A Deep Learning-Based PCD Instrument Changing Decision Method and System

Abstract

This invention provides a deep learning-based method and system for PCD insert tool changing decisions, belonging to the field of tool changing decision technology. The method includes: collecting multi-dimensional monitoring data and extracting features; constructing a wear assessment model based on CNN-GRU fusion to evaluate the wear state of the PCD insert in real time; constructing a correlation graph of the multi-dimensional monitoring data and establishing a multi-criteria tool changing decision model; solving the multi-criteria tool changing decision model to obtain a solution set of tool changing strategies, and selecting the optimal tool changing decision strategy for the current machining condition; and converting the optimal tool changing decision strategy into execution instructions to realize the PCD insert tool changing decision. This invention achieves accurate assessment of the PCD insert wear state by deeply processing and analyzing multi-dimensional data during the machining process, combined with an intelligent model, constructing a scientific tool changing decision system and dynamically optimizing it, effectively solving the subjectivity and limitations of traditional tool changing decisions, and improving the intelligence and accuracy of the decision.

Description

Technical Field

[0001] This invention belongs to the field of tool change decision technology, specifically a PCD tool change decision method and system based on deep learning. Background Technology

[0002] In the machining of PCD (Polycrystalline Diamond) inserts, the rationality of tool change decisions directly affects machining quality, production efficiency, and production costs, making it a key technical challenge in the machining field. Currently, traditional PCD insert tool change decisions rely heavily on operator experience and judgment, which is highly subjective, easily influenced by human factors, and difficult to adapt to complex and changing machining conditions, often resulting in tool changes being made too early or too late. Premature tool changes increase tool usage costs and tool change time, reducing production efficiency; while late tool changes lead to decreased machining accuracy, generate a large number of scrap products, and may even damage equipment and workpieces.

[0003] While existing tool change decision-making methods attempt to incorporate simple monitoring data, they lack effective fusion and in-depth analysis of multi-dimensional monitoring data, making it difficult to accurately assess tool wear conditions. Furthermore, existing methods fail to construct a tool change decision-making model that considers multiple objectives, thus failing to balance the contradictions between machining accuracy, machining efficiency, and tool cost. This results in insufficient scientific rigor and practicality, making it difficult to meet the high-precision and intelligent requirements of modern precision machining for tool change decisions. Therefore, there is an urgent need for an intelligent tool change decision-making method based on deep learning to address the shortcomings of existing technologies. Summary of the Invention

[0004] This invention provides a deep learning-based PCD insert changing decision method and system to address the shortcomings of existing technologies.

[0005] On the one hand, this invention provides a deep learning-based PCD insert changing decision method, including: Real-time acquisition of multi-dimensional monitoring data during the PCD insert machining process, including machining process data, tool status data, and machining condition data.

[0006] Empirical mode decomposition (EMD) was used to denoise the multi-dimensional monitoring data, and the t-distributed random neighborhood embedding algorithm was used for feature extraction to obtain process features, tool features, and operating condition features.

[0007] A wear assessment model based on CNN-GRU fusion is constructed to evaluate the wear status of PCD inserts in real time according to process characteristics, tool characteristics and working condition characteristics, and to make preliminary tool replacement warning judgment based on the wear status and a preset threshold.

[0008] A particle filter algorithm is used to fuse multi-dimensional monitoring data, and a graph neural network is used to construct a correlation map of the multi-dimensional monitoring data.

[0009] Based on the correlation graph, a multi-criteria tool changing decision model including machining accuracy, machining efficiency, and tool usage cost is established. The particle swarm optimization algorithm is used to solve the multi-criteria tool changing decision model to obtain the solution set of tool changing strategies. Finally, the optimal tool changing decision strategy for the current machining condition is selected through grey relational decision theory.

[0010] The optimal tool change decision strategy is transformed into execution instructions, and PCD tool change decision is realized by combining real-time feedback from the machining process.

[0011] The deep learning-based PCD insert changing decision method provided by this invention includes machining process data such as cutting speed, feed rate, depth of cut, and cutting fluid flow rate. Tool condition data includes insert tip wear, flank wear width, edge chipping size, and tool vibration amplitude. Machining condition data includes workpiece material hardness, ambient temperature, spindle speed fluctuation, and cutting force magnitude.

