basis for adjustment, reducing the error of manual visual judgment, and improving the consistency of sintered products.

[0044] 4. This invention automatically adjusts the valve opening based on the temperature deviation ratio, changes the supply of gas and air, and combines a delayed feedback mechanism to robustly and gradually approach the target stable and qualified temperature field, thereby improving the adjustment efficiency while avoiding system oscillation and maintaining the stability of the sintering system during the optimization process.

[0045] 5. Based on the stabilized sintering system, this invention further optimizes and locks the optimal air-fuel ratio using a progressive air-fuel ratio search algorithm, thereby maximizing combustion efficiency, further improving sintering quality and reducing energy costs. Attached Figure Description

[0046] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0047] 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.

[0048] like Figure 1 As shown, this invention discloses an air-fuel ratio optimization control method based on machine vision recognition. The method steps are as follows:

[0049] Step 1. Historical Data Regression

[0050] By analyzing historical manual data, and based on the air-fuel ratio between blast furnace gas and coke oven gas, a more reasonable value is found for use when adjusting valves in the sintering system.

[0051] Based on historical data (gas flow rate, air flow rate, opening degree, pressure, etc.) and the target (temperature prediction), the regression model employs a non-intercept multiple linear regression because the combustion temperature should theoretically be zero when both fuel and air flow rates are zero. The formula is as follows:

[0052]

[0053] in: G is the predicted temperature; G is the gas flow rate (independent variable); A is the air flow rate; α is the coefficient of influence of gas on the calorific value of combustion; β is the coefficient of influence of air on the calorific value of combustion.

[0054] The coefficients α and β were estimated using the least squares method based on historical data.

[0055] Suppose there are n sets of historical data samples, each set containing gas flow rate Gi, air flow rate Ai, and actual measured temperature Ti (i=1,2,…,n).

[0056] Design matrix X: n×2 in size, containing gas and air flow data.

[0057]

[0058] Temperature vector y: of size n×1, containing actual temperature values:

[0059]

[0060] The coefficient vector θ: θ = [α, β]T, can be solved using the normal equation:

[0061]

[0062] Once α and β are derived from historical data, the formula becomes fixed as follows:

[0063]

[0064] Step 2. Flame Temperature Identification

[0065] Because thermocouples are frequently damaged at the production site, it is essential to use a color CCD camera to continuously capture video of the flame inside the sintering furnace for temperature identification. In practice, each frame of the RGB image can be extracted, Gaussian filtering can be used to eliminate image noise, and the image pixel values ​​can be normalized to the [0,1] range.

[0066] For the preprocessed image, the R, G, and B component values ​​are acquired pixel by pixel. Under stable operating conditions, the temperature value T fed back by the thermocouple and the corresponding RGB image of the flame are recorded simultaneously. To ensure calibration accuracy, data acquisition needs to be performed under multiple different temperature conditions to cover a wider temperature range.

[0067] In the image, select a pixel region (e.g., a small rectangular area) corresponding to the location of the thermocouple measurement point. Calculate the average of the R, G, and B components of all pixels within this region to obtain a set of RGB mean values ​​(Ravg, Gavg, Bavg) representing that temperature point. This operation aims to reduce the influence of random noise. Then, combine the data collected at multiple temperature points into (Ravg, Gavg, Bavg)... This forms the calibration dataset.

[0068] Establish RGB values ​​and radiation intensity using a calibration dataset. , The linear relationship between the R, G, and B channels of the CCD sensor and the representative wavelengths λR=700nm, λG=546.1nm, and λB=435.8nm is as follows: The radiation intensity and the RGB values ​​satisfy the following:

[0069]

[0070] Where cR, cG, and cB are proportionality coefficients obtained through thermocouple calibration.

[0071] After this step is completed, based on the temperature calculation principle of colorimetry, the temperature calculation formula can be expressed as:

[0072]

[0073] in, The target temperature is A, B, and C, which are model parameters obtained by fitting a calibration dataset.

