Cylindrical surface error process tracing method based on multi-scale features and neural network

Abstract

The invention provides a cylindrical surface error process tracing method based on multi-scale features and a neural network, and the method comprises the steps: carrying out the J-layer decomposition of an error signal through a multi-scale feature extraction technology, precisely separating the low-frequency machine tool spindle / clamping error, the medium-frequency tool wear / cutting vibration, and the high-frequency surface roughness error components, and carrying out the processing of the cylindrical surface error. And each frequency component characteristic is quantified through energy, statistics and shape characteristics, so that the technical bottleneck that a traditional method cannot distinguish error sources is solved, the error traceability accuracy is improved, the limitation that the traditional method can only obtain error values but cannot trace the source is overcome, error traceability is upgraded from qualitative judgment to quantitative analysis, and the error traceability efficiency is improved. The accurate basis is provided for targeted optimization of process parameters, meanwhile, the trial and error cost is reduced, the process development period is shortened, and a systematic solution is provided for efficient and high-quality manufacturing of precise cylindrical parts.

Description

Technical Field

[0001] This invention relates to a method for tracing the process source of errors on cylindrical surfaces based on multi-scale features and neural networks, belonging to the field of machining quality control and process optimization technology. Background Technology

[0002] Precision cylindrical parts, as core components in mechanical transmission and hydraulic systems, have surface errors that directly affect the product's fitting accuracy, motion stability, and service life, especially in critical fields such as automotive engine cylinder liners and aerospace bearings. Traditional cylindrical surface error detection and process traceability technologies are no longer sufficient to meet the demands of high-precision manufacturing: On the one hand, existing detection methods mostly focus on quantifying error amplitudes, such as maximum radial error and roundness error, failing to analyze the multi-frequency characteristics contained in the error signal. This makes it impossible to distinguish between errors from different sources, such as low-frequency machine tool spindle errors, medium-frequency tool wear, and high-frequency surface roughness, making it difficult to accurately pinpoint the root cause of quality problems. On the other hand, the correlation analysis between errors and process parameters heavily relies on expert experience. When faced with scenarios involving the coupling of multiple parameters such as cutting speed, feed rate, and tool angle, it is difficult to establish a quantitative mapping relationship, leading to blind process optimization, high trial-and-error costs, and long cycles.

[0003] In existing related technologies, some methods attempt to predict processing errors through machine learning. However, because they do not perform multi-scale decomposition of the error signals, they cannot effectively extract features of different frequency components, resulting in limited ability of the models to identify complex error patterns and an accuracy rate of less than 70% in tracing the source. At the same time, traditional methods lack systematic analysis of the correlation between error features and fit performance, making it difficult to identify key influencing factors, which further restricts the efficiency of process optimization.

[0004] To address the aforementioned issues, there is an urgent need for an intelligent traceability method capable of accurately separating error frequency components and quantifying the relationship between features and process parameters. Therefore, this paper proposes a process traceability method for cylindrical surface errors based on multi-scale features and neural networks. This method analyzes the sources of error through multi-scale decomposition and constructs a mapping model between features and process parameters using neural networks, achieving closed-loop control from error detection to process optimization to meet the technical requirements of high-precision manufacturing. Summary of the Invention

[0005] In view of this, the present invention provides a method for tracing the process of cylindrical surface errors based on multi-scale features and neural networks, so as to solve or alleviate the technical problems existing in the prior art, and at least provide a beneficial option.