[0012] The PCD blade changing decision method based on deep learning provided by the present invention includes the following process for feature extraction using the t-distributed random neighborhood embedding algorithm: Construct a sample feature matrix of multi-dimensional monitoring data, and normalize each feature dimension in the sample feature matrix.

[0013] Calculate the Gaussian conditional probability between samples and construct the conditional probability of the corresponding samples in the low-dimensional space. Use the Student's t-distribution to characterize the similarity between samples in the low-dimensional space.

[0014] The optimal solution for the feature map is obtained by minimizing the KL divergence between the conditional probabilities in the low-dimensional space and the high-dimensional space.

[0015] The features mapped from the low-dimensional space are classified according to the dimensions of process, tool, and working condition to obtain process features, tool features, and working condition features.

[0016] The process of constructing a wear assessment model based on the fusion of convolutional neural networks and gated recurrent units, according to the deep learning-based PCD insert replacement decision method provided by the present invention, includes: Collect multi-dimensional monitoring data samples of PCD cutting tools throughout their entire lifecycle and label the tool wear status label corresponding to the feature vector of each sample. The tool wear status labels include no wear, slight wear, moderate wear, heavy wear, and scrap.

[0017] A CNN-GRU fusion model is constructed, consisting of a CNN feature extraction layer, a GRU temporal capture layer, and an output layer. The CNN feature extraction layer employs three one-dimensional convolutional and pooling layers to extract local spatial features from the feature vector and perform dimensionality reduction. The GRU temporal capture layer receives local spatial features and captures their temporal dependencies. The output layer uses a fully connected layer combined with a softmax function to output the probability of each wear state of the blade.

[0018] The cross-entropy loss function is used as the loss function of the model, and the model parameters are optimized by combining the stochastic gradient descent algorithm. The model parameters that meet the preset accuracy are retained to obtain the wear assessment model of PCD blade.

[0019] The PCD blade changing decision method based on deep learning provided by this invention includes the following process for constructing a correlation graph of multi-dimensional monitoring data using a graph neural network: The multi-dimensional monitoring data after particle filtering and fusion is processed into nodes, with each feature dimension of machining process, tool status, and machining condition as the map nodes, and the node features as the time series data of the corresponding features.

[0020] The particle filter algorithm is used to fuse multi-dimensional monitoring data to obtain fused node feature data.

[0021] Construct the adjacency matrix of the graph neural network and determine the connection weights between nodes based on the Pearson correlation coefficient between features.

[0022] The node features and adjacency matrix are input into a graph convolutional neural network to extract the graph features of the nodes.

[0023] By training a graph neural network to learn the relationships between nodes, the trained graph nodes are used as vertices, and the attention weights between nodes are used as edge weights to obtain a multi-dimensional monitoring data relationship graph.

[0024] The PCD insert changing decision method based on deep learning provided by the present invention includes the following process for establishing a multi-criteria tool changing decision model based on the correlation graph, which includes machining accuracy, machining efficiency, and tool usage cost: Machining accuracy is defined as the deviation rate between the actual machined dimension and the design dimension. Machining accuracy is calculated based on the dimensional deviation value and the design dimension, expressed by the following formula:

[0025] In the formula, For machining accuracy, These are the actual machining dimensions. For design dimensions.

[0026] Processing efficiency is defined as the number of workpieces processed effectively per unit time. The processing efficiency is calculated based on the number of workpieces processed and the total processing time, expressed by the following formula:

[0027] In the formula, For processing efficiency, To determine the effective number of workpieces processed, This represents the total processing time.

[0028] Tool usage cost is defined as the comprehensive tool cost per unit of workpiece processed, including tool purchase cost, tool change time cost, and cost of scrap. The formula is as follows:

[0029] In the formula, For the cost of using cutting tools, For PCD blade procurement costs, This refers to the labor cost corresponding to the tool changer. Costs of machining scrap due to blade wear, The number of workpieces that can be processed effectively.

