WO2023226322A1

WO2023226322A1 - Mechanism- and data driving-based engine cylinder head milled surface quality prediction method

Info

Publication number: WO2023226322A1
Application number: PCT/CN2022/131914
Authority: WO
Inventors: 姜兴宇; 李家振; 徐思迪; 刘同明; 陈克强; 邓建超; 杨国哲; 刘伟军
Original assignee: 沈阳工业大学
Priority date: 2022-05-25
Filing date: 2022-11-15
Publication date: 2023-11-30
Also published as: CN114818201A

Abstract

The present invention relates to a mechanism- and data driving-based engine cylinder head milled surface quality prediction method, belonging to the technical field of automated prediction. The method comprises the following steps: determining a key evaluation index and key influence factors for cylinder head milled surface quality; collecting process parameters and a surface roughness value; on the basis of a cylinder head milling mechanical state, constructing milling force and heat mechanism models based on a semi-analytical method and a heat source method; acquiring state variable data, preprocessing same and storing same in a historical database; constructing a surface roughness prediction model based on SVR optimized by an ADE algorithm, the process parameters and the state variable data outputted by the mechanism models being taken as an input of a data driving model and a surface roughness value being taken as an output thereof, and performing training by means of historical data to obtain an SVR optimal parameter combination; and, by means of real-time process parameters and the state variable data outputted by the mechanism models, predicting a milled surface roughness. The method has the advantages of high model state characterization capability and low state variable data acquisition cost, and can realize accurate prediction of cylinder head milled surface quality.

Description

A mechanism-based and data-driven surface quality prediction method for engine cylinder head milling

Technical field

The invention relates to a mechanism- and data-driven surface quality prediction method for engine cylinder head milling, relates to the technical field of cylinder head milling mechanism and data-driven milling surface quality prediction, and belongs to the field of automated prediction technology.

Background technique

As the core component of the diesel engine, the cylinder head is made of vermicular graphite cast iron RuT400 material. It is difficult to process and requires high precision. It is a porous, thin-walled box component. Its processing quality determines the sealing performance and service life of the engine. Milling is the main processing method for cylinder head production, and the surface milling process accounts for more than 60% of the man-hours. Therefore, exploring the surface quality prediction method of cylinder head milling is of great engineering significance to improve the assembly accuracy and sealing performance of the cylinder head and ensure the long-term stable operation of the engine.

Existing research and applications show that the surface quality prediction technology for cylinder head milling has reached a certain level, but the scope of application of existing methods still has certain limitations, and it has not yet reached the level of automated and efficient prediction. In actual production, there are model input variable data types. Single and improper selection of internal parameters of the model lead to low accuracy and long time-consuming prediction results, resulting in cylinder head milling surface quality that does not meet production requirements, poor engine sealing performance, high scrap rate, and serious economic losses.

Contents of the invention

In response to the above technical needs and problems, the present invention aims to provide a mechanism- and data-driven surface quality prediction method for engine cylinder head milling, which can avoid the accuracy of prediction results caused by a single input variable data type of the prediction model and improper selection of internal parameters of the model. Low cost and time consuming problem. First, based on the actual processing conditions of the cylinder head, surface roughness is determined to be the key quality characteristic of cylinder head surface milling, and based on the metal milling surface formation mechanism, the three elements of milling and process state variable data are determined to be the key influencing factors of cylinder head milling surface quality; secondly, consider In view of the difficulty and high cost in obtaining process state variables milling force and milling heat data, a milling force mechanism model based on semi-analytical method and a milling thermal mechanism model based on heat source method were constructed, and the process parameter data, process state variables and their The corresponding milling surface roughness value constitutes a data set and is input to the historical database; again, in the data-driven cylinder head milling surface roughness prediction part, the milling force and milling heat data, process parameters and surface roughness measurements output by the mechanism model are comprehensively considered. value, construct a cylinder head milling surface roughness prediction model based on adaptive differential evolution algorithm to optimize SVR, use process parameter data, milling force, and milling heat data as the input of the surface roughness prediction model, and the corresponding surface roughness value is used as the model output. The prediction model is trained using the data set in the historical database; finally, the real-time milling process parameters and state variable data are substituted into the trained surface roughness prediction model, and the real-time predicted surface roughness value is output, and the combination of process parameters with poor prediction results is output Make timely corrections and adjustments, import data set combinations with higher prediction accuracy into the historical database, and continuously train and update the prediction model parameters to improve the accuracy of the prediction model. By increasing the input variable types of the data-driven prediction model, narrowing the regression model parameter search range, and adaptively adjusting the parameter changes of the intelligent algorithm, the present invention achieves accurate and real-time prediction of the cylinder head milling surface roughness value, and greatly reduces the cost and cost of obtaining state variable data in the milling process. The prediction model running time has the advantages of high prediction accuracy, short model time and low cost.

