Method and system for optimizing process parameters of die casting island based on machine learning

Abstract

The application provides a die casting island process parameter optimization method and system based on machine learning, relates to the die casting process technical field, and comprises the following steps: acquiring process parameters in a die casting process, calculating energy loss values and entropy weights of each process stage, identifying a target process stage and performing local optimization; constructing a low-dimensional space based on historical process parameters, and clustering based on geodesic distance; selecting the most optimal historical parameters as a search starting point, performing parallel parameter search to obtain an optimal process parameter combination; and issuing an adjustment instruction to a die casting island control system. The application can improve die casting product quality, reduce energy consumption, and realize self-adaptive optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of die casting process, and particularly relates to a die casting island process parameter optimization method and system based on machine learning. BACKGROUND

[0002] Die casting process is an important metal forming technology, widely used in automobile, electronic, aerospace and other industries. With the development of Industry 4.0, die casting island as a typical equipment combination of intelligent manufacturing, integrates die casting machine, furnace, spray device, picking robot and other equipment, with the characteristics of multiple process parameters, complex control and high product quality requirements. Traditional die casting process parameter optimization mainly relies on experience accumulation and repeated experiments, which is time-consuming and has limited effect. In recent years, the application of machine learning technology in the manufacturing field provides a new idea for process parameter optimization.

[0003] The existing die casting process parameter optimization method based on machine learning mainly uses neural network, genetic algorithm and other technologies to construct the mapping relationship between process parameters and product quality, and obtains the optimal process parameter combination through reverse deduction. Some researches combine physical model with data-driven method to establish a multi-physical field model of the die casting process, and realize parameter optimization. In addition, some researchers try to use reinforcement learning and other artificial intelligence technologies to realize online optimization control of the die casting process to adapt to the dynamic changes of the production process.

[0004] The existing technology lacks systematic analysis of energy utilization efficiency in different process stages of the die casting process, and it is difficult to identify the key stages with high energy loss, which leads to lack of clear optimization direction for parameter optimization and reduces the optimization efficiency. The traditional method cannot effectively handle the high-dimensional characteristics of the die casting process parameters, and the complex coupling relationship between the parameters makes the optimization search space large, resulting in high computational complexity and low optimization efficiency, which makes it difficult to realize rapid optimization. Therefore, it is of great significance to design a fast, efficient and accurate die casting island process parameter optimization method. SUMMARY

[0005] The embodiment of the present application provides a die casting island process parameter optimization method and system based on machine learning, which can solve the problems in the prior art.

[0006] In a first aspect, the embodiment of the present application provides a die casting island process parameter optimization method based on machine learning, comprising:

[0007] The process parameters in the die casting process are obtained, the process parameters are divided into multiple process stages according to the die casting process, the energy loss value of each process stage is calculated according to the energy conversion relationship in the die casting process, the entropy weight of each process stage is calculated according to the product quality data, the target process stage affecting the quality is identified, when the energy loss value of the target process stage is higher than the loss threshold, the process parameters of the process stage are optimized, and the local optimization process parameters are obtained.

[0008] constructing historical process parameters as a historical parameter matrix, performing eigenvalue decomposition on a covariance matrix of the historical parameter matrix, constructing a low-dimensional parameter space, mapping the historical process parameters to the low-dimensional parameter space, calculating geodesic distances based on the locally optimized process parameters, and combining historical process parameter combinations with geodesic distances less than a distance threshold into the same class; marking a historical process parameter combination with the highest product quality evaluation value in each class as a search starting point, performing parallel parameter search on the search starting point, obtaining a candidate process parameter combination, and selecting a candidate process parameter combination with the highest quality evaluation value as an optimal process parameter combination;

[0009] According to the optimal process parameter combination, an adjustment instruction for the die casting island process parameters is generated, and the adjustment instruction is issued to the die casting island control system to adjust the process parameters of each device in real time.

[0010] According to product quality data, the entropy weight of each process stage is calculated, and the target process stage affecting the quality is identified. When the energy loss value of the target process stage is higher than the loss threshold, the process parameters of the process stage are optimized to obtain locally optimized process parameters, including:

[0011] Collecting multi-dimensional quality detection index data of die casting products, and performing normalization processing on the multi-dimensional quality detection index data to obtain normalized detection data;

[0012] Divide the normalized detection data by the sum of all normalized data of the corresponding detection index to obtain a proportion value, and divide the sum of the product of the proportion value and the natural logarithm thereof by the sample number to obtain the information entropy of each detection index;

[0013] The Spearman correlation coefficient is used to calculate the correlation coefficient of each process parameter to each detection index, and the correlation coefficient is weighted and summed to obtain the correlation degree of the process stage and each detection index. The product of the correlation degree and the corresponding information entropy is added to obtain the entropy weight of each process stage. The process stage with the largest entropy weight is selected as the target process stage;

[0014] When the energy loss value of the target process stage is higher than the preset loss threshold, the first ratio of the energy loss value to the preset loss threshold and the second ratio of the average value of each detection index to the standard value are calculated. The first ratio and the second ratio are weighted and summed to obtain a local optimization target value;

[0015] The gradient value of the local optimization target value to each process parameter is calculated, and the process parameters of the target process stage are optimized based on the gradient value. When the deviation value of the local optimization target value of adjacent two iterations is less than a convergence threshold, the current process parameters are recorded as the locally optimized process parameters.

[0016] constructing historical process parameters as a historical parameter matrix, performing eigenvalue decomposition on a covariance matrix of the historical parameter matrix, and constructing a low-dimensional parameter space to map the historical process parameters to the low-dimensional parameter space, calculating geodesic distance based on the locally optimized process parameters, and combining historical process parameters with geodesic distance less than a distance threshold into the same class, comprising:

[0017] obtaining a historical process parameter combination in a die casting process and constructing the historical process parameter combination as a historical parameter matrix; multiplying the historical parameter matrix by a transpose matrix thereof and dividing by a sample number to obtain a covariance matrix;

[0018] performing eigenvalue decomposition on the covariance matrix to obtain a plurality of eigenvalues and eigenvectors, performing inner product operation on the historical parameter matrix and the eigenvectors with eigenvalues greater than a feature threshold to obtain projection values of the historical process parameter combination in each feature direction; multiplying the projection values by contribution rates corresponding to the eigenvalues to obtain mapping coordinates of the historical process parameters in the low-dimensional parameter space, the contribution rates being calculated by dividing the eigenvalues by a sum of the eigenvalues;

[0019] calculating partial derivatives of the locally optimized process parameters with respect to the low-dimensional parameter space coordinates; summing the partial derivatives multiplied by randomly initialized weight parameters and adding a unit matrix to obtain a Riemann metric matrix; using the Riemann metric matrix, solving a minimum path length of any two historical process parameters in the low-dimensional parameter space by a variational method, and taking the minimum path length as geodesic distance;

[0020] combining historical process parameters with geodesic distance less than a preset distance threshold into the same class to obtain a historical process parameter classification result.

[0021] using the Riemann metric matrix, solving an optimal geodesic path of any two historical process parameters in the low-dimensional parameter space by a variational method, and taking a path length of the optimal geodesic path as geodesic distance, comprising:

[0022] for any two historical process parameters, calculating a geodesic path thereof in the low-dimensional parameter space by a linear interpolation method, calculating a gradient norm of the Riemann metric matrix on the geodesic path, obtaining a path weight by exponential mapping of the gradient norm, and multiplying the path weight by a square root of a derivative of the geodesic path under the Riemann metric matrix to obtain a path length;

[0023] calculating a variational derivative of the path length with respect to the geodesic path, adding a regularization term with exponential decay to the variational derivative to obtain a path update direction; calculating a first derivative norm and a second derivative norm of the geodesic path, dividing the second derivative norm by a square of the first derivative norm to obtain a local curvature of the path; multiplying a reference iteration step size by a negative exponential function of the local curvature of the path to obtain an iteration step size;

[0024] In each iteration, a length relative change rate is calculated by dividing the difference between the path lengths of two adjacent iterations by the path length of the last iteration, and a curvature difference value is calculated by the difference between the local curvatures of two adjacent iterations; when the length relative change rate is less than a length stability threshold value, and the curvature difference value is less than a curvature stability threshold value, the geodesic path of the current iteration is determined as the optimal geodesic path, and the path length of the optimal geodesic path is taken as the geodesic distance between two historical process parameter points.