[0030] A comprehensive optimization function for tool change decision based on multiple criteria is established, taking into account machining accuracy, machining efficiency, and tool usage cost. The formula is expressed as follows:

[0031] In the formula, To optimize the function, For the decision variable vector, For cutting speed, For feed rate, For cutting depth, This is the tool change time threshold.

[0032] The PCD tool changing decision method based on deep learning provided by this invention, which uses particle swarm optimization algorithm to solve the multi-criteria tool changing decision model and obtain the solution set of tool changing strategies, includes the following steps: Based on the process constraints of PCD cutting tool machining, randomly generate A vector of decision variables is used as the initial particle swarm, each particle corresponds to a tool-changing strategy, the maximum number of iterations of the particle swarm is set, and the position and velocity of the particles are initialized.

[0033] For each particle in the particle swarm, based on the correlation graph and the machining process model, the corresponding machining accuracy, machining efficiency, and tool usage cost are calculated to obtain the comprehensive optimization function value.

[0034] Determine the individual optimal position and the global optimal position for each particle. The individual optimal position is the position with the best overall optimization effect during the particle's own iteration process, while the global optimal position is the position with the best overall optimization effect in the entire particle swarm.

[0035] An external archive set is introduced to store non-dominated solutions during the iteration process, and a congestion calculation method is used to remove redundant solutions with a congestion level less than a preset threshold.

[0036] The particle's position and velocity are updated based on individual optimality and global optimality, and boundary constraints are applied to the updated particle position.

[0037] When the number of particle swarm iterations reaches the maximum number of particle swarm iterations, all non-dominated solutions in the external archive set are used as the solution set for the tool-changing strategy.

[0038] The PCD insert changing decision method based on deep learning provided by this invention, through grey relational decision theory, includes the following process for selecting the optimal tool changing strategy for the current machining condition: The set of factors and the set of solutions for tool change decisions are determined. The set of factors corresponds to machining accuracy, machining efficiency, and tool usage cost. The set of solutions is the solution set for the tool change strategy.

[0039] Based on the requirements of the processing conditions, the weight vector of each factor is determined using the analytic hierarchy process (AHP).

[0040] Construct the grey relational coefficient matrix of each scheme to the ideal scheme, and calculate the grey relational degree of each tool-changing strategy.

[0041] The tool-changing strategy with the highest grey relational degree is selected as the optimal tool-changing decision strategy for the current machining condition.

[0042] The PCD insert changing decision method based on deep learning provided by the present invention, the process of realizing PCD insert changing decision by combining real-time feedback of the machining process includes: The optimal tool changing decision strategy is transmitted to the PLC controller of the CNC machining system.

[0043] A real-time feedback model for the machining process is established using the blade wear state, machining accuracy deviation, and cutting force variation as feedback variables.

[0044] The PLC controller receives feedback data from the machining process in real time. When the feedback variable reaches a preset warning threshold, it issues a tool change warning signal. When the feedback variable reaches the tool change execution threshold, it triggers the automatic tool changer of the CNC machining system or issues a manual tool change command.

[0045] On the other hand, the present invention also provides a deep learning-based PCD blade changing decision system, comprising: The data acquisition module is used to collect multi-dimensional monitoring data in real time during the PCD cutting tool machining process. The multi-dimensional monitoring data includes machining process data, tool status data, and machining condition data.

[0046] The feature extraction module is used to denoise multi-dimensional monitoring data using empirical mode decomposition methods and to extract features using the t-distribution random neighborhood embedding algorithm to obtain process features, tool features, and working condition features.

[0047] The model building module is used to build a wear assessment model based on the fusion of convolutional neural networks and gated loop units. It evaluates the wear status of PCD inserts in real time according to process characteristics, tool characteristics and working condition characteristics, and makes preliminary tool replacement warning judgments based on the wear status and preset thresholds.

[0048] The data association module is used to fuse multi-dimensional monitoring data using a particle filter algorithm and to construct a relationship map of the multi-dimensional monitoring data using a graph neural network.

