Machining process optimization control method for supporting shaft of drive axle of mining truck

Abstract

The invention belongs to the technical field of process optimization control, and particularly relates to a machining process optimization control method for a supporting shaft of a drive axle of a mining truck. The problems that existing machining depends on empirical parameters, quality fluctuation is large, and traditional PID control cannot dynamically adapt to working conditions are solved. The method comprises the following steps: collecting control parameters such as cutting speed and feeding amount and corresponding supporting shaft mass fractions, and constructing a sample data set; preprocessing the data to eliminate abnormal values and complement missing values; calculating influence weights of the parameters on quality, distinguishing key and non-key parameters, and identifying sensitive intervals of the key parameters; solving an optimal parameter combination by using a non-dominated sorting genetic algorithm in combination with the sensitive interval; and the optimal parameter is used as a PID set value, and the improved PID controller dynamic setting fused with the real-time quality deviation is designed. According to the method, experience dependence is replaced by data driving, the machining quality stability and consistency are improved, and the method adapts to complex working conditions.

Description

Technical Field

[0001] This invention belongs to the field of process optimization and control technology, and in particular relates to a method for optimizing and controlling the machining process of a drive axle support shaft for mining trucks. Background Technology

[0002] The drive axle support shaft of mining trucks is a core load-bearing component, requiring long-term resistance to heavy loads, impacts, and complex operating conditions. Its dimensional accuracy, surface quality, and mechanical properties directly determine the truck's operational safety and service life. Currently, the industry relies heavily on empirical parameter settings for the processing of this component, lacking systematic analysis of the correlation between multiple control parameters and quality indicators. This leads to significant quality fluctuations during production, making it difficult to consistently meet stringent service requirements and hindering improvements in the operational efficiency and reliability of mining trucks. Existing process methods lack a scientific quality quantification evaluation system, making it difficult to comprehensively characterize the impact of multiple dimensions on product quality. Furthermore, traditional PID control with fixed parameter tuning cannot dynamically adjust based on real-time quality deviations, making it difficult to adapt to parameter optimization needs under complex operating conditions. Ultimately, this results in insufficient processing quality stability and poor process optimization effects. Summary of the Invention

[0003] In view of the technical problems existing in the background art, the present invention proposes a process optimization and control method for the support shaft of the drive axle of mining truck.

[0004] To achieve the above objectives, the technical solution adopted by the present invention includes the following steps:

[0005] S1. Collect the cutting speed, feed rate, depth of cut, and grinding wheel speed control parameters of the production equipment in the production of drive axle support shafts for mining trucks; and obtain the corresponding mass fraction of drive axle support shafts during the production process to construct a sample dataset of control parameters and mass fractions.

[0006] S2. Preprocess the sample dataset during the production process to obtain the processed sample dataset;

[0007] S3. Based on the processed sample dataset, quantitatively calculate the influence weight of each control parameter on the quality score, determine the distinction between key and non-key parameters, fix the non-key control parameters as the normal optimal value, adjust the individual key parameters by a certain step size, and synchronously record the corresponding quality score changes. Using the rate of change of quality score and gradient as constraints, identify the sensitive interval of key parameters.

[0008] S4. Taking the mass fraction of the drive axle support shaft as the target, and combining the sensitive range of key parameters and the conventional optimal value range of non-key parameters, an intelligent algorithm is used to iteratively solve the optimal combination of control parameters that adapts to the production conditions in the sensitive range in real time.

[0009] S5. Using the real-time optimal control parameters as the PID setpoint, design an improved PID controller that integrates real-time quality deviation, dynamically tunes the parameters, and corrects the control quantity in real time.

[0010] Preferably, the realization of the corresponding drive axle support shaft mass fraction during the production process in step S1 is as follows:

[0011] S11. Based on the service requirements of the drive axle support shaft, select indicators including dimensional accuracy, surface quality, and mechanical performance correlation indicators; collect data from no less than five test sections for each part, and take three sets of parallel data for each section to form the original indicator dataset;

[0012] S12. Based on the obtained original indicator dataset, the weights of each indicator are obtained according to the degree of dispersion of each indicator through clustering method. Then, the data of each indicator are normalized according to the interval [0,1], and the quality score is obtained by weighted summation.

