Aluminum profile machining precision optimization method and system

Abstract

The application discloses an aluminum profile machining precision optimization method and system. The method comprises the following steps: obtaining an aluminum profile cross-section target contour; collecting aluminum profile cross-section actual contour point cloud data and preprocessing; calculating the profile error AP of the component contour; using the long short-term memory network algorithm model LSTM combined with the attention mechanism to predict the error trend and obtain the predicted error; calculating the process parameter compensation amount according to the predicted error; and obtaining the final applied process parameter according to the process parameter compensation amount and the basic process parameter set value. The application adopts an optical scanner to non-contact online scanning to obtain complete contour point cloud data, combines B-spline curve fitting with normal distance accurate calculation of profile error, realizes comprehensive collection of contour data and accurate quantization of error, can control the tolerance in a smaller range, and meets the strict requirements of high-end applications such as chip heat sinks, automobile anti-collision beams and precision guide rails on size precision and shape precision.

Description

Technical Field

[0001] This application relates to the field of aluminum profile processing technology, specifically to a method and system for optimizing the processing accuracy of aluminum profiles. Background Technology

[0002] Aluminum profile extrusion is a highly efficient process for manufacturing complex cross-section metal components, widely used in key fields such as construction, transportation, aerospace, and electronics. With industrial development, the requirements for dimensional accuracy, shape accuracy, and batch consistency of aluminum profiles are becoming increasingly stringent. However, current aluminum profile extrusion production suffers from the following shortcomings:

[0003] (1) The wall thickness tolerance set by the existing national standards is acceptable for routine applications, but in high-end applications such as chip heat sinks, automotive anti-collision beams, and precision guide rails, especially in the manufacturing of precision electronic components such as chip heat sinks, higher precision requirements are needed, and the tolerance needs to be controlled within a smaller range. However, the existing technology lacks effective molding control methods, making it difficult to achieve higher precision requirements.

[0004] (2) The process parameters of the existing production line are usually fixed after being set, and cannot be adaptively adjusted in real time according to changes such as mold wear and ambient temperature fluctuations, resulting in poor precision and consistency of batch-produced profiles, high scrap rate and production cost.

[0005] (3) Traditional profile contour detection methods mostly rely on contact measuring tools such as micrometers and calipers or offline coordinate measuring machines with phased sampling. They are inefficient, have limited sampling points, cannot obtain complete profile information, and have serious feedback lag.

[0006] (4) Traditional detection methods are only used for quality inspection and cannot be used for real-time control of the production process.

[0007] Therefore, it is necessary to improve the existing aluminum profile processing technology. Summary of the Invention

[0008] In view of the technical problems mentioned in the background section, the purpose of this invention is to provide a method and system for optimizing the processing accuracy of aluminum profiles.

[0009] To achieve the objectives of this invention, the technical solution provided by this invention is as follows:

[0010] First aspect

[0011] This application provides a method for optimizing the processing accuracy of aluminum profiles, including the following steps:

[0012] Step S1: Obtain the target profile of the aluminum profile cross section; the target profile of the aluminum profile cross section includes multiple basic shape component units;

[0013] Step S2: Collect the actual contour point cloud data of the aluminum profile cross section and perform preprocessing to obtain the preprocessed actual contour point cloud data;

[0014] Step S3: Based on the preprocessed actual contour point cloud data, construct a component fitting contour that reflects multiple basic shape component units; based on the component fitting contour and the target contour of the aluminum profile section, calculate the error value from all data points on the component fitting contour to their corresponding target contour; based on the error value, obtain the contour error ΔP of the component contour.

[0015] Step S4: Repeat steps S2-S3 to obtain the contour error ΔP of the component contour at multiple time points for a preset number of moments;

[0016] Step S5: Based on the contour error ΔP of the component contour at multiple time points and process parameters, construct a temporal feature matrix; according to the temporal feature matrix, use the Long Short-Term Memory (LSTM) network algorithm model combined with an attention mechanism to predict the error trend and obtain the prediction error; the process parameters include the aluminum rod heating temperature of the processing equipment, the extrusion speed of the processing equipment, the traction force of the processing equipment, the die wear index, and the ambient temperature of the processing environment;

[0017] Step S6: Calculate the process parameter compensation amount based on the prediction error; obtain the final applied process parameters based on the process parameter compensation amount and the basic process parameter setting value.

[0018] Step S7: Process according to the final applied process parameters.

[0019] Second aspect

[0020] This application provides an aluminum profile processing accuracy optimization system for executing the aluminum profile processing accuracy optimization method, including the following units: target contour acquisition unit, point cloud data acquisition unit, contour error acquisition unit, repetitive execution unit, error prediction unit, process parameter acquisition unit, and execution unit;

[0021] The target contour acquisition unit is used to acquire the target contour of the aluminum profile cross section; the target contour of the aluminum profile cross section includes multiple basic shape component units;

[0022] The point cloud data acquisition unit is used to collect the actual contour point cloud data of the aluminum profile cross section and perform preprocessing to obtain the preprocessed actual contour point cloud data.