[0049] The decision selection module is used to establish a multi-criteria tool changing decision model based on the correlation graph, including machining accuracy, machining efficiency and tool usage cost. The multi-criteria tool changing decision model is solved by particle swarm optimization algorithm to obtain the solution set of tool changing strategies. Finally, the optimal tool changing decision strategy for the current machining condition is selected by grey relational decision theory.

[0050] The decision execution module is used to convert the optimal tool change decision strategy into execution instructions, and realize PCD tool change decision by combining real-time feedback from the machining process.

[0051] The PCD insert replacement decision method and system based on deep learning provided by this invention achieves real-time and accurate assessment of tool wear status by denoising and extracting features from multi-dimensional monitoring data and combining it with a CNN-GRU fusion model. This avoids the subjectivity of human judgment, improves the accuracy and reliability of wear assessment, and provides a scientific basis for tool replacement decisions.

[0052] By fusing multi-dimensional data using particle filtering algorithms and constructing a relational graph using graph neural networks, the intrinsic correlations between various monitoring parameters can be fully explored, laying the foundation for establishing a multi-criteria tool-changing decision model. Combining multi-objective optimization algorithms and grey relational decision theory, a multi-objective balance between machining accuracy, machining efficiency, and tool usage costs is achieved, yielding the optimal tool-changing strategy adapted to the current machining conditions. Simultaneously, a real-time feedback mechanism dynamically fine-tunes the decision parameters, further improving the accuracy and adaptability of tool-changing decisions, effectively reducing scrap generation, lowering tool usage costs, and increasing production efficiency. This aligns with the intelligent and efficient demands of modern precision machining and has broad application prospects. Attached Figure Description

[0053] The invention will now be further described with reference to the accompanying drawings.

[0054] Figure 1 This is a flowchart illustrating the PCD blade changing decision method based on deep learning in this invention. Figure 2 This is a schematic diagram of the feature extraction process using the t-distributed random neighborhood embedding algorithm in this invention; Figure 3 This is a schematic diagram of the process of constructing a correlation map of multi-dimensional monitoring data using graph neural networks in this invention; Figure 4 This is a schematic diagram of the process of using the particle swarm optimization algorithm to solve the multi-criteria tool changing decision model and obtain the solution set of the tool changing strategy in this invention; Figure 5 This is a schematic diagram of the PCD cutting tool changing decision system based on deep learning in this invention. Detailed Implementation

[0055] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0056] like Figures 1 to 5 As shown, the PCD tool changing decision method and system based on deep learning provided in this embodiment of the invention can be executed by a PCD tool changing decision method based on deep learning, the method including: Real-time acquisition of multi-dimensional monitoring data during the PCD insert machining process, including machining process data, tool status data, and machining condition data.

[0057] Machining process data includes cutting speed, feed rate, depth of cut, and cutting fluid flow rate. Cutting speed is calculated using a spindle speed sensor in conjunction with the tool diameter. Feed rate is obtained by analyzing pulse signals from the feed axis servo system. Depth of cut is determined after calibration of the CNC system's process parameter settings and actual values. Cutting fluid flow rate is monitored in real-time by a flow sensor. Tool status data includes insert tip wear, flank wear width, edge chipping size, and tool vibration amplitude. Tip wear and flank wear width are calculated by image recognition and pixel calibration algorithms obtained from images of the insert's working surface captured by a machine vision inspection module. Edge chipping size is obtained through visual edge detection and contour comparison analysis. Tool vibration amplitude is collected by a three-axis accelerometer fixed to the tool holder, and the effective value of the vibration signal is extracted as a feature value. The machining condition data includes workpiece material hardness, machining environment temperature, spindle speed fluctuation, and cutting force. The workpiece material hardness is the test calibration value before machining and is recorded. The machining environment temperature is collected in real time by the temperature and humidity sensor of the machining area. The spindle speed fluctuation is calculated by the difference between the real-time value of the spindle speed sensor and the set value. The cutting force is collected by the three-dimensional force sensor installed on the tool holder. The cutting force and feed force are obtained separately and integrated into comprehensive cutting force data.