[0013] Preferably, in step S12, based on the obtained original index dataset, weights are obtained according to the dispersion of each index using a clustering method. Subsequently, the index data is normalized according to the [0,1] interval, and the quality score is obtained by weighted summation. The specific implementation includes:

[0014] S121. First, for each quality indicator in the original indicator dataset, calculate the coefficient of variation to eliminate the influence of dimensions and characterize its degree of dispersion. Then, use the density peak clustering algorithm to group all indicators according to the similarity of discrete features, and screen effective clusters to remove interference from abnormal indicators. First, calculate the coefficient of variation: ,in, Let be the coefficient of variation of the i-th index. Given the standard deviation and mean of the i-th index, determine the local density of the cluster: ,in, Let be the Euclidean distance between the discrete coefficients of indices i and j. To cut off the distance, Using the indicator function; with the discrete coefficient of each indicator as the unique feature dimension, by calculating the local density of each indicator, the density peak point is automatically identified as the core center of each cluster, and non-peak indicators are assigned to the cluster to which the nearest density peak point belongs, thus completing the automatic clustering and grouping of all indicators. Then, the clustering profile coefficient of each indicator is calculated. ,in, The average distance between index i and its cluster indices. Given the average distance between index i and heterogeneous cluster indices, retain the clusters whose cluster profile coefficients are greater than a set threshold to obtain K effective clusters;

[0015] S122. Determine the cluster weights based on the average dispersion of the indices within the cluster: ,in, The average dispersion coefficient of the k-th cluster The cluster weights are then determined based on the dispersion coefficients of individual indicators within the cluster. ,in, Let the i-th indicator belong to the k-th cluster; the combined weights of each indicator are obtained by multiplying the two: ;

[0016] S123. Indicators are categorized into positive and negative indicators based on their type. Normalization is performed within the [0,1] interval based on the extreme values ​​of their respective clusters. For positive indicators, the normalization process is as follows: ,in, For the normalized data, The original data, These are the maximum and minimum values ​​of the positive index within the k-th cluster, respectively; for the negative index, normalization is performed: ,in, These are the maximum and minimum values ​​of the negative index within the k-th cluster, respectively; finally, a weighted sum is performed to obtain the quality score: .

[0017] Preferably, step S2 involves preprocessing the sample dataset from the production process to obtain the processed sample dataset.

[0018] S21. For the control parameters in the sample dataset, kernel density estimation is used to calculate the normal distribution range of each control parameter for different processing stages. If a data point exceeds the corresponding range, it is judged as an outlier and discarded. For the quality score in the sample dataset, the least squares method is used to construct the quality score curve over time, and the slope of the curve is calculated in real time. When the slope of the curve exceeds the set slope threshold at a certain moment, it is judged as data anomaly and discarded.

[0019] S22. For missing data, weighted interpolation is used for adjacent acquisition times under the same working conditions. The weights are inversely proportional to time. The values ​​of the first three acquisition times are taken and weighted in chronological order. The weight of the most recent acquisition time is 0.5, the next is 0.3, and the furthest is 0.2. The missing values ​​of this acquisition time are obtained by weighting and completing the missing values ​​to obtain the processed sample dataset.

[0020] Preferably, step S3, based on the processed sample dataset, quantitatively calculates the influence weight of each control parameter on the quality score, determines the distinction between key and non-key parameters, fixes the non-key control parameters as conventional optimal values, adjusts individual key parameters by a certain step size, synchronously records the corresponding quality score changes, and identifies the sensitive intervals of key parameters using the rate of change and gradient of quality score as constraints. This includes:

[0021] S31. Based on the processed sample dataset, the linear correlation weight of each control parameter with the quality score is calculated using the Pearson correlation coefficient. ;in The representative is the first One control parameter;

[0022] S32. Construct a random forest regression model, using control parameters as input and quality scores as output. Train the model to obtain the feature importance of each control parameter, and then normalize the feature importance to obtain nonlinear weights. ;

[0023] S33. Using dynamic weighting factors The final weights are obtained by fusion: The dynamic weighting factor The adjustment is adaptive based on the degree of nonlinearity of the sample data. The degree of nonlinearity is determined by the coefficient of determination; the closer the coefficient of determination is to 0, the better. The larger the value, the range is [0.3, 0.7]; based on the final weight. It is compared with the preset final weight threshold to distinguish between key parameters and non-key parameters;

[0024] S34. Then, determine the optimal conventional values ​​for non-critical parameters. In the current processing scenario, select historical samples with quality scores greater than a set quality score and parameter fluctuations less than a set threshold, denoted as a high-quality sample subset. For each non-critical parameter, first calculate the median of that parameter in the high-quality subset samples, then use 3... The interval is defined as the range within which the parameter value that maximizes the quality fraction is taken as the conventional optimal value.

[0025] S35. According to the final weight of key parameters Prioritize the key parameters, and adjust one key parameter at a time according to the priority of each key parameter after sorting. The remaining key parameters are fixed at the minimum value of the allowable range of the project.

