A metal quantitative detection method and device based on pressure feedback

Abstract

The application discloses a metal quantitative detection method based on pressure feedback, comprising the following steps: obtaining the pressure curve of the metal to be detected in unit time, extracting the upper envelope and the lower envelope of the pressure curve through Hilbert-Huang transform, determining the indicator diagram of the pressure curve at the current moment, and performing normalization processing on the indicator diagram; extracting the geometric features of the indicator diagram, extracting the moment features of the indicator diagram, extracting the statistical features of the indicator diagram, performing set processing on the statistical features, the geometric features and the moment features as a feature set, performing polynomial feature analysis on the feature set, inputting the result after the polynomial feature analysis into a regression prediction model, determining the soup output of the metal to be detected at the current moment in unit time, and comparing the accumulated soup output with a set value to quantitatively detect the metal. The application realizes accurate and rapid prediction of the soup output of the metal, simplifies the control process, and reduces the control cost.

Description

Technical Field

[0001] This invention belongs to the field of automation equipment technology, specifically relating to a method and device for quantitative metal detection based on pressure feedback. Background Technology

[0002] Patent CN202122300782.9 discloses an aluminum liquid metering mechanism, assembled on a holding furnace. The metering mechanism uses a weight detection device to detect the weight of the aluminum liquid, and simultaneously uses a flow guiding mechanism to discharge the detected aluminum liquid from the metering mechanism. During the processing of the aluminum liquid, a precision weighing sensor accurately controls the weight of the discharged liquid. Because the weighing sensor is used as a reference, the accuracy can reach within + / -1%, ensuring the stability of the die-casting process and thus effectively improving product quality.

[0003] Patent CN202021446622.4 discloses a quantitative aluminum liquid discharge device for casting in aluminum rod production. It is equipped with a laser liquid level sensor and a PLC controller to detect the liquid level of aluminum liquid inside the device. At the same time, the PLC controller can perform quantitative discharge of liquid from the device, which increases the practicality of the device and reduces its limitations.

[0004] Patent CN202122787478.1 discloses a gravity-feedback aluminum liquid quantitative holding furnace device. The gravity liquid level feedback system, through a suitable installation position, can detect parameters related to the quality of the aluminum liquid. These parameters, after conversion, can provide feedback on the current liquid level. A high-precision proportional valve in the pressure control system provides precise pressure control based on the current liquid level, pressing out a fixed amount of aluminum liquid. Based on the detected amount of pressed out aluminum liquid, the system automatically makes minor adjustments to the data detected by the gravity liquid level detection system, achieving fully closed-loop control.

[0005] Currently, there are three main known methods for precise metal metering control: 1. Accurately controlling the weight of the discharged material using precision weighing sensors; 2. Using a metering pump to meter the molten metal; 3. Installing a laser level sensor inside the furnace in conjunction with a PLC controller to complete metal metering. All three methods require high-precision, high-temperature resistant sensors as core components for data feedback. Although this data is more intuitive than pressure feedback, the high-temperature environment of molten metal makes these sensors very expensive, thus drastically increasing the manufacturing cost of the metering furnace.

[0006] While existing technologies use precise weight feedback sensors or level sensors to directly measure the dispensing volume of molten metal, these sensors also increase the manufacturing and control costs of the dispensing equipment. For example, weight feedback sensors lack real-time performance; if the dispensing flow rate is too high, delayed gravity feedback can lead to excessive dispensing volume and insufficient accuracy. Reducing the dispensing flow rate to accommodate the weight sensor, while maintaining accuracy, sacrifices efficiency. To balance accuracy and efficiency, a gravity-pressure interactive control method is often chosen, which further increases the control complexity.

[0007] Therefore, in response to the above problems, how to achieve accurate and rapid prediction of the amount of metal produced, simplify the control process, and reduce control costs are technical issues that need to be addressed. Summary of the Invention

[0008] The purpose of this invention is to provide a method for quantitative detection of metals based on pressure feedback, comprising:

[0009] The pressure curve of the metal to be tested is obtained within a unit time, wherein the pressure signal of the metal to be tested is acquired by a pressure sensor at the top of the quantitative furnace to obtain the pressure curve.

[0010] The upper and lower envelopes of the pressure curve are extracted by Hilbert-Huang transform. Based on the upper and lower envelopes, the dynamometer diagram of the pressure curve at the current moment is determined, and the dynamometer diagram is normalized.

[0011] Extract the geometric features of the normalized indicator diagram. The geometric features include at least the edge length, number of edges, coordinates of inflection points, number of inflection points, area, perimeter, projection length, and projection width of the indicator diagram.

