Tool magazine tool changing method and system

Abstract

The application relates to a tool magazine tool changing method of a numerical control machining center. According to collected tool operation data, data preprocessing is carried out, preset feature extraction is carried out, correlation analysis and importance evaluation are carried out, and a data set is generated. A bidirectional recurrent neural network model is used to continuously train the data set, a tool magazine optimization model is generated, the tool magazine optimization model is deployed, real-time operation data of the numerical control machining center is continuously collected, the real-time operation data is input into the tool magazine optimization model, and a prediction result is generated. According to the prediction result, the tool magazine tool changing strategy is dynamically optimized, and the tool magazine tool changing strategy is used to replace the initial tool changing strategy of the numerical control machining center in real time. The tool magazine tool changing method can predict tool replacement of the tool magazine by using a bidirectional recurrent neural network model, optimize the tool magazine tool changing process of the numerical control machining center, reduce tool changing time, and improve the machining efficiency of the numerical control center.

Description

Technical Field

[0001] This application relates to the field of tool operation technology for CNC machining centers, and more specifically to a tool changing method and system for CNC machining centers. Background Technology

[0002] A CNC machining center is a highly efficient, automated, and high-precision machine tool that is not only used in the machinery manufacturing industry but also widely used in aerospace, automotive, and electronics industries. A CNC machining center integrates multiple machining processes such as drilling, milling, boring, reaming, and tapping, enabling the completion of complex machining of multiple surfaces in a single setup, aiming to improve production efficiency and reduce labor intensity.

[0003] Currently, CNC machining centers typically consist of several key components: the machine tool body, the CNC system, the automatic tool changer (ATC), and auxiliary devices. The machine tool body includes the mechanical structure such as the bed, worktable, and spindle. The CNC system is the core of the machining center, controlling the equipment's operation and coordination through programming and algorithms. The automatic tool changer enables the automatic replacement of various tools, improving machining efficiency. The automatic tool changer usually includes a tool magazine and a robotic arm. The tool magazine stores various tools, arranged according to pre-set rules, and is automatically changed by the automatic tool changer, enabling tool switching between different machining processes. Auxiliary devices, such as cooling systems and chip removal systems, ensure the smooth operation of the machining process.

[0004] During tool changing, the CNC system issues a tool changing command according to the machining program. The tool magazine selects the appropriate tool according to the command, and the automatic tool changer removes the selected tool from the tool magazine and installs it onto the spindle. At the same time, the original tool is returned to the tool magazine. After the tool changing is completed, the CNC system continues to execute the machining program. Therefore, in CNC machining, tool changing time is one of the important factors affecting machining efficiency. Frequent tool changing operations not only increase non-productive time but also affect the machining cycle time and overall production efficiency.

[0005] Therefore, there is an urgent need for a tool changer method that can optimize the tool changing process during the production of CNC machining centers. Summary of the Invention

[0006] This application provides a tool changing method and system for CNC machining centers. It uses a bidirectional recurrent neural network model to predict tool changes in the tool magazine, thereby optimizing the tool changing process of the CNC machining center, reducing tool changing time, and thus improving the machining efficiency of the CNC center.

[0007] In a first aspect, this application provides a tool changing method for a CNC machining center, the method comprising the following steps: preprocessing data based on collected tool operation data; extracting preset features based on the preprocessed tool operation data, and then performing correlation analysis and importance assessment based on the extracted preset features to generate a dataset; continuously training the dataset using a bidirectional recurrent neural network model to generate a tool magazine optimization model, the training process comprising: dividing the dataset into a training set and a test set, updating the weights of the training set using forward propagation, loss calculation, and backpropagation, and then using the test set for performance evaluation; deploying the tool magazine optimization model while continuously collecting real-time operation data of the CNC machining center, and then inputting the real-time operation data into the tool magazine optimization model to generate prediction results; dynamically optimizing the tool changing strategy based on the prediction results, and then replacing the initial tool changing strategy of the CNC machining center with the tool changing strategy in real time.

[0008] In one alternative of the first aspect, the method for data preprocessing includes the following steps: reading collected tool operation data, checking and deleting duplicate data, and then performing data cleaning: handling missing values ​​and outliers; performing data standardization based on the cleaned tool operation data; and performing data verification based on the standardized tool operation data.

[0009] In one alternative embodiment of the first aspect, when generating the dataset, the method includes the following steps: extracting preset features contained in the tool operation data, the preset features including one or more of machining sequence, machining time, tool type, machining material, tool life, and tool wear state; and performing correlation analysis using Pearson correlation coefficient. Generate features that are strongly correlated with the target features; where x i Let y be the feature value of the i-th sample. i Let i be the target value for the i-th sample. The mean of the eigenvalues, The mean of the target value is n, and the number of samples is n. Gradient boosting trees are used to evaluate the importance of each feature, the features are ranked by importance, and then the features with the preset ranking are selected to generate the dataset.

[0010] In one alternative of the first aspect, when forward propagation is used on the training set, the method includes the following steps: for the input l at time step t t Update the hidden state of the forward LSTM layer: in, This represents the positive hidden state at time step t. This represents the forward hidden state at time step t-1; for the input l at time step t... t Update the hidden state of the inverse LSTM layer: in, The reverse hidden state at time step t. The hidden state at time step t+1 is the forward hidden state; the hidden states of the forward LSTM layer and the backward LSTM layer are concatenated as the output: in, To connect the forward and backward hidden states along the feature dimension, h t This represents the hidden state after forward and reverse connections at time step t.

[0011] In one alternative to the first aspect, when performing loss calculation on the training set, the method includes the following steps:

[0012] The output at each time step t is calculated based on forward propagation. in, h is the predicted value of the i-th sample at time step t; t,i Let f be the hidden state of the i-th sample after forward and backward connections at time step t, and let f be the activation function of the output layer; calculate the loss at each time step t: Among them, L t The loss at each time step t, where n is the number of samples, y t,i Let L be the true value of the i-th sample at time step t; let L be the loss at each time step t. t Accumulate to generate the total loss L total : Where T is the time step length.