[0026] S36, Engineering allowable range of key parameters Based on this, set key parameters and adjust the step size. Using the rate of change of quality fraction as a dynamic constraint, the adjustment sequence of a single key parameter is calculated. The rate of change of quality fraction between two adjacent adjustments is calculated. If it is greater than or equal to the rate of change threshold, the corresponding interval is determined to be a potential sensitive region. Consecutive potential sensitive regions are merged. The two endpoints of the merged interval are used as the initial boundaries, and then the interval is expanded by one step in both directions. The quality fraction gradient of the expanded interval is calculated. ,in, These represent the mass fractions of the drive axle support shafts corresponding to the end and start points of the expanded interval, respectively. These are the control parameter values ​​for the end and start points of the expanded interval, respectively. If the absolute value of the gradient is greater than 0.8 times the maximum gradient within the interval, the expanded boundary is retained; otherwise, it is shrunk back to the initial boundary to obtain the final sensitive interval.

[0027] Preferably, step S4 employs an intelligent algorithm to iteratively solve for the optimal combination of control parameters that adapts to the production conditions within the sensitive interval. The intelligent algorithm uses a non-dominated sorting genetic algorithm, which maps the combination of control parameters to chromosome individuals using real-number encoding. This satisfies the requirement that key parameters take values ​​within the sensitive interval, and non-key parameters take values ​​within ±3% of the optimal conventional values. The fitness function is the reciprocal of the difference in quality scores. The non-dominated sorting genetic algorithm outputs the individual with the best fitness, thus obtaining the optimal combination of control parameters.

[0028] Preferably, in step S5, using the real-time optimal control parameters as the PID setpoint, an improved PID controller incorporating real-time quality deviation is designed, and the parameters are dynamically tuned to dynamically correct the control quantity in real time. The specific implementation of this is as follows:

[0029] S51. First, calculate the real-time quality fraction deviation. And normalize to the [0,1] interval to obtain Calculate the rate of change of quality deviation Where t is the real-time point in the processing. The sampling period for the quality deviation signal;

[0030] S52, Proportional coefficient based on deviation magnitude and differential coefficients Adjustments will be made: , The integral coefficient is adjusted in conjunction with the magnitude and rate of change of the deviation. : ;in, These are the initial values ​​for the proportional, integral, and derivative coefficients, respectively, and they are adjusted to meet the pre-defined coefficient constraint range.

[0031] S53, based on the quality fraction deviation of the first three sampling periods , , A linear prediction model is used to predict the deviation in the next period. The feedforward compensation amount is designed based on the predicted deviation for the next cycle: ,in The compensation coefficient for the control quantity in relation to the predicted deviation;

[0032] S54. Based on the optimized real-time PID coefficients and mass fraction deviation, calculate the feedback control quantity: ,in, For integration variables;

[0033] S55, Integrating feedforward and feedback control quantities, the final control quantity output is: .

[0034] Compared with existing technologies, the advantages and positive effects of this invention are as follows: It establishes a scientific quality quantification system, accurately characterizing product quality through multi-dimensional indicator collection and cluster weighted normalization, thus solving the problem of fuzziness in traditional evaluation; it accurately identifies key parameters and sensitive intervals, integrating linear and nonlinear weight analysis to avoid the blindness of empirical parameters; it employs intelligent algorithms to iteratively solve for the optimal parameter combination, adapting to complex working conditions; and it improves the PID controller by combining real-time quality deviation dynamic tuning parameters with feedforward compensation to achieve closed-loop precise control. Overall, it replaces experience-based dependence with data-driven approaches, significantly improving the stability and consistency of processing quality and reducing the scrap rate. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic diagram of the structural process of a method for optimizing and controlling the machining process of a drive axle support shaft for mining trucks. Detailed Implementation

[0037] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0038] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0039] In this example, the drive axle of a mining truck, as a core component in the vehicle's power transmission chain, requires its support shaft to maintain high stability and reliability under extreme loads, complex working conditions, and long-term continuous operation. However, existing manufacturing processes generally rely on experience to set processing parameters, lacking systematic optimization, resulting in significant fluctuations in processing quality. Insufficient understanding of parameter sensitivity makes the process window unclear, hindering the formation of stable, high-quality batch production. Furthermore, traditional PID control cannot dynamically tune parameters based on real-time processing quality feedback, nor can it be linked with process optimization parameters, leading to lags in production process control and unnecessary scrap and rework. To steadily improve the processing quality of the drive axle support shaft and establish a scientifically quantifiable process quality evaluation mechanism, a data-driven processing optimization control method is proposed. This method constructs a quality score system, identifies the sensitive range of control parameters, uses intelligent optimization to find the optimal parameter combination, and finally compensates through real-time improvement of the PID control closed loop, achieving intelligent, refined, and predictive control of the processing process. The specific implementation process is as follows: Figure 1 As shown.