[0006] The technical solution of this invention is implemented as follows: a method for tracing the process source of errors on cylindrical surfaces based on multi-scale features and neural networks, comprising the following steps: Step 1: Data Acquisition and Preprocessing. Use a coordinate measuring machine to acquire the coordinate data of the measurement points on the cylindrical surface; use the least squares method to fit the ideal cylindrical surface to solve for the cylinder's axis coordinates and radius; calculate the radial error of each measurement point and generate an error sequence. Step 2: Multi-scale feature extraction. Daubechies wavelet is selected as the mother wavelet function. J-level wavelet transform decomposition is performed on the error sequence to obtain components of different frequency bands. Multi-scale features of each component are calculated. The multi-scale features include energy features, statistical features and shape features. All features are integrated to construct a comprehensive feature vector. Step 3: Construct a neural network process traceability model. Collect machining samples under different process parameters, including cutting speed, feed rate, depth of cut, side depth of cut, rake angle, and clearance angle. Construct a three-layer feedforward neural network, using the comprehensive feature vector as input, and output the predicted values ​​of the process parameters. Train the neural network model using the backpropagation algorithm. Step 4: Error sensitivity analysis, calculate the single-factor sensitivity coefficient and interaction sensitivity coefficient of each multi-scale feature on the fit performance, and determine the key error features based on the sensitivity coefficients; Step 5: Optimize process parameters. Establish a multi-objective optimization model with quality, cost, and time as objective functions. Set process parameter constraints and quality constraints, and use a genetic algorithm to solve the optimization model to output the optimal process parameters.

[0007] More preferably, in step one, a coordinate measuring machine is used to precisely measure the surface of the machined cylinder to obtain the spatial coordinate information of a series of discrete measurement points. Each measurement point is represented by a three-dimensional coordinate system. This represents the total number of measurement points, where i = 1, 2, ..., N. The ideal cylindrical surface is fitted using the least squares method, and its equation is: ,in Let R be the coordinates of the axis of the ideal cylinder, and R be the ideal radius. To make the fitting result closest to the actual measurement point, the objective function is set as follows: The objective function is to minimize the sum of squares of the deviations from the ideal cylindrical surface at each measurement point. By solving this objective function, the coordinates of the ideal cylinder's axis can be determined. and radius R.

[0008] More preferably, in step one, the radial error at each measurement point is calculated using the fitted ideal cylindrical surface as a reference. ;in It is the actual distance from the measured point to the ideal axis, minus the ideal radius R, to obtain the result. This refers to the radial deviation of the point from the ideal cylindrical surface. The radial errors of all measured points are arranged sequentially to form an error sequence. .

[0009] Further preferably, in step two, the Daubechies wavelet is selected as the mother wavelet function to perform J-level decomposition on the radial error sequence ε(t), and the decomposition result is expressed as: ; in The low-frequency components of the error sequence corresponding to the J-th layer approximation components are mainly caused by macroscopic factors such as machine tool spindle error and clamping error. The high-frequency components corresponding to the j-th layer detail components are mainly caused by microscopic factors such as tool wear, cutting vibration, and surface roughness. During the decomposition process, the calculation of each layer component depends on approximation coefficients and detail coefficients, where the approximation coefficients... Detail coefficient The scaling function and wavelet function By scaling and translation operations, signals of different frequencies and positions can be captured; based on the number of decomposition layers and the sampling frequency fs, the frequency band range of each layer component can be determined.

[0010] Further preferably, in step two, after obtaining each frequency component through multi-scale decomposition, error features need to be extracted to quantify the key characteristics of each component and provide structured input for subsequent process traceability. Specifically, energy features are used to quantify the intensity of each frequency band component, calculated as the energy of the j-th layer detail. , and approximate component energy , Higher energy indicates a more significant impact from the corresponding error source. Statistical characteristics describe the patterns of each component from the perspective of probability distribution, and the calculations include the mean. Reflecting the central tendency and variance of the data , Measuring the dispersion and skewness of data Characterizing the asymmetry and kurtosis of data distribution Describes the degree of peaks or flattening in the data distribution; Shape features are used to describe the impact characteristics and local anomalies of waveforms. Including peak factor Used to evaluate the impulse characteristics and impulse factor of a signal. Further quantization of the signal's pulse strength, margin factor It is more sensitive to minute changes in signals and is often used for early fault diagnosis; finally, all the above features are integrated to form a comprehensive feature vector. This vector comprehensively covers the intensity, distribution pattern, and waveform characteristics of each frequency component, providing accurate input for constructing the mapping relationship between error characteristics and process parameters.

[0011] More preferably, in step three, the process parameter vector is: Where v is the cutting speed and f is the feed rate. To measure the amount of knife cuts on the back, γ represents the depth of cut, α represents the rake angle, and α represents the clearance angle.