[0013] In one alternative to the first aspect, when loss calculation and backpropagation are applied to the training set, the method includes the following steps: calculating the error term for each time step t: in, The error term for each time step t, For the predicted output at time step t, y t Given the true value of time step t; calculate the error term for cell state: in, The error term is for the cell state; the error term for the output gate is calculated: in, The error term for the output gate is calculated; the error term for the input gate is calculated as follows: in, The error term for the input gate; the error term for the candidate cells is calculated: in, For candidate cells, calculate the error term; for the forgetting gate, calculate the error term: in, The error term for the forget gate; the error term for the hidden state: in, Let the error term be the hidden state; based on the error term, calculate the gradients of the forget gate weights and biases: in, Let be the gradient of the weights of the forget gate. h is the gradient of the bias of the forget gate. t-1 Let be the hidden state of the i-th sample after forward and backward connections at time step t-1; calculate the gradients of the input gate weights and biases based on the error term: in, Let be the gradient of the weights of the input gate. The gradient of the input gate bias; based on the error term, calculate the gradient of the candidate cell state weights and bias: in, Let be the gradient of the weights of the forget gate. The gradient of the bias of the forget gate; based on the error term, calculate the gradients of the output gate weights and bias: in, Let be the gradient of the weights of the forget gate. The gradient of the bias for the forget gate; the weights and biases are updated using the gradient descent optimization method.

[0014] In one alternative of the first aspect, when updating the weights and biases using a gradient descent optimization method, the method includes the following steps: updating the gradients of the weights and biases of the forget gate according to the learning rate η: Among them, W f ′ represents the updated weight of the forget gate, b f Let ' be the updated bias of the forget gate; based on the learning rate η, update the gradients of the input gate weights and the bias: Among them, W i ′ represents the updated weight of the forget gate, b i Let ' be the updated bias of the forget gate; based on the learning rate η, update the gradients of the weights and biases of the candidate cell states: Among them, W C ′ represents the updated weight of the forget gate, b C Let ' be the updated bias of the forget gate; based on the learning rate η, update the gradients of the weights and bias of the output gate: Among them, W O ′ represents the updated weight of the forget gate, b O ′ represents the offset of the updated Forgot Gate.

[0015] In one alternative of the first aspect, when using a test set for performance evaluation, the method includes the following steps: using a trained bidirectional recurrent neural network model to predict the test set, obtaining predicted labels and true labels; constructing a confusion matrix based on the prediction results and true labels: Where TP is the number of samples correctly predicted as positive, TN is the number of samples correctly predicted as negative, FP is the number of samples incorrectly predicted as positive, and FN is the number of samples incorrectly predicted as negative; Accuracy is calculated as follows: Calculate recall rate: Calculation accuracy: Calculate the F1 score:

[0016] In one alternative of the first aspect, when dynamically optimizing the tool changer strategy, the method includes the following steps: extracting tool machining sequence data based on the prediction results: {D1,D2,D3,...,D...} N Based on the tool machining sequence data, define the tool change path between tools under the current tool magazine layout and calculate the tool change time: Among them, D switch The total time for all tool changes is given by N, where N is the number of tools and D is the total time for all tool changes. a D is the tool used in step a. a+1 For the tool used in step a+1, d(D) a D a+1 ) for cutting tool D a Switch to tool D a+1 The time between; constructing the tool change time matrix: Then, the nearest neighbor method is used to generate the optimal tool change sequence: {D′1,D′2,D′3,...,D′ N The tool magazine tool changing strategy is generated based on the optimal tool changing order, and the tools in the tool magazine are dynamically sorted.

[0017] Secondly, this application provides a tool changer system for a CNC machining center, comprising: a data processing module for preprocessing collected tool operation data; a feature processing module for extracting preset features from the preprocessed tool operation data, and then performing correlation analysis and importance assessment based on the extracted preset features to generate a dataset; a model training module for continuously training the dataset using a bidirectional recurrent neural network model to generate a tool changer optimization model, the training process including: dividing the dataset into a training set and a test set, updating the weights of the training set using forward propagation, loss calculation, and backpropagation, and then using the test set for performance evaluation; a model deployment module for deploying the tool changer optimization model; a data acquisition module for continuously acquiring real-time operation data of the CNC machining center; and a tool changer optimization module for inputting the real-time operation data into the tool changer optimization model to generate prediction results; and dynamically optimizing the tool changer strategy based on the prediction results, thereby replacing the initial tool changer strategy of the CNC machining center in real time with the tool changer strategy.

[0018] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0019] The accompanying drawings, which are incorporated herein and form part of this specification, illustrate one or more embodiments of the present application and, together with the description, serve to explain the principles of the present application and to enable those skilled in the art to make and use the present application.

[0020] Figure 1 This is a flowchart of an exemplary tool changer method for a CNC machining center according to some embodiments of this application.

[0021] Figure 2 This is a flowchart of an exemplary data preprocessing method according to some embodiments of this application.

[0022] Figure 3 This is a flowchart of an exemplary dataset generation method according to some embodiments of this application.

[0023] Figure 4 This is a flowchart of an exemplary forward propagation and loss calculation method according to some embodiments of this application.

[0024] Figure 5 This is a flowchart of an exemplary backpropagation method according to some embodiments of this application.

[0025] Figure 6 This is a flowchart of an exemplary weight and bias update method according to some embodiments of this application.

[0026] Figure 7 This is a flowchart of an exemplary performance evaluation method according to some embodiments of this application.

[0027] Figure 8 This is a line graph illustrating how an exemplary evaluation metric changes with training rounds, according to some embodiments of this application.

[0028] Figure 9 This is a flowchart of an exemplary method for optimizing tool magazine tool changing strategy according to some embodiments of this application.

[0029] Figure 10 This is a flowchart of an exemplary automatic tool management method according to some embodiments of this application.

[0030] Figure 11 This is a flowchart of an exemplary tool adjustment method according to some embodiments of this application.

[0031] Figure 12This is a schematic diagram of the module connections of an exemplary CNC machining center tool magazine changing system according to some embodiments of this application.

[0032] Figure 13 This is a schematic diagram of the connection of an exemplary control unit according to some embodiments of this application.

[0033] Figure 14 This is a schematic diagram illustrating the configuration of an exemplary terminal device according to some embodiments of this application. Detailed Implementation

[0034] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided to make this application more complete and comprehensive, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of this application.

[0035] To make it easier to understand, let's first give a brief introduction to CNC machining centers.

[0036] A CNC machining center is a highly automated, multi-functional CNC machine tool widely used for high-precision machining of complex parts. Its main components include a CNC system, machine tool body, automatic tool changer (ATC), and various auxiliary devices.