In order to achieve the above objectives, the present invention provides a mechanism-based and data-driven surface quality prediction method for engine cylinder head milling, which includes the following steps:

(1) According to the production conditions of the cylinder head, surface roughness is determined as the key evaluation index of milling surface quality. The surface roughness of the cylinder head milling is measured using a surface roughness meter, and the three elements of milling are identified based on the metal milling surface formation mechanism: milling speed, each tooth Feed rate, back cutting amount and process state variables milling force and milling heat are the key influencing factors;

(2) Considering the high cost of using sensors to collect milling force and milling heat data in cylinder head production, a milling force mechanism model based on the semi-analytical method and a milling thermal mechanism model based on the heat source method were constructed to explore the cylinder head milling mechanism. The model output is instantaneous milling force and milling heat data based on the current process parameters, and a data set is constructed and stored in the historical database based on the process parameters collected during the milling process, the process state variable data output by the mechanism model, and the measured surface roughness values;

(3) The milling force and milling heat data output by the mechanism model are combined with the current three corresponding milling elements as the input of the data-driven model. The measured surface roughness value of the cylinder head milling is the model output, and a milling optimization SVR based on the adaptive differential evolution algorithm is constructed. Surface roughness prediction model, and initially determine the prediction model parameter range;

(4) Use the milling process to obtain real-time data to predict the cylinder head milling surface roughness value, and compare whether the error between the model output surface roughness value and the actual measured value meets the requirements. If it does not meet the accuracy requirements, return to step (2) and rebuild Mechanism model and data-driven model; if the accuracy requirements are met, it will be further judged whether the surface roughness value meets the production quality requirements, and process parameter combinations with poor prediction results will be adjusted in a timely manner. At the same time, the real-time status data set will be imported into the historical database , used for prediction model training.

In the above-mentioned mechanism- and data-driven engine cylinder head milling surface quality prediction method, the surface roughness is determined as the cylinder head milling surface quality evaluation index in step (1). The reason is that the cylinder head surface roughness measurement cost is low and the state characterization ability is Strong and quantifiable indicators. At the same time, based on the metal surface formation mechanism, the three elements of milling and milling force and milling heat data are identified as the key influencing factors of milling surface quality. Step (2) includes the following sub-steps:

(21) Comprehensively consider the shearing mechanism and plowing mechanism of milling, differentiate and discretize the cutting edges of N blades into M micro-elements along the axial direction, and obtain the k-th cutting edge on the j-th cutting edge in the blade coordinate system. The instantaneous cutting force expression of micro-element;

(22) Convert the micro-element milling force in the milling cutter blade coordinate system to the tool coordinate system, integrate it along the axis direction, and sum the milling forces on the N milling edges to obtain the instantaneous milling force acting on the milling cutter in three directions. force;

(23) Based on the rapid calibration method, the milling force coefficient is identified, and the average value of the three-dimensional milling force measured through several milling experiments is used to calculate the milling force coefficient;

(24) When parameters such as the feed per tooth and the amount of back tool engagement in milling are known, the three-dimensional instantaneous milling force and the resultant milling force on the workpiece surface can be quickly calculated;

(25) Solve the temperature field generated by the instantaneous point heat source through the thermal conduction differential equation, and calculate the temperature field generated when the line heat source is differentially dispersed into countless micro-element points;

(26) The temperature field generated by the moving finite-large surface heat source is derived, and the temperature fields of each surface heat source are superimposed. The mirror method is used to obtain the temperature field boundary condition solution, and the instantaneous temperature at any point of the milling surface temperature field is obtained. Step (3) includes the following sub-steps:

(31) The milling force and milling heat data output from the mechanism model are combined with the current corresponding milling three-element data and normalized using the max-min method. The processed variable data is distributed in the range of [0,1];