[0025] The historical process parameter combination with the highest product quality evaluation value in each category is marked as a search starting point, and parallel parameter search is performed on the search starting point to obtain a candidate process parameter combination, and the candidate process parameter combination with the highest quality evaluation value is selected as the optimal process parameter combination, comprising:

[0026] The multiple quality detection index data corresponding to the process parameters are multiplied by corresponding weight coefficients and summed to obtain a quality evaluation value;

[0027] In each category, the historical process parameter with the highest quality evaluation value is selected as a search starting point, mutual information entropy of the search starting point and the quality evaluation value is calculated, for the search starting point with mutual information entropy higher than an importance threshold value, the parameter value range of the search starting point is equally divided into multiple grid units of a reference number to construct a local search space, and for the search starting point with mutual information entropy lower than the importance threshold value, the parameter value range of the search starting point is equally divided into grid units of a reference number to construct a global search space;

[0028] A process parameter difference vector between the current iteration position and the last iteration position of each search starting point is calculated, and the difference vector is divided by its module length to obtain a search direction vector; a quality evaluation value difference between adjacent iteration positions of each search starting point is calculated to obtain a quality improvement value;

[0029] When the maximum quality improvement value of all search starting points is less than a quality improvement threshold value, process parameter combinations corresponding to the current iteration positions of all search starting points are obtained, and the process parameter combination with the highest quality evaluation value is selected as the optimal process parameter combination.

[0030] After obtaining the quality improvement value, the method further comprises:

[0031] Based on the search direction vector and the quality improvement value of each search starting point, a direction-revenue mapping matrix for parallel search is established;

[0032] Based on the direction-revenue mapping matrix, an expected revenue of each candidate search direction is calculated, and the candidate search direction with the highest expected revenue is selected as the search direction of each search starting point;

[0033] When the expected income of a certain search starting point is lower than the expected value for a preset number of times in succession, the search direction with the highest expected income is obtained from the adjacent search starting point.

[0034] The expected income of each candidate search direction is calculated based on the direction-income mapping matrix, and the candidate search direction with the highest expected income is selected as the search direction of each search starting point, including:

[0035] Based on the direction-income mapping matrix, the included angle between adjacent search direction vectors is calculated and subjected to negative exponential operation to obtain a direction similarity, the correlation coefficient between adjacent quality improvement values is calculated, and the direction similarity and the correlation coefficient are weighted and summed to obtain a correlation degree.

[0036] For a candidate search direction, the included angle between the candidate search direction and each search direction vector in the direction-income mapping matrix is calculated and subjected to negative exponential operation to obtain a candidate direction similarity; the candidate direction similarity is multiplied by the sum of the corresponding quality improvement value and the correlation degree, and the calculation results of all search direction vectors are normalized and weighted to obtain an initial expected income of the candidate search direction.

[0037] The negative exponential product of the difference between the current time and the historical time is calculated as a time decay weight; the initial expected income is modified based on the time decay weight to obtain an expected income, and the candidate search direction with the highest expected income is selected as the search direction of each search starting point.

[0038] In a second aspect of the embodiment of the present application, a die casting island process parameter optimization system based on machine learning is provided, including:

[0039] A first unit is configured to obtain process parameters in a die casting process, divide the process parameters into a plurality of process stages according to a die casting flow, calculate energy loss values of each process stage according to an energy conversion relationship in the die casting process, calculate entropy weights of each process stage according to product quality data, identify a target process stage affecting quality, optimize process parameters of the target process stage when the energy loss value of the target process stage is higher than a loss threshold, and obtain locally optimized process parameters.

[0040] A second unit is configured to construct historical process parameters into a historical parameter matrix, perform eigenvalue decomposition on a covariance matrix thereof, and construct a low-dimensional parameter space; map the historical process parameters to the low-dimensional parameter space; calculate geodesic distances based on the locally optimized process parameters; combine historical process parameter combinations with geodesic distances less than a distance threshold into the same class; mark a historical process parameter combination with the highest product quality evaluation value in each class as a search starting point; perform parallel parameter search on the search starting point to obtain candidate process parameter combinations; and select a candidate process parameter combination with the highest product quality evaluation value as an optimal process parameter combination.

[0041] The third unit is configured to generate adjustment instructions of the die casting island process parameters according to the optimal process parameter combination, and send the adjustment instructions to a die casting island control system to adjust the process parameters of each device in real time.

[0042] The third aspect of the embodiment of the present application,

[0043] An electronic device is provided, comprising:

[0044] A processor;

[0045] A memory for storing processor-executable instructions;

[0046] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0047] The fourth aspect of the embodiment of the present application,

[0048] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0049] The beneficial effects of the present application are as follows:

[0050] The die casting island process parameter optimization method based on machine learning can accurately identify the key process stages that affect product quality, and through the combination of entropy weight calculation and energy loss analysis, the precise local optimization of process parameters is realized, unnecessary parameter adjustment is reduced, and the optimization efficiency is improved.

[0051] The present application adopts eigenvalue decomposition and low-dimensional parameter space construction technology, combined with geodesic distance classification method, effectively reduces the search complexity of process parameters, can quickly locate the optimal process parameter combination in complex multi-parameter space, realizes the effective transition from local optimization to global optimization, and improves the accuracy and efficiency of parameter optimization.

[0052] The present application realizes dynamic optimization and automatic adjustment of process parameters by sending adjustment instructions to the die casting island control system in real time, reduces manual intervention, reduces the risk of operation errors, improves the stability and consistency of die casting production, effectively improves the product quality, and reduces energy consumption and production cost. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 The flowchart of the die casting island process parameter optimization method based on machine learning of the embodiment of the present application is shown in the figure.

[0054] Figure 2 The system architecture diagram for optimizing die casting process parameters based on energy loss and product quality data is shown in the figure. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0056] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and some embodiments may not be described again for the same or similar concepts or processes.

[0057] Figure 1 The flowchart of the method for optimizing die casting island process parameters based on machine learning according to the embodiments of the present application is shown in FIG. 1, which comprises the following steps. Figure 1

[0058] Obtaining process parameters in a die casting process, dividing the process parameters into multiple process stages according to a die casting flow, calculating energy loss values of each process stage according to energy conversion relationships in the die casting process, calculating entropy weights of each process stage according to product quality data, identifying a target process stage affecting quality, optimizing process parameters of the target process stage when the energy loss value of the target process stage is higher than a loss threshold, and obtaining locally optimized process parameters.

[0059] Constructing historical process parameters into a historical parameter matrix, performing eigenvalue decomposition on a covariance matrix thereof, constructing a low-dimensional parameter space, mapping the historical process parameters to the low-dimensional parameter space, calculating geodesic distances based on the locally optimized process parameters, combining historical process parameter combinations with geodesic distances less than a distance threshold into the same class, marking a historical process parameter combination with the highest product quality evaluation value in each class as a search starting point, performing parallel parameter search on the search starting point, obtaining candidate process parameter combinations, and selecting a candidate process parameter combination with the highest quality evaluation value as an optimal process parameter combination.

[0060] Generating adjustment instructions for die casting island process parameters according to the optimal process parameter combination, issuing the adjustment instructions to a die casting island control system, and adjusting device process parameters in real time.

[0061] Figure 2 ​A schematic diagram of a system architecture for optimizing die casting process parameters based on energy consumption and product quality data. In an optional embodiment, the entropy weight of each process stage is calculated according to the product quality data, the target process stage that affects the quality is identified, and when the energy consumption value of the target process stage is higher than the loss threshold, the process parameters of the process stage are optimized to obtain local optimization process parameters, including:

[0062] Collecting multi-dimensional quality detection index data of die casting products, and normalizing the multi-dimensional quality detection index data to obtain normalized detection data;

[0063] Divide the normalized detection data by the sum of all normalized data of the corresponding detection index to obtain a proportion value, and divide the sum of the product of the proportion value and its natural logarithm by the sample number to obtain the information entropy of each detection index;

[0064] Calculate the correlation coefficient of each process parameter to each detection index using the Spearman correlation coefficient, and obtain the correlation degree of the process stage and each detection index by weighted sum of the correlation coefficient, and add the product of the correlation degree and the corresponding information entropy to obtain the entropy weight of each process stage, and select the process stage with the largest entropy weight as the target process stage;

[0065] When the energy consumption value of the target process stage is higher than the preset loss threshold, calculate the first ratio of the energy consumption value and the preset loss threshold and the second ratio of the average value and the standard value of each detection index, and obtain the local optimization target value by weighted sum of the first ratio and the second ratio;

[0066] Calculate the gradient value of the local optimization target value to each process parameter, and optimize the process parameters of the target process stage based on the gradient value; when the deviation value of the local optimization target value of adjacent two iterations is less than the convergence threshold, record the current process parameters as the local optimization process parameters.

[0067] The present application relates to a method for optimizing die casting process parameters based on energy consumption and product quality data. In actual implementation, by calculating the entropy weight of each process stage, the target process stage that most significantly affects the product quality is identified, and the process parameters are optimized when the energy consumption of the process stage is too high, thereby improving the product quality and reducing the energy consumption.