[0037] Specifically, the CNC system is the core of a CNC machining center. It is responsible for interpreting the machining program and issuing control commands. It connects with the machine tool body, automatic tool changer, and various auxiliary devices included in the CNC machining center to control the various movements and functions of the machine tool and ensure the precise execution of the entire production process. The CNC system typically includes a computer, various control modules, and a human-machine interface (HMI). The computer and various control modules are used to process CNC codes and generate corresponding control signals. The HMI provides a human-machine interface through which operators input machining programs, monitor the machining process, and adjust machining parameters.

[0038] Specifically, the machine tool body is the mechanical structure of a CNC machining center, providing a stable machining platform and necessary motion functions. The machine tool body typically includes a bed, worktable, spindle, and servo mechanism. The bed is the foundation of the machine tool, providing a robust and stable support structure, usually made of cast iron or high-strength steel to ensure rigidity and stability during machining. The worktable is the plane on which the workpiece is mounted, capable of moving along multiple axes to perform multi-directional machining in conjunction with the spindle. The worktable can be fixed or rotating to adapt to different machining needs. The spindle is the core component that drives the tool rotation; its speed and power directly affect machining efficiency and accuracy. The servo mechanism typically includes servo drivers and servo motors for precise control of the movement of each axis.

[0039] Specifically, the automatic tool changer (ATC) is a key component of a CNC machining center, used to automatically change various tools, improving machining efficiency and automation. An ATC typically includes a tool magazine and a robotic arm. The tool magazine stores and manages tools, arranged according to specific rules, and works with the robotic arm to automatically change tools, enabling rapid switching between different machining operations. Tool magazines hold various tools, with capacities ranging from a few to dozens. Common types include chain-type and disc-type. In a disc-type magazine, tools are arranged on discs with multiple tool positions, each holding one tool. In a chain-type magazine, tools are arranged on a chain with multiple tool holders, each holding one tool. This type can hold more tools and is suitable for complex machining scenarios requiring a large number of tools. The robotic arm is the automatic tool changing device, capable of quickly and accurately transferring tools from the tool magazine to the spindle or vice versa. The specific tool changing process involves the CNC system controlling the robotic arm to perform tool changing operations according to a preset program, including tool identification, gripping, transfer, and installation.

[0040] Specifically, auxiliary devices ensure the smooth operation of the machining process and improve machining quality and efficiency. These devices typically include cooling systems, chip removal systems, lubrication systems, protective devices, and measuring devices. Specifically: cooling systems usually include coolant pumps, coolant nozzles, and coolant recovery devices to cool the tool and workpiece during cutting, reducing heat buildup and preventing tool overheating and workpiece thermal deformation; chip removal systems typically include chip conveyors, chip belts, and chip boxes to promptly remove metal chips generated during cutting, keeping the machining area clean and preventing chip accumulation from affecting machining quality and equipment operation; lubrication systems typically include centralized lubrication devices and automatic lubrication devices to lubricate various moving parts of the machine tool, reducing friction and extending equipment life; protective devices typically include machine tool guards, door locks, and safety light curtains to protect operator safety, prevent chip and coolant splashing, and ensure a safe machining environment; measuring devices include online measuring systems and tool setters to monitor workpiece dimensions and tool position in real time, perform accuracy correction and error compensation, and improve machining accuracy.

[0041] Therefore, for reference Figure 1 As shown, in some embodiments, the tool changing method for a CNC machining center according to this application includes the following steps:

[0042] S1: Perform data preprocessing based on the collected tool operation data.

[0043] Specifically, tool operation data includes data such as tool usage records, machining sequence, machining time, and tool change records obtained from CNC machining centers. The specific tool operation data can be added or reduced by the operator according to actual needs. Among them, the collected data is stored in a database to ensure the integrity and availability of the data.

[0044] In actual implementation, refer to Figure 2 As shown, data preprocessing includes the following steps:

[0045] S11: Read the collected tool operation data, check and delete duplicate data, and then perform data cleaning: handle missing values ​​and outliers.

[0046] Specifically, duplicate records are checked and deleted to prevent duplicate data from biasing the model. Methods for handling missing values ​​mainly include deletion, imputation, and interpolation. If the proportion of missing values ​​is very small, data containing missing values ​​is directly deleted. If missing values ​​need to be imputed, existing mean, median, and mode methods are used. If interpolation is required for missing values, existing linear interpolation methods are used. Outlier handling can be performed through statistical methods, typically using the existing Z-Score method to detect outliers and then deleting them.

[0047] S12: Perform data standardization processing based on the cleaned tool operation data.

[0048] Specifically, data standardization can give data with different characteristics the same dimensions. Common standardization methods include Z-Score standardization and Min-Max normalization. Z-Score standardization converts data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Min-Max normalization scales data to a specified range (usually 0 to 1).

[0049] This application uses Min-Max normalization as an example. The formula for data standardization using the Min-Max normalization method is as follows:

[0050]

[0051] Where R′ represents the tool operation data after data standardization, R represents the tool operation data after data cleaning, and R min R is the minimum value of the tool operation data. max This represents the maximum value of the tool operation data.

[0052] S13: Perform data verification based on standardized tool operation data.

[0053] Specifically, after the tool operation data is standardized, statistical measures such as mean, standard deviation, and quartiles are calculated to assess the central tendency and dispersion of the data distribution; and / or skewness and kurtosis are calculated to assess the deviation and central tendency of the data distribution. The standardized data are checked to ensure that the data distribution is uniform and to eliminate the impact of outliers and missing values ​​on model training.

[0054] Among them, the skewness is calculated as follows:

[0055]

[0056] Where S is the skewness, E is the expected value (mean), X is a sample value in the data, μ is the mean, and υ is the standard deviation. Skewness = 0 indicates that the data distribution is symmetrical; skewness > 0 indicates that the data distribution is positively skewed (right-skewed), meaning that the right tail of the distribution is longer and most data are concentrated on the left side; skewness < 0 indicates that the data distribution is negatively skewed (left-skewed), meaning that the left tail of the distribution is longer and most data are concentrated on the right side. Skewness is a statistic that describes the degree of asymmetry in the data distribution and is used to measure the symmetry of the data distribution relative to the mean. The larger the absolute value of the skewness value, the greater the degree of deviation of the data distribution from symmetry.