[0012] Extract the moment features of the normalized dynamometer diagram, wherein the moment features include at least the center moment of the dynamometer diagram and the Hu moment of the dynamometer diagram;

[0013] The statistical features of the normalized indicator diagram are extracted. The statistical features include at least the Holt exponential smoothing feature, information entropy, crossover feature, sequence patch feature, block feature based on tiling window, autocorrelation feature, partial autocorrelation feature, sparsity, stability, periodicity, Kwiatkowski unit root test statistic and Phillips-Peren unit root test statistic.

[0014] The statistical features, geometric features, and moment features of the indicator diagram are combined and processed to form a feature set. Polynomial feature analysis is performed on the feature set, and the result of the polynomial feature analysis is input into the regression prediction model to determine the amount of the metal to be measured at the current time per unit time.

[0015] The cumulative soup output is compared with the set value to obtain the quantitative detection result of the metal to be tested. The cumulative soup output is the output accumulated from the start of soup output to the current time.

[0016] Optionally, the indicator diagram can be normalized, specifically as follows:

[0017] Extract the maximum and minimum pressure values ​​of the metal to be tested from the start of the dispensing process to the current time within a unit of time. Use the pressure value as the vertical axis and time as the horizontal axis to determine the calibration background of the indicator diagram. Draw an indicator diagram with coordinates on the calibration background, and use the indicator diagram with coordinates as the normalized indicator diagram.

[0018] Optionally, the geometric features of the indicator diagram can be determined using a contour detection algorithm.

[0019] Optionally, extracting the center moment of the indicator diagram includes:

[0020] The centroid coordinates (x0, y0) of the target region are determined by using the zeroth-order and first-order moments.

[0021] The centroid coordinates (x0, y0) are determined according to the following formula:

[0022]

[0023]

[0024] Among them, m 00 The zeroth-order origin moment is the pixel intensity at ((0,0)) of the normalized indicator map, m. 01 and m 10 Let m be the first-order moment at the origin. 01 is the pixel intensity of the normalized indicator map at (1,0), is the pixel intensity of the normalized indicator map at (0,1), x0 is the centroid of the normalized indicator map in the x-axis direction, and y0 is the centroid of the normalized indicator map in the y-axis direction.

[0025] Based on the centroid coordinates (x0, y0), construct the center moment of the indicator diagram;

[0026] The central moment of the indicator diagram is determined according to the following formula:

[0027]

[0028] Among them, u pqLet p and q be the center moments of the dynamometer image, p and q be the order of the dynamometer image, (x, y) be the coordinates of the current pixel position, (x0, y0) be the centroid coordinates, C be the height of the dynamometer image, and R be the width of the dynamometer image.

[0029] Optionally, extracting the indicator diagram Hu moment includes:

[0030] Seven sets of invariant moments are derived using the second-order and third-order central moments of the image. The second-order and third-order central moments are obtained by normalizing the central moments of the dynamometer diagram. The second-order and third-order central moments have scale invariance and translation invariance.

[0031] The indicator diagram Hu moment is determined according to the following formula:

[0032] Φ1=η 20 +η 02

[0033]

[0034] Φ3=(η 20 -3η 12 ) 2 +3(η 21 -η 03 ) 2

[0035] Φ4=(η 30 +η 12 ) 2 +(η 21 +η 03 ) 2

[0036] Φ5=(η 30 +3η 12 )(η 30 +η 12 )[(η 30 +η 12 ) 2 -3(η 21 +η 03 ) 2 ]

[0037] +(3η 21 -η 03 )(η 21 +η 03 )[3(η 30 +η 12 ) 2 -(η 21 +η 03 ) 2 ]

[0038] Φ6=(η 20 -η 02 )[(η 30 +η 12 ) 2 -(η 21 +η 03 ) 2 ]+4η 11 (η 30 +η 12 )(η 21 +η 03 )

[0039] Φ7=(3η 21 -η 03 )(η 3o +η 12 )[(η 30 +η 12 ) 2 -3(η 21 +η 03 ) 2 ]+]+(3η 12 -η 30 )(η 21 +η 03 )[3(η 30 +η 12 ) 2 -(η 21 +η 03 ) 2 ]

[0040] Where, η ij The normalized center moment is of order i+j. When it is the second-order gauge center moment, the value of i is 0≤i≤2, where i is a positive integer, and the value of j is 0≤j≤2, where j is a positive integer, and i+j=2. When it is the third-order gauge center moment, the value of i is 0≤i≤3, where i is a positive integer, and the value of j is 0≤j≤3, where j is a positive integer, and i+j=3. Φ1 to Φ7 are 7 sets of invariant moments, and the 7 sets of invariant moments are the Hu moments of the indicator diagram.