[0068] In the implementation of the present application, it is necessary to collect multi-dimensional quality detection index data of die casting products. Taking a certain die casting production line as an example, 5 quality detection index data of 100 pieces of aluminum alloy die castings, such as surface roughness, dimensional accuracy, porosity, hardness and tensile strength, are collected. Since these indicators have different dimensions and numerical ranges, they need to be normalized to eliminate the influence of dimensions. For indicators that are better the smaller (such as surface roughness, dimensional deviation, porosity), the minimum maximum normalization method is adopted, and the calculation formula is normalized value = maximum value - current value ÷ maximum value - minimum value; for indicators that are better the larger (such as hardness, tensile strength), the formula normalized value = current value - minimum value ÷ maximum value - minimum value is adopted. After normalization, the values of all detection indicators are mapped to the interval of 0 to 1.

[0069] After normalization, the information entropy of each detection index is calculated. Taking the surface roughness index as an example, the proportion of the normalized value of the surface roughness of the first product in the sum of the normalized values of the surface roughness of all 100 products is calculated, and the result is 0.0103. Similarly, the proportion values of the surface roughness of all 100 products are calculated. Then the product of each proportion value and its natural logarithm is calculated, such as 0.0103 multiplied by ln(0.0103) to get -0.0472. Add all the products to get -4.5861, and then divide by the sample size 100 to get the information entropy of the surface roughness index as 0.4586. The information entropy of other detection indexes is calculated in the same way. Assuming that the information entropy of the dimensional accuracy is 0.3921, the information entropy of the porosity is 0.5217, the information entropy of the hardness is 0.4102, and the information entropy of the tensile strength is 0.4738.

[0070] In the die casting process, the die casting process parameters can be divided into five process stages, including melting, conveying, injection, cooling, and demolding. Each stage contains multiple process parameters. In the melting stage, the key process parameters include furnace power, heating curve, stirring speed, degassing time, and slag removal amount. In the conveying stage, the key process parameters include conveying temperature, holding power, metal flow rate, conveying pressure, and filtration efficiency. In the injection stage, the key process parameters include injection speed, injection pressure, speed switching point position, injection speed, and nozzle temperature. In the cooling stage, the key process parameters include cooling time, die casting temperature, holding time, cooling channel distribution, and cooling efficiency. In the demolding stage, the key process parameters include opening speed, opening distance, ejection speed, release agent spraying pressure, and spraying uniformity. Based on the above process parameters, the energy loss values of different process stages can be calculated for subsequent data analysis. In a certain die casting production line, a batch of aluminum alloy parts with a single piece weight of 2 kg is produced. In the melting stage, 200 kg of aluminum alloy is processed, and 100 products can be produced. According to the energy conversion relationship of each stage, the stage energy loss value can be calculated. In the melting stage, the rated power of the resistance furnace is 85 kW, the actual running power is 78 kW, and the melting time is 45 minutes. The input energy is calculated as: 78 kW x 0.75 h = 58.5 kWh. The theoretical melting energy of aluminum alloy is calculated as: 200 kg x [0.9 kJ / (kg·℃) x (690℃-25℃) + 397 kJ / kg] = 33.12 kWh. Among them, 0.9 kJ / (kg·℃) is the specific heat capacity of aluminum alloy, 690℃ is the melting temperature, 25℃ is the initial temperature, and 397 kJ / kg is the latent heat of fusion of aluminum alloy. The total energy loss in the melting stage is 58.5-33.12=25.38 kWh, and the energy loss per product is 25.38÷100=0.2538 kWh. The energy loss is subdivided into: 12.69 kWh of furnace heat loss; 7.61 kWh of heat loss during furnace startup; and 5.08 kWh of efficiency loss of electric energy conversion to heat energy.

[0071] In the conveying stage, the holding conveyor device power is 15 kW, the actual running power is 12 kW, and the conveying time is 3.5 minutes. The input energy is calculated as: 12 kW x 0.058 h = 0.7 kWh. During the conveying process, the metal temperature decreases from 690℃ to 665℃, and the heat loss is calculated as: 200 kg x 0.9 kJ / (kg·℃) x (690℃-665℃) ÷ 3600 = 1.25 kWh. Considering the energy supply of the holding equipment of 0.7 kWh, the actual energy loss in the conveying stage is 1.25-0.7=0.55 kWh, and the energy loss per product is 0.55÷100=0.0055 kWh. The energy loss is subdivided into: 0.275 kWh of heat loss due to metal and air contact; 0.165 kWh of heat loss due to conveyor wall conduction; and 0.11 kWh of heat loss due to radiation.

[0072] During the injection phase, the injection system hydraulic pump power is 55 kW, and the injection process is divided into two sub-phases: low-speed injection and high-speed injection. During the low-speed injection phase, the actual hydraulic system power is 25 kW, the duration is 2.5 seconds, and the energy input is: 25 kW x (2.5 ÷ 3600) hours = 0.0174 kWh. During the high-speed injection phase, the actual hydraulic system power is 48 kW, the duration is 0.8 seconds, and the energy input is: 48 kW x (0.8 ÷ 3600) hours = 0.0107 kWh. The total energy input during the injection phase is 0.0174 + 0.0107 = 0.0281 kWh. By measuring the pressure and displacement of the injection cylinder, the effective mechanical work is calculated: 0.0063 kWh during the low-speed injection phase, 0.0046 kWh during the high-speed injection phase, and the total effective mechanical work is 0.0109 kWh. The total energy loss during the injection phase is 0.0281 - 0.0109 = 0.0172 kWh, and the energy loss per product is 0.0172 kWh. The energy loss is further divided into: hydraulic system mechanical friction loss 0.0069 kWh; oil flow resistance loss 0.0052 kWh; hydraulic element internal leakage loss 0.0034 kWh; and system vibration loss 0.0017 kWh.

[0073] During the cooling phase, the mold cooling system water pump power is 12 kW, the actual operating power is 10 kW, and the cooling time is 25 seconds. The energy input calculation is: 10 kW x (25 ÷ 3600) hours = 0.0694 kWh. The cooling water flow rate is 85 L / min, the inlet water temperature is 25°C, and the outlet water temperature is 38°C. The cooling water heat absorption calculation is: 85 x (25 ÷ 60) x 4.2 x (38 - 25) ÷ 3600 = 0.2316 kWh. Among them, 4.2 kJ / (kg·℃) is the specific heat capacity of water. The metal is cooled from 665°C to 180°C, and the released heat calculation is: 2 kg x [0.9 kJ / (kg·℃) x (665°C - 180°C)] + 2 kg x 397 kJ / kg = 1.7154 kWh. The cooling system absorbs heat of 0.2316 kWh, plus the energy input of 0.0694 kWh, totaling 0.301 kWh. The energy loss during the cooling phase is 1.7154 - 0.301 = 1.4144 kWh, and the energy loss per product is 0.2015 kWh (considering that the mold can produce 7 products at the same time). The energy loss is further divided into: mold and environment heat exchange loss 0.7072 kWh; cooling system pipe heat loss 0.4243 kWh; and cooling water and pipe heat conduction loss 0.2829 kWh.

[0074] In the demolding stage, the hydraulic system power is 18 kW, which is divided into two actions of mold opening and ejection. During the mold opening process, the actual power of the hydraulic system is 15 kW, the duration is 4 seconds, and the energy input is: 15 kW x (4 ÷ 3600) hours = 0.0167 kWh. During the ejection process, the actual power of the hydraulic system is 10 kW, the duration is 2 seconds, and the energy input is: 10 kW x (2 ÷ 3600) hours = 0.0056 kWh. The total energy input in the demolding stage is 0.0167 + 0.0056 = 0.0223 kWh. Through the force sensor and displacement sensor measurement, the effective mechanical work is calculated: 0.0085 kWh in the mold opening process, 0.0028 kWh in the ejection process, and the total effective mechanical work is 0.0113 kWh. The total energy loss in the demolding stage is 0.0223 - 0.0113 = 0.011 kWh, and the energy loss per product is 0.0016 kWh (considering that the mold simultaneously demolds 7 products). The energy loss is subdivided into: hydraulic system mechanical friction loss 0.0044 kWh (40% of the loss); oil flow resistance loss 0.0033 kWh; mold guide column friction resistance loss 0.0022 kWh; system vibration loss 0.0011 kWh.

[0075] Based on the above calculations, the energy loss in each stage of producing one aluminum alloy steering knuckle is: 0.2538 kWh in the melting stage, 0.0055 kWh in the conveying stage, 0.0172 kWh in the injection stage, 0.2015 kWh in the cooling stage, and 0.0016 kWh in the demolding stage. The total energy loss is 0.4796 kWh. Through energy loss analysis, it is known that the energy loss in the melting stage and the cooling stage is the largest, accounting for 52.9% and 42.0% of the total loss, respectively. The conveying, injection and demolding stages account for only 5.1% of the total loss.

[0076] Determine the correlation degree of each process stage and quality index. Use Spearman correlation coefficient to calculate the correlation between each process parameter and each quality detection index. For example, for the injection speed parameter in the injection stage, the correlation coefficient with surface roughness is 0.82, the correlation coefficient with size accuracy is 0.65, the correlation coefficient with porosity is 0.79, the correlation coefficient with hardness is 0.42, and the correlation coefficient with tensile strength is 0.56. Similarly, calculate the correlation coefficients of all process parameters and detection indexes.