[0040] First, the cutting speed, feed rate, depth of cut, and grinding wheel speed control parameters of the production equipment in the production of drive axle support shafts for mining trucks are collected. The corresponding quality score of the drive axle support shaft during the production process is then obtained, and a sample dataset of control parameters and quality scores is constructed. The quality score of the drive axle support shaft during the production process is realized based on the service requirements of the drive axle support shaft, selecting indicators including dimensional accuracy, surface quality, and mechanical performance correlation indicators. Data is collected from at least five test sections for each part, with three sets of parallel data taken from each section to form an original indicator dataset. Then, based on the obtained original indicator dataset, a clustering method is used to obtain weights according to the dispersion of each indicator. Subsequently, the indicator data is normalized according to the [0,1] interval, and the quality score is obtained through weighted summation. Specifically, a quality indicator system covering multi-dimensional performance is first constructed based on the actual service requirements of the component. Specifically, the system simultaneously covers dimensional accuracy, surface quality, and mechanical performance correlation indicators related to the final load-bearing capacity. To ensure the representativeness and statistical stability of the quality data, at least five inspection sections were set up for each support shaft, and three sets of parallel measurements were collected at each section, thus forming a structured and dense raw indicator dataset. Then, for each quality indicator in the raw indicator dataset, the coefficient of variation (coefficient of variation) was calculated to characterize its dispersion. Next, a density peak clustering algorithm was used to group all indicators according to the similarity of their discrete features, and effective clusters were selected to eliminate interference from abnormal indicators. The coefficient of variation was calculated first: ,in, Let be the coefficient of variation of the i-th index. Given the standard deviation and mean of the i-th index, determine the local density of the cluster: ,in, Let be the Euclidean distance between the discrete coefficients of indices i and j. To cut off the distance, Using the indicator function; with the discrete coefficient of each indicator as the unique feature dimension, by calculating the local density of each indicator, the density peak point is automatically identified as the core center of each cluster, and non-peak indicators are assigned to the cluster to which the nearest density peak point belongs, thus completing the automatic clustering and grouping of all indicators. Then, the clustering profile coefficient of each indicator is calculated. ,in, The average distance between index i and its cluster indices. Given the average distance between index i and the indices of other clusters, we retain the clusters whose cluster profile coefficients are greater than a set threshold, thus obtaining K effective clusters; the cluster weights are determined based on the average dispersion of the indices within each cluster. ,in, The average dispersion coefficient of the k-th cluster The cluster weights are then determined based on the dispersion coefficients of individual indicators within the cluster. ,in, Let the i-th indicator belong to the k-th cluster; the combined weights of each indicator are obtained by multiplying the two: Indicators are categorized into positive and negative indicators based on their type. Normalization is performed within the [0,1] interval based on the extreme values ​​of their respective clusters. For positive indicators, the normalization process is as follows: ,in, For the normalized data, The original data, These are the maximum and minimum values ​​of the positive index within the k-th cluster, respectively; for the negative index, normalization is performed: ,in, These are the maximum and minimum values ​​of the negative index within the k-th cluster, respectively; finally, a weighted sum is performed to obtain the quality score: This provides a reliable quantitative basis for subsequent process parameter optimization and control system construction.

[0041] Considering the existence of sensor errors, unexpected noise, workpiece anomalies, and individual detection distortions in actual production processes, direct modeling without processing would lead to deviations in parameter sensitivity assessment and optimization. Therefore, the sample dataset from the production process is preprocessed to obtain a processed sample dataset. This is achieved by: for control parameters in the sample dataset, kernel density estimation is used to calculate the normal distribution range of each control parameter for different processing stages; if a data point exceeds the corresponding range, it is identified as an outlier and discarded; for quality scores in the sample dataset, the least squares method is used to construct a curve of the quality score over time, and the slope of the curve is calculated in real time; if the slope of the curve exceeds a set slope threshold at a certain moment, the data is identified as an outlier and discarded; for missing data, weighted interpolation is used based on adjacent acquisition times under the same working condition, with the weights inversely proportional to time. The values ​​from the first three acquisition times are weighted in chronological order, with the most recent acquisition time having a weight of 0.5, the next most recent 0.3, and the furthest 0.2, to obtain the missing values ​​at that acquisition time, thus completing the missing value calculation and obtaining the processed sample dataset. Specifically, firstly, for the various control parameters in the sample dataset, considering that the processing involves different stages such as roughing, semi-finishing, and finishing, and the parameter distribution characteristics of each stage are not consistent, the kernel density estimation method is used to calculate the normal distribution range of the control parameters in each processing stage. When a data point exceeds the probability density main interval of its corresponding stage, the data point is determined to be an anomalous parameter and is directly discarded to avoid abnormal fluctuations interfering with subsequent modeling. Secondly, for the quality score data in the sample dataset, this embodiment constructs a fitting curve of the quality score changing over time based on the least squares method and calculates the slope of the curve in real time. When the slope exceeds a preset threshold at a certain moment and shows an abnormal sudden change trend, it is determined that the quality data has become abnormal at that moment, such as detection error or change in operating conditions, and is thus removed. Thirdly, for locally missing data caused by acquisition jumps, sensor jitter, or temporary communication delays, this embodiment introduces a weighted interpolation completion method. Specifically, based on the data collected at the three previous acquisition times consistent with the missing data point, an inverse weighting is applied according to the principle of "the closer the time, the stronger the correlation," with the most recent time valued at 0.5, the next at 0.3, and the furthest at 0.2. A reasonable estimate of the missing point is obtained through weighted summation. Through this preprocessing mechanism, this embodiment effectively eliminates the interference of noisy, outlier, and missing data on model training, making the sample data more stable, continuous, and reliable, thereby significantly improving the accuracy and robustness of subsequent parameter sensitivity analysis and optimization algorithms.