[0074] Step 3. Color identification of sintered material surface

[0075] The image preprocessing in this step is similar to step 2. After preprocessing, the image is converted from the default BGR color space to the HSV color space because the HSV space is more robust to changes in lighting. The conversion from BGR to HSV can be achieved using OpenCV's `cv2.cvtColor()` function. After the conversion, the threshold range for the HSV space can be manually set. For example, the threshold range for overly raw material can be set as follows: Hue (H) range 25-40, Saturation (S) ≥ 100, and Lightness (V) ≥ 70. Reaching this standard indicates that the flame temperature needs to be increased.

[0076] Step 4. Valve opening adjustment amount and execution

[0077] The system reads the target temperature input by the operator and calculates the temperature deviation ratio based on the model regressed from equation (5):

[0078]

[0079] After the calculation is completed, adjust the valve opening according to the ratio.

[0080] If Ratio > 1 (target temperature is higher than predicted value), increase the gas and air valves proportionally. If Ratio < 1 (target temperature is lower than predicted value), decrease the valves proportionally.

[0081]

[0082] Where Kcurrent is the actual valve opening degree, and Knew is the calculated valve opening degree.

[0083] After the calculation is completed, the system sends a command to the actuator (such as an electric or pneumatic regulating valve) to perform valve regulation operation and change the supply of gas and air.

[0084] After the valve adjustment is executed, the ignition status of the material surface can be observed after a few minutes.

[0085] If the fuel surface is not ignited, increase the gas and air valves proportionally.

[0086] If the material surface is too molten, reduce the valve size proportionally.

[0087] Formula 9 can be applied directly here, but it should be noted that... Adjustments can be made based on the actual material level, but the valve adjustment should be limited to avoid excessive adjustment ranges that could cause system oscillations. This process can be repeated: after each adjustment, wait, assess the material level, calculate the new ratio, limit the range, and adjust again. Continue until the material level is within acceptable limits.

[0088] The core of this invention lies in its robust and gradual approach to the target. It effectively addresses the delayed characteristics of material levels through a closed-loop logic of "adjustment-waiting-evaluation-readjustment." Limiting the amplitude of the regulating valve is crucial to ensuring system stability.

[0089] Step 5. Adaptive air-fuel ratio optimization

[0090] In this step, after confirming successful ignition and a stable temperature field, the system initiates an adaptive air-fuel ratio optimization program. Its core function is to dynamically find and lock the optimal air-fuel ratio under the current operating conditions by fine-tuning the air valve opening, thereby maximizing combustion efficiency. The optimization process follows these instructions:

[0091] If the system continuously detects the following state for more than a set time (set to 5 minutes in this embodiment), it determines that the optimized startup conditions are met:

[0092] The material surface ignites well, and the flame coverage is uniform and stable.

[0093] Thermocouple measurements show that the temperature fluctuation rate inside the furnace is below the threshold δ (e.g., δ = ±2%), that is:

[0094]

[0095] Where T_settled is the stable temperature reference value, and ΔT is the temperature fluctuation amplitude within the time window.

[0096] Progressive air-fuel ratio search algorithm:

[0097] After optimization and startup, the system executes the following closed-loop search algorithm:

[0098] Parameter fine-tuning: The system increases the opening degree V_air of the air valve in fixed steps (e.g., ΔV=1%), that is:

[0099]

[0100] Where k is the optimization step number.

[0101] Status monitoring and waiting: After each adjustment, the system waits for a preset process response time (5 minutes) to allow the combustion state to reach a new equilibrium.

[0102] Combustion efficiency assessment:

[0103] After the waiting period, the system returns to step 3, acquires the surface image features using the sintering material color recognition model, and simultaneously reads the thermocouple temperature values. Then, it calculates the current comprehensive combustion efficiency index η(k):

[0104] Where: F_image(I_k) is a combustion state function quantified based on the image recognition results from step 3 (e.g., surface brightness, uniformity, etc.). T_thermocouple(k) is the current thermocouple temperature. T_target is the target temperature. and It is the weighting coefficient ( + =1), used to balance the contributions of image features and temperature field in efficiency evaluation.