[0077] To obtain the correlation degree of each process stage and detection index, the correlation coefficients of all process parameters under each process stage and a certain detection index are weighted and summed. The weights can be determined according to the importance of each parameter. Assuming that the injection stage includes injection speed, injection pressure and locking force, the weights are 0.5, 0.3 and 0.2 respectively. The correlation degree of the injection stage and the surface roughness is 0.74, the correlation degree of the size accuracy is 0.61, the correlation degree of the porosity is 0.73, the correlation degree of the hardness is 0.48, and the correlation degree of the tensile strength is 0.58.

[0078] To calculate the entropy weight of each process stage, the correlation degree of each process stage and each detection index is multiplied by the information entropy of the corresponding detection index and then added. Taking the injection stage as an example, the entropy weight is calculated as 0.74x0.4586 + 0.61x0.3921 + 0.73x0.5217 + 0.48x0.4102 + 0.58x0.4738 = 1.6436. Similarly, the entropy weights of other process stages are calculated, the melting stage is 1.2341, the conveying stage is 0.9872, the cooling stage is 1.4510, and the demolding stage is 0.8765. By comparing the entropy weights of each process stage, it is found that the entropy weight of the injection stage is the largest, which is 1.6436, so the injection stage is selected as the target process stage.

[0079] The energy loss value of the injection stage is monitored. Assuming that the current energy loss is 85 kWh, and the preset loss threshold is 75 kWh, the energy loss value is higher than the preset loss threshold, and process parameter optimization is needed. The first ratio of energy loss value to loss threshold is 85 / 75 = 1.133, indicating that the energy exceeds 13.3%. The second ratio of the average value of each detection index to the standard value is calculated, for example, the average value of the surface roughness is 4.2 microns, and the standard value is 3.5 microns, the second ratio is 4.2 / 3.5 = 1.2, indicating that the quality exceeds 20%. Similarly, the second ratio of other detection indexes is calculated.

[0080] The energy loss and quality indexes are respectively given weights of 0.4 and 0.6, and the local optimization target value is calculated. For the surface roughness, the local optimization target value is 0.4x1.133 + 0.6x1.2 = 1.1732. Similarly, the local optimization target values of other detection indexes are calculated.

[0081] Based on the local optimization target value, the gradient descent method is used to optimize the process parameters. The gradient value of the local optimization target value to each process parameter is calculated, that is, the sensitivity of the target value to the parameter change. Assuming that the gradient value of the injection speed is 0.25, which means that increasing the injection speed will cause the target value to rise by 0.25; the gradient value of the injection pressure is -0.18, which means that increasing the injection pressure will cause the target value to decrease by 0.18. According to the gradient value, adjust the process parameters, that is, reduce the parameters corresponding to the positive gradient, and increase the parameters corresponding to the negative gradient.

[0082] After multiple iterations of optimization, when the deviation value of the local optimization target value of the adjacent two iterations is less than the preset convergence threshold 0.001, stop iteration and record the current process parameters as the local optimization process parameters. After optimization, the injection speed is reduced from the original 2.8 meters / second to 2.5 meters / second, the injection pressure is increased from 120 megapascals to 135 megapascals, and the clamping force is adjusted from 4500 tons to 4200 tons. After using this set of optimized parameters, the energy consumption is reduced to 72 kilowatt-hours, which is lower than the preset threshold 75 kilowatt-hours, and the product quality detection index is also improved, with the surface roughness reduced to 3.7 microns, closer to the standard value 3.5 microns.

[0083] In an optional embodiment, the historical process parameters are constructed as a historical parameter matrix, the covariance matrix thereof is subjected to eigenvalue decomposition, a low-dimensional parameter space is constructed, the historical process parameters are mapped to the low-dimensional parameter space, the geodesic distance is calculated based on the local optimization process parameters, the historical process parameter combinations with geodesic distance less than a distance threshold are combined into the same class, comprising:

[0084] The historical process parameter combinations in the die casting process are obtained and constructed as a historical parameter matrix; the historical parameter matrix is multiplied by its transpose matrix and divided by the sample number to obtain a covariance matrix;

[0085] The covariance matrix is subjected to eigenvalue decomposition to obtain a plurality of eigenvalues and eigenvectors; the historical parameter matrix is subjected to inner product operation with the eigenvectors whose eigenvalues are greater than a feature threshold to obtain the projection values of the historical process parameter combinations in each feature direction; the mapping coordinates of the historical process parameters in the low-dimensional parameter space are obtained by multiplying the projection values by the contribution rates corresponding to the eigenvalues, and the contribution rates are calculated by dividing the eigenvalues by the total sum of the eigenvalues;

[0086] The partial derivative of the local optimization process parameters to the low-dimensional parameter space coordinates is calculated; the sum of the partial derivative multiplied by the randomly initialized weight parameters and the unit matrix is added to obtain a Riemann metric matrix; using the Riemann metric matrix, the minimum path length between any two historical process parameters in the low-dimensional parameter space is solved by the variational method, and the minimum path length is taken as the geodesic distance;

[0087] The historical process parameters with geodesic distances less than a preset distance threshold are classified into the same class to obtain a historical process parameter classification result.

[0088] The embodiment provides a historical process parameter classification method based on geodesic distance. The method constructs a historical parameter matrix, performs dimension reduction processing through eigenvalue decomposition, then calculates geodesic distances in a low-dimensional space, and finally realizes classification of historical process parameters.

[0089] In the die casting production process, a large amount of historical process parameter combination data is collected. These process parameters include but are not limited to: mold temperature, alloy temperature, injection speed, boost pressure, low-speed section speed, holding time, etc. For example, 100 sets of historical process parameter combinations are collected, each set containing 10 different process parameters. These parameters are arranged in a row-column manner to construct a 100-row 10-column historical parameter matrix P. Each row of the matrix P represents a set of process parameter combinations, and each column represents a process parameter type.

[0090] In order to calculate the covariance matrix, the historical parameter matrix P is multiplied by its transpose matrix P T , and then divided by the sample number 100 to obtain a 10x10 covariance matrix C. The covariance matrix C reflects the correlation between different process parameters.

[0091] Eigenvalue decomposition is performed on the covariance matrix C to obtain 10 eigenvalues λ1, λ2,..., λ 10 and their corresponding eigenvectors v1, v2,..., v 10 . The eigenvalues are arranged in descending order, for example, λ1=5.8, λ2=2.3, λ3=1.2, λ4=0.7, λ5=0.4, etc. Set the eigenvalue threshold to 0.8, and select the eigenvectors v1, v2, and v3 with eigenvalues greater than 0.8 as the main characteristic directions.

[0092] The projection values of the historical process parameters on the main characteristic directions are calculated. For each row of data p i in the historical parameter matrix P, the inner product operation is performed with the eigenvectors v1, v2, and v3 respectively to obtain the projection values proj i1 , proj i2 , and proj i3 . For example, for the first set of process parameters p1, the calculation results are proj 11 =25.6, proj 12 =10.2, and proj 13 =-5.8.

[0093] The contribution rates corresponding to the eigenvalues are calculated. The contribution rate is equal to the eigenvalue divided by the sum of the eigenvalues. The sum of the eigenvalues is λ1+λ2+...+λ 10= 11.3. Therefore, the contribution rate of λ1 is 5.8 / 11.3 = 0.513, the contribution rate of λ2 is 2.3 / 11.3 = 0.204, and the contribution rate of λ3 is 1.2 / 11.3 = 0.106.

[0094] The projection value is multiplied by the contribution rate of the corresponding eigenvalue to obtain the mapping coordinates of the historical process parameters in the low-dimensional parameter space. Taking the first set of process parameters p1 as an example, the low-dimensional mapping coordinates are (25.6x0.513, 10.2x0.204, -5.8x0.106) = (13.13, 2.08, -0.61). Through similar calculations, the mapping coordinates of all historical process parameters in the three-dimensional low-dimensional parameter space can be obtained.

[0095] The locally optimized process parameters are obtained, which are mapped into the low-dimensional parameter space to obtain the coordinates (15.5, 3.2, -0.8). The partial derivatives of the locally optimized process parameters with respect to the low-dimensional parameter space coordinates are calculated. Specifically, the partial derivatives of the optimization objective function f (such as the product quality index) with respect to the low-dimensional coordinates x, y, and z are calculated to obtain the partial derivative vector grad f = (∂f / ∂x, ∂f / ∂y, ∂f / ∂z) = (0.23, 0.15, -0.08).

[0096] The weight parameters w1 = 0.5, w2 = 0.3, and w3 = 0.2 are randomly initialized. The partial derivatives are multiplied by the weight parameters and summed to obtain w1x(∂f / ∂x) 2 + w2x(∂f / ∂y) 2 + w3x(∂f / ∂z) 2 = 0.5x0.23 2 + 0.3x0.15 2 + 0.2x(-0.08) 2 = 0.03. This value is added to the 3x3 unit matrix to obtain the Riemann metric matrix G.