[0041] Optionally, extract statistical features of the indicator diagram, including:

[0042] Use the time series feature extraction toolkit tsfeatures to extract all time series feature values;

[0043] Pearson correlation is performed between each time series feature value and the label column. After removing features with an absolute Pearson correlation value less than 0.8, the statistical features of the indicator chart are determined. The label column is the actual liquid weight of the metal to be tested output per unit time.

[0044] Optionally, polynomial feature parsing is performed on the feature set, specifically as follows:

[0045] Square each feature value in the feature set to form a new feature set;

[0046] The new feature set is subjected to quadratic polynomial feature analysis.

[0047] Optionally, after performing quadratic polynomial feature analysis on the new feature set, the analysis results are input into the linear regression model to determine the amount of the metal to be tested in soup at the current time per unit time.

[0048] Optionally, the cumulative amount of soup output can be compared with a set value to quantify the metal to be tested, specifically:

[0049] If the cumulative soup volume equals the set value, then soup dispensing will stop;

[0050] If the cumulative soup volume is less than the set value, continue dispensing soup until the cumulative soup volume at the current time equals the set value.

[0051] Accordingly, the present invention also proposes a pressure feedback-based metal quantitative detection device, comprising:

[0052] A quantitative furnace, comprising a furnace body, an outlet level gauge, a compressed gas inlet, a compressed gas inlet valve, and a pressure relief port;

[0053] When compressed gas enters the furnace through the compressed gas inlet valve, the molten metal to be tested is squeezed out of the furnace to form a molten liquid.

[0054] When the liquid metal to be tested comes into contact with the outlet liquid level, the dispensing begins. When the dispensing volume reaches the quantitative requirement, the dispensing is terminated by releasing pressure through the pressure relief port.

[0055] Beneficial effects:

[0056] Compared to existing technologies, this new method reduces the computational and control workload by replacing multiple sensors with a single pressure sensor, saving on sensor hardware costs and reducing control complexity. It enables accurate and rapid prediction of the amount of metal produced, simplifies the control process, and lowers control costs. Attached Figure Description

[0057] Figure 1 This is a schematic flowchart of a pressure feedback-based quantitative metal detection method provided according to an embodiment of this application;

[0058] Figure 2 This is a schematic diagram of a quantitative furnace according to an embodiment of this application;

[0059] Figure 3This is another schematic flowchart of a pressure feedback-based quantitative metal detection method provided according to an embodiment of this application. Detailed Implementation

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0061] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0062] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0063] Example 1:

[0064] Terminology Explanation:

[0065] 1. Pressure feedback: Pressure signals collected in real time by the pressure sensor at the top of the quantitative furnace.

[0066] 2. Metal quantity: Whether the amount of metal used in the metal casting process matches the required amount.

[0067] like Figure 1 The diagram shows a flow chart of a pressure feedback-based quantitative metal detection method proposed in an embodiment of the present invention. The method includes:

[0068] S201, Obtain the pressure curve of the metal to be tested within a unit time, wherein the pressure signal of the metal to be tested is acquired by a pressure sensor at the top of the quantitative furnace to obtain the pressure curve.

[0069] Specifically, typically, a pressure point is taken every 4 mm, and a pressure signal feedback is performed every second, extracting 250 pressure points per second. The PLC controller acquires the pressure curve of the metal under test within a unit time (i.e., one second). The pressure signal of the metal under test is collected by a pressure sensor at the top of the quantitative furnace and sent to the PLC controller to obtain the pressure curve. The valve opening degree of the quantitative furnace limits the upper limit of the pressure curve on the indicator diagram, thus standardizing the pressure curve.

[0070] This invention enables rapid prediction of soup output using only the valve opening and pressure feedback of the quantitative furnace, simplifying the control process and reducing control and hardware costs.

[0071] It should be noted that the unit time can be adjusted according to the actual time situation, and the extracted pressure point can be adjusted according to the actual situation.

[0072] S202, extract the upper and lower envelopes of the pressure curve using Hilbert-Huang transform, determine the dynamometer diagram of the pressure curve at the current moment based on the upper and lower envelopes, and normalize the dynamometer diagram.

[0073] Specifically, the upper and lower envelopes of the pressure curve are extracted using the Hilbert-Huang transform, and an indicator diagram of the pressure curve is formed based on the upper and lower envelopes.