(32) Select SVR for regression prediction, and initially determine the three parameter ranges within the support vector regression machine to narrow the search space;

(33) The internal parameter scaling factor F and crossover probability CR of the DE algorithm are adaptively changed according to the following formula, and the adaptive differential evolution algorithm is used to optimize the support vector regression model;

In the formula, the function domain is (-∞, +∞), and the value range is (-1,1). In the formula, CR _max is the maximum crossover probability, CR _min is the minimum crossover probability, and the crossover probability CR is between CR _max and CR changes within _{the min} range,

(34) The internal parameters of the SVR model obtained by optimizing the training data are determined as the best parameter combination, and the cylinder head milling surface roughness value is predicted by inputting real-time process parameters and state variable data.

In the above-mentioned mechanism- and data-driven surface quality prediction method for engine cylinder head milling, the actual sample data of cylinder head production is used to verify the output accuracy of the prediction model, and the prediction is evaluated by analyzing the algorithm operation time and the residual difference between the actual value and the predicted value. Performance of the model. If the accuracy of the prediction model does not meet the error requirements, return to steps (2) and (3) to recalculate the milling mechanism model and train the data-driven model to find the best combination of model parameters.

According to one aspect of the present invention, a mechanism- and data-driven surface quality prediction method for engine cylinder head milling is provided. The prediction method mainly includes the following steps:

(1) According to the engine cylinder head production conditions, surface roughness is determined as the key evaluation index of milling surface quality. The surface roughness measuring instrument is used to measure the cylinder head milling surface roughness, and the three elements of milling are identified based on the metal milling surface formation mechanism: milling speed, The feed per tooth, the amount of back tool engagement, process state variable data, milling force, and milling heat are the key influencing factors;

(2) Considering the high cost of using sensors to collect milling force and milling heat data in cylinder head production, a milling force mechanism model based on the semi-analytical method and a milling thermal mechanism model based on the heat source method were constructed to explore the cylinder head milling mechanism. The model output is the instantaneous milling force and milling heat data calculated based on the current process parameter values. The process parameters, state variable data and measured surface roughness values collected during the milling process are constructed to construct a data set and stored in the historical database;

(3) The milling force and milling heat data output from the mechanism model are combined with the current three corresponding milling elements as the input of the data-driven model. The measured surface roughness value of the cylinder head milling is the model output, and a cylinder optimized SVR based on the adaptive differential evolution algorithm is constructed. Cover milling surface roughness prediction model, determine the model parameter range, and use the data set in the historical database to train the prediction model;

(4) Use real-time milling process data to predict the cylinder head milling surface roughness, and analyze whether the error between the model output surface roughness value and the actual measured value meets the requirements. If it does not meet the accuracy requirements, return to step (2) and rebuild the mechanism. Model and data-driven model; if the accuracy requirements are met, it will be further judged whether the surface roughness value meets the production quality requirements, and the process parameters will be adjusted in a timely manner if the prediction results are not good.

Furthermore, when constructing the milling force mechanism model based on the semi-analytical method, since the milling cutter blade is not absolutely sharp and has a certain width, the shearing mechanism and the plowing mechanism are comprehensively considered, and the milling force is composed of the shearing force and the plowing force. . The cutting edges of N inserts are differentiated and discretized into M micro-elements along the axis direction. Then the instantaneous cutting force of the k-th cutting micro-element on the j-th cutting edge is calculated as:

In the formula, df _tjk , df _rjk , and df _ajk are respectively the tangential, radial, and axial cutting forces of the k-th cutting element on the j-th cutting edge, and K _ts , K _rs , and K _as are the tangential, The radial and axial shear force coefficients, in N/mm ² , are usually related to milling process parameters; K _te , K _re , and _Kae are the tangential, radial, and axial plow shear force coefficients, respectively. The unit is N/mm. dh is the projection of the cutting depth in the axial direction, ds is the arc length of the cutting edge element, f _z is the feed per tooth. g(θ _jk ) is the unit transition function, θ is the cutting angle of the milling cutter blade, when θ _s <θ <θ _e , g(θ _jk )=1, and in other cases g(θ _jk )=0.