[0023] The contour error acquisition unit is used to construct a component fitting contour that reflects multiple basic shape component units based on the preprocessed actual contour point cloud data; calculate the error value from all data points on the component fitting contour to their corresponding target contour based on the component fitting contour and the target contour of the aluminum profile section; and obtain the contour error ΔP of the component contour based on the error value.

[0024] The repeat execution unit is used to repeatedly execute the point cloud data acquisition unit and the contour error acquisition unit to obtain the contour error ΔP of the component contour at multiple time points of a preset number of time periods.

[0025] The error prediction unit is used to construct a temporal feature matrix based on the contour error ΔP of the component contour at multiple time points and process parameters; according to the temporal feature matrix, LSTM combined with an attention mechanism is used to predict the error trend to obtain the prediction error; the process parameters include the heating temperature of the aluminum rod of the processing equipment, the extrusion speed of the processing equipment, the traction force of the processing equipment, the die wear index, and the processing environment temperature;

[0026] The process parameter acquisition unit is used to calculate the process parameter compensation amount based on the prediction error; and to obtain the final applied process parameters based on the process parameter compensation amount and the basic process parameter setting value.

[0027] The execution unit is used to perform processing according to the final applied process parameters.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] (1) This application uses an optical scanner to obtain complete contour point cloud data through non-contact online scanning. Combined with B-spline curve fitting and normal distance to accurately calculate contour error ΔP, it realizes comprehensive acquisition of contour data and accurate quantification of error. The tolerance can be controlled within a smaller range to meet the stringent requirements of high-end applications such as chip heat sinks, automotive anti-collision beams, and precision guide rails for dimensional and shape accuracy.

[0030] (2) This application is based on an error trend prediction model that combines LSTM with attention mechanism. It integrates the temporal characteristics of key process parameters such as aluminum rod heating temperature, extrusion speed, traction force, ambient temperature, and die wear index, which can predict the profile error in the future. Then, through adaptive gradient descent algorithm and PID control, the compensation amount of process parameters is calculated in real time, which solves the problem that the process parameters of traditional production lines are fixed and cannot cope with real-time changes such as die wear and environmental fluctuations. It significantly reduces the precision dispersion of batch profiles and improves batch consistency.

[0031] (3) This application adopts non-contact online scanning and outlier removal technology based on statistical distribution characteristics to achieve efficient acquisition and preprocessing of contour data and avoid feedback lag; at the same time, it constructs a closed-loop control process of "detection-error calculation-error prediction-parameter compensation-processing", upgrading the detection function from simple quality inspection to real-time control of the production process, greatly reducing the scrap rate and reducing production costs. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the aluminum profile processing accuracy optimization method provided in an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of the actual contour point cloud data of the aluminum profile cross-section in the embodiments of this application;

[0034] Figure 3 Comparison of the effects before and after optimizing the processing precision of aluminum profiles. Detailed Implementation

[0035] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0036] like Figure 1 As shown in the figure, this embodiment provides a method for optimizing the processing accuracy of aluminum profiles, including the following steps:

[0037] Step S1: Obtain the target profile of the aluminum profile cross section; the target profile of the aluminum profile cross section includes multiple basic shape component units;

[0038] It should be noted that in this step, firstly, a unified measurement and evaluation reference coordinate system is established to ensure the consistency and accuracy of the contour evaluation. The target contour of the aluminum profile cross-section is then obtained; this target contour includes multiple basic shape component units, including the outer contour, center hole, corner holes, T-slots, etc.

[0039] Step S2: Collect the actual contour point cloud data of the aluminum profile cross-section and perform preprocessing to obtain the preprocessed actual contour point cloud data; specifically including the following:

[0040] Step S21: After the aluminum profile has cooled and solidified, an optical scanner is used to perform a non-contact online scan of the aluminum profile cross-section to obtain the actual contour point cloud data of the aluminum profile cross-section; such as Figure 2 As shown.

[0041] It should be noted that the scanning process uses a step length of 0.2-0.5mm for discretization sampling. The sampling step length of 0.2-0.5mm can accurately capture the roundness characteristics of the center hole and corner holes, ensuring accurate representation of complex geometric features.

[0042] Step S22: An outlier detection algorithm based on statistical distribution features is used to automatically identify and remove measurement noise points to obtain preprocessed actual contour point cloud data.

[0043] It should be noted that an outlier detection algorithm based on statistical distribution characteristics is used to automatically identify and remove measurement noise caused by factors such as surface reflection, oxidation residue, and cutting fluid stains, and to verify the integrity of the point cloud data to ensure the accuracy and reliability of subsequent analysis.