[0097] Using the Riemann metric matrix G, the minimum path length, i.e., the geodesic distance, between any two historical process parameters in the low-dimensional parameter space is calculated by the variational method. For example, the geodesic distance between the locally optimized process parameters and the 25th set of historical process parameters is 2.6, the geodesic distance between the locally optimized process parameters and the 37th set of historical process parameters is 1.8, and the geodesic distance between the locally optimized process parameters and the 42nd set of historical process parameters is 3.5.

[0098] The preset distance threshold is set as 3.0, and the historical process parameters with a geodesic distance less than 3.0 are classified into the same class. Therefore, the historical process parameters in the 25th group and the 37th group are classified into the same class as the locally optimized process parameters, and the historical process parameters in the 42nd group are classified into different classes. In this way, all the historical process parameters can be classified into several classes, and the process parameter combinations in each class have similar performance characteristics.

[0099] The classification method based on the geodesic distance fully considers the geometric structure of the process parameters in the low-dimensional parameter space and the gradient information of the locally optimized process parameters, and can more accurately reflect the internal correlation between the parameter combinations, thereby providing effective support for the optimization of the die casting process parameters. By analyzing the common characteristics of the process parameters in the same class, the key factors and adjustment directions for improving the process can be further refined, thereby guiding the optimization and adjustment of the die casting production process.

[0100] In an optional implementation, an optimal geodesic path between any two historical process parameters in the low-dimensional parameter space is solved by using the Riemann metric matrix through a variational method, and a path length of the optimal geodesic path is taken as the geodesic distance, including:

[0101] For any two historical process parameters, a linear interpolation method is used to calculate a geodesic path thereof in the low-dimensional parameter space, a gradient norm of the Riemann metric matrix on the geodesic path is calculated, a path weight is obtained by performing exponential mapping on the gradient norm, and a path length is obtained by multiplying the path weight and a square root of a derivative of the geodesic path under the Riemann metric matrix.

[0102] A variational derivative of the path length with respect to the geodesic path is calculated, a path update direction is obtained by adding a regularization term with exponential decay to the variational derivative, a first-order derivative norm and a second-order derivative norm of the geodesic path are calculated, the second-order derivative norm is divided by a square of the first-order derivative norm to obtain a local curvature of the path, and an iteration step size is obtained by multiplying a reference iteration step size and a negative exponential function of the local curvature of the path.

[0103] In each iteration, a length relative change rate is calculated by subtracting a path length of a previous iteration from a difference between path lengths of adjacent two iterations, and a curvature difference value is calculated by subtracting a path local curvature of the previous iteration from a difference between path local curvatures of the adjacent two iterations. When the length relative change rate is less than a length stability threshold and the curvature difference value is less than a curvature stability threshold, a geodesic path of a current iteration is determined as an optimal geodesic path, and a path length of the optimal geodesic path is taken as the geodesic distance between the two historical process parameter points.

[0104] The technical solution described herein provides a method for calculating the optimal geodesic path between any two historical process parameters in a low-dimensional parameter space based on the Riemann metric matrix and the variational method, and taking the length of the path as the geodesic distance. In order to achieve this goal, the following specific implementation steps are adopted in the technical solution.

[0105] In the low-dimensional parameter space, for any two historical process parameter points p1 and p2, an initial geodesic path γ(t) is constructed by linear interpolation, where t ∈ [0, 1]. Assuming that p1 corresponds to the parameter value (2.5, 3.7) and p2 corresponds to the parameter value (8.3, 6.2), the initial path can be represented as γ(t) = (2.5 + 5.8t, 3.7 + 2.5t). The gradient norm of the established Riemann metric matrix G on the path is calculated. For example, at the path point γ(0.5) = (5.4, 4.95), four points are selected near this point for sampling: f(5.4+h, 4.95) = 8.24, f(5.4-h, 4.95) = 8.06, f(5.4, 4.95+h) = 8.32, f(5.4, 4.95-h) = 8.07, where h = 0.2 is the sampling step. The gradient is calculated using the central difference formula: ∂f / ∂x1= [f(5.4+h, 4.95) - f(5.4-h, 4.95)] / (2h)= (8.24 - 8.06) / (2×0.2)= 0.18 / 0.4= 0.45, ∂f / ∂x2= [f(5.4, 4.95+h) - f(5.4, 4.95-h)] / (2h)= (8.32 - 8.07) / (2×0.2)= 0.25 / 0.4= 0.625 ≈ 0.62, therefore, the gradient vector at the path point γ(0.5) is: ▽f|(5.4, 4.95) =[0.45, 0.62]. Under the Riemann metric matrix G, the gradient norm of the path γ(0.5) is ||▽f G = √(▽f T ·G -1 ·▽f), the specific calculation steps are as follows: G -1 ·▽f = [0.8724, -0.1758; -0.1758, 0.7577]·[0.45; 0.62]= [0.8724×0.45 + (-0.1758)×0.62, (-0.1758)×0.45 + 0.7577×0.62]= [0.3926 - 0.1090, -0.0791 + 0.4698]= [0.2836, 0.3907], then calculate ▽f T ·(G -1 ·▽f): ▽f T ·(G-1 • ∇f) = [0.45, 0.62] · [0.2836; 0.3907] = 0.45 x 0.2836 + 0.62 x 0.3907 = 0.1276 + 0.2422 = 0.3698. Finally, the gradient norm is calculated as ||∇f|| = √0.3698 = 0.608. This gradient norm is converted to a path weight by the exponential mapping function exp(gradient norm), which here has a weight value of exp(0.608) ≈ 1.84. Multiplying this path weight by the square root of the geodesic path derivative under the Riemannian metric matrix gives the path infinitesimal length. By integrating over the entire path, the total length of the initial path is calculated to be 12.36. G

[0106] To find the optimal geodesic path, the variational derivative of the path length with respect to the geodesic path needs to be calculated. For each path point γ(t i ), its variational derivative value is calculated. For example, the variational derivative value at γ(0.5) is ∂E / ∂γ(0.5) = (∂w / ∂γ) · ds + w · (∂ds / ∂γ) = [0.552, 0.92] · 7.56 + 1.84 · [0.536, 0.573] = [4.173, 6.955] + [0.986, 1.054] = [5.159, 8.009], and the unit vector of the variational derivative is [5.159, 8.009] / 9.514 = [0.542, 0.842], after scaling and normalization, the variational derivative value is approximately [0.43, 0.67].

[0107] Add this variational derivative to the regularization term with exponential decay to get the path update direction. If the exponential decay factor is 0.95, the regularization coefficient in the first iteration is 0.05, and the path update direction after regularization is (1-0.05) · [0.43, 0.67] = 0.95 · [0.43, 0.67] = [0.41, 0.64].

[0108] The step size needs to be adaptively adjusted during the iteration process. The first and second derivative norms of the geodesic path γ(t) are calculated. Suppose that in a certain iteration, the first derivative norm is 6.32 and the second derivative norm is 2.15, then the path local curvature is calculated as K = 6.32 / 2.15 ≈ 2.93. If the reference iteration step size is set to 0.1, then the actual iteration step size is 0.1 x exp(-0.054) ≈ 0.095.

[0109] ​Using the calculated update direction and iteration step, the geodesic path is updated. For example, if the current path point γ(0.5) = (5.4, 4.95), the update direction is (0.41, 0.64), and the iteration step is 0.095, then the updated path point is (5.4, 4.95) + 0.095 x (0.41, 0.64) = (5.439, 5.011). All points on the path are updated in this way to obtain a new path γ'(t).

[0110] After each iteration, it is necessary to evaluate whether the iteration converges. The difference between the path lengths of the adjacent two iterations is calculated, and the length relative change rate is obtained by dividing the path length of the last iteration. Assuming that the path length of the last iteration is 12.36 and the path length of the current iteration is 12.31, the length relative change rate is |12.31-12.36| / 12.36≈0.004. At the same time, the difference value of the local curvatures of the adjacent two iterations is calculated, and assuming that the average local curvature of the last iteration is 0.054 and the average local curvature of the current iteration is 0.052, the curvature difference value is |0.052-0.054|=0.002.

[0111] The length stability threshold is set to 0.005 and the curvature stability threshold is set to 0.003. When the length relative change rate is less than the length stability threshold and the curvature difference value is less than the curvature stability threshold, it is considered that the iteration converges, and the geodesic path of the current iteration is determined as the optimal geodesic path. In this example, the length relative change rate 0.004 is less than the threshold 0.005, and the curvature difference value 0.002 is less than the threshold 0.003, so the iteration converges, and the current path is the optimal geodesic path sought, and the path length 12.31 is taken as the geodesic distance between the two historical process parameter points.