[0074] In order to normalize the indicator diagram, in a preferred embodiment of this solution, the indicator diagram is normalized as follows:

[0075] Extract the maximum and minimum pressure values ​​of the metal to be tested from the start of the dispensing process to the current time within a unit of time. Use the pressure value as the vertical axis and time as the horizontal axis to determine the calibration background of the indicator diagram. Draw an indicator diagram with coordinates on the calibration background, and use the indicator diagram with coordinates as the normalized indicator diagram.

[0076] Specifically, the dynamometer diagram is normalized by combining the valve opening value: the normalization method is to extract the maximum and minimum values ​​of the global metal pressure to be measured, and use the maximum and minimum values ​​as the vertical axis, where the minimum pressure is the origin of the coordinate system, and time is the horizontal axis. A unified calibration background is specified for the dynamometer diagram, and the dynamometer diagram is drawn on this background. This can reflect the differences in the size of the dynamometer diagram, and the dynamometer diagram can be drawn within a unified background size.

[0077] S203, extract the geometric features of the normalized indicator diagram, the geometric features including at least the side length of the indicator diagram, the number of sides of the indicator diagram, the coordinates of the inflection points of the indicator diagram, the number of inflection points of the indicator diagram, the area of ​​the indicator diagram, the perimeter of the indicator diagram, the projection length of the indicator diagram, and the projection width of the indicator diagram.

[0078] Specifically, the geometric features of the indicator diagram are extracted, including but not limited to: indicator diagram side length, indicator diagram side number, indicator diagram inflection point coordinates, indicator diagram inflection point number, indicator diagram area, indicator diagram perimeter, indicator diagram projection length, and indicator diagram projection width; among which the geometric features of the indicator diagram are obtained by a contour detection algorithm, which is an algorithm in OpenCV for processing structural analysis and shape description.

[0079] S204, extract the moment features of the normalized indicator diagram, the moment features including at least the center moment of the indicator diagram and the Hu moment of the indicator diagram.

[0080] Specifically, the moment features of the dynamometer card are extracted, including but not limited to: the central moment of the dynamometer card and the Hu moment of the dynamometer card. The process for calculating the central moment is as follows: From the zeroth-order and first-order origin moments, we can obtain the centroid coordinates of the target region: In the indicator diagram, a valid pixel is 1, and an invalid pixel is 0; the zero-order origin moment m 00 The sum of pixel values ​​in the indicator diagram represents the area of ​​the effective pixels; the first-order origin moment m 01 The sum of (pixel values ​​multiplied by their corresponding y-axis coordinates) is used to find the region of concentrated effective pixels on the y-axis; first-order origin moment m 10 The sum of (pixel values ​​multiplied by their corresponding x-axis coordinates) is used to find the region with the most effective pixels on the x-axis; m 10 / m 00 , which is the center point of the effective pixel concentration area along the x-axis, also known as the centroid of the x-axis of the dynamometer diagram; using m 01 / m 00 The center point of the effective pixel concentration area along the y-axis is the centroid of the y-axis of the indicator diagram; the target area is the indicator diagram. The center point of the effective pixel concentration area along the x-axis is the centroid of the x-axis of the indicator diagram; the center point of the effective pixel concentration area along the y-axis is the centroid of the y-axis of the indicator diagram; (x0, y0) is the centroid of the indicator diagram, which is also the center point of the effective pixel concentration area.

[0081] The central moments can be constructed using the obtained centroid coordinates:

[0082]

[0083] Among them, u pq Let p and q be the center moments of the dynamometer image, p and q be the order of the image, (x, y) be the pixel position coordinates, (x0, y0) be the centroid coordinates, C be the height of the dynamometer image, and R be the width of the dynamometer image.

[0084] Where p and q represent the order of the image, x and y represent the coordinates of the current pixel position, x0 and y0 represent the centroid coordinates, and C and R represent the height and width of the image, respectively.

[0085] The method for solving the Hu moment on the indicator diagram is as follows:

[0086] Seven sets of invariant moments are derived using the second-order and third-order normalized central moments of the image. The second-order and third-order normalized central moments are obtained by normalizing the central moments of the dynamometer image. These remain invariant during image translation, rotation, and scaling. The normalized central moments are the same as the scale-invariant central moments. Normalization ensures the central moments are scale-invariant. The central moments are translation-invariant, meaning the central moment characteristics remain unchanged regardless of the dynamometer image's translation. The normalized central moments, in addition to translation invariance, are scale-invariant, meaning scaling does not affect the normalized central moment characteristic values. The Hu moments derived from the second-order and third-order normalized central moments further possess rotation invariance, meaning their Hu moment characteristics remain unchanged regardless of image translation, scaling, or rotation.