[0044] Step S3: Based on the preprocessed actual contour point cloud data, construct a component fitting contour that reflects multiple basic shape component units; based on the component fitting contour and the target contour of the aluminum profile section, calculate the error value from all data points on the component fitting contour to their corresponding target contour; based on the error value, obtain the contour accuracy error ΔP of the component contour; specifically including the following:

[0045] Step S3.1: Use the B-spline curve fitting formula to fit the preprocessed actual contour point cloud data to construct a component fitting contour that reflects multiple basic shape component units;

[0046] Step S3.2: Based on the component fitting profile and the aluminum profile cross-section target profile, calculate the normal distance from all data points on the component fitting profile to their corresponding target profile, and the obtained normal distance is the error value;

[0047] The formula for calculating the normal distance is as follows:

[0048] ;

[0049] in, This represents the normal distance to the i-th data point on the component's fitted contour; ) indicates the component fitting profile on the th i The coordinates of each data point; ) represents the first contour on the target contour that matches the component's fitted contour. i The coordinates of the points corresponding to each data point; Indicates the component fitting profile on the first i At the nth data point, the direction of the fitted contour normal is parallel to the nth data point. i The angle between the directions of the lines connecting each data point to the corresponding point on the target contour; Indicates the component fitting profile on the first i Euclidean distance from each data point to the corresponding point on the target contour;

[0050] Step S3.3: Based on the error value, obtain the contour error ΔP of the component contour. The calculation formula is as follows:

[0051] ;

[0052] in, Indicates the profile error of the component outline; tm This represents the total number of fitted data points on the component outline that participated in the calculation. Indicates the first i The absolute value of the normal distance to each data point.

[0053] Step S4: Repeat steps S2-S3 to obtain the contour error ΔP of the component contour at multiple time points for a preset number of moments;

[0054] Step S5: Based on the contour error ΔP of the component contour at multiple time points and process parameters, construct a temporal feature matrix; according to the temporal feature matrix, use the Long Short-Term Memory (LSTM) network algorithm model combined with an attention mechanism to predict the error trend and obtain the prediction error; the process parameters include the aluminum rod heating temperature of the processing equipment, the extrusion speed of the processing equipment, the traction force of the processing equipment, the die wear index, and the ambient temperature of the processing environment; specifically including the following:

[0055] Step S5.1: Based on the contour error ΔP of the component contour at multiple time points and the process parameters collected on-site, perform time-series alignment to construct a time-series feature matrix. ;

[0056] ;

[0057] in, This is the contour error sequence from the current time t to the historical time k; This is the contour error sequence from the current time t to the historical time k+1; This refers to the heating temperature of the aluminum rod. This refers to the extrusion speed; For traction force; The ambient temperature; The mold wear index; This is the contour error sequence from the current time t;

[0058] Step S5.2: Based on the aforementioned temporal feature matrix, process it using the Long Short-Term Memory (LSTM) network algorithm model, including:

[0059] Step S5.2.1: Forget gate processing, as follows:

[0060] ;

[0061] in, The output of the forget gate indicates the current time. t The activation value of the forget gate is between [0,1]. Use the Sigmoid activation function; This is the forget gate weight matrix; To concatenate vectors, It is the hidden state from the previous moment; Forget gate bias term;

[0062] Step S5.2.2: Input gate processing, as follows:

[0063] ;

[0064] in, The input gate output represents the activation value of the input gate at the current time t, and its value ranges from [0,1]. The input gate weight matrix; Forget gate bias term;

[0065] Step S5.2.3: Output gate processing, as detailed below:

[0066] ;

[0067] in, The output value is the value of the output gate at the current time t, and its value ranges from [0,1]. This is the output gate weight matrix; This is the output gate bias term;

[0068] Step S5.2.4: Calculation of cell state and hidden state, as detailed below:

[0069] ;

[0070] in, The candidate cell state represents the new information to be added to memory at the current time t. The candidate state weight matrix; Candidate state bias terms;

[0071] ;

[0072] in, Let be the cell state at the current time t, and the cell state at the current time t is long-term memory; This represents the cell state at the previous moment; This is element-wise multiplication;

[0073] ;

[0074] in, Let be the hidden state at the current time t, and let be the short-term memory / output.

[0075] Step S5.3: Calculate the attention mechanism weights, including:

[0076] ;

[0077] in, This represents the hidden state of historical time k at the current time t. The attention weights assigned range from [0,1]. At a certain point in history; T1 is the length of the time series; To measure the difference between the current time t and historical times k Attention energy scores indicating their correlation or importance; An attention energy score that measures the relevance or importance of the current time t to a historical time j; where, This is the attention vector; This is the historical state weight matrix; This is the current state weight matrix; It is in a state of historical concealment; The current hidden state is T, where T is the matrix transpose.

[0078] Step S5.4: Calculate the prediction error;

[0079] ;

[0080] in, For prediction error, the predicted future time. Contour error; T 1 represents the length of the time series.

[0081] Step S6: Calculate the process parameter compensation amount based on the prediction error; obtain the final applied process parameters based on the process parameter compensation amount and the basic process parameter setting value; specifically including the following:

[0082] Step S6.1: Construct an optimization objective function with the sole objective of minimizing the prediction error, as follows:

[0083] ;

[0084] in, To optimize the objective, minimize the absolute value of the prediction error;

[0085] Decision variables: ;

[0086] Set constraints:

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] in, This represents the minimum threshold temperature for heating the aluminum rod. This represents the maximum threshold temperature for heating the aluminum rod. The minimum threshold for extrusion speed, The maximum threshold for extrusion speed; The minimum threshold for traction force. This represents the maximum threshold value for traction force. The minimum threshold for ambient temperature. The maximum threshold for ambient temperature; The minimum threshold for mold wear index, The maximum threshold for mold wear index; , , , , This indicates 5 constraints;

[0093] Step S6.2: For the objective function constructed in step S6.1, the process parameters are solved using the adaptive gradient descent algorithm, as shown in the following formula:

[0094] ;

[0095] in, Number of iterations n Process parameters at that time; Number of iterations n+1 Process parameters at that time; The learning rate; For the objective function in step S6.1 The gradient at the point is implicitly satisfied by the constraints of step S6.1 through the iterative process of gradient descent.