[0105] Optimize termination logic and settings:

[0106] Further optimization conditions: If the calculated combustion efficiency satisfies η(k)>η(k-1) after the adjustment in step k, it indicates that the efficiency has improved, and the system will automatically execute the next adjustment (k+1).

[0107] Optimization Termination Conditions: The air-fuel ratio optimization process will be terminated immediately if the system detects any of the following conditions:

[0108] Material surface is too raw: Based on the image recognition results in step 3, it is determined that the material surface has the characteristic of being "over-raw".

[0109] Efficiency decrease: Calculation shows that η(k)≤η(k-1).

[0110] Once any termination condition is met, the system sets the air valve opening to the previous value, i.e., V_air(optimal) = V_air(k-1), and locks the air-fuel ratio as the optimal parameter for the current production condition.

[0111] Step 6. Production Data Storage

[0112] After step 5 is completed, the operating conditions corresponding to step 5 are recorded in the database. The next time air-fuel ratio optimization is performed, if the operating conditions are the same, the corresponding air-fuel ratio can be directly adopted.

[0113] The process of this invention is summarized as follows:

[0114] Step 1: Model Building

[0115] Based on historical data, a temperature prediction model was established by using non-intercept multiple linear regression to solve for the coefficients α and β of gas and air flow rates with respect to temperature. =α·G+β·A.

[0116] Step 2: Visual temperature measurement calibration

[0117] By using a CCD camera to photograph the flame and simultaneously measuring the temperature with thermocouples, the relationship between RGB values ​​and radiation intensity was calibrated under different operating conditions. A colorimetric temperature measurement model was then established to obtain the target temperature.

[0118] Step 3: Material Surface Status Identification

[0119] The flame image is converted to the HSV color space, and the threshold ranges of H, S, and V are set to determine the over- or over-melted state of the material surface in real time, providing a basis for control.

[0120] Step 4: Basic Temperature Control

[0121] The system reads the target temperature, calculates the ratio between the target temperature and the predicted temperature, and adjusts the opening of the gas and air valves synchronously according to this ratio to achieve preliminary temperature control.

[0122] Step 5: Air-fuel ratio optimization

[0123] After the system stabilizes, the air valve is finely adjusted, and the combustion efficiency η is calculated by combining the material surface image characteristics and temperature. The optimization is carried out in cycles until η is maximized, and the optimal air-fuel ratio is locked.

[0124] Step 6: Parameter Recording and Self-Learning

[0125] The optimized air-fuel ratio and corresponding operating parameters are stored in the database. When encountering similar operating conditions later, these parameters can be prioritized, enabling the system to learn and optimize itself.

[0126] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for air-fuel ratio optimal control based on machine vision recognition, characterized by, The method comprises the following steps: S1: Based on the sintering system historical data, the influence coefficient is estimated by least square method, the initial temperature prediction model is constructed by using multiple linear regression without intercept, and the predicted temperature is obtained ; S2: Extract RGB frame images from the video of the sintering furnace flame and preprocess, pair with real-time temperature of thermocouple to form calibration dataset, use the calibration dataset to establish linear relationship between RGB value and radiation intensity, construct colorimetric temperature measurement model, and obtain target temperature ; S3: Convert the RGB frame image to the HSV color space, set the H, S, and V threshold range, and judge the over-burning or over-melting state of the material surface in real time; S4: calculate target temperature with predicted temperature the ratio as a temperature deviation ratio, and adjust the valve opening degree of the actuator according to the ratio and the material surface condition in turn; S5: Based on the progressive air-fuel ratio search algorithm, the air-fuel ratio is adaptively optimized, and the current comprehensive combustion efficiency index η is calculated and optimized to be maximum, and the current optimal air-fuel ratio is locked.