[0087] The indicator diagram Hu moment is determined according to the following formula:

[0088] Φ1=η 20 +η 02

[0089]

[0090] Φ3=(η 20 -3η 12 ) 2 +3(η 21 -η 03 ) 2

[0091] Φ4=(η 30 +η 12 ) 2 +(η 21 +η 03 ) 2

[0092] Φ5=(η 30 +3η 12 )(η 30 +η 12 )[(η 30 +η 12 ) 2 -3(η 21 +η 03 ) 2 ]+(3η 21 -η 03 )(η 21 +η 03 )[3(η 30 +η 12 ) 2 -(η21 +η 03 ) 2 ]

[0093] Φ6=(η 20 -η 02 )[(η 30 +η 12 ) 2 -(η 21 +η 03 ) 2 ]+4η 11 (η 30 +η 12 )(η 21 +η 03 )

[0094] Φ7=(3η 21 -η 03 )(η 30 +η 12 )[(η 30 +η 12 ) 2 -3(η 21 +η 03 ) 2 ]+]+(3η 12 -η 30 )(η 21 +η 03 )[3(η 30 +η 12 ) 2 -(η 21 +η 03 ) 2 ]

[0095] Where, η ij The normalized center distance is of order i+j, and Φ1 to Φ7 are seven sets of invariant moments (these seven sets of invariant moments are the Hu moments of the indicator diagram). η ij The range of values ​​for is as follows: 2nd moment: i ∈ [0,2], j ∈ [0,2], i+j=2; 3rd moment: i ∈ [0,3], j ∈ [0,3], i+j=3. Both i and are positive integers.

[0096] S205, extract the statistical features of the normalized indicator diagram, the statistical features including at least the Holt exponential smoothing feature, information entropy, crossover feature, sequence patch feature, block feature based on tiling window, autocorrelation feature, partial autocorrelation feature, sparsity, stability, periodicity, Kwiatkowski unit root test statistic and Phillips-Peren unit root test statistic.

[0097] Specifically, the statistical features of the indicator plot are extracted, including but not limited to: Holt exponential smoothing features, information entropy, crossover features, sequence patchy features, blocky features based on tiling windows, autocorrelation features, partial autocorrelation features, sparsity, stability, periodicity, Kwiatkowski unit root test statistic, Phillips-Peren unit root test statistic, etc. The feature extraction process is as follows: 1. Use the time series feature extraction toolkit tsfeatures to extract all time series feature values; the "all" refers to all feature values ​​that have not been filtered for correlation > 0.8, including: autocorrelation... The features include numerical characteristics, heterogeneity characteristics, sequence length, autoregressive conditional heteroscedasticity statistical characteristics, Holt exponential smoothing parameters, sparsity characteristics, information entropy characteristics, Hearst exponential characteristics, stability characteristics, crossover characteristics, triple exponential smoothing parameters, trend and periodic decomposition characteristics, interval characteristics, Kwiatkowski unit root test statistical characteristics, Phillips-Peren unit root test statistical characteristics, patchy characteristics, time series characteristics based on tiling windows, frequency domain characteristics, nonlinear characteristics, power transform characteristics, and partial autocorrelation coefficient characteristics, etc.; the time series refers to a sequence composed of single-point values ​​collected continuously over time. 2. Perform Pearson correlation analysis on each feature sequence and the label column, excluding features with an absolute correlation value less than 0.8. The features remaining after the above steps are the features mentioned above.

[0098] S206, the statistical features, geometric features, and moment features of the dynamometer are combined and processed into a feature set. The feature set is then subjected to polynomial feature analysis. The result of the polynomial feature analysis is input into the regression prediction model to determine the amount of the metal to be measured at the current time per unit time.

[0099] In order to perform polynomial feature parsing on the feature set, in a preferred embodiment of this scheme, polynomial feature parsing is performed on the feature set, specifically as follows:

[0100] Square each feature value in the feature set to form a new feature set;

[0101] The new feature set is subjected to quadratic polynomial feature analysis.

[0102] Specifically, polynomial feature analysis is performed on the feature set. Polynomial feature analysis adds the power of each feature as a new feature, forming an expanded feature set (i.e., a new feature set). This invention uses second-order polynomial feature analysis, which means that in addition to the original features, the square of each value of the original features is also taken to form a new feature value. The purpose is to assume the existence of nonlinear relationships in advance and directly extract the nonlinear powers. Subsequently, linear regression algorithms can be used for data learning, which shortens the computation time, ensures the real-time performance of control feedback, and increases the accuracy of soup dispensing control.