[0096] Among them, the learning rate Adaptive learning rate adjustment is performed to obtain :

[0097] ;

[0098] in, The initial learning rate; The attenuation coefficient;

[0099] Step S6.3: Real-time compensation of process parameters based on PID control, as detailed below:

[0100] ;

[0101] in, For the first The amount of compensation at any given moment; , , These are the proportional, integral, and differential coefficients, respectively. Let be the error at time t; For target contour error; The actual profile error at the current time t; This represents the error at time t-1; Indicates from the initial time j =1 to the current time in the accumulated error sequence of time n j One error value;

[0102] Final applied process parameters The formula is as follows:

[0103] ;

[0104] in, These are the basic process parameter settings. .

[0105] Step S7: Process according to the final applied process parameters.

[0106] like Figure 3 The figure shows a comparison of the effects of aluminum profile processing accuracy optimization before and after. As shown, before optimization, the profile error ΔP remained at a high level, exhibiting stable fluctuations. After applying the optimized process parameters, the profile error ΔP gradually converged and stabilized near the target error. Figure 3 The method is visually demonstrated to significantly improve machining accuracy.

[0107] In addition, this embodiment provides an aluminum profile processing accuracy optimization system for executing the aluminum profile processing accuracy optimization method, including the following units: target contour acquisition unit, point cloud data acquisition unit, contour error acquisition unit, repetitive execution unit, error prediction unit, process parameter acquisition unit, and execution unit;

[0108] The target contour acquisition unit is used to acquire the target contour of the aluminum profile cross section; the target contour of the aluminum profile cross section includes multiple basic shape component units;

[0109] The point cloud data acquisition unit is used to collect the actual contour point cloud data of the aluminum profile cross section and perform preprocessing to obtain the preprocessed actual contour point cloud data.

[0110] The contour error acquisition unit is used to construct a component fitting contour that reflects multiple basic shape component units based on the preprocessed actual contour point cloud data; calculate the error value from all data points on the component fitting contour to their corresponding target contour based on the component fitting contour and the target contour of the aluminum profile section; and obtain the contour error ΔP of the component contour based on the error value.

[0111] The repeat execution unit is used to repeatedly execute the point cloud data acquisition unit and the contour error acquisition unit to obtain the contour error ΔP of the component contour at multiple time points of a preset number of time periods.

[0112] The error prediction unit is used to construct a temporal feature matrix based on the contour error ΔP of the component contour at multiple time points and process parameters; according to the temporal feature matrix, LSTM combined with an attention mechanism is used to predict the error trend to obtain the prediction error; the process parameters include the heating temperature of the aluminum rod of the processing equipment, the extrusion speed of the processing equipment, the traction force of the processing equipment, the die wear index, and the processing environment temperature;

[0113] The process parameter acquisition unit is used to calculate the process parameter compensation amount based on the prediction error; and to obtain the final applied process parameters based on the process parameter compensation amount and the basic process parameter setting value.

[0114] The execution unit is used to perform processing according to the final applied process parameters.

[0115] Specifically, the point cloud data acquisition unit is used to perform the following steps:

[0116] Step S21: After the aluminum profile is cooled and shaped, an optical scanner is used to perform a non-contact online scan of the aluminum profile cross section to obtain the actual contour point cloud data of the aluminum profile cross section;

[0117] Step S22: An outlier detection algorithm based on statistical distribution features is used to automatically identify and remove measurement noise points to obtain preprocessed actual contour point cloud data.

[0118] Specifically, the contour error acquisition unit is used to perform the following steps:

[0119] Step S3.1: Use the B-spline curve fitting formula to fit the preprocessed actual contour point cloud data to construct a component fitting contour that reflects multiple basic shape component units;

[0120] Step S3.2: Based on the component fitting profile and the aluminum profile cross-section target profile, calculate the normal distance from all data points on the component fitting profile to their corresponding target profile, and the obtained normal distance is the error value;

[0121] The formula for calculating the normal distance is as follows:

[0122] ;

[0123] in, This represents the normal distance to the i-th data point on the component's fitted contour; ) indicates the component fitting profile on the th i The coordinates of each data point; ) represents the first contour on the target contour that matches the component's fitted contour. i The coordinates of the points corresponding to each data point; Indicates the component fitting profile on the first i At the nth data point, the direction of the fitted contour normal is parallel to the nth data point. i The angle between the directions of the lines connecting each data point to the corresponding point on the target contour; Indicates the component fitting profile on the first i Euclidean distance from each data point to the corresponding point on the target contour;

[0124] Step S3.3: Based on the error value, obtain the contour error ΔP of the component contour. The calculation formula is as follows:

[0125] ;

[0126] in, Indicates the profile error of the component outline; tm This represents the total number of fitted data points on the component outline that participated in the calculation. Indicates the first i The absolute value of the normal distance to each data point.