[0057] Among them, kurtosis is calculated as follows:

[0058]

[0059] Where K is kurtosis. Kurtosis = 3 indicates that the peak of the data distribution is the same as that of the normal distribution. In this case, 3 is usually subtracted to adjust the kurtosis of the normal distribution to 0. Kurtosis > 0 indicates that the peak of the data distribution is higher than that of the normal distribution, indicating that more data are concentrated around the mean. Kurtosis < 0 indicates that the data distribution is flatter than that of the normal distribution, indicating that the data are more dispersed around the mean. Kurtosis is a statistic that describes the peak and flatness of the data distribution and is used to measure the peak or flatness of the data distribution relative to the normal distribution.

[0060] S2: Based on the preprocessed tool operation data, perform preset feature extraction, and then perform correlation analysis and importance assessment based on the extracted preset features to generate a dataset.

[0061] In actual implementation, refer to Figure 3 As shown, the following steps are included when generating the dataset:

[0062] S21: Extract preset features contained in the tool operation data. The preset features include one or more of the following: machining sequence, machining time, tool type, machining material, tool life, and tool wear status.

[0063] Feature extraction involves extracting meaningful information from the tool operation data after data preprocessing so that the model can effectively learn and predict; the machining sequence is the sequential encoding of the machining process, which can be done using integer encoding or one-hot encoding; machining time is the time spent in the machining process and can be directly used as a numerical feature; tool type encoding and machining material type encoding can both be done using integer encoding or one-hot encoding; tool life can be used as a numerical feature; and tool wear status can be measured by the operator operating the machine or obtained directly from sensor data and used as a numerical feature.

[0064] S22: Correlation analysis using Pearson correlation coefficient:

[0065]

[0066] Generate features that are strongly correlated with the target features; where x i Let y be the feature value of the i-th sample. i Let i be the target value for the i-th sample. The mean of the eigenvalues, Let n be the mean of the target values, and n be the number of samples.

[0067] In correlation analysis, the Spearman correlation coefficient can also be used.

[0068]

[0069] Γ i Let be the rank difference between the feature value and the target value of the i-th sample.

[0070] S23: Use gradient boosting trees to evaluate the importance of each feature, rank the features by importance, and then select the features with the preset ranking to generate the dataset.

[0071] Gradient boosting trees are an ensemble learning method that combines multiple decision trees, each trained on the residuals of all preceding trees. Feature importance is typically determined by calculating the contribution of each feature's split points across all trees to the model's performance. Common metrics include Gini importance and loss reduction. Gini importance measures the importance of each feature in reducing Gini impurity, while loss reduction measures the importance of each feature in reducing the target loss function.

[0072] Among them, Gini impurity is used to measure the impurity of a node. For node m, the formula for calculating its Gini impurity G(m) is:

[0073]

[0074] C represents the number of categories, p c This represents the proportion of samples belonging to category c.

[0075] Among them, for the split of feature j at node m, the formula for calculating the Gini importance ΔI is:

[0076]

[0077] M is the number of samples at node m. left and M right G(left) represents the number of samples for the left and right child nodes after the split, and G(right) represents the Gini impurity of the left and right child nodes, respectively.

[0078] The formula for calculating the loss reduction ΔL for feature j splitting at node m is as follows:

[0079]

[0080] L(m) is the loss function value of node m, and L(left) and L(right) are the loss function values ​​of the left and right child nodes, respectively.

[0081] By following the steps and formulas above, we can assess the importance of each feature in the gradient boosting tree model. Based on the preset ranking threshold, we can select the corresponding important features to generate a new dataset, thereby reducing model complexity and improving prediction accuracy.

[0082] S3: A bidirectional recurrent neural network model is used to continuously train the dataset to generate a tool magazine optimization model. The training process includes: dividing the dataset into a training set and a test set, updating the weights of the training set using forward propagation, loss calculation and back propagation, and then using the test set for performance evaluation.

[0083] Bidirectional Recurrent Neural Networks (Bi-RNNs) are a type of neural network capable of processing sequential data. They are a variant of Temporal Recurrent Neural Networks (Long Short-Term Memory networks) and can simultaneously consider past and future information within a sequence of data. Unlike traditional Recurrent Neural Networks (RNNs), BRNNs use two independent recurrent structures at each time step: one for transmitting information from the past to the future, and the other for transmitting information from the future to the past. Specifically, when processing sequential data, BRNNs introduce two independent hidden layers to handle the forward and backward information of the input sequence respectively. Typically, the input data is fed into the model during forward propagation, and the bidirectional LSTM layers process the forward and backward information separately, resulting in two hidden state sequences, which are then concatenated.

[0084] In actual implementation, refer to Figure 4 As shown, when using forward propagation on the training set, the method includes the following steps:

[0085] In a bidirectional recurrent neural network, the forward propagation of the forward and backward LSTMs is performed in the bidirectional LSTM layer. Specifically, the bidirectional LSTM layer consists of two independent LSTM networks: a forward LSTM and a backward LSTM. They process the forward and backward information of the input sequence, respectively. The forward LSTM processes the information from the beginning to the end of the input sequence, while the backward LSTM processes the information from the end to the beginning of the input sequence. Finally, the hidden states of the forward and backward LSTMs are concatenated as the final output.

[0086] During forward propagation, the input sequence l = {l1, l2, ..., l...} t The input sequence l is simultaneously passed to both the forward LSTM and the backward LSTM. The forward LSTM processes the forward information of the input sequence l, i.e., from l1 to l2. t The inverse LSTM processes the reverse information of the input sequence l, that is, from l t To L1.

[0087] thus:

[0088] S301: Input l for time step t t Update the hidden state of the forward LSTM layer:

[0089]

[0090] in, This represents the positive hidden state at time step t. This represents the positive hidden state at time step t-1.

[0091] During implementation, the specific process of updating the hidden state of the forward LSTM layer is as follows:

[0092] Computational forget gate:

[0093]

[0094] Among them, f t The output of the forget gate is σ, where σ is the sigmoid function and W is the output of the forget gate. f Here is the weight matrix for the forget gate. For the connection between the positive hidden state and the input, l t For the input of time step t, b f This is the bias term for the forget gate.

[0095] The forget gate determines how much of the cell state from the previous time step is retained in the current time step, and σ is the sigmoid function.

[0096] Calculate the input gate i t :

[0097]

[0098] Generate candidate cell states

[0099]

[0100] Among them, i t W is the output of the input gate. i Let b be the weight matrix of the input gate. i For the bias term of the input gate, For candidate cell states, W C Let b be the weight matrix of the candidate cell states. C Here, is the bias term for the candidate cell state, and tanh is the hyperbolic tangent function.