[0103] This application is for the analysis of features of second power polynomials. It should be noted that any scheme that can achieve the analysis of features of multiple power polynomials and enable data learning using linear regression algorithms is within the scope of protection of this application.

[0104] In order to determine the amount of the metal to be tested in soup at the current time per unit time, in a preferred embodiment of this scheme, after performing quadratic polynomial feature analysis on the new feature set, the analysis result is input into a linear regression model to determine the amount of the metal to be tested in soup at the current time per unit time.

[0105] The statistical features, geometric features, and moment features of the indicator diagram are combined and processed to form a feature set.

[0106] S207, compare the cumulative soup output with the set value to obtain the quantitative detection result of the metal to be tested, wherein the cumulative soup output is the cumulative output from the start of soup output to the current time.

[0107] Specifically, the cumulative amount of soup dispensed is compared with a set value to obtain the quantitative detection result of the metal to be tested. If the cumulative amount of soup dispensed is equal to the set value, the dispensing stops; if the cumulative amount of soup dispensed is less than the set value, the dispensing continues until the cumulative amount of soup dispensed at the current time is equal to the set value.

[0108] Key technical points:

[0109] 1. This application extracts the pressure feedback waveform from the start of soup dispensing to the current time, rather than the pressure value at a single moment, and extracts the upper and lower envelopes of the pressure feedback waveform to form a dynamometer diagram;

[0110] 2. Computational-control interaction: Using real-time calculation results to influence control rather than sensor signal values.

[0111] The advantages of this invention are:

[0112] Compared to existing technologies, this method reduces the computational and control workload from multiple sensors to a single pressure sensor, saving on sensor hardware costs and reducing control complexity.

[0113] This invention uses big data analysis to transform the dimension of precision control from multi-dimensional to single-dimensional. Currently, there is no other alternative solution to achieve the same purpose for single-dimensional signals.

[0114] This invention enables rapid prediction of soup output using only the valve opening and pressure feedback of the quantitative furnace, simplifying the control process and reducing control and hardware costs.

[0115] This invention can effectively simplify the structure of the quantitative furnace, reduce the manufacturing cost of the quantitative furnace, and ensure that the quantitative effect is not reduced.

[0116] Accordingly, this application also proposes a pressure feedback-based metal quantitative detection device.

[0117] The apparatus includes a metering furnace, such as Figure 2 The diagram shows the structure of a quantitative dispensing furnace, which consists of a furnace body, an outlet level gauge, a compressed gas inlet, a compressed gas inlet valve (solenoid valve + proportional valve), a furnace body pressure sensor, and a pressure relief port. When compressed gas enters the furnace, the gas expands, changing its volume and the furnace pressure. As the pressure increases, the molten metal is forced out of the furnace, forming dispensing liquid. The compressed gas inlet valve controls the gas intake. The outlet level gauge is designed to maintain pressure; when the molten metal reaches the outlet level, it alerts the user that dispensing is about to begin. When the dispensing reaches the required quantity, pressure is released through the pressure relief port to stop dispensing. This quantitative dispensing furnace eliminates the gravity measurement module and high-precision liquid level and temperature measurement module required by other quantitative dispensing furnaces, reducing its size and manufacturing costs.

[0118] To further illustrate the technical concept of this invention, the technical solution of this invention will now be described in conjunction with specific application scenarios.

[0119] Figure 3 This is another schematic diagram of a pressure feedback-based method for quantitative metal detection.

[0120] 1) First, obtain the pressure curve of the metal to be tested per unit time. The pressure curve is obtained by the pressure sensor at the top of the quantitative furnace.

[0121] 2) Use the Hilbert-Huang transform to extract the upper and lower envelopes of the pressure curve and form the dynamometer diagram of the pressure curve.

[0122] 3) Normalize the indicator diagram by combining the valve opening value: The normalization method is to extract the global pressure maximum and minimum values, and use the maximum and minimum values ​​as the vertical axis and time as the horizontal axis to define a unified calibration background for the indicator diagram.

[0123] 4) Extract the geometric features of the indicator diagram, including but not limited to: indicator diagram side length, indicator diagram side number, indicator diagram inflection point coordinates, indicator diagram inflection point number, indicator diagram area, indicator diagram perimeter, indicator diagram projection length, indicator diagram projection width, etc.; wherein the geometric features of the indicator diagram are obtained by the contour detection algorithm, which is an algorithm in OpenCV for processing structural analysis and shape description.