[0127] Specifically, the error prediction unit is used to perform the following steps:

[0128] Step S5.1: Based on the contour error ΔP of the component contour at multiple time points and the process parameters collected on-site, perform time-series alignment to construct a time-series feature matrix. ;

[0129] ;

[0130] in, This is the contour error sequence from the current time t to the historical time k; This is the contour error sequence from the current time t to the historical time k+1; This refers to the heating temperature of the aluminum rod. This refers to the extrusion speed; For traction force; The ambient temperature; The mold wear index;

[0131] Step S5.2: Based on the aforementioned temporal feature matrix, process it using the Long Short-Term Memory (LSTM) network algorithm model, including:

[0132] Step S5.2.1: Forget gate processing, as follows:

[0133] ;

[0134] in, The output of the forget gate indicates the current time. t The activation value of the forget gate is between [0,1]. Use the Sigmoid activation function; This is the forget gate weight matrix; To concatenate vectors, It is the hidden state from the previous moment; Forget gate bias term;

[0135] Step S5.2.2: Input gate processing, as follows:

[0136] ;

[0137] in, The input gate output represents the activation value of the input gate at the current time t, and its value ranges from [0,1]. The input gate weight matrix; Forget gate bias term;

[0138] Step S5.2.3: Output gate processing, as detailed below:

[0139] ;

[0140] in, The output value is the value of the output gate at the current time t, and its value ranges from [0,1]. This is the output gate weight matrix; This is the output gate bias term;

[0141] Step S5.2.4: Calculation of cell state and hidden state, as detailed below:

[0142] ;

[0143] in, The candidate cell state represents the new information to be added to memory at the current time t. The candidate state weight matrix; Candidate state bias terms;

[0144] ;

[0145] in, Let be the cell state at the current time t, and the cell state at the current time t is long-term memory; This represents the cell state at the previous moment; This is element-wise multiplication;

[0146] ;

[0147] in, Let be the hidden state at the current time t, and let be the short-term memory / output.

[0148] Step S5.3: Calculate the attention mechanism weights, including:

[0149] ;

[0150] in, This represents the hidden state of historical time k at the current time t. The attention weights assigned range from [0,1]. At a certain point in history; T1 is the length of the time series; To measure the difference between the current time t and historical times k Attention energy scores indicating their correlation or importance; An attention energy score that measures the relevance or importance of the current time t to a historical time j; where, This is the attention vector; This is the historical state weight matrix; This is the current state weight matrix; It is in a state of historical concealment; The current hidden state is T, where T is the matrix transpose.

[0151] Step S5.4: Calculate the prediction error;

[0152] ;

[0153] in, For prediction error, the predicted future time. Contour error; T 1 represents the length of the time series.

[0154] The process parameter acquisition unit is specifically used to perform the following steps:

[0155] Step S6.1: Construct an optimization objective function with the sole objective of minimizing the prediction error, as follows:

[0156] ;

[0157] in, To optimize the objective, minimize the absolute value of the prediction error;

[0158] Decision variables: ;

[0159] Set constraints:

[0160] ;

[0161] ;

[0162] ;

[0163] ;

[0164] ;

[0165] in, This represents the minimum threshold temperature for heating the aluminum rod. This represents the maximum threshold temperature for heating the aluminum rod. The minimum threshold for extrusion speed, The maximum threshold for extrusion speed; The minimum threshold for traction force. This represents the maximum threshold value for traction force. The minimum threshold for ambient temperature. The maximum threshold for ambient temperature; The minimum threshold for mold wear index, The maximum threshold for mold wear index; , , , , This indicates 5 constraints;

[0166] Step S6.2: For the objective function constructed in step S6.1, the process parameters are solved using the adaptive gradient descent algorithm, as shown in the following formula:

[0167] ;

[0168] in, Number of iterations n Process parameters at that time; Number of iterations n+1 Process parameters at that time; The learning rate; For the objective function in step S6.1 The gradient at the point is implicitly satisfied by the constraints of step S6.1 through the iterative process of gradient descent.

[0169] Among them, the learning rate Adaptive learning rate adjustment is performed to obtain :

[0170] ;

[0171] in, The initial learning rate; The attenuation coefficient;

[0172] Step S6.3: Real-time compensation of process parameters based on PID control, as detailed below:

[0173] ;

[0174] in, For the first The amount of compensation at any given moment; , , These are the proportional, integral, and differential coefficients, respectively. Let be the error at time t; For target contour error; The actual profile error at the current time t; This represents the error at time t-1; Indicates from the initial time j =1 to the current time in the accumulated error sequence of time n j One error value;

[0175] Final applied process parameters The formula is as follows:

[0176] ;

[0177] in, These are the basic process parameter settings. .

[0178] Finally, it should be noted that the above embodiments are merely illustrative and explanatory of the present invention, and are not intended to limit the present invention to the scope of the described embodiments. Furthermore, those skilled in the art will understand that the present invention is not limited to the above embodiments, and many more variations and modifications can be made based on the teachings of the present invention, all of which fall within the scope of protection claimed by the present invention.