[0012] More preferably, in step three, the hidden layer expression of the three-layer feedforward neural network is: The output layer expression is , The loss function is ,in This is the weight matrix. Let M be the bias vector and M be the number of samples. This is a vector of actual process parameters. This is for predicting process parameter vectors.

[0013] More preferably, in step four, the compatibility performance index... ,in For the maximum radial error, Surface roughness, and roundness error; the single-factor sensitivity coefficient Interaction sensitivity coefficient The key error characteristics satisfy .

[0014] More preferably, in step five, the objective function of the multi-objective optimization model is: ,in , , The weighting coefficients and The initial population for the genetic algorithm is: fitness is Crossover probability Probability of mutation .

[0015] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions: I. This invention utilizes multi-scale feature extraction technology and Daubechies wavelet to perform J-level decomposition on the error signal, accurately separating low-frequency machine tool spindle / clamping errors, mid-frequency tool wear / cutting vibrations, and high-frequency surface roughness error components. Furthermore, it quantifies the characteristics of each frequency component through energy, statistics, and shape features, overcoming the technical bottleneck of traditional methods' inability to distinguish error sources. This improves the accuracy of error tracing, achieving a leap from error amplitude detection to precise error source location. It overcomes the limitation of traditional methods that can only obtain error values ​​but cannot trace the root cause, upgrading error tracing from qualitative judgment to quantitative analysis, and providing precise basis for targeted optimization of process parameters.

[0016] Second, this invention combines a three-layer feedforward neural network to construct a nonlinear mapping model between error features and process parameters, and uses sensitivity analysis to identify key influencing factors. Finally, it achieves multi-objective optimization of quality, cost, and time through a genetic algorithm, improving the efficiency of process parameter optimization, increasing product qualification rate, reducing trial and error costs, and shortening the process development cycle. This forms a closed-loop control system for error detection, feature extraction, and source tracing optimization, providing a systematic solution for the efficient and high-quality manufacturing of precision cylindrical parts.

[0017] The above overview is for illustrative purposes only and is not intended to be limiting in any way. Further aspects, embodiments, and features of the invention will become readily apparent from the following detailed description, in addition to the illustrative aspects, embodiments, and features described above. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation

[0020] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0021] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] like Figure 1As shown, this embodiment of the invention provides a method for tracing the process source of errors on cylindrical surfaces based on multi-scale features and neural networks, including the following steps: Step 1: Data Acquisition and Preprocessing. Use a coordinate measuring machine to acquire the coordinate data of the measurement points on the cylindrical surface; use the least squares method to fit the ideal cylindrical surface to solve for the cylinder's axis coordinates and radius; calculate the radial error of each measurement point and generate an error sequence. Step 2: Multi-scale feature extraction. Daubechies wavelet is selected as the mother wavelet function. J-level wavelet transform decomposition is performed on the error sequence to obtain components of different frequency bands. Multi-scale features of each component are calculated. The multi-scale features include energy features, statistical features and shape features. All features are integrated to construct a comprehensive feature vector. Step 3: Construct a neural network process traceability model. Collect machining samples under different process parameters, including cutting speed, feed rate, depth of cut, side depth of cut, rake angle, and clearance angle. Construct a three-layer feedforward neural network, using the comprehensive feature vector as input, and output the predicted values ​​of the process parameters. Train the neural network model using the backpropagation algorithm. Step 4: Error sensitivity analysis, calculate the single-factor sensitivity coefficient and interaction sensitivity coefficient of each multi-scale feature on the fit performance, and determine the key error features based on the sensitivity coefficients; Step 5: Optimize process parameters. Establish a multi-objective optimization model with quality, cost, and time as objective functions. Set process parameter constraints and quality constraints, and use a genetic algorithm to solve the optimization model to output the optimal process parameters.