[0042] Then, based on the processed sample dataset, the influence weights of each control parameter on the quality score are quantitatively calculated to distinguish between key and non-key parameters. Non-key control parameters are fixed at their conventional optimal values, while individual key parameters are adjusted at certain step sizes, with corresponding quality score changes recorded synchronously. The sensitive intervals of key parameters are identified using the rate of change and gradient of the quality score as constraints. Specifically, based on the processed sample dataset, the linear correlation weight of each control parameter on the quality score is calculated using the Pearson correlation coefficient. ;in The representative is the first A random forest regression model is constructed, with the control parameters as input and the quality score as output. The feature importance of each control parameter is obtained through model training, and the feature importance is normalized to obtain nonlinear weights. ; Dynamic weighting factor is used The final weights are obtained by fusion: The dynamic weighting factor The adjustment is adaptive based on the degree of nonlinearity of the sample data. The degree of nonlinearity is determined by the coefficient of determination; the closer the coefficient of determination is to 0, the better. The larger the value, the range is [0.3, 0.7]; based on the final weight. The parameters are compared with a preset final weight threshold to distinguish between key and non-key parameters. Then, the optimal normal value for the non-key parameters is determined. In the current processing scenario, historical samples with quality scores greater than a set quality score and parameter fluctuations less than a set threshold are selected and denoted as a high-quality sample subset. For each non-key parameter, the median of that parameter in the high-quality subset is first calculated, and then... The interval is defined as the range within which the parameter value that maximizes the quality score is taken as the conventional optimal value; the final weights of the key parameters are then used. Prioritize the key parameters, and for each prioritized key parameter, adjust one key parameter at a time according to priority, while fixing the remaining key parameters at the minimum of the project's allowable range; the project's allowable range for the key parameters is then used as the basis for adjustment. Based on this, set key parameters and adjust the step size. Using the rate of change of quality fraction as a dynamic constraint, the adjustment sequence of a single key parameter is calculated. The rate of change of quality fraction between two adjacent adjustments is calculated. If it is greater than or equal to the rate of change threshold, the corresponding interval is determined to be a potential sensitive region. Consecutive potential sensitive regions are merged. The two endpoints of the merged interval are used as the initial boundaries, and then the interval is expanded by one step in both directions. The quality fraction gradient of the expanded interval is calculated. ,in, These represent the mass fractions of the drive axle support shafts corresponding to the end and start points of the expanded interval, respectively. These are the control parameter values ​​for the end and start points of the expanded interval, respectively. If the absolute value of the gradient is greater than 0.8 times the maximum gradient within the interval, the expanded boundary is retained; otherwise, it shrinks to the initial boundary to obtain the final sensitive interval. This step constructs a high-precision and highly robust method for identifying key parameters and extracting sensitive intervals through a four-pronged mechanism of linear-nonlinear fusion, dynamic weighting, adaptive filtering, and gradient correction, laying a solid data foundation for subsequent optimal control parameter solving.