[0112] In practical applications, multiple iterations are required to reach the convergence condition. For example, when dealing with more complex parameter spaces, 20-30 iterations are required. To prevent infinite iterations, the maximum number of iterations is set to 50, and if the maximum number of iterations is reached and the iteration does not converge, the path of the last iteration is taken as the approximate optimal solution.

[0113] After the calculation of the geodesic distance, it can be used for subsequent process parameter similarity analysis. For example, for a set of historical process parameters {p1, p2, p3, p4, p5}, the calculated geodesic distance matrix shows that the distance between p1 and p3 is the smallest, which is 8.76, indicating that these two sets of process parameters are the most similar; while the distance between p2 and p5 is the largest, which is 25.42, indicating that their process characteristics differ significantly. Based on these geodesic distances, reference is provided for the design of new process parameters.

[0114] Through the above embodiments, the technical solution can accurately calculate the optimal geodesic path and geodesic distance between any two historical process parameters in a low-dimensional parameter space, and provides a theoretical basis and practical tool for similarity measurement of process parameters.

[0115] In an optional embodiment, the historical process parameter combination with the highest product quality evaluation value in each category is marked as a search starting point, parallel parameter search is performed on the search starting point to obtain candidate process parameter combinations, and the candidate process parameter combination with the highest quality evaluation value is selected as the optimal process parameter combination, comprising:

[0116] The multiple quality detection index data corresponding to the process parameters are multiplied by the corresponding weight coefficients and summed to obtain the quality evaluation value;

[0117] In each category, the historical process parameter with the highest quality evaluation value is selected as a search starting point, mutual information entropy of the search starting point and the quality evaluation value is calculated, for the search starting point with mutual information entropy higher than an importance threshold, the parameter value range of the search starting point is equally divided into multiple grid units of the reference number to construct a local search space, and for the search starting point with mutual information entropy lower than the importance threshold, the parameter value range of the search starting point is equally divided into grid units of the reference number to construct a global search space;

[0118] The process parameter difference vector of the current iteration position and the previous iteration position of each search starting point is calculated, the difference vector is divided by the module length to obtain a search direction vector, the quality evaluation value difference of the adjacent iteration positions of each search starting point is calculated to obtain a quality improvement value,

[0119] When the maximum quality improvement value of all search starting points is less than a quality improvement threshold, the process parameter combination corresponding to the current iteration position of each search starting point is obtained, and the process parameter combination with the highest quality evaluation value is selected as the optimal process parameter combination.

[0120] After obtaining the quality improvement value, the method further comprises:

[0121] Based on the search direction vector and the quality improvement value of each search starting point, a direction-revenue mapping matrix for parallel search is established;

[0122] Based on the direction-revenue mapping matrix, the expected revenue of each candidate search direction is calculated, and the candidate search direction with the highest expected revenue is selected as the search direction of each search starting point;

[0123] When the expected revenue of a certain search starting point is lower than the expected value for a continuous preset number of times, the search direction with the highest expected revenue is obtained from the adjacent search starting point.

[0124] In the specific embodiment, a process parameter optimization method based on multi-start point parallel search can be realized by the following technical means.

[0125] The method collects a large amount of historical production data, including process parameter combinations and corresponding multiple quality detection index data. For the historical data, an unsupervised clustering algorithm is used to classify the process parameter combinations. For example, the K-means clustering method can be used to divide the parameter space into multiple categories. In practice, the parameters can be divided into 5 to 8 categories, each of which represents a different parameter configuration mode.

[0126] For each process parameter combination, the system calculates its quality evaluation value by weighted summation. Assuming that the quality detection index corresponding to a certain process parameter combination is [98.2, 0.05, 142.3], and the preset weight coefficient is [0.5, -0.3, 0.2], the quality evaluation value of the parameter combination is calculated as follows: 98.2 x 0.5 + 0.05 x (-0.3) + 142.3 x 0.2 = 49.1 + (-0.015) + 28.46 = 77.545. The weight coefficient can be set according to the actual production needs of the enterprise. A positive weight indicates that the higher the index, the better, and a negative weight indicates that the lower the index, the better.

[0127] In each category, the system selects the historical process parameter combination with the highest quality evaluation value as the search starting point. For example, in 5 categories, the process parameter combinations with quality evaluation values of [82.3, 79.5, 84.1, 78.6, 81.2] are selected as the search starting points.

[0128] For each search starting point, the system calculates the mutual information entropy between it and the quality evaluation value. The mutual information entropy calculation is based on the degree of influence of parameter changes on the quality evaluation value, which is obtained through parameter sensitivity analysis. When the parameter change has a significant impact on the quality, the mutual information entropy value is higher. The system sets the importance threshold to 0.65. For search starting points with mutual information entropy higher than 0.65, the parameter value range is equally divided into multiple grid units of the reference number, such as 30 grid units when the reference number is 10, to construct a local search space. For search starting points with mutual information entropy lower than 0.65, the parameter value range is equally divided into 10 grid units to construct a global search space.

[0129] The search process uses an iterative optimization method. In each iteration, the system calculates the parameter difference vector between the current position and the previous iteration position of the search starting point. For example, the previous iteration position parameter of a search starting point is [210, 55, 0.8], and the current position parameter is [215, 58, 0.75], then the difference vector is [5, 3, -0.05]. Divide the difference vector by its modulus to get the search direction vector. The modulus is calculated as the square root of (5^2 + 3^2 + (-0.05)^2) = 5.83, so the search direction vector is [0.857, 0.514, -0.0086].

[0130] At the same time, the quality evaluation value difference of adjacent iteration positions is calculated. For example, if the quality evaluation value of the previous position is 79.5 and that of the current position is 82.3, the quality improvement value is 2.8. Based on the search direction vector of each search starting point and the quality improvement value, the system establishes a direction-reward mapping matrix for parallel search.

[0131] The direction-reward mapping matrix records each search direction and its corresponding quality improvement value. The system uses this matrix to calculate the expected reward of different candidate search directions. For example, if a search starting point has three candidate search directions, the corresponding expected rewards are [2.1, 1.8, 3.2], and the system will select the candidate direction with an expected reward of 3.2 as the search direction of the search starting point.

[0132] During the search process, if the expected reward of a search starting point is continuously lower than the expected value of 0.5 for three consecutive iterations, the system will obtain the search direction with the highest expected reward from the adjacent search starting points. For example, when the expected reward of search starting point 1 is continuously lower than 0.5, and the expected rewards of adjacent search starting points 2 and 3 are 1.2 and 0.8 respectively, the system will make search starting point 1 use the search direction of search starting point 2 to improve search efficiency.

[0133] The search process continues until the maximum quality improvement value of all search starting points is less than the quality improvement threshold of 0.1, indicating that each search path has converged. At this time, the system obtains the process parameter combinations corresponding to the current iteration positions of all search starting points, compares their quality evaluation values, and selects the highest one as the optimal process parameter combination.

[0134] In an actual application case, a manufacturing enterprise originally had five typical process parameter combinations, with quality evaluation values of [82.3, 79.5, 84.1, 78.6, 81.2]. After 20 rounds of iteration optimization, the quality evaluation values of the optimal parameter combinations evolved by these search starting points were improved to [86.7, 84.2, 89.5, 83.1, 85.8]. The system selected the process parameter combination [225, 62, 0.72] corresponding to the quality evaluation value of 89.5 as the final optimal process parameter combination, which was improved by 5.4 compared with the original parameter combination [215, 58, 0.75], achieving significant quality improvement.

[0135] This method fully utilizes historical data and overcomes the problem of single-point search easily falling into local optimum through a multi-start point parallel search strategy, improving the efficiency and effectiveness of process parameter optimization and being suitable for process parameter optimization in complex industrial production environments.

[0136] In an alternative embodiment, the expected return of each candidate search direction is calculated based on the direction-return mapping matrix, and the candidate search direction with the highest expected return is selected as the search direction of each search starting point, comprising:

[0137] Based on the direction-return mapping matrix, the included angle between adjacent search direction vectors is calculated and subjected to negative exponential operation to obtain a direction similarity, the correlation coefficient between adjacent quality improvement values is calculated, and the direction similarity and the correlation coefficient are weighted and summed to obtain a correlation degree;

[0138] For a candidate search direction, the included angle between it and each search direction vector in the direction-return mapping matrix is calculated and subjected to negative exponential operation to obtain a candidate direction similarity; the candidate direction similarity is multiplied by the sum of the corresponding quality improvement value and the correlation degree, and the calculation results of all search direction vectors are normalized and weighted to obtain the initial expected return of the candidate search direction;

[0139] The negative exponential product of the difference between the current time and the historical time is calculated as a time decay weight; the initial expected return is modified based on the time decay weight to obtain an expected return, and the candidate search direction with the highest expected return is selected as the search direction of each search starting point.

[0140] The embodiment provides a method for calculating the expected return of each candidate search direction based on a direction-return mapping matrix and selecting the candidate search direction with the highest expected return as the search direction of each search starting point. The specific implementation steps of the method will be described in detail below.