Claims

1. A method for optimizing machining accuracy of an aluminum profile, characterized by, The method comprises the following steps: Step S1: obtaining an aluminum profile cross-section target profile; the aluminum profile cross-section target profile comprises a plurality of basic shape component units; Step S2: collecting aluminum profile cross-section actual profile point cloud data and preprocessing to obtain preprocessed actual profile point cloud data; Step S3: based on the preprocessed actual profile point cloud data, a component fitting profile reflecting a plurality of basic shape component units is constructed; based on the component fitting profile and the aluminum profile cross-section target profile, the error value of all data points on the component fitting profile to the corresponding target profile is calculated; based on the error value, the profile error ΔP of the component profile is obtained; Step S4: repeating steps S2-S3 to obtain the profile error ΔP of the component profile at a plurality of time points; Step S5: based on the profile error ΔP of the component profile at a plurality of time points and the process parameters, a time sequence feature matrix is constructed; According to the time sequence feature matrix, the long short-term memory network algorithm model LSTM is combined with the attention mechanism to predict the error trend and obtain the predicted error; the process parameters include the aluminum rod heating temperature of the processing equipment, the extrusion speed of the processing equipment, the traction of the processing equipment, the mold wear index, and the processing environment temperature; Step S6: according to the predicted error, the process parameter compensation amount is calculated; according to the process parameter compensation amount and the basic process parameter set value, the final applied process parameter is obtained; Step S7: processing according to the final applied process parameter.

2. The method for optimizing the machining accuracy of aluminum profiles according to claim 1, characterized in that, Step S2 specifically comprises the following: Step S21: after the aluminum profile is cooled and shaped, an optical scanner is used to perform non-contact online scanning on the aluminum profile cross-section to obtain aluminum profile cross-section actual profile point cloud data; Step S22: using an abnormal point detection algorithm based on statistical distribution characteristics, automatically identifying and removing measurement noise points, to obtain preprocessed actual profile point cloud data.

3. The method according to claim 2, wherein Step S3 specifically comprises the following: Step S3.1: using a B-spline curve fitting formula to fit the preprocessed actual profile point cloud data to construct a component fitting profile reflecting a plurality of basic shape component units; Step S3.2: based on the component fitting profile and the aluminum profile cross-section target profile, the normal distance of all data points on the component fitting profile to the corresponding target profile is calculated, and the normal distance is the error value; The normal distance calculation formula is as follows: ； in, This represents the normal distance to the i-th data point on the component's fitted contour; ) indicates the component fitting profile on the th i The coordinates of each data point; ) represents the first contour on the target contour that matches the component's fitted contour. i The coordinates of the points corresponding to each data point; Indicates the component fitting profile on the first i At the nth data point, the direction of the fitted contour normal is parallel to the nth data point. i The angle between the directions of the lines connecting each data point to the corresponding point on the target contour; Indicates the component fitting profile on the first i Euclidean distance from each data point to the corresponding point on the target contour; Step S3.3: based on the error value, the profile error ΔP of the component profile is obtained, and the calculation formula is as follows: ； wherein, represents a profile error of a component profile; tm represents the number of all fitted data points on the component profile involved in the calculation; represents the absolute value of the normal distance of the i th data point.

4. The aluminum profile machining precision optimization method according to claim 3, characterized in that, Step S5 specifically comprises the following: Step S5.1: constructing a time series feature matrix based on the profile error ΔP of the preset number of profile components at multiple time points and the process parameters collected on site and performing time series alignment ; ； wherein, is a profile error sequence from a current time t to a historical time k; is a profile error sequence from a current time t to a historical time k+1; is an aluminum bar heating temperature; is an extrusion speed; is a traction force; is an ambient temperature; is a die wear index; is a profile error sequence from a current time t; Step S5.2: according to the time sequence feature matrix, the long short-term memory network algorithm model LSTM is used for processing, including: Step S5.2.1: forgetting gate processing, specifically as follows: ； wherein, is the forget gate output, representing the activation value of the forget gate at the current time step t, taking values in the range [0, 1]; t is the forget gate output, representing the activation value of the forget gate at the current time step t, taking values in the range [0, 1]; is the sigmoid activation function; is the forget gate weight matrix; is the concatenation vector, is the hidden state at the previous time step t - 1; is the forget gate bias term; Step S5.2.2: input gate processing, specifically as follows: ； wherein, is the input gate output, representing the activation value of the input gate at the current time t, with a value range of [0, 1]; is the input gate weight matrix; is the forget gate bias term; Step S5.2.3: output gate processing, specifically as follows: ； wherein, is the output of the output gate, representing the activation value of the output gate at the current time t, taking values in the range [0, 1]; is the weight matrix of the output gate; is the bias term of the output gate; Step S5.2.4: cell state and hidden state calculation, specifically as follows: ； wherein, is a candidate cell state, representing new information to be added to the memory at the current time instant t; is a candidate state weight matrix; is a candidate state bias term; ； wherein, is the cell state at the current time t, the cell state at the current time t being long-term memory; is the cell state at the previous time; is an element-wise multiplication; ； wherein, is the hidden state at the current time t, and the hidden state at the current time t is the short-term memory / output; Step S5.3: attention mechanism weight calculation, including: ； wherein, denotes the hidden state of the history time k at the current time t assigned attention weight, which ranges from 0 to 1; history time; T1 is the length of the time series; is the attention energy score measuring the relevance or importance between the current time t and the history time k ; and is the attention energy score measuring the relevance or importance between the current time t and the history time j; wherein, is the attention vector; is the history state weight matrix; is the current state weight matrix; is the history hidden state; is the current hidden state, and T is the matrix transpose. Step S5.4: calculating the predicted error; ； wherein, is the prediction error, the predicted future time of the profile error; T 1 is the time series length.