The calculation formula for converting the micro-element milling force in the insert coordinate system to the tool coordinate system is:

Integrate the micro-element milling force in the tool coordinate system along the axis direction, and sum the milling forces on the N milling edges to obtain the instantaneous milling force acting on the milling cutter in three directions. The calculation formula is:

According to the milling resultant force calculation formula, the milling resultant force

The larger the number M of milling elements that discretely divides the cutting edge of the milling tool, the more accurate the milling force calculation results will be. The milling force coefficient is an important parameter in the milling force model. Here, the experimental fast calibration method is used to solve the milling force coefficient. When the milling speed and back cutting amount are fixed, the feed parameters of each tooth are changed, the three-dimensional average milling force of a single tooth in one cycle is measured, and the milling force coefficient is identified using linear regression analysis.

Furthermore, when constructing the milling thermal mechanism model based on the heat source method, the cutting heat generated by the work of the shear surface milling force in the first deformation zone is regarded as the only heat source, the shape of the shear surface is regarded as a regular rectangle, and the size of the rectangular surface is related to the tool The diameter is related to the milling depth, and the milling thermal modeling is transformed into a problem of solving the workpiece surface temperature field of a moving finite large-area heat source. According to the idea of the mirror method, another temperature field is used to represent the boundary effect within the investigation boundary, and the temperature fields of the two surface heat sources are superimposed. The uniqueness of the solution indicates that the temperature field is equal to the solution of the boundary value problem. When the mirror heat source is applied to the milling process, the temperature at any point on the surface of the milled workpiece is the superposition of two equal-strength surface heat sources, the real and the mirror. The calculation formula of the surface temperature field is:

Furthermore, in order to improve the accuracy of the support vector regression model and minimize the empirical risk of model training error, ω and b are evaluated by the following formula:

In the formula, C and L are the penalty coefficient and cost function, that is, the C value represents the degree of punishment for data exceeding the L value. The Gaussian radial basis kernel function is usually selected as the kernel function of the prediction model, which can be expressed as:

The support vector regression model penalty coefficient C affects the model complexity and error approximation, and determines the model learning ability. The parameter ε of the cost function L controls the width of the fitted strip of the regression function. The kernel width σ of the Gaussian radial basis kernel function determines the radial action range. Therefore, the three parameter (C, ε, σ) values inside the support vector regression model are used as optimization targets.

Furthermore, in order to reduce the model parameter search range and shorten the parameter optimization time, when determining the parameter C range, first estimate the value of parameter C based on the target value expression, and then compare the estimated value of C with 50% of the average value of the target variable. As a standard deviation-mean normal distribution, the range of parameter C is finally determined on both sides of the estimated value.

C＝max(|y _bar +3σ _t |,|y _bar -3σ _t |) (7)

In the formula, y _bar and σ _t are the standard deviation and mean of the target variable respectively.

The ranges of the remaining two parameters ε and σ are calculated according to the following formula:

In the formula, Z is the number of parameters that affect surface roughness.

Furthermore, in order to improve the performance of the differential evolution algorithm, the scaling factor F value in the algorithm can be expressed as:

In the formula, t is the current number of iterations of the population, and T is the total number of iterations set by the population.

The crossover probability CR parameter value in the algorithm can be expressed as:

In the formula, CR _max is the maximum crossover probability, CR _min is the minimum crossover probability, and the crossover probability CR changes within the range of CR _max and CR _min .

Generally speaking, by comparing the above technical solutions of the present invention with the existing technology, it can be seen that the mechanism-based and data-driven surface quality prediction method for engine cylinder head milling provided by the present invention mainly has the following beneficial effects:

1. Combining the advantages of the mechanism model and the data-driven model, the cylinder head milling surface quality prediction model is more accurate and more interpretable, which has guiding significance for the actual production of cylinder heads.

2. Constructing a milling mechanism model solves the difficulty and high cost of obtaining state variable data in the milling process, making the calculation of milling force and milling heat data more convenient and faster.

3. Input the state variable data calculated and output by the mechanism model combined with the milling process parameters into the support vector regression model for training. Compared with the state data of a single input variable type, the fusion representation ability of multi-type state data input is stronger and can make predictions The results are more accurate.

4. In terms of optimizing the parameters of the support vector regression model, the parameter optimization range has been initially determined, the search time has been shortened, and the adaptive differential evolution algorithm has been used to optimize the model parameters. In the early stage of algorithm operation, it can ensure the diversity of individuals in the population and the Improved global search capabilities can speed up algorithm convergence in the later stages of algorithm operation.