[0043] Furthermore, taking the mass fraction of the drive axle support shaft as the target, and combining the sensitive range of key parameters and the conventional optimal value range of non-key parameters, an intelligent algorithm is used to iteratively solve for the optimal combination of control parameters that adapts to the production conditions within the sensitive range. The intelligent algorithm employs a non-dominated sorting genetic algorithm, mapping the combination of control parameters to chromosome individuals using real-number encoding. This satisfies the condition that key parameters take values ​​within the sensitive range, and non-key parameters take values ​​within ±3% of their conventional optimal values. The fitness function is the reciprocal of the difference in mass fractions. The non-dominated sorting genetic algorithm outputs the individual with the best fitness, thus obtaining the optimal combination of control parameters. Specifically, the implementation process is as follows: First, all process control parameters to be optimized are mapped to chromosome individuals using real-number encoding. The values ​​of key parameters are limited to their sensitive range, while the values ​​of non-key parameters are limited to a fluctuation range of ±3% of their conventional optimal values, ensuring that the search space has sufficient exploration capability while remaining within a reasonably feasible range in engineering. Then, an initial chromosome population is randomly generated according to the set population size, and the predicted mass fraction value corresponding to each chromosome individual is calculated based on the actual working conditions of the current processing. To enhance convergence towards high-quality solutions, this embodiment uses the reciprocal of the quality score difference as the fitness function; that is, the higher the quality score, the greater the fitness value, making it easier to be preserved during genetic evolution. The algorithm then performs selection, crossover, and mutation operations, dividing the population into different Pareto levels through a non-dominated sorting mechanism to simultaneously maintain solution diversity and a trend towards quality improvement. During each iteration, the algorithm continuously compares the fitness of the old and new populations and includes chromosomes with higher quality prediction results in the next generation. As the number of iterations increases, the combination of control parameters gradually converges towards the optimal region. Finally, when the convergence condition is met or the maximum number of iterations is reached, the algorithm outputs the chromosome with the highest fitness, corresponding to the optimal combination of control parameters under the current operating condition. This process possesses significant advantages such as strong adaptability, high convergence efficiency, and the ability to avoid local optima, providing a highly reliable parameter benchmark for subsequent real-time control.

[0044] Finally, using the real-time optimal control parameters as the PID setpoint, an improved PID controller integrating real-time quality deviation is designed, dynamically tuned, and the control quantity is corrected in real time. This is implemented by first calculating the real-time deviation based on the quality score at the current processing moment and normalizing it to the [0,1] interval, while further calculating the rate of change of this deviation. Then, the proportional and derivative coefficients of the PID controller are adaptively adjusted according to the deviation amplitude, making the P and D coefficients stronger as the deviation increases, thereby enhancing the system's rapid response capability; simultaneously, the integral coefficient is coordinated with the deviation change rate to suppress long-term accumulated errors and ensure that all coefficient adjustments are within preset constraints. Based on this, a linear prediction model is constructed using the deviations of the previous three sampling periods to predict the quality deviation of the next period, and a feedforward compensation quantity is generated accordingly to achieve advance correction of future deviation trends. Subsequently, the feedback control quantity is calculated based on the adjusted PID parameters and the current deviation, and the final feedback term is obtained by combining the integral variable. Finally, the feedforward compensation and feedback control quantity are fused to output the final control quantity. Specifically, the real-time quality score deviation is first calculated. And normalize to the [0,1] interval to obtain Calculate the rate of change of quality deviation Where t is the real-time point in the processing. The sampling period of the quality deviation signal; based on the magnitude of the deviation and the scaling factor. and differential coefficients Adjustments will be made: , The integral coefficient is adjusted in conjunction with the magnitude and rate of change of the deviation. : ;in, These are the initial values ​​for the proportional, integral, and derivative coefficients, respectively, and are adjusted to meet pre-defined coefficient constraint ranges; based on the quality fraction deviation of the first three sampling periods... , , A linear prediction model is used to predict the deviation in the next period. The feedforward compensation amount is designed based on the predicted deviation for the next cycle: ,in The compensation coefficient for the predicted deviation on the control quantity is used; based on the optimized real-time PID coefficients and the mass fraction deviation, the feedback control quantity is calculated: ,in, For integral variables; combining feedforward and feedback control quantities, the final control output is: This enables real-time and precise adjustment of the machining process, thereby significantly improving the stability and consistency of the machining quality of the drive axle support shaft.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for optimizing and controlling the machining process of a support shaft for a drive axle of a mining truck, characterized in that, Includes the following steps: S1. Collect the cutting speed, feed rate, depth of cut, and grinding wheel speed control parameters of the production equipment in the production of drive axle support shafts for mining trucks; and obtain the corresponding mass fraction of drive axle support shafts during the production process to construct a sample dataset of control parameters and mass fractions. S2. Preprocess the sample dataset during the production process to obtain the processed sample dataset; S3. Based on the processed sample dataset, quantitatively calculate the influence weight of each control parameter on the quality score, determine the distinction between key and non-key parameters, fix the non-key control parameters as the normal optimal value, adjust the individual key parameters by a certain step size, and synchronously record the corresponding quality score changes. Using the rate of change of quality score and gradient as constraints, identify the sensitive interval of key parameters. S4. Taking the mass fraction of the drive axle support shaft as the target, and combining the sensitive range of key parameters and the conventional optimal value range of non-key parameters, an intelligent algorithm is used to iteratively solve the optimal combination of control parameters that adapts to the production conditions in the sensitive range in real time. S5. Using the real-time optimal control parameters as the PID setpoint, design an improved PID controller that integrates real-time quality deviation, dynamically tunes the parameters, and corrects the control quantity in real time.