[0023] In one embodiment, in step one, a coordinate measuring machine is used to precisely measure the surface of the machined cylinder to obtain the spatial coordinate information of a series of discrete measurement points. Each measurement point is represented by a three-dimensional coordinate system. This represents the total number of measurement points, where i = 1, 2, ..., N. The ideal cylindrical surface is fitted using the least squares method, and its equation is: ,in Let R be the coordinates of the axis of the ideal cylinder, and R be the ideal radius. To make the fitting result closest to the actual measurement point, the objective function is set as follows: The objective function is to minimize the sum of squares of the deviations from the ideal cylindrical surface at each measurement point. By solving this objective function, the coordinates of the ideal cylinder's axis can be determined. And radius R. In step one, using the fitted ideal cylindrical surface as a reference, the radial error of each measurement point is calculated. ;in It is the actual distance from the measured point to the ideal axis, minus the ideal radius R, to obtain the result. This refers to the radial deviation of the point from the ideal cylindrical surface. The radial errors of all measured points are arranged sequentially to form an error sequence. ; Taking the machining of engine cylinder liner material, gray cast iron HT250, as an example, the specific implementation of step one is as follows: 200 measurement points are uniformly collected on the outer cylindrical surface of the cylinder liner using a coordinate measuring machine. The coordinates are represented as follows: The ideal cylindrical surface is fitted using the least squares method, and the objective function is: Solving for the axis yields Radius R = 50 mm; Calculate radial error To form an error sequence .

[0024] In one embodiment, in step two, the Daubechies wavelet is selected as the mother wavelet function to perform J-level decomposition on the radial error sequence ε(t), and the decomposition result is expressed as: ; in The low-frequency components of the error sequence corresponding to the J-th layer approximation components are mainly caused by macroscopic factors such as machine tool spindle error and clamping error. The high-frequency components corresponding to the j-th layer detail components are mainly caused by microscopic factors such as tool wear, cutting vibration, and surface roughness. During the decomposition process, the calculation of each layer component depends on approximation coefficients and detail coefficients, where the approximation coefficients... Detail coefficient The scaling function and wavelet function By scaling and translation operations, signals of different frequencies and locations can be captured. Based on the number of decomposition layers and the sampling frequency fs, the frequency band range of each component can be determined. In step two, after obtaining each frequency component through multi-scale decomposition, error features need to be further extracted to quantify the key characteristics of each component, providing structured input for subsequent process traceability. Specifically, energy features are used to quantify the intensity of each frequency band component, calculated as the detail energy of the j-th layer. , and approximate component energy , Higher energy indicates a more significant impact from the corresponding error source. Statistical characteristics describe the patterns of each component from the perspective of probability distribution, and the calculations include the mean. Reflecting the central tendency and variance of the data , Measuring the dispersion and skewness of data Characterizing the asymmetry and kurtosis of data distribution Describes the degree of peaks or flattening in the data distribution; Shape features are used to describe the impact characteristics and local anomalies of waveforms. Including peak factor Used to evaluate the impulse characteristics and impulse factor of a signal. Further quantization of the signal's pulse strength, margin factor It is more sensitive to minute changes in signals and is often used for early fault diagnosis; finally, all the above features are integrated to form a comprehensive feature vector. This vector comprehensively covers the intensity, distribution pattern, and waveform characteristics of each frequency component, providing accurate input for constructing the mapping relationship between error characteristics and process parameters.

[0025] Taking the machining of gray cast iron HT250 as an example, the multi-scale feature extraction in step two specifically involves selecting the Daubechies 4 wavelet to perform a 5-level decomposition on the error sequence to obtain the low-frequency approximate components. and high-frequency detail components Frequency band division For 200-400Hz, For 100-200Hz, For 50-100Hz, For 25-50Hz, 12.5−25Hz The frequency is 0-12.5 Hz. Calculate energy characteristics, ; Statistical characteristics ; Shape characteristics, ; Constructing a 30-dimensional comprehensive feature vector .

[0026] In one embodiment, in step three, the process parameter vector is: Where v is the cutting speed and f is the feed rate. To measure the amount of knife cuts on the back, Let γ be the depth of cut, α be the rake angle, and α be the rake angle. In step three, the hidden layer expression of the three-layer feedforward neural network is: The output layer expression is , The loss function is ,in This is the weight matrix. Let M be the bias vector and M be the number of samples. This is a vector of actual process parameters. This is for predicting process parameter vectors.