[0101] The input gate determines how much information from the input at the current time step and the hidden state at the previous time step is updated to the cell state.

[0102] Update cell state C t :

[0103]

[0104] Among them, C t Let be the cell state at the current time step. The cell state update is a weighted sum of the forget gate and the input gate.

[0105] Calculate the output gate o t :

[0106]

[0107] Among them, o t W is the output of the output gate. o Let b be the weight matrix of the output gate. o This is the bias term for the output gate.

[0108] The output gate controls the hidden state of the current time step.

[0109] Update positive hidden state

[0110]

[0111] in, Let tanh(C) be the positive hidden state at time step t. t ) represents the nonlinear transformation of the cell state at the current time step.

[0112] S302: Input l for time step t t Update the hidden state of the inverse LSTM layer:

[0113]

[0114] in, The reverse hidden state at time step t. This represents the positive hidden state at time step t+1.

[0115] During implementation, the specific process of updating the hidden state of the inverse LSTM layer is as follows:

[0116] Computational forget gate:

[0117]

[0118] in, This is a connection between the reverse hidden state and the input.

[0119] Calculate the input gate i t :

[0120]

[0121] Generate candidate cell states

[0122]

[0123] Update cell state C t :

[0124]

[0125] Calculate the output gate o t :

[0126]

[0127] Update reverse hidden state

[0128]

[0129] in, This represents the reverse hidden state at time step t.

[0130] S303: Concatenate the hidden states of the forward LSTM layer and the backward LSTM layer as the final output at time step t:

[0131]

[0132] in, To connect the forward and backward hidden states along the feature dimension, h t This represents the hidden state after forward and reverse connections at time step t.

[0133] Through the above steps, bidirectional LSTM, by concatenating the forward and backward hidden states at each time step, can make fuller use of the contextual information of the input sequence, improve the model's ability to understand time series data, and thus improve the accuracy and reliability of prediction.

[0134] In actual implementation, refer to Figure 4 As shown, the loss calculation for the training set includes the following steps:

[0135] S304: Calculate the output at each time step t based on forward propagation.

[0136]

[0137] in, h is the predicted value of the i-th sample at time step t; t,iLet f be the hidden state of the i-th sample after forward and backward connections at time step t, and f be the activation function of the output layer (e.g., a linear function for regression tasks and a softmax function for classification tasks).

[0138] S305: Calculate the loss at each time step t:

[0139]

[0140] Among them, L t The loss at each time step t, where n is the number of samples, y t,i Let be the true value of the i-th sample at time step t.

[0141] For classification tasks, cross-entropy loss can be used to measure the difference between predicted probabilities and true labels.

[0142]

[0143] S306: The loss L at each time step t t Accumulate to generate the total loss L total :

[0144]

[0145] Where T is the time step length.

[0146] Loss calculation is one of the key steps in training a bidirectional recurrent neural network model. The loss function measures the difference between the model's predicted value and the true value. By minimizing the loss function, the model parameters are optimized.

[0147] In actual implementation, refer to Figure 5 As shown, when performing loss calculation and backpropagation on the training set, the following steps are included:

[0148] Backpropagation updates weights by calculating the gradient of the loss function with respect to the network parameters. Backpropagation consists of two stages: error propagation and gradient calculation.

[0149] Error propagation is as follows:

[0150] S311: Calculate the error term for each time step t and backpropagate it starting from the output layer:

[0151]

[0152] in, The error term for each time step t, For the predicted output at time step t, y t This represents the actual value at time step t.

[0153] S312: Error term for calculating cell state:

[0154]

[0155] in, For the error term related to cell state, error propagation between time steps needs to be considered:

[0156]

[0157] S313: Calculate the error term of the output gate:

[0158]

[0159] in, This is the error term for the output gate.

[0160] S314: Calculate the error term of the input gate:

[0161]

[0162] in, This is the error term for the input gate.

[0163] S315: Calculate the error term for candidate cells:

[0164]

[0165] in, This represents the error term for candidate cells.

[0166] S316: Calculate the error term for the forgetting gate:

[0167]

[0168] in, This is the error term for the forget gate.

[0169] S317: Calculate the error term for the hidden state:

[0170]

[0171] in, This is the error term for the hidden state.

[0172] The gradient is calculated as follows:

[0173] S318: Based on the error term, calculate the gradients of the weights and biases of the forget gate:

[0174]

[0175] in, Let be the gradient of the weights of the forget gate. h is the gradient of the bias of the forget gate. t-1 The hidden state after forward and reverse connections at time step t.

[0176] S319: Calculate the gradients of the input gate weights and biases based on the error term:

[0177]

[0178] in, Let be the gradient of the weights of the input gate. The gradient is the bias of the input gate.

[0179] S320: Based on the error term, calculate the gradient of the weights and biases of the candidate cell states:

[0180]

[0181] in, Let be the gradient of the weights of the forget gate. The gradient of the bias for the forget gate.

[0182] S321: Calculate the gradients of the output gate weights and biases based on the error term:

[0183]

[0184] in, Let be the gradient of the weights of the forget gate. The gradient of the bias for the forget gate.

[0185] S322: Update weights and biases using gradient descent optimization.

[0186] In actual implementation, refer to Figure 6 As shown, updating weights and biases using the gradient descent optimization method includes the following steps:

[0187] S3221: Update the gradients of the forget gate weights and biases based on the learning rate η:

[0188]

[0189] Among them, W f ′ represents the updated weight of the forget gate, b f ′ represents the offset of the updated Forgot Gate.

[0190] S3222: Update the gradients of the input gate weights and biases based on the learning rate η:

[0191]

[0192]

[0193] Among them, W i ′ represents the updated weight of the forget gate, b i ′ represents the offset of the updated Forgot Gate.

[0194] S3223: Update the gradients of the weights and biases of the candidate cell states based on the learning rate η:

[0195]

[0196] Among them, W C ′ represents the updated weight of the forget gate, b C ′ represents the offset of the updated Forgot Gate.

[0197] S3224: Update the gradients of the output gate weights and biases based on the learning rate η:

[0198]

[0199] Among them, W O ′ represents the updated weight of the forget gate, b O ′ represents the offset of the updated Forgot Gate.