5. The method according to claim 4, wherein Step S6 specifically comprises the following: Step S6.1: Construct an optimization objective function with the minimization of prediction error as the only objective, specifically as follows: ； wherein To optimize the objective, the absolute value of the prediction error is minimized. Decision variables: ; Set the constraint condition: ； ； ； ； ； wherein, is a minimum threshold value for the aluminum bar heating temperature, is a maximum threshold value for the aluminum bar heating temperature; is a minimum threshold value for the extrusion speed, is a maximum threshold value for the extrusion speed; is a minimum threshold value for the draft force, is a maximum threshold value for the draft force; is a minimum threshold value for the ambient temperature, is a maximum threshold value for the ambient temperature; is a minimum threshold value for the die wear index, is a maximum threshold value for the die wear index; , , , , represent 5 constraints; Step S6.2: Solve the process parameters by using an adaptive gradient descent algorithm for the objective function constructed in step S6.1, as follows: ； wherein is the iteration number n is the process parameter at iteration number is the iteration number n+1 is the process parameter at iteration number is the learning rate is the gradient of the objective function in step S6.1 at iteration number is satisfied implicitly by the iterative process of gradient descent. where the learning rate is adapted by an adaptive learning rate adjustment to obtain : ； wherein, is an initial learning rate; is a decay coefficient; Step S6.3: Real-time compensation of process parameters based on PID control, specifically as follows: ； wherein, is the tth compensation amount at the tth , , are proportional, integral, and derivative coefficients, respectively; is the error at the tth is the target profile error; is the actual profile error at the tth denotes the error at the t-1th denotes the tth j error value in the cumulative error sequence from the initial time j = 1 to the tth Final applied process parameters The formula is as follows: ； wherein setting values for base process parameters, .

6. An aluminum extrusion machining accuracy optimization system for performing the aluminum extrusion machining accuracy optimization method according to any one of claims 1 to 5, characterized by, It includes the following units: target profile acquisition unit, point cloud data acquisition unit, profile error acquisition unit, repeated execution unit, error prediction unit, process parameter acquisition unit and execution unit; The target profile acquisition unit is configured to acquire an aluminum profile section target profile; the aluminum profile section target profile includes a plurality of basic shape component units; The point cloud data acquisition unit is configured to collect aluminum profile section actual profile point cloud data and pre-process the aluminum profile section actual profile point cloud data to obtain pre-processed actual profile point cloud data; The profile error acquisition unit is configured to construct a component fitting profile reflecting a plurality of basic shape component units based on the pre-processed actual profile point cloud data; calculate the error values of all data points on the component fitting profile to their corresponding target profiles based on the component fitting profile and the aluminum profile section target profile; and obtain the profile error ΔP of the component profile based on the error values; The repeated execution unit is configured to repeatedly execute the point cloud data acquisition unit and the profile error acquisition unit to obtain a preset number of profile errors ΔP of component profiles at different times; The error prediction unit is configured to construct a time series feature matrix based on the preset number of profile errors ΔP of component profiles at different times and process parameters; According to the time series feature matrix, an error trend is predicted by using LSTM combined with an attention mechanism to obtain a prediction error; the process parameters include aluminum rod heating temperature of a processing device, extrusion speed of the processing device, traction of the processing device, die wear index, and processing environment temperature; The process parameter acquisition unit is configured to calculate a process parameter compensation amount based on the prediction error; According to the process parameter compensation amount and the basic process parameter set value, a final applied process parameter is obtained; The execution unit is configured to perform processing according to the final applied process parameter.

7. The system for optimizing the processing precision of aluminum profiles according to claim 6, characterized in that, The point cloud data acquisition unit is specifically configured to perform the following steps: Step S21: After the aluminum profile is cooled and shaped, an optical scanner is used to perform non-contact online scanning on the aluminum profile section to obtain aluminum profile section actual profile point cloud data; Step S22: An abnormal point detection algorithm based on statistical distribution characteristics is used to automatically identify and remove measurement noise points, to obtain pre-processed actual profile point cloud data.

8. The system for optimizing the processing precision of aluminum profiles according to claim 7, characterized in that, The profile error acquisition unit is specifically configured to perform the following steps: Step S3.1: A B-spline curve fitting formula is used to fit the pre-processed actual profile point cloud data to construct a component fitting profile reflecting a plurality of basic shape component units; Step S3.2: Based on the component fitting profile and the aluminum profile section target profile, the normal distance of all data points on the component fitting profile to their corresponding target profiles is calculated to obtain error values; The normal distance calculation formula is as follows: ； in, This represents the normal distance to the i-th data point on the component's fitted contour; ) indicates the component fitting profile on the th i The coordinates of each data point; ) represents the first contour on the target contour that matches the component's fitted contour. i The coordinates of the points corresponding to each data point; Indicates the component fitting profile on the first i At the nth data point, the direction of the fitted contour normal is parallel to the nth data point. i The angle between the directions of the lines connecting each data point to the corresponding point on the target contour; Indicates the component fitting profile on the first i Euclidean distance from each data point to the corresponding point on the target contour; Step S3.3: based on the error value, the profile error ΔP of the component profile is obtained, and the calculation formula is as follows: ； wherein, represents a profile error of the component profile; tm represents the number of all fitted data points on the component profile that participate in the calculation; represents the absolute value of the normal distance of the i th data point.