[0124] 5) Extract the moment features of the dynamometer diagram, including but not limited to: the center moment of the dynamometer diagram, the Hu moment of the dynamometer diagram, etc.

[0125] 6) Extract the statistical features of the indicator plot. The feature extraction process is as follows: 1. Use the time series feature extraction toolkit tsfeatures to extract all time series feature values; 2. Perform Pearson correlation analysis on each feature sequence and the label column, and exclude features with an absolute correlation value less than 0.8. The features remaining after the above steps are the features mentioned above.

[0126] 7) Collect the features described in steps 4), 5), and 6) to form a feature set; perform polynomial feature analysis on the feature set; polynomial feature analysis: add the power of each feature as a new feature to form an extended feature set. This invention uses second-order polynomial feature analysis, that is, in addition to the original features described in steps 4), 5), and 6), the square of each value of the original feature is also calculated to form a new feature value. The purpose is to assume the existence of nonlinear relationships in advance and directly obtain the nonlinear powers. Subsequently, a linear regression algorithm can be used for data learning, which shortens the calculation time, ensures the real-time performance of control feedback, and increases the accuracy of soup dispensing control.

[0127] A linear regression model is used to predict the current soup output using an extended feature set. Linear regression is a statistical analysis method that uses regression analysis in mathematical statistics to determine the quantitative relationship of interdependence between two or more variables.

[0128] 8) Compare the cumulative soup output with the set value. If they match, stop dispensing soup. If the set value is not reached, repeat process 1-9.

[0129] Example: Assuming the mold requires 50kg of molten aluminum to fill, the set value is 50. Using 1 second as a unit of time, an indicator diagram is extracted every unit of time, along with the aforementioned descriptive features. These features are then analyzed using polynomial feature parsing and input into a linear regression model to obtain the amount of molten aluminum dispensed per unit of time. For example, if 4kg of molten aluminum is dispensed per unit of time, which is less than the required 50kg, the process of determining the amount dispensed per unit of time continues until the final determination shows that the cumulative amount dispensed meets 50kg. At this point, dispensing is terminated, and the aluminum casting process is complete.

[0130] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for quantitative detection of metals based on pressure feedback, characterized in that, include: The pressure curve of the metal to be tested is obtained within a unit time, wherein the pressure signal of the metal to be tested is acquired by a pressure sensor at the top of the quantitative furnace to obtain the pressure curve. The upper and lower envelopes of the pressure curve are extracted by Hilbert-Huang transform. Based on the upper and lower envelopes, the dynamometer diagram of the pressure curve at the current moment is determined, and the dynamometer diagram is normalized. Extract the geometric features of the normalized indicator diagram. The geometric features include at least the edge length, number of edges, coordinates of inflection points, number of inflection points, area, perimeter, projection length, and projection width of the indicator diagram. Extract the moment features of the normalized dynamometer diagram, wherein the moment features include at least the center moment of the dynamometer diagram and the Hu moment of the dynamometer diagram; The statistical features of the normalized indicator diagram are extracted. The statistical features include at least the Holt exponential smoothing feature, information entropy, crossover feature, sequence patch feature, block feature based on tiling window, autocorrelation feature, partial autocorrelation feature, sparsity, stability, periodicity, Kwiatkowski unit root test statistic and Phillips-Peren unit root test statistic. The statistical features, geometric features, and moment features of the indicator diagram are combined and processed to form a feature set. Polynomial feature analysis is performed on the feature set, and the result of the polynomial feature analysis is input into the regression prediction model to determine the amount of the metal to be measured at the current time per unit time. The cumulative soup output is compared with the set value to obtain the quantitative detection result of the metal to be tested. The cumulative soup output is the output accumulated from the start of soup output to the current time.

2. The method according to claim 1, characterized in that, The indicator diagram is normalized as follows: Extract the maximum and minimum pressure values ​​of the metal to be tested from the start of the dispensing process to the current time within a unit of time. Use the pressure value as the vertical axis and time as the horizontal axis to determine the calibration background of the indicator diagram. Draw an indicator diagram with coordinates on the calibration background, and use the indicator diagram with coordinates as the normalized indicator diagram.

3. The method according to claim 1, characterized in that, The geometric features of the indicator diagram are determined by a contour detection algorithm.