[0200] Backpropagation is the core step in training a neural network. Through the steps described above, gradients can be calculated during backpropagation, and these gradients can be used to update the weights of the bidirectional recurrent neural network, thereby minimizing the loss function and improving the model's prediction accuracy.

[0201] In actual implementation, refer to Figure 7 and Figure 8 As shown, the following steps are included when using a test set for performance evaluation:

[0202] S323: Use a trained bidirectional recurrent neural network model to predict the test set and obtain the predicted label and the true label;

[0203] S324: Construct a confusion matrix based on the prediction results and the true labels:

[0204]

[0205] Where TP is the number of samples correctly predicted as positive, TN is the number of samples correctly predicted as negative, FP is the number of samples incorrectly predicted as positive, and FN is the number of samples incorrectly predicted as negative.

[0206] S325: Calculation accuracy:

[0207]

[0208] Accuracy is the proportion of correctly predicted samples out of the total sample.

[0209] S326: Calculate recall rate:

[0210]

[0211] Recall rate refers to the proportion of samples that are correctly predicted as positive out of all samples that are actually positive; it is also known as sensitivity.

[0212] S327: Calculation accuracy:

[0213]

[0214] Precision rate refers to the proportion of samples that are actually positive out of all samples predicted as positive.

[0215] S328: Calculate the F1 score:

[0216]

[0217] The F1 score is the harmonic mean of precision and recall, taking into account both metrics.

[0218] When evaluating the performance of a bidirectional recurrent neural network model, a test set is typically used, and a series of metrics are calculated to measure the model's predictive ability. These metrics include accuracy, recall, precision, and F1 score to ensure that the model has good predictive ability.

[0219] S4: Deploy the tool magazine optimization model and continuously collect real-time operation data of the CNC machining center. Then, input the real-time operation data into the tool magazine optimization model to generate prediction results.

[0220] S5: Based on the prediction results, dynamically optimize the tool magazine tool changing strategy, and then replace the initial tool changing strategy of the CNC machining center with the tool magazine tool changing strategy in real time.

[0221] In actual implementation, refer to Figure 9 As shown, the dynamic optimization of the tool magazine tool changing strategy includes the following steps:

[0222] S51: Based on the prediction results, extract the tool machining sequence data: {D1,D2,D3,...,D...} N}

[0223] S52: Based on the tool machining sequence data, define the tool change path between tools under the current tool magazine layout and calculate the tool change time:

[0224]

[0225] Among them, D switch The total time for all tool changes is given by N, where N is the number of tools and D is the total time for all tool changes. a D is the tool used in step a. a+1 For the tool used in step a+1, d(D) a D a+1 ) for cutting tool D a Switch to tool D a+1 The time between.

[0226] S53: Construct the tool change time matrix:

[0227]

[0228] Then, the nearest neighbor method is used to generate the optimal tool change sequence: {D′1,D′2,D′3,...,D′ N}

[0229] The nearest neighbor method involves the following steps:

[0230] S531: Select an initial node (usually starting with D1).

[0231] S532: Find the nearest unvisited node to the current node and move to that node.

[0232] S533: Repeat S532 until all nodes have been visited.

[0233] S534: Return to the starting node.

[0234] S54: Generate a tool magazine tool changing strategy based on the optimal tool changing order, and dynamically sort the tools in the tool magazine.

[0235] Among these measures, the tool magazine changing strategy controls the dynamic sorting of tools in the tool magazine to minimize tool change time.

[0236] In some embodiments, reference Figure 10 As shown, after the CNC machining center is started, the method of this application also includes the following steps:

[0237] S101: The RFID reading device 200 installed in the CNC machining center is used to collect the RFID tag information of the cutting tool in real time, and then identify the cutting tool data contained in the RFID tag information.

[0238] In practical implementation, high-temperature resistant, impact-resistant, and waterproof RFID tags are selected to ensure their stability and durability in the processing environment. The RFID tags are then fixed to non-working parts of the cutting tools to ensure they do not interfere with normal tool operation while still being identifiable by the RFID reader 200. Based on the layout of the CNC machining center and the tool usage environment, a suitable RFID reader 200 is selected to ensure coverage of the entire work area. For example, the RFID reader 200 is installed in key locations such as the tool magazine and the entrance to the processing area to ensure accurate identification when tools enter and exit. Furthermore, the RFID reader 200 is integrated with the control unit 30 of the CNC machining center to ensure real-time data sharing and synchronization.

[0239] S102: Based on tool data, automatically identify each tool and update its position in the tool magazine in real time, thereby achieving automatic tool management, reducing manual intervention, and improving work efficiency.

[0240] In some embodiments, reference Figure 11 As shown, after machining on a CNC machining center, the method of this application further includes the following steps:

[0241] S111: Collect data on the frequency, sequence, and duration of use of various cutting tools in each production task, and then statistically analyze the frequency of use of various cutting tools.

[0242] S112: Define a preset threshold based on the tool usage frequency, define tools with a usage frequency higher than the threshold as frequently used tools, and then dynamically adjust the position of the tools in the tool magazine according to the tool usage frequency, placing frequently used tools in the most easily accessible position.

[0243] The preset threshold is set by the operator according to actual needs.

[0244] In some embodiments, reference Figure 12 and Figure 13 As shown, this application also relates to a CNC machining center tool magazine changing system, including the aforementioned CNC machining center 10, and further including an RFID device and a control unit 30. The RFID device includes an RFID reader 200 and an RFID tag. The RFID reader 200 is installed at the tool magazine or the entrance of the machining area and is used to read the information stored in the tool RFID tag. The RFID tag is fixed to the non-working part of the tool and is used to record the information of the tool.

[0245] The control unit 30 includes:

[0246] The data processing module 301 is used to perform data preprocessing based on the collected tool operation data. During data preprocessing, the collected tool operation data is read, duplicate data is checked and deleted, and then data cleaning is performed: missing values ​​and outliers are handled; data standardization is performed based on the cleaned tool operation data; and data verification is performed based on the standardized tool operation data.

[0247] The feature processing module 302 is used to extract preset features from the preprocessed tool operation data, and then perform correlation analysis and importance assessment based on the extracted preset features to generate a dataset. When generating the dataset, preset features contained in the tool operation data are extracted. The preset features include one or more of the following: machining sequence, machining time, tool type, machining material, tool life, and tool wear state. Pearson correlation coefficient is used for correlation analysis, gradient boosting tree is used to assess the importance of each feature, the feature importance is ranked, and then the features with the preset ranking are selected to generate the dataset.