9. The system for optimizing the processing precision of aluminum profiles according to claim 8, characterized in that, The error prediction unit is specifically configured to perform the following steps: Step S5.1: constructing a time-series feature matrix based on the profile error ΔP of the preset number of profile components at multiple time points and the process parameters collected on site and performing time-series alignment ; ； wherein, is a profile error sequence from the current time t to the historical time k; is a profile error sequence from the current time t to the historical time k+1; is an aluminum bar heating temperature; is an extrusion speed; is a traction force; is an ambient temperature; is a die wear index; Step S5.2: according to the timing feature matrix, the long short-term memory network algorithm model LSTM is used for processing, including: Step S5.2.1: forget gate processing, specifically as follows: ； wherein, is the output of the forget gate, representing the activation value of the forget gate at the current time step t is the output of the forget gate, representing the activation value of the forget gate at the current time step is the output of the forget gate, representing the activation value of the forget gate at the current time step is the output of the forget gate, representing the activation value of the forget gate at the current time step is the output of the forget gate, representing the activation value of the forget gate at the current time step is the output of the forget gate, representing the activation value of the forget gate at the current time step is the output of the forget gate, representing the activation value of the forget gate at the current time step Step S5.2.2: input gate processing, specifically as follows: ； wherein, is the input gate output, representing the activation value of the input gate at the current time t, with a value range of [0, 1]; is the input gate weight matrix; is the forget gate bias term; Step S5.2.3: output gate processing, specifically as follows: ； wherein, is the output of the output gate, representing the activation value of the output gate at the current time t, taking values in the range [0, 1]; is the weight matrix of the output gate; is the bias term of the output gate; Step S5.2.4: cell state and hidden state calculation, specifically as follows: ； wherein, is a candidate cell state, representing new information to be added to the memory at the current time instant t; is a candidate state weight matrix; is a candidate state bias term; ； wherein, is the cell state at the current time t, which is the long-term memory at the current time t; is the cell state at the previous time; is an element-wise multiplication; ； wherein, is the hidden state at the current time t, and the hidden state at the current time t is the short-term memory / output; Step S5.3: attention mechanism weight calculation, including: ； wherein, denotes the hidden state of the history time k at the current time t assigned attention weight, whose value ranges between [0, 1]; history time; T1 is the length of the time series; is the attention energy score measuring the relevance or importance between the current time t and the history time k ; wherein, is the attention energy score measuring the relevance or importance between the current time t and the history time j; wherein, is the attention vector; is the history state weight matrix; is the current state weight matrix; is the history hidden state; is the current hidden state, and T is the matrix transpose. Step S5.4: calculate the prediction error; ； wherein, is the profile error at future time predicted by the prediction model; T 1 is the length of the time series.

10. The system for optimizing the processing precision of aluminum profiles according to claim 9, characterized in that, The process parameter acquisition unit is specifically configured to perform the following steps: Step S6.1: an optimization objective function is constructed with the minimization of the prediction error as the only objective, specifically as follows: ； wherein to optimize the objective, minimize the absolute value of the prediction error; Decision variables: ; Set the constraint condition: ； ； ； ； ； wherein, is a minimum threshold value for the aluminum bar heating temperature, is a maximum threshold value for the aluminum bar heating temperature; is a minimum threshold value for the extrusion speed, is a maximum threshold value for the extrusion speed; is a minimum threshold value for the draft force, is a maximum threshold value for the draft force; is a minimum threshold value for the ambient temperature, is a maximum threshold value for the ambient temperature; is a minimum threshold value for the die wear index, is a maximum threshold value for the die wear index; , , , , representing 5 constraints; Step S6.2: for the objective function constructed in step S6.1, the adaptive gradient descent algorithm is used to solve the process parameters, and the formula is as follows: ； wherein is the iteration number n is the process parameter at iteration number is the iteration number n+1 is the process parameter at iteration number is the learning rate is the gradient of the objective function in step S6.1 at iteration number is satisfied implicitly by the iterative process of gradient descent. where the learning rate is adapted by an adaptive learning rate adjustment to obtain : ； wherein, is an initial learning rate; is a decay coefficient; Step S6.3: PID control-based real-time compensation of process parameters, specifically as follows: ； wherein, is the tth compensation amount at the tth , , are proportional, integral, and derivative coefficients, respectively; is the error at the tth is the target profile error; is the actual profile error at the tth denotes the error at the t-1th denotes the tth j error value in the cumulative error sequence from the initial time j to the tth Final applied process parameters The formula is as follows: ； wherein set values for the base process parameters, .

Images