[0141] In the embodiment, a direction-return mapping matrix is obtained, which contains search direction vectors and corresponding quality improvement values recorded in the historical search process. For example, it is assumed that five search direction vectors D1(0.5, 0.5, 0.7), D2(0.6, 0.2, 0.8), D3(0.3, 0.7, 0.6), D4(0.8, 0.1, 0.6), and D5(0.4, 0.5, 0.7) are recorded in the historical search process, and the corresponding quality improvement values are Q1=0.25, Q2=0.18, Q3=0.22, Q4=0.30, and Q5=0.20.

[0142] Based on the above direction-return mapping matrix, the included angle between adjacent search direction vectors is calculated and subjected to negative exponential operation to obtain a direction similarity. Specifically, the included angle between D1 and D2 is about 25 degrees, and the direction similarity S 12 =0.78 is obtained by negative exponential operation; the included angle between D2 and D3 is about 39 degrees, and the direction similarity S 23 =0.65 is obtained; the included angle between D3 and D4 is about 44 degrees, and the direction similarity S34 = 0.60; the angle between D4 and D5 is calculated to be about 32 degrees, resulting in a direction similarity S 45 = 0.70.

[0143] The correlation coefficients between adjacent quality boost values are then calculated. For example, the correlation coefficient C 12 = 0.85 between Q1 and Q2, the correlation coefficient C 23 = 0.78 between Q2 and Q3, the correlation coefficient C 34 = 0.82 between Q3 and Q4, and the correlation coefficient C 45 = 0.79 between Q4 and Q5. The direction similarity and the correlation coefficients are weighted and summed to obtain the relevance. The weight of the direction similarity is set to be 0.6 and the weight of the correlation coefficient is set to be 0.4. Then, the relevance R 12 = 0.6 x 0.78 + 0.4 x 0.85 = 0.81 between D1 and D2, the relevance R 23 = 0.6 x 0.65 + 0.4 x 0.78 = 0.70 between D2 and D3, the relevance R 34 = 0.6 x 0.60 + 0.4 x 0.82 = 0.69 between D3 and D4, and the relevance R 45 = 0.6 x 0.70 + 0.4 x 0.79 = 0.74 between D4 and D5.

[0144] For a new candidate search direction D c (0.55, 0.35, 0.75), the angles between it and the historical search direction vectors are calculated and the negative exponent operation is performed to obtain the candidate direction similarity. The angle between D c and D1 is calculated to be about 15 degrees, and the negative exponent operation is performed to obtain the candidate direction similarity S c1 = 0.87; similarly, S c2 = 0.90, S c3 = 0.72, S c4 = 0.65, and S c5 = 0.80.

[0145] The candidate direction similarity is multiplied by the sum of the corresponding quality boost value and the correlation. For D1, the correlation calculation result is 0.87x(0.25+0.81)=0.92; for D2, the correlation calculation result is 0.90x(0.18+0.70+0.81)=1.52; for D3, the correlation calculation result is 0.72x(0.22+0.70+0.69)=1.16; for D4, the correlation calculation result is 0.65x(0.30+0.69+0.74)=1.13; for D5, the correlation calculation result is 0.80x(0.20+0.74)=0.75. The calculation results of all search direction vectors are normalized and weighted to obtain the initial expected income E of the candidate search direction D c c =(0.92+1.52+1.16+1.13+0.75) / 5=1.10.

[0146] Then, the negative exponential product of the current time and the historical time difference is calculated as the time decay weight. Assuming that the current time is T, and the recorded times of the historical search directions are T1, T2, T3, T4, and T5, respectively. The time difference between T and T1 is 5 hours, and the negative exponential operation obtains the time decay weight W1=0.78; the time difference between T and T2 is 4 hours, and the time decay weight W2=0.82 is obtained; the time difference between T and T3 is 2 hours, and the time decay weight W3=0.90 is obtained; the time difference between T and T4 is 8 hours, and the time decay weight W4=0.65 is obtained; the time difference between T and T5 is 3 hours, and the time decay weight W5=0.86 is obtained.

[0147] The initial expected income is modified based on the time decay weight to obtain the expected income. The calculation of the modified expected income E' c =E c x(W1+W2+W3+W4+W5) / 5=1.10x(0.78+0.82+0.90+0.65+0.86) / 5=1.10x0.80=0.88.

[0148] For other candidate search directions, such as D a (0.62, 0.45, 0.64) and D b (0.48, 0.52, 0.71), their expected incomes are also calculated according to the above steps. Assuming that the expected income of D a is 0.76, and the expected income of D b is 0.92. Since the expected income of D b is the highest, D b is selected as the search direction of the search starting point.

[0149] ​In the case of multiple search starting points, the expected return of each candidate search direction is calculated according to the above steps for each search starting point, and the candidate search direction with the highest expected return is selected as the search direction of the search starting point. For example, for the search starting point P1, the expected return of the candidate search direction D b is the highest at 0.92, and D b is selected as the search direction of P1; for the search starting point P2, the expected return of the candidate search direction D c is the highest at 0.88, and D c is selected as the search direction of P2.

[0150] The above embodiment describes in detail the method of calculating the expected return of each candidate search direction based on the direction-return mapping matrix, and selecting the candidate search direction with the highest expected return as the search direction of each search starting point. This method realizes the evaluation and selection of candidate search directions by calculating parameters such as direction similarity, correlation coefficient, association degree, candidate direction similarity, initial expected return, time decay weight, and expected return, making the search process more efficient and accurate, and improving the overall search quality. This method can be applied to various scenarios that require search optimization, such as path planning, resource scheduling, parameter optimization, and other fields.

[0151] The die casting island process parameter optimization system based on machine learning includes:

[0152] The first unit is used to obtain process parameters in the die casting process, divide the process parameters into multiple process stages according to the die casting process, calculate the energy loss value of each process stage according to the energy conversion relationship in the die casting process, calculate the entropy weight of each process stage according to the product quality data, identify the target process stage that affects the quality, optimize the process parameters of the target process stage when the energy loss value is higher than the loss threshold, and obtain the locally optimized process parameters.

[0153] The second unit is used to construct historical process parameters into a historical parameter matrix, perform eigenvalue decomposition on the covariance matrix, construct a low-dimensional parameter space, map the historical process parameters to the low-dimensional parameter space, calculate the geodesic distance based on the locally optimized process parameters, and combine the historical process parameters with a geodesic distance less than a distance threshold into the same class. The historical process parameter combination with the highest product quality evaluation value in each class is marked as a search starting point, and parallel parameter search is performed on the search starting point to obtain candidate process parameter combinations. The candidate process parameter combination with the highest quality evaluation value is selected as the optimal process parameter combination.

[0154] The third unit is used to generate an adjustment instruction for the die casting island process parameters based on the optimal process parameter combination, issue the adjustment instruction to the die casting island control system, and adjust the process parameters of each device in real time.

[0155] In a third aspect, the present application provides an electronic device, comprising:

[0156] a processor;

[0157] a memory for storing processor-executable instructions;

[0158] wherein the processor is configured to invoke the instructions stored in the memory to perform the method described above.

[0159] In a fourth aspect, the present application provides a computer-readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the method described above.

[0160] The present application can be a method, apparatus, system, and / or computer program product. Computer program products can include computer-readable storage media having computer-readable program instructions loaded thereon for performing various aspects of the present application.

[0161] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for die casting island process parameter optimization based on machine learning, characterized in that, The method comprises the following steps: acquiring process parameters in the die casting process, dividing the process parameters into multiple process stages according to the die casting process, and calculating energy loss values of the process stages according to energy conversion relationships in the die casting process; calculating entropy weights of the process stages according to product quality data, identifying a target process stage that affects quality, and optimizing process parameters of the target process stage when the energy loss value of the target process stage is higher than a loss threshold value to obtain locally optimized process parameters; constructing historical process parameters into a historical parameter matrix, performing eigenvalue decomposition on a covariance matrix of the historical parameter matrix, constructing a low-dimensional parameter space, mapping the historical process parameters to the low-dimensional parameter space, calculating geodesic distances based on the locally optimized process parameters, and combining historical process parameter combinations with geodesic distances less than a distance threshold value into the same class; labeling a historical process parameter combination with the highest product quality evaluation value in each class as a search starting point, performing parallel parameter search on the search starting point to obtain candidate process parameter combinations, and selecting a candidate process parameter combination with the highest quality evaluation value as an optimal process parameter combination; generating adjustment instructions for die casting island process parameters according to the optimal process parameter combination, issuing the adjustment instructions to a die casting island control system, and adjusting device process parameters in real time.