Description of the drawings

Figure 1 is a schematic flowchart of a mechanism- and data-driven surface quality prediction method for engine cylinder head milling provided by an embodiment of the present invention.

Figure 2 shows the indexable milling cutter face milling model diagram.

Figure 3 is a linear regression diagram of the milling force coefficient solved by the experimental calibration method.

Figure 4 is a schematic diagram of the temperature field of a moving finite large-area heat source.

Figure 5 shows the milling experiment plus platform.

Figure 6 is a schematic diagram of the acquisition of force and thermal data in milling experiments.

Figures 7(1) and 7(2) are the milling force and milling heat measurement data graphs respectively.

Figure 8 shows the structure of the surface roughness prediction model.

Figure 9 is a support vector regression diagram.

Figure 10 shows the surface roughness prediction model process.

Figure 11 is a comparison chart between the output value of the surface roughness prediction model and the actual value.

Figure 12 is a residual diagram between the output value of the surface roughness prediction model and the actual value.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings and implementation examples. However, it should be understood that the examples are used to explain the present invention and are not intended to limit the present invention.

As shown in the figure, Figure 1 is a schematic flow chart of a mechanism- and data-driven surface quality prediction method for engine cylinder head milling provided by an embodiment of the present invention. Figure 2 shows the indexable milling cutter face milling model diagram. Figure 3 is a linear regression diagram of the milling force coefficient solved by the experimental calibration method. Figure 4 is a schematic diagram of the temperature field of a moving finite large-area heat source. Figure 5 shows the milling experiment plus platform. Figure 6 is a schematic diagram of the acquisition of force and thermal data in milling experiments. Figures 7(1) and 7(2) are the milling force and milling heat measurement data graphs respectively. Figure 8 shows the structure of the surface roughness prediction model. Figure 9 is a support vector regression diagram. Figure 10 shows the surface roughness prediction model process. Figure 11 is a comparison chart between the output value of the surface roughness prediction model and the actual value. Figure 12 is a residual diagram between the output value of the surface roughness prediction model and the actual value.

The present invention provides a mechanism- and data-driven surface quality prediction method for engine cylinder head milling, which includes the following steps:

According to one aspect of the present invention, the present invention provides a mechanism- and data-driven surface quality prediction method for engine cylinder head milling. The provided monitoring method is completed in the following steps: First, determine the surface roughness according to the production conditions of the engine cylinder head. As a key evaluation index of milling surface quality, the cylinder head milling surface roughness is measured, and the three elements of milling are identified based on the metal milling surface formation mechanism: milling speed, feed per tooth, back cutting amount and process state variable data milling force, milling Heat is the key influencing factor; secondly, considering the high cost of using sensors to collect milling force and milling heat data in cylinder head production, a milling force mechanism model based on the semi-analytical method and a milling thermal mechanism model based on the heat source method were constructed to explore the cylinder Cover milling mechanism, the output of the mechanism model is instantaneous milling force and milling heat data calculated based on the current process parameter values, and a data set of process parameters, state variable data and measured surface roughness values collected during the milling process is stored in the historical database. Middle; Thirdly, the milling force and milling heat data output from the mechanism model are combined with the current three corresponding milling elements as the data-driven model input, and the cylinder head milling surface roughness value is the model output to construct a cylinder head optimized SVR based on the adaptive differential evolution algorithm. Milling surface roughness prediction model, initially determine the model parameter range, and use the data set in the historical database to train the prediction model; finally, use real-time milling process data to predict the cylinder head milling surface roughness, and compare the model output surface roughness value with the actual measured value Whether the error between them meets the requirements. If it does not meet the accuracy requirements, the mechanism model and the data-driven model will be rebuilt. If it meets the accuracy requirements, it will be further judged whether the surface roughness value meets the production quality requirements, and the poor prediction results will be reported in a timely manner. Adjust process parameters.

Specifically, the present invention's mechanism- and data-driven surface quality prediction method for engine cylinder head milling mainly includes the following steps:

C＝max(|y _bar +3σ _t |,|y _bar -3σ _t |) (7)

In the formula, Z is the number of parameters that affect surface roughness.