[0248] The model training module 303 is used to continuously train the dataset using a bidirectional recurrent neural network model to generate a tool magazine optimization model. The training process includes: dividing the dataset into a training set and a test set; updating the weights of the training set using forward propagation, loss calculation, and back propagation; and then using the test set for performance evaluation. When using forward propagation on the training set, the input of the current time step and the hidden state and cell state of the previous time step are input; the forget gate, input gate, candidate cell state, cell state update, output gate, and hidden state update are calculated. When using loss calculation on the training set, the output of each time step is calculated based on forward propagation; the loss of each time step t is calculated; and the losses of each time step t are accumulated to generate the total loss. When using loss calculation and back propagation on the training set, the output of each time step is calculated. Error terms for step-by-step calculation, cell state calculation, output gate calculation, input gate calculation, candidate cell calculation, forget gate calculation, and hidden state calculation are used. Based on these error terms, the gradients of the weights and biases of the forget gate, input gate, candidate cell state, and output gate are calculated, and the weights and biases are updated using gradient descent optimization. When updating the weights and biases using gradient descent optimization, the gradients of the weights and biases of the forget gate, input gate, candidate cell state, and output gate are updated according to the learning rate. When evaluating performance using a test set, the trained bidirectional recurrent neural network model is used to predict the test set to obtain predicted labels and true labels. A confusion matrix is ​​constructed based on the prediction results and true labels, and then accuracy, recall, precision, and F1 score are calculated.

[0249] The model deployment module 304 is used to deploy the tool magazine optimization model.

[0250] The data acquisition module 305 is used to continuously acquire real-time operation data of the CNC machining center.

[0251] The tool magazine optimization module 306 is used to input the real-time operation data into the tool magazine optimization model to generate prediction results; based on the prediction results, it dynamically optimizes the tool magazine tool changing strategy, and then replaces the initial tool changing strategy of the CNC machining center with the tool magazine tool changing strategy in real time; when dynamically optimizing the tool magazine tool changing strategy, it extracts tool processing sequence data based on the prediction results, defines the tool changing path between tools under the current tool magazine layout based on the tool processing sequence data, calculates the tool changing time, constructs the tool changing time matrix, and then uses the nearest neighbor method to generate the optimal tool changing sequence, generates the tool magazine tool changing strategy based on the optimal tool changing sequence, and dynamically sorts the tools in the tool magazine.

[0252] Figure 14 A block diagram of one configuration of the terminal device 40 of this application is shown. In some embodiments, the terminal device 40 includes a processor 401 and a memory 402, wherein the number of processors 401 and memory 402 may be one or more. The memory 402 is coupled to the processor 401 and is used to store a computer program or instruction 403 executed by the processor 401, which, when executed by the processor 401, causes the terminal device 40 to perform the tool changer method described in any of the preceding descriptions.

[0253] Specifically, processor 401 communicates with memory 402, which may include read-only memory and random access memory, providing instructions and data to processor 401. Furthermore, a portion of memory 402 may also include non-volatile random access memory (NVRAM). Memory 402 stores operation instructions, executable modules, data structures, or subsets thereof, or even extended sets thereof. These operation instructions cover a variety of operations for implementing various tasks. The tool changer method described in this application embodiment can be applied to processor 401, or implemented by processor 401. Processor 401 can be any suitable computer processor, such as a central processing unit (CPU), graphics processing unit (GPU), programmable logic device (FPGA), etc. In the embodiments of this application, processor 401 is responsible for executing the various steps of the tool changer method.

[0254] The computer program or instructions 403 include a computer program for executing computer processes on a computing device. In some embodiments, the computer program or instructions 403 are provided using a signal-bearing medium. The signal-bearing medium may include one or more program instructions that, when executed by one or more processors 401, can provide the functions or parts thereof described in the above embodiments of the tool changer method. Thus, for example, one or more features in the tool changer method may be embodied by one or more instructions associated with the signal-bearing medium.

[0255] In some embodiments, this application also provides a computer-readable storage medium storing a computer program that, when executed by processor 401, implements the tool magazine changing method described in any of the preceding claims.

[0256] Specifically, a computer-readable storage medium refers to a medium that can be read by a computer system, such as a hard disk, solid-state drive, optical disk, flash drive, etc. In some embodiments of this application, the computer-readable storage medium stores a set of computer programs, which are executed by processor 401 to implement the various steps and functions described in the tool changer method. These computer programs may include operating systems, embedded software, application programs, etc., for controlling and managing the process of the method. By reading and executing the programs stored on the computer-readable storage medium, the computer system can effectively implement the tool changer method of this application.

[0257] In some embodiments, this application also provides a computer program product, which includes computer-executable instructions that, when executed by a processor, cause a computer to implement the tool changer method described in any of the preceding claims.

[0258] Specifically, a computer program product is a product storing computer-executable instructions, the purpose of which is to implement the various steps and functions described in the tool changer method when executed by the processor 401 of a computer system. The computer-executable instructions may include operating systems, application programs, embedded software, etc., to control and manage the process of the method. By using such a computer program product, a user can execute the tool changer method of this application on a computer system.