[0027] Taking the machining of gray cast iron HT250 as an example, the neural network model construction in step three involves collecting 100 sets of process parameter samples, where v = 100−150m / min and f = 0.1−0.2mm / r. A three-layer neural network was constructed: an input layer with a 30-dimensional feature vector F, a hidden layer with 50 neurons and activation functions tanh, and an output layer with 6-dimensional process parameters. The model was trained using the backpropagation algorithm with a learning rate η=0.01. After 1000 iterations, the loss function L=0.008, and the model converged.

[0028] In one embodiment, in step four, the compatibility performance index ,in For the maximum radial error, Surface roughness, and roundness error; the single-factor sensitivity coefficient Interaction sensitivity coefficient The key error characteristics satisfy .

[0029] Taking the machining of gray cast iron HT250 as an example, the sensitivity analysis results of step four are as follows: Layer energy Single-factor sensitivity coefficient This is the key feature corresponding to the tool wear error source; Interaction sensitivity coefficient with cutting speed v .

[0030] In one embodiment, in step five, the objective function of the multi-objective optimization model is: , , The weighting coefficients and The initial population for the genetic algorithm is: fitness is Crossover probability Probability of mutation .

[0031] Taking the machining of gray cast iron HT250 as an example, the optimization result of step five is as follows: Objective function weights, quality weights Cost weight Time weight ; After 100 iterations of the genetic algorithm, the optimal process parameters are v = 135 m / min and f = 0.12 mm / r. =0.8mm; After optimization, the cylinder liner roundness error decreased from 8μm to 5μm, and the surface roughness... The thickness was reduced from 3.2μm to 2.5μm, resulting in a 15% increase in the pass rate.

[0032] In operation, this invention separates the frequency components of the error signal through multi-scale decomposition, extracts energy, statistical, and shape features to construct vectors, establishes a mapping between features and process parameters using neural networks, locates key error sources by combining sensitivity analysis, and finally outputs the optimal process parameters through a genetic algorithm, thus realizing intelligent operation of the entire process from error detection to process optimization.

[0033] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for tracing the process source of errors on cylindrical surfaces based on multi-scale features and neural networks, characterized in that: Includes the following steps, Step 1: Data Acquisition and Preprocessing. Use a coordinate measuring machine to acquire the coordinate data of the measurement points on the cylindrical surface; use the least squares method to fit the ideal cylindrical surface to solve for the cylinder's axis coordinates and radius; calculate the radial error of each measurement point and generate an error sequence. Step 2: Multi-scale feature extraction. Daubechies wavelet is selected as the mother wavelet function. J-level wavelet transform decomposition is performed on the error sequence to obtain components of different frequency bands. Multi-scale features of each component are calculated. The multi-scale features include energy features, statistical features and shape features. All features are integrated to construct a comprehensive feature vector. Step 3: Construct a neural network process traceability model. Collect machining samples under different process parameters, including cutting speed, feed rate, depth of cut, side depth of cut, rake angle, and clearance angle. Construct a three-layer feedforward neural network, using the comprehensive feature vector as input, and output the predicted values ​​of the process parameters. Train the neural network model using the backpropagation algorithm. Step 4: Error sensitivity analysis, calculate the single-factor sensitivity coefficient and interaction sensitivity coefficient of each multi-scale feature on the fit performance, and determine the key error features based on the sensitivity coefficients; Step 5: Optimize process parameters. Establish a multi-objective optimization model with quality, cost, and time as objective functions. Set process parameter constraints and quality constraints, and use a genetic algorithm to solve the optimization model to output the optimal process parameters.

2. The method for tracing the process source of cylindrical surface errors based on multi-scale features and neural networks according to claim 1, characterized in that: In step one, a coordinate measuring machine is used to precisely measure the surface of the machined cylinder, obtaining the spatial coordinate information of a series of discrete measurement points. Each measurement point is represented by a three-dimensional coordinate system. This represents the total number of measurement points, where i = 1, 2, ..., N. The ideal cylindrical surface is fitted using the least squares method, and its equation is: ,in Let R be the coordinates of the axis of the ideal cylinder, and R be the ideal radius. To make the fitting result closest to the actual measurement point, the objective function is set as follows: The objective function is to minimize the sum of squares of the deviations from the ideal cylindrical surface at each measurement point. By solving this objective function, the coordinates of the ideal cylinder's axis can be determined. and radius R.