2. The method for optimizing and controlling the machining process of a drive axle support shaft for a mining truck according to claim 1, characterized in that, The realization of the corresponding drive axle support shaft mass fraction in step S1 during the production process is as follows: S11. Based on the service requirements of the drive axle support shaft, select indicators including dimensional accuracy, surface quality, and mechanical performance correlation indicators; collect data from no less than five test sections for each part, and take three sets of parallel data for each section to form the original indicator dataset; S12. Based on the obtained original indicator dataset, the weights of each indicator are obtained according to the degree of dispersion of each indicator through clustering method. Then, the data of each indicator are normalized according to the interval [0,1], and the quality score is obtained by weighted summation.

3. The method for optimizing and controlling the machining process of a drive axle support shaft for a mining truck according to claim 2, characterized in that, In step S12, based on the obtained original indicator dataset, weights are obtained according to the dispersion of each indicator using a clustering method. Then, the indicator data is normalized according to the [0,1] interval, and the quality score is obtained through weighted summation. The specific implementation includes: S121. First, for each quality indicator in the original indicator dataset, calculate the coefficient of variation to eliminate the influence of dimensions and characterize its degree of dispersion. Then, use the density peak clustering algorithm to group all indicators according to the similarity of discrete features, and screen effective clusters to remove interference from abnormal indicators. First, calculate the coefficient of variation: ,in, Let be the coefficient of variation of the i-th index. Given the standard deviation and mean of the i-th index, determine the local density of the cluster: ,in, Let be the Euclidean distance between the discrete coefficients of indices i and j. To cut off the distance, Using the indicator function; with the discrete coefficient of each indicator as the unique feature dimension, the local density of each indicator is calculated, and the density peak point is automatically identified as the core center of each cluster. Non-peak indicators are assigned to the cluster to which the nearest density peak point belongs, completing the automatic clustering and grouping of all indicators. Then, the clustering profile coefficient of each indicator is calculated. ,in, Let i be the average distance between index i and its cluster indices. Given the average distance between index i and heterogeneous cluster indices, retain the clusters whose cluster profile coefficients are greater than a set threshold to obtain K effective clusters; S122. Determine the cluster weights based on the average dispersion of the indices within the cluster: ,in, The average dispersion coefficient of the k-th cluster The cluster weights are then determined based on the dispersion coefficients of individual indicators within the cluster. ,in, Let the i-th indicator belong to the k-th cluster; the combined weights of each indicator are obtained by multiplying the two: ; S123. Indicators are categorized into positive and negative indicators based on their type. Normalization is performed within the [0,1] interval based on the extreme values ​​of their respective clusters. For positive indicators, the normalization process is as follows: ,in, For the normalized data, The original data, These are the maximum and minimum values ​​of the positive index within the k-th cluster, respectively; for the negative index, normalization is performed: ,in, These are the maximum and minimum values ​​of the negative index within the k-th cluster, respectively; finally, a weighted sum is performed to obtain the quality score: .

4. The method for optimizing and controlling the machining process of a drive axle support shaft for a mining truck according to claim 1, characterized in that, Step S2, which preprocesses the sample dataset during the production process to obtain the processed sample dataset, includes the following steps: S21. For the control parameters in the sample dataset, kernel density estimation is used to calculate the normal distribution range of each control parameter for different processing stages. If a data point exceeds the corresponding range, it is judged as an outlier and discarded. For the quality score in the sample dataset, the least squares method is used to construct the quality score curve over time, and the slope of the curve is calculated in real time. When the slope of the curve exceeds the set slope threshold at a certain moment, it is judged as data anomaly and discarded. S22. For missing data, weighted interpolation is used for adjacent acquisition times under the same working conditions. The weights are inversely proportional to time. The values ​​of the first three acquisition times are taken and weighted in chronological order. The weight of the most recent acquisition time is 0.5, the next is 0.3, and the furthest is 0.

2. The missing values ​​of this acquisition time are obtained by weighting and completing the missing values ​​to obtain the processed sample dataset.