[0259] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A tool changing method for a CNC machining center tool magazine, characterized in that, The method includes the following steps: Based on the collected tool operation data, data preprocessing is performed; Based on the preprocessed tool operation data, preset features are extracted, and then correlation analysis and importance assessment are performed based on the extracted preset features to generate a dataset. A bidirectional recurrent neural network model is continuously trained on the dataset to generate a tool magazine optimization model. The training process includes: dividing the dataset into a training set and a test set; updating the weights of the training set using forward propagation, loss calculation, and back propagation; and then using the test set for performance evaluation. The tool magazine optimization model is deployed, and real-time operation data of the CNC machining center is continuously collected. The real-time operation data is then input into the tool magazine optimization model to generate prediction results. Based on the prediction results, the tool magazine tool changing strategy is dynamically optimized, and then the tool magazine tool changing strategy replaces the initial tool changing strategy of the CNC machining center in real time. The method for generating the dataset includes the following steps: Extract preset features contained in the tool operation data, the preset features including one or more of the following: machining process sequence, machining time, tool type, machining material, tool life, and tool wear state; Correlation analysis was performed using the Pearson correlation coefficient. ， Generate features that are strongly correlated with the target features; among them, For the first Feature values ​​of each sample For the first The target value for each sample The mean of the eigenvalues, The mean of the target value, The number of samples; Gradient boosting trees are used to evaluate the importance of each feature, the features are ranked by importance, and then the features with the preset ranking are selected to generate the dataset. The method for dynamically optimizing the tool magazine tool changing strategy includes the following steps: Based on the prediction results, extract the tool processing sequence data: ; Based on the tool machining sequence data, define the tool change path between tools under the current tool magazine layout and calculate the tool change time: ， in, The total time for all tool changes. For the number of cutting tools, For the first The tools used in the step For the first The tools used in the step For from the knife Switch to tool The time between; Construct the tool change time matrix: , Then, the nearest neighbor method is used to generate the optimal tool change sequence: ; The tool magazine tool changing strategy is generated based on the optimal tool changing order, and the tools in the tool magazine are dynamically sorted. The method described in this application further includes the following steps after the machining is completed in a CNC machining center: Collect data on the frequency, sequence, and duration of use of various types of cutting tools in each production task, and then statistically analyze the frequency of use of various types of cutting tools. Based on the frequency of tool use, a preset threshold is defined, and tools with a usage frequency higher than the threshold are defined as frequently used tools. Then, the position of the tools in the tool magazine is dynamically adjusted according to the tool usage frequency, and frequently used tools are placed in the most easily accessible position.

2. The tool changer method according to claim 1, characterized in that, When performing data preprocessing, the method includes the following steps: Read the collected tool operation data, check and delete duplicate data, and then perform data cleaning: handle missing values ​​and outliers; Based on the cleaned tool operation data, data standardization processing is performed; Data verification is performed based on standardized tool operation data.

3. The tool changer method according to claim 1, characterized in that, When using forward propagation on the training set, the method includes the following steps: For time step Input Update the hidden state of the forward LSTM layer: ， in, For time steps The positive hidden state, For time steps The positive hidden state; For time step Input Update the hidden state of the inverse LSTM layer: ， in, For time steps The reverse hidden state, For time steps The positive hidden state; The hidden states of the forward LSTM layer and the backward LSTM layer are concatenated and used as the output: ， in, To connect the forward and backward hidden states along the feature dimension, For time steps The hidden state after forward and reverse connections.

4. The tool changer method according to claim 3, characterized in that, When performing loss calculation on the training set, the method includes the following steps: Each time step is calculated based on the forward propagation. Output : ， in, For the first Each sample at time step The predicted value; For the first Each sample at time step Hidden states after forward and reverse connections The activation function for the output layer; Calculate each time step Losses: ， in, For each time step The loss, For the sample size, For the first Each sample at time step The true value; Each time step loss Accumulate to generate the total loss. : ， in, This represents the time step length.

5. The tool changer method according to claim 4, characterized in that, When applying loss calculation and backpropagation to the training set, the method includes the following steps: Calculate each time step Error terms: ， in, For each time step Error term, For time steps The predicted output, For time steps The true value; Error term for calculating cell state: ， in, This represents the error term related to cell state. Calculate the error term of the output gate: ， in, This is the error term for the output gate; Calculate the error term of the input gate: ， in, This is the error term for the input gate; Calculate the error term for candidate cells: ， in, For candidate cells; Calculate the error term of the forget gate: ， in, This is the error term for the forget gate; Calculate the error term for the hidden state: ， in, For the error term of the hidden state, For the first Each sample at time step Hidden states after forward and reverse connections; Based on the error term, calculate the gradients of the weights and biases of the forget gate: ， ， in, Let be the gradient of the weights of the forget gate. The gradient of the bias for the forget gate; Based on the error term, calculate the gradients of the input gate weights and biases: ， ， in, Let be the gradient of the weights of the input gate. The gradient of the input gate bias; Based on the error term, calculate the gradients of the weights and biases of the candidate cell states: ， ， in, Let be the gradient of the weights of the forget gate. The gradient of the bias for the forget gate; Based on the error term, calculate the gradients of the output gate weights and biases: ， ， in, Let be the gradient of the weights of the forget gate. The gradient of the bias for the forget gate; Use gradient descent to optimize the updating of weights and biases.

6. The tool changer method according to claim 5, characterized in that, When updating weights and biases using the gradient descent optimization method, the method includes the following steps: Based on learning rate Update the gradients of the forget gate weights and biases: ， ， in, For the updated weights of the Forgotten Gate, The offset for the updated Forgotten Gate; Based on learning rate Update the gradients of the input gate weights and biases: ， ， in, For the updated weights of the Forgotten Gate, For the offset of the updated Forgotten Gate; Based on learning rate Update the gradients of the weights and biases of the candidate cell states: ， ， in, For the updated weights of the Forgotten Gate, The offset for the updated Forgotten Gate; Based on learning rate Update the gradients of the output gate weights and biases: ， ， in, For the updated weights of the Forgotten Gate, This is the offset for the updated Forgotten Gate.

7. The tool changer method according to claim 1, characterized in that, When using a test set for performance evaluation, the method includes the following steps: The trained bidirectional recurrent neural network model is used to predict the test set and obtain the predicted label and the true label. Construct a confusion matrix based on the prediction results and the actual labels: , in, To correctly predict the number of samples that are classified as positive, To correctly predict the number of samples that are negative, The number of samples incorrectly predicted as positive. The number of samples incorrectly predicted as negative; Calculation accuracy: ; Calculate recall rate: ； Calculation accuracy: ; Calculate the F1 score: 。 8. A tool changer system for a CNC machining center that implements the method of any one of claims 1-7, characterized in that, include: The data processing module is used to preprocess the collected tool operation data; The feature processing module is used to extract preset features from the preprocessed tool operation data, and then perform correlation analysis and importance assessment based on the extracted preset features to generate a dataset. The model training module is used to continuously train the dataset using a bidirectional recurrent neural network model to generate a tool magazine optimization model. The training process includes: dividing the dataset into a training set and a test set; updating the weights of the training set using forward propagation, loss calculation, and back propagation; and then using the test set for performance evaluation. The model deployment module is used to deploy the tool magazine optimization model; The data acquisition module is used to continuously collect real-time operation data from the CNC machining center. The tool magazine optimization module is used to input the real-time operation data into the tool magazine optimization model to generate prediction results; based on the prediction results, the tool magazine tool changing strategy is dynamically optimized, and then the tool magazine tool changing strategy replaces the initial tool changing strategy of the CNC machining center in real time.

Images