2. The method of claim 1, wherein, The method comprises the following steps: acquiring process parameters in the die casting process, dividing the process parameters into multiple process stages according to the die casting process, and calculating energy loss values of the process stages according to energy conversion relationships in the die casting process; calculating entropy weights of the process stages according to product quality data, identifying a target process stage that affects quality, and optimizing process parameters of the target process stage when the energy loss value of the target process stage is higher than a loss threshold value to obtain locally optimized process parameters, comprising: collecting multi-dimensional quality detection index data of the die casting product, normalizing the multi-dimensional quality detection index data to obtain normalized detection data; dividing the normalized detection data by the sum of all normalized data of the corresponding detection index to obtain a proportion value, and dividing the sum of the product of the proportion value and the natural logarithm thereof by the sample quantity to obtain the information entropy of each detection index; calculating the correlation coefficient of each process parameter with each detection index using the Spearman correlation coefficient, performing weighted summation on the correlation coefficients to obtain the correlation degree of the process stage and each detection index, and adding the product of the correlation degree and the corresponding information entropy to obtain the entropy weight of each process stage, and selecting the process stage with the largest entropy weight as the target process stage; 3. The method of claim 1, wherein, when the energy loss value of the target process stage is higher than a preset loss threshold value, calculating a first ratio of the energy loss value to the preset loss threshold value and a second ratio of the average value of each detection index to the standard value, and performing weighted summation on the first ratio and the second ratio to obtain a local optimization target value; calculating the gradient value of the local optimization target value with respect to each process parameter, and optimizing the process parameters of the target process stage based on the gradient value; when the deviation value of the local optimization target value of two adjacent iterations is less than a convergence threshold value, the current process parameters are recorded as the locally optimized process parameters. The method comprises the following steps: constructing historical process parameters into a historical parameter matrix, performing eigenvalue decomposition on a covariance matrix of the historical parameter matrix, constructing a low-dimensional parameter space, mapping the historical process parameters to the low-dimensional parameter space, calculating geodesic distances based on the locally optimized process parameters, and combining historical process parameter combinations with geodesic distances less than a distance threshold value into the same class, comprising: Obtain a historical process parameter combination in a die casting process and construct a historical parameter matrix; multiply the historical parameter matrix by its transpose matrix and divide by the sample number to obtain a covariance matrix; Perform eigenvalue decomposition on the covariance matrix to obtain a plurality of eigenvalues and eigenvectors, perform inner product operation on the historical parameter matrix and the eigenvectors whose eigenvalues are greater than a feature threshold value to obtain projection values of the historical process parameter combination in each feature direction; multiply the projection values by contribution rates corresponding to the eigenvalues to obtain mapping coordinates of the historical process parameters in a low-dimensional parameter space, and the contribution rates are calculated by dividing the eigenvalues by the sum of the eigenvalues; Calculate the partial derivative of the locally optimized process parameter with respect to the low-dimensional parameter space coordinates; sum the partial derivative multiplied by the randomly initialized weight parameter and add a unit matrix to obtain a Riemann metric matrix; use the Riemann metric matrix to solve the minimum path length of any two historical process parameters in the low-dimensional parameter space by variational method, and take the minimum path length as the geodesic distance; Classify the historical process parameter combinations with geodesic distance less than a preset distance threshold into the same class to obtain a historical process parameter classification result.

4. The method of claim 3, wherein, Use the Riemann metric matrix to solve the optimal geodesic path of any two historical process parameters in the low-dimensional parameter space by variational method, and take the path length of the optimal geodesic path as the geodesic distance, including: For any two historical process parameters, calculate the geodesic path in the low-dimensional parameter space by linear interpolation method, calculate the gradient norm of the Riemann metric matrix on the geodesic path, obtain the path weight by exponential mapping of the gradient norm, and multiply the path weight by the square root of the derivative of the geodesic path under the Riemann metric matrix to obtain the path length; Calculate the variational derivative of the path length with respect to the geodesic path, add the regularization term with exponential decay to the variational derivative to obtain the path update direction; calculate the first derivative norm and the second derivative norm of the geodesic path, divide the second derivative norm by the square of the first derivative norm to obtain the local curvature of the path; multiply the reference iteration step size by the negative exponential function of the local curvature of the path to obtain the iteration step size; In each iteration, calculate the difference between the path lengths of adjacent two iterations divided by the path length of the previous iteration to obtain the length relative change rate, and calculate the difference between the local curvatures of adjacent two iterations to obtain the curvature difference value; when the length relative change rate is less than a length stability threshold value and the curvature difference value is less than a curvature stability threshold value, the geodesic path of the current iteration is determined as the optimal geodesic path, and the path length of the optimal geodesic path is taken as the geodesic distance between the two historical process parameter points.

5. The method of claim 1, wherein, Mark the historical process parameter combination with the highest product quality evaluation value in each category as a search starting point, and perform parallel parameter search on the search starting point to obtain a candidate process parameter combination, and select the candidate process parameter combination with the highest quality evaluation value as the optimal process parameter combination, including: Multiply a plurality of quality detection index data corresponding to the process parameters by corresponding weight coefficients and sum them up to obtain a quality evaluation value; selecting a historical process parameter with the highest quality evaluation value in each category as a search starting point, calculating mutual information entropy of the search starting point and the quality evaluation value, dividing the parameter value range of the search starting point with the mutual information entropy higher than an importance threshold into multiple grid units of a multiple reference number to construct a local search space, and dividing the parameter value range of the search starting point with the mutual information entropy lower than the importance threshold into grid units of a reference number to construct a global search space; calculating a process parameter difference vector between a current iteration position and a last iteration position of each search starting point, dividing the difference vector by a module length to obtain a search direction vector, and calculating a quality evaluation value difference between adjacent iteration positions of each search starting point to obtain a quality improvement value; when the maximum quality improvement value of all search starting points is less than a quality improvement threshold, obtaining process parameter combinations corresponding to current iteration positions of all search starting points, and selecting a process parameter combination with the highest quality evaluation value as an optimal process parameter combination.

6. The method of claim 5, wherein, After obtaining the quality improvement value, the method further includes: establishing a direction-reward mapping matrix of parallel search based on the search direction vector and the quality improvement value of each search starting point; calculating an expected reward of each candidate search direction based on the direction-reward mapping matrix, and selecting a candidate search direction with the highest expected reward as the search direction of each search starting point; when the expected reward of a search starting point is lower than an expected value for a preset number of times in succession, obtaining a search direction with the highest expected reward from an adjacent search starting point.

7. The method of claim 6, wherein, calculating an expected reward of each candidate search direction based on the direction-reward mapping matrix, and selecting a candidate search direction with the highest expected reward as the search direction of each search starting point, includes: calculating an included angle between adjacent search direction vectors based on the direction-reward mapping matrix and performing negative exponential operation to obtain a direction similarity, calculating a correlation coefficient between adjacent quality improvement values, and performing weighted summation on the direction similarity and the correlation coefficient to obtain a correlation degree; for a candidate search direction, calculating an included angle between the candidate search direction and each search direction vector in the direction-reward mapping matrix and performing negative exponential operation to obtain a candidate direction similarity; multiplying the candidate direction similarity by a sum of a corresponding quality improvement value and the correlation degree, and performing normalized weighting on calculation results of all search direction vectors to obtain an initial expected reward of the candidate search direction; calculating a negative exponential product of a current time and a historical time difference as a time decay weight; modifying the initial expected reward based on the time decay weight to obtain an expected reward, and selecting a candidate search direction with the highest expected reward as the search direction of each search starting point.

8. A machine learning based die casting island process parameter optimization system for implementing the method of any one of claims 1-7, characterized in that, includes: a first unit configured to obtain process parameters in a die casting process, divide the process parameters into a plurality of process stages according to a die casting process, and calculate energy loss values of the process stages according to an energy conversion relationship in the die casting process; calculate entropy weights of the process stages according to product quality data, identify a target process stage affecting quality, and optimize process parameters of the target process stage when the energy loss value of the target process stage is higher than a loss threshold to obtain locally optimized process parameters. The second unit is configured to construct historical process parameters into a historical parameter matrix, perform eigenvalue decomposition on a covariance matrix of the historical parameter matrix, construct a low-dimensional parameter space, map the historical process parameters to the low-dimensional parameter space, calculate geodesic distances based on the locally optimized process parameters, and combine historical process parameters with geodesic distances less than a distance threshold into the same class; The historical process parameter combination with the highest product quality evaluation value in each class is marked as a search starting point, parallel parameter searching is performed on the search starting point, candidate process parameter combinations are obtained, and the candidate process parameter combination with the highest quality evaluation value is selected as an optimal process parameter combination; The third unit is configured to generate an adjustment instruction of the die casting island process parameters according to the optimal process parameter combination, issue the adjustment instruction to a die casting island control system, and adjust the process parameters of each device in real time.

9. An electronic device, comprising: The computer program instructions are executed by the processor to implement the method in any one of claims 1 to 7. The computer program instructions are executed by the processor to implement the method in any one of claims 1 to 7. ​ ​ 10. A computer-readable storage medium having stored thereon computer program instructions, wherein, ​

Images