In order to verify the method proposed in this invention, taking the milling of the bottom surface of the engine cylinder head as an example, the surface roughness is determined as the key evaluation index of milling surface quality, and the three elements of milling are identified: milling speed, feed per tooth, back cutting amount and process state variables. Data milling force and milling heat are the key factors affecting the quality of cylinder head bottom surface milling.

The surface roughness value of the cylinder head milling corresponding to the process parameters was measured using a surface roughness meter, and the real-time milling force and milling heat data were calculated through the constructed milling force and milling thermal mechanism model. Store 100 sets of milling process state variable data, milling three element parameter values and surface roughness measurement values obtained from the collection and mechanism model output to the historical database. In order to ensure the availability of data in the prediction model, the data is preprocessed after obtaining the data. The preprocessing results are shown in Table 1.

Table 1 Example of sample data of prediction model after preprocessing

The preprocessed data set is used to train the surface roughness prediction model to obtain the best internal parameter combination of the model. In the mechanism- and data-driven surface quality prediction method for engine cylinder head milling adopted in this invention, the SVR optimized by the adaptive differential evolution algorithm is used to predict the surface roughness value of the cylinder head bottom surface milling. The model training and verification platform is MATLAB. In 2021a, the computer parameters are i5-6200U processor and the memory is 2.30GHz.

For the surface milling of the cylinder head, the number of factors affecting surface roughness Z = 5, that is, the milling speed v _c , the feed per tooth f _z , the amount of back cutting a _p and the milling force and milling heat data. According to equations (7) and (8), when initially determining the values of SVR internal parameters C, ε, and σ, the calculated estimated value of C is 4.9561. The preliminary range of SVR internal parameters is determined as shown in Table 2.

Table 2 SVR parameter range

In terms of determining the internal parameter scaling factor F value of the adaptive differential evolution algorithm, equation (9) is used to fix the nonlinear F value range within 0.4-1, retaining the advantage of smooth transition of the change curve, so that the scaling factor F can be used in the early stage of the algorithm iteration. Take a larger value to maintain the diversity of individuals in the population, and take a smaller value in the later local optimization process to speed up the convergence of the algorithm.

In terms of determining the cross probability CR value of the internal parameter of the adaptive differential evolution algorithm, according to Equation (10), the minimum crossover probability CR _min =0.3 and the maximum crossover probability CR _max =0.9 are initially set to ensure that the crossover probability CR is within the range of 0.3 to 0.9. In the early stage of iteration, CR takes a smaller value to improve the global search capability. As the number of iterations t increases, the CR value increases nonlinearly, speeding up the convergence speed and helping to determine the optimal individual.

When using SVR optimized based on the adaptive differential evolution algorithm to predict the surface roughness value, the algorithm population size NP is set to 50 and the number of iterations M is set to 200. After training 100 sets of historical data sets, the optimal internal parameter combination of SVR was obtained through global search optimization using the adaptive differential evolution algorithm as C=8.6375, ε=0.0682, and σ=0.7351.

The trained milling surface roughness prediction model was input into real-time process parameters and mechanism model output data. The output results showed that the prediction model achieved good prediction results on real-time data. The error between the output results of the prediction model and the actual milling surface roughness value is shown in Figures 11 and 12.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the protection scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should analyze , the technical solution of the present invention may be modified or equivalently substituted without departing from the essence and scope of the technical solution of the present invention.

Claims

A mechanism- and data-driven surface quality prediction method for engine cylinder head milling, which is characterized by including the following steps:

(1) According to the production conditions of the cylinder head, surface roughness is determined as the key evaluation index of milling surface quality. The surface roughness of the cylinder head milling is measured using a surface roughness meter, and the three elements of milling are identified based on the metal milling surface formation mechanism: milling speed, each tooth Feed rate, back cutting amount and process state variables milling force and milling heat are the key influencing factors;

(2) Considering the high cost of using sensors to collect milling force and milling heat data in cylinder head production, a milling force mechanism model based on the semi-analytical method and a milling thermal mechanism model based on the heat source method were constructed to explore the cylinder head milling mechanism. The model output is instantaneous milling force and milling heat data based on the current process parameters, and a data set is constructed and stored in the historical database based on the process parameters collected during the milling process, the process state variable data output by the mechanism model, and the measured surface roughness values;