3. The method for tracing the process source of cylindrical surface errors based on multi-scale features and neural networks according to claim 2, characterized in that: In step one, using the fitted ideal cylindrical surface as a reference, the radial error at each measurement point is calculated. ;in It is the actual distance from the measured point to the ideal axis, minus the ideal radius R, to obtain the result. This refers to the radial deviation of the point from the ideal cylindrical surface. The radial errors of all measured points are arranged sequentially to form an error sequence. .

4. The method for tracing the process source of cylindrical surface errors based on multi-scale features and neural networks according to claim 1, characterized in that: In step two, the Daubechies wavelet is selected as the mother wavelet function to perform J-level decomposition on the radial error sequence ε(t). The decomposition result is expressed as: ； in The low-frequency components of the error sequence corresponding to the J-th layer approximation components are mainly caused by macroscopic factors such as machine tool spindle error and clamping error. The high-frequency components corresponding to the j-th layer detail components are mainly caused by microscopic factors such as tool wear, cutting vibration, and surface roughness. During the decomposition process, the calculation of each layer component depends on approximation coefficients and detail coefficients, where the approximation coefficients... Detail coefficient The scaling function and wavelet function By scaling and translation operations, signals of different frequencies and positions can be captured; based on the number of decomposition layers and the sampling frequency fs, the frequency band range of each layer component can be determined.

5. The method for tracing the process source of cylindrical surface errors based on multi-scale features and neural networks according to claim 4, characterized in that: In step two, after obtaining each frequency component through multi-scale decomposition, it is necessary to further extract error features to quantify the key characteristics of each component and provide structured input for subsequent process traceability. Specifically, this includes energy features used to quantify the intensity of each frequency band component, calculated as the energy of the j-th layer detail. ， and approximate component energy ， Higher energy indicates a more significant impact from the corresponding error source. Statistical characteristics describe the patterns of each component from the perspective of probability distribution, and the calculations include the mean. Reflecting the central tendency and variance of the data ， Measuring the dispersion and skewness of data Characterizing the asymmetry and kurtosis of data distribution Describes the degree of peaks or flatness in the data distribution; Shape features are used to describe the impact characteristics and local anomalies of waveforms. Including peak factor Used to evaluate the impulse characteristics and impulse factor of a signal. Further quantization of the signal's pulse strength, margin factor It is more sensitive to minute changes in signals and is often used for early fault diagnosis; finally, all the above features are integrated to form a comprehensive feature vector. This vector comprehensively covers the intensity, distribution pattern, and waveform characteristics of each frequency component, providing accurate input for constructing the mapping relationship between error characteristics and process parameters.

6. The method for tracing the process source of cylindrical surface errors based on multi-scale features and neural networks according to claim 1, characterized in that: In step three, the process parameter vector is: Where v is the cutting speed and f is the feed rate. To measure the amount of knife cuts on the back, γ represents the depth of cut, α represents the rake angle, and α represents the clearance angle.

7. The method for tracing the process source of cylindrical surface errors based on multi-scale features and neural networks according to claim 1, characterized in that: In step three, the hidden layer expression of the three-layer feedforward neural network is: The output layer expression is , The loss function is ,in This is the weight matrix. Let M be the bias vector and M be the number of samples. This is a vector of actual process parameters. This is for predicting process parameter vectors.

8. The method for tracing the process source of cylindrical surface errors based on multi-scale features and neural networks according to claim 1, characterized in that: In step four, the matching performance index ,in For the maximum radial error, Surface roughness is defined as Surface roughness, and Roundness is defined as roundness error; the single-factor sensitivity coefficient is... Interaction sensitivity coefficient The key error characteristics satisfy .

9. The method for tracing the process error of a cylindrical surface based on multi-scale features and neural networks according to claim 1, characterized in that: In step five, the objective function of the multi-objective optimization model is: ，. in , ， The weighting coefficients and The initial population for the genetic algorithm is: fitness is Crossover probability Probability of mutation .

Images