2. The method according to claim 1, characterized by, The historical data in S1 include coal gas flow, air flow, opening degree, pressure, and initial temperature prediction model formula as follows: wherein, is the predicted temperature; G is the gas flow rate; A is the air flow rate; a is the influence coefficient of gas on the combustion heat value; β is the influence coefficient of air on the combustion heat value; the coefficients a and β are estimated by least squares method from historical data.

3. The method according to claim 1, characterized by, The RGB frame image preprocessing process in S2 comprises the following steps: extracting the RGB image of each frame, removing image noise by using Gaussian filtering, and normalizing the image pixel value to the [0, 1] interval.

4. The method according to claim 3, characterized by, The colorimetric temperature calculation model is specifically constructed as follows: In the pre-processed image, a pixel region corresponding to the position of the thermocouple measuring point is selected, the average values of R, G, B components of all pixels in the region are calculated to obtain a set of RGB average values (R avg ,G avg ,B avg ) representing the temperature point, and the data pairs (R avg ,G avg ,B avg )→ acquired at multiple temperature points are combined to form a calibration data set. The linear relationship between the RGB values and the radiance intensity is established by a calibration data set , The radiance intensity and the RGB values satisfy: wherein , , is a proportionality factor obtained by calibrating the thermocouple; Based on the temperature calculation principle of the colorimetric method, the temperature calculation formula is represented as: Wherein, A, B, and C are model parameters, which are obtained by fitting the calibration data set.

5. The method according to claim 1, characterized by, The specific adjustment of the sintering furnace valve opening degree according to the Ratio and the material surface condition in S4 is as follows: If the Ratio is greater than 1, that is, the target temperature is higher than the predicted value, the gas and air valves are increased in proportion; if the Ratio is less than 1, that is, the target temperature is lower than the predicted value, the valves are reduced in proportion; Wherein, K current is the actual valve opening degree, and K new is the calculated valve opening degree. The sintering system sends a command to the actuator to execute the valve adjustment operation, changes the supply amount of gas and air, and observes and evaluates the ignition condition of the material surface: If the material surface is over-burning, the gas and air valves are increased in proportion; if the material surface is over-melting, the valves are reduced in proportion, until the material surface is qualified.

6. The method according to claim 1, characterized by, The starting condition of the adaptive optimization of the air-fuel ratio in S5 is as follows: The sintering system continuously monitors that the material surface ignition state is good, the flame coverage is uniform and stable, and the thermocouple measures that the temperature fluctuation rate in the furnace is lower than the temperature threshold.

7. The method according to claim 6, wherein The adaptive optimization of the air-fuel ratio based on the progressive air-fuel ratio search algorithm in S5 is specifically as follows: The sintering system increases the opening degree parameter V air of the air valve by a fixed step, waits for a preset process response time, and makes the combustion state reach a new balance; After the response time ends, the material surface image features are obtained in S3, and the thermocouple temperature value is read synchronously, and the current comprehensive combustion efficiency index η(k) is calculated: where F_image(I_k) is a combustion state function, T_thermocouple(k) is the current thermocouple temperature, T_target is the target temperature, and are weight coefficients; After the kth adjustment, the calculated combustion efficiency satisfies η(k)>η(k-1), indicating that the efficiency is improved, and the system will automatically execute the next adjustment, that is, k+1; When the S3 image recognition result judges that the material surface is over-burning or η(k)≤η(k-1) is calculated, the sintering system sets the air valve opening degree to the value of the previous time, that is, V_air(optimal)=V_air(k-1), and locks the air-fuel ratio as the best parameter of the current production condition.

8. The method according to claim 1, characterized by, After the optimal air-fuel ratio is locked in S5, the corresponding working condition parameters are recorded in the database, and the sintering system can directly use the corresponding air-fuel ratio when the air-fuel ratio is optimized next time, such as the same working condition.

Images