5. The method for optimizing and controlling the machining process of a drive axle support shaft for a mining truck according to claim 1, characterized in that, Step S3, based on the processed sample dataset, quantitatively calculates the influence weight of each control parameter on the quality score, determines the distinction between key and non-key parameters, fixes the non-key control parameters as the conventional optimal values, adjusts individual key parameters by a certain step size, and synchronously records the corresponding quality score changes. The implementation of identifying the sensitive intervals of key parameters, constrained by the rate of change and gradient of the quality score, includes: S31. Based on the processed sample dataset, the linear correlation weight of each control parameter with the quality score is calculated using the Pearson correlation coefficient. ;in The representative is the first One control parameter; S32. Construct a random forest regression model, using control parameters as input and quality scores as output. Train the model to obtain the feature importance of each control parameter, and then normalize the feature importance to obtain nonlinear weights. ; S33. Using dynamic weighting factors The final weights are obtained by fusion: The dynamic weighting factor The adjustment is adaptive based on the degree of nonlinearity of the sample data. The degree of nonlinearity is determined by the coefficient of determination; the closer the coefficient of determination is to 0, the better. The larger the value, the range is [0.3, 0.7]; based on the final weight. It is compared with the preset final weight threshold to distinguish between key parameters and non-key parameters; S34. Then, determine the optimal conventional values ​​for non-critical parameters. In the current processing scenario, select historical samples with quality scores greater than a set quality score and parameter fluctuations less than a set threshold, denoted as a high-quality sample subset. For each non-critical parameter, first calculate the median of that parameter in the high-quality subset samples, then use 3... The interval is defined as the range within which the parameter value that maximizes the quality fraction is taken as the conventional optimal value. S35. According to the final weight of key parameters Prioritize the key parameters, and adjust one key parameter at a time according to the priority of each key parameter after sorting. The remaining key parameters are fixed at the minimum value of the allowable range of the project. S36, Engineering allowable range of key parameters Based on this, set key parameters and adjust the step size. Using the rate of change of quality fraction as a dynamic constraint, the adjustment sequence of a single key parameter is calculated. The rate of change of quality fraction between two adjacent adjustments is calculated. If it is greater than or equal to the rate of change threshold, the corresponding interval is determined to be a potential sensitive region. Consecutive potential sensitive regions are merged. The two endpoints of the merged interval are used as the initial boundaries, and then the interval is expanded by one step in both directions. The quality fraction gradient of the expanded interval is calculated. ,in, These represent the mass fractions of the drive axle support shafts corresponding to the end and start points of the expanded interval, respectively. These are the control parameter values ​​for the end and start points of the expanded interval, respectively. If the absolute value of the gradient is greater than 0.8 times the maximum gradient within the interval, the expanded boundary is retained; otherwise, it is shrunk back to the initial boundary to obtain the final sensitive interval.

6. The method for optimizing and controlling the machining process of a drive axle support shaft for a mining truck according to claim 1, characterized in that, Step S4 employs an intelligent algorithm to iteratively solve for the optimal combination of control parameters that adapts to the production conditions within the sensitive interval. The intelligent algorithm uses a non-dominated sorting genetic algorithm, which maps the combination of control parameters to chromosome individuals using real-number encoding. This satisfies the requirement that key parameters take values ​​within the sensitive interval, and non-key parameters take values ​​within ±3% of the optimal conventional values. The fitness function is the reciprocal of the difference in quality scores. The non-dominated sorting genetic algorithm outputs the individual with the best fitness, thus obtaining the optimal combination of control parameters.

7. The method for optimizing and controlling the machining process of a drive axle support shaft for a mining truck according to claim 1, characterized in that, In step S5, using the real-time optimal control parameters as the PID setpoint, an improved PID controller integrating real-time quality deviation is designed, and the parameters are dynamically tuned to dynamically correct the control quantity in real time. The specific implementation is as follows: S51. First, calculate the real-time quality fraction deviation. And normalize to the [0,1] interval to obtain Calculate the rate of change of quality deviation Where t is the real-time point in the processing. The sampling period for the quality deviation signal; S52, Proportional coefficient based on deviation magnitude and differential coefficients Adjustments will be made: , The integral coefficient is adjusted in conjunction with the magnitude and rate of change of the deviation. : ;in, These are the initial values ​​for the proportional, integral, and derivative coefficients, respectively, and they are adjusted to meet the pre-defined coefficient constraint range. S53, based on the quality fraction deviation of the first three sampling periods , , A linear prediction model is used to predict the deviation in the next period. The feedforward compensation amount is designed based on the predicted deviation for the next cycle: ,in The compensation coefficient for the control quantity in relation to the predicted deviation; S54. Based on the optimized real-time PID coefficients and mass fraction deviation, calculate the feedback control quantity: ,in, For integration variables; S55, Integrating feedforward and feedback control quantities, the final control quantity output is: .

Images