(3) The milling force and milling heat data output by the mechanism model are combined with the current three corresponding milling elements as the input of the data-driven model. The measured surface roughness value of the cylinder head milling is the model output, and a milling optimization SVR based on the adaptive differential evolution algorithm is constructed. Surface roughness prediction model, and initially determine the prediction model parameter range;

(4) Use the milling process to obtain real-time data to predict the cylinder head milling surface roughness value, and compare whether the error between the model output surface roughness value and the actual measured value meets the requirements. If it does not meet the accuracy requirements, return to step (2) and rebuild Mechanism model and data-driven model; if the accuracy requirements are met, it will be further judged whether the surface roughness value meets the production quality requirements, and process parameter combinations with poor prediction results will be adjusted in a timely manner. At the same time, the real-time status data set will be imported into the historical database , used for prediction model training.
The mechanism- and data-driven surface quality prediction method for engine cylinder head milling as claimed in claim 1, characterized in that: in step (1), the surface roughness is determined as the cylinder head milling surface quality evaluation index because the cylinder head surface is rough. The measurement cost is low, the state characterization ability is strong, and it is a quantifiable indicator. At the same time, the identification of the three elements of milling and milling force and milling heat data based on the metal surface formation mechanism are the key influencing factors of milling surface quality.
The mechanism- and data-driven surface quality prediction method for engine cylinder head milling as claimed in claim 1, wherein step (2) includes the following sub-steps:

(21) Comprehensively consider the shearing mechanism and plowing mechanism of milling, differentiate and discretize the cutting edges of N blades into M micro-elements along the axial direction, and obtain the k-th cutting edge on the j-th cutting edge in the blade coordinate system. The instantaneous cutting force expression of micro-element;

(22) Convert the micro-element milling force in the milling cutter blade coordinate system to the tool coordinate system, integrate it along the axis direction, and sum the milling forces on the N milling edges to obtain the instantaneous milling force acting on the milling cutter in three directions. force;

(23) Based on the rapid calibration method, the milling force coefficient is identified, and the average value of the three-dimensional milling force measured through several milling experiments is used to calculate the milling force coefficient;

(24) When parameters such as the feed per tooth and the amount of back tool engagement in milling are known, the three-dimensional instantaneous milling force and the resultant milling force on the workpiece surface can be quickly calculated;

(25) Solve the temperature field generated by the instantaneous point heat source through the thermal conduction differential equation, and calculate the temperature field generated when the line heat source is differentially dispersed into countless micro-element points;

(26) The temperature field generated by the moving finite-large surface heat source is derived, and the temperature fields of each surface heat source are superimposed. The mirror method is used to obtain the temperature field boundary condition solution, and the instantaneous temperature at any point of the milling surface temperature field is obtained.
The mechanism- and data-driven surface quality prediction method for engine cylinder head milling as claimed in claim 1, wherein step (3) includes the following sub-steps:

(31) The milling force and milling heat data output from the mechanism model are combined with the current corresponding milling three-element data and normalized using the max-min method. The processed variable data is distributed in the range of [0,1];

(32) Select SVR for regression prediction, and initially determine the three parameter ranges within the support vector regression machine to narrow the search space;

(33) The internal parameter scaling factor F and crossover probability CR of the DE algorithm are adaptively changed according to the following formula, and the adaptive differential evolution algorithm is used to optimize the support vector regression model;

In the formula, the function domain is (-∞, +∞), and the value range is (-1,1). In the formula, CR max is the maximum crossover probability, CR min is the minimum crossover probability, and the crossover probability CR is between CR max and CR changes within the min range,

(34) The internal parameters of the SVR model obtained by optimizing the training data are determined as the best parameter combination, and the cylinder head milling surface roughness value is predicted by inputting real-time process parameters and state variable data.
The mechanism- and data-driven surface quality prediction method for engine cylinder head milling as claimed in claim 1, characterized in that: the actual sample data of cylinder head production is used to verify the output accuracy of the prediction model, and the calculation time and actual value of the algorithm are compared with the prediction The residual error between values is used to evaluate the performance of the prediction model. If the accuracy of the prediction model does not meet the error requirements, return to steps (2) and (3) to recalculate the milling mechanism model and train the data-driven model to find the best combination of model parameters. .