4. The method according to claim 1, characterized in that, Extracting the center moment of the indicator diagram includes: The centroid coordinates (x0, y0) of the target region are determined by using the zeroth-order and first-order moments. The centroid coordinates (x0, y0) are determined according to the following formula: Where, m 00 The zeroth-order origin moment is the pixel intensity at ((0,0)) of the normalized indicator map, m. 01 and m 10 Let m be the first-order moment at the origin. 10 m represents the pixel intensity at (1,0) in the normalized indicator map. 01 y0 represents the pixel intensity at (0,1) of the normalized indicator image, x0 represents the centroid of the normalized indicator image along the x-axis, and y0 represents the centroid of the normalized indicator image along the y-axis. Based on the centroid coordinates (x0, y0), construct the center moment of the indicator diagram; The central moment of the indicator diagram is determined according to the following formula: Among them, u pq Let p and q be the center moments of the dynamometer, p and q be the orders of the dynamometer, (x, y) be the coordinates of the current pixel position, (x0, y0) be the centroid coordinates, C be the height of the dynamometer image, and R be the width of the dynamometer image.

5. The method according to claim 4, characterized in that, Extracting the indicator diagram Hu moment includes: Seven sets of invariant moments are derived using the second-order and third-order nominal central moments of the image. The second-order and third-order nominal central moments are obtained by normalizing the central moments of the dynamometer diagram. The indicator diagram Hu moment is determined according to the following formula: Φ1=η 20 +n 02 Φ3=(η 20 -3rd 12 ) 2 +3(h 21 -or 03 ) 2 Φ4=(η 30 +n 12 ) 2 +(the 21 +n 03 ) 2 Φ5=(η 30 +3rd 12 )(or 30 +n 12 )[(or 30 +n 12 ) 2 -3(h 21 +n 03 ) 2 ] +(3rd 21 -or 03 )(or 21 +n 03 )[3(h 30 +n 12 ) 2 -(or 21 +n 03 ) 2 ] Φ6=(η 20 -or 02 )[(or 30 +n 12 ) 2 -(or 21 +n 03 ) 2 ]+4th 11 (or 30 +n 12 )(or 21 +n 03 ) Φ7=(3rd) 21 -or 03 )(or 30 +n 12 )[(or 30 +n 12 ) 2 -3(h 21 +n 03 ) 2 ]+] +(3rd 12 -or 30 )(or 21 +n 03 )[3(h 30 +n 12 ) 2 -(or 21 +n 03 ) 2 ] Where, η ij The normalized center moment is of order i+j. When it is the second-order gauge center moment, the value of i is 0≤i≤2, where i is a positive integer, and the value of j is 0≤j≤2, where j is a positive integer, and i+j=2. When it is the third-order gauge center moment, the value of i is 0≤i≤3, where i is a positive integer, and the value of j is 0≤j≤3, where j is a positive integer, and i+j=3. Φ1 to Φ7 are 7 sets of invariant moments, and the 7 sets of invariant moments are the Hu moments of the indicator diagram.

6. The method according to claim 1, characterized in that, Extract the statistical features of the indicator diagram, including: Use the time series feature extraction toolkit tsfeatures to extract feature values ​​from the entire time series; Pearson correlation is performed between each time series feature value and the label column. Feature values ​​with an absolute Pearson correlation value less than 0.8 are removed to determine the statistical characteristics of the indicator chart. The label column is the actual liquid weight of the metal to be tested output per unit time.

7. The method according to claim 1, characterized in that, The feature set is subjected to polynomial feature parsing, specifically as follows: Square each feature value in the feature set to form a new feature set; The new feature set is subjected to quadratic polynomial feature analysis.

8. The method according to claim 7, characterized in that, After performing quadratic polynomial feature analysis on the new feature set, the analysis results are input into the linear regression model to determine the amount of the metal to be tested in soup at the current time per unit time.

9. The method according to claim 8, characterized in that, The cumulative output volume is compared with a set value to quantify the metal to be tested, specifically as follows: If the cumulative soup volume equals the set value, then soup dispensing will stop; If the cumulative soup volume is less than the set value, continue dispensing soup until the cumulative soup volume at the current time equals the set value.

10. An apparatus for use in the method of any one of claims 1 to 9, comprising: A quantitative furnace, comprising a furnace body, an outlet level gauge, a compressed gas inlet, a compressed gas inlet valve, and a pressure relief port; When compressed gas enters the furnace through the compressed gas inlet valve, the molten metal to be tested is squeezed out of the furnace to form a molten liquid. When the liquid metal to be tested comes into contact with the outlet liquid level, the dispensing begins. When the dispensing volume reaches the quantitative requirement, the dispensing is terminated by releasing pressure through the pressure relief port.

Images