[0058] Empirical Mode Decomposition (EMD) was used to denoise the multi-dimensional monitoring data, and the t-distributed random neighborhood embedding algorithm was employed for feature extraction to obtain process features, tool features, and operating condition features. The process of feature extraction using the t-distributed random neighborhood embedding algorithm includes: A sample feature matrix of multi-dimensional monitoring data is constructed. Each feature dimension in the sample feature matrix is ​​normalized to eliminate the difference in units and obtain normalized data. The row vector of the sample feature matrix is ​​the multi-dimensional feature data at a single sampling time, and the column vector is the time series data of the same feature at different sampling times. The normalization process adopts the minimum-maximum normalization method to map all feature values ​​to the interval [0,1] to avoid the problem of feature weight imbalance caused by the difference in feature units and numerical ranges.

[0059] The Gaussian conditional probability between samples is calculated, representing the conditional distribution of sample i in the neighborhood centered at sample j. The formula is as follows:

[0060] In the formula, These are the feature vectors of the i-th and j-th samples, respectively. The Gaussian kernel bandwidth for sample i is determined by the nearest neighbor distance. To exclude samples Other samples, It is a natural exponential function. Let be the square norm of the vector, and let be the Gaussian kernel bandwidth. The method combines a binary search approach with a nearest neighbor constraint to ensure that each sample contains a preset number of nearest neighbor samples within its neighborhood.

[0061] Constructing the conditional probabilities of corresponding samples in a low-dimensional space The Student's t-distribution is used to characterize the similarity between samples in the low-dimensional space, and the formula is expressed as:

[0062] In the formula, Let i be the feature vector of the i-th and j-th samples in the low-dimensional space. To exclude samples Other samples, Let be the square norm of the vector. For reciprocal operations, the dimension of the low-dimensional space is set to 2-5 dimensions according to the feature analysis requirements. This ensures the dimensionality reduction effect of the features while avoiding the loss of feature information due to excessively low dimensionality. The degrees of freedom of the student t-distribution are set to 1 to adapt to the similarity representation requirements of samples in the low-dimensional space.

[0063] The optimal solution for the feature mapping is obtained by minimizing the KL divergence between the conditional probabilities in the low-dimensional space and the high-dimensional space, as expressed by the formula:

[0064] In the formula, Let be the KL divergence between the conditional probability distribution P in high-dimensional space and the conditional probability distribution Q in low-dimensional space. Let be the conditional probability of sample i centered at sample j in a high-dimensional space. Let be the conditional probability of sample i centered at sample j in a low-dimensional space. The KL divergence is a natural logarithmic function. The solution to minimize the KL divergence is obtained by using the gradient descent algorithm until the KL divergence value converges to below a preset threshold, thus completing the mapping from high-dimensional features to low-dimensional features.

[0065] The features mapped from the low-dimensional space are classified according to the dimensions of process, tool, and working condition to obtain process features, tool features, and working condition features.

[0066] A wear assessment model based on the fusion of CNN (Convolutional Neural Network) and GRU (Gated Recurrent Unit) is constructed. This model assesses the wear state of PCD inserts in real time based on process characteristics, tool characteristics, and operating conditions. Preliminary tool change warnings are then issued based on preset thresholds for the wear state. These thresholds are set according to the machining process requirements of the PCD inserts and the tool life standard. Specifically, the threshold for transitioning from slight to moderate wear is 0.1mm-0.2mm of tool tip wear; the threshold for transitioning from moderate to severe wear is 0.2mm-0.3mm of tool tip wear; and the threshold for transitioning from severe wear to scrap is 0.3mm-0.4mm of tool tip wear. When the assessment result reaches the severe wear threshold, a preliminary tool change warning is triggered.

[0067] The process of constructing a wear assessment model based on the fusion of convolutional neural networks and gated recurrent units includes: Collect multi-dimensional monitoring data samples of PCD cutting tools throughout their entire lifecycle, assuming the total amount of historical data collected is [number missing]. Each sample contains 10 samples, and each sample includes 10 samples. There are 10 features, which are represented as a feature matrix. ,in It is the first The feature vector of each sample corresponds to the blade wear state label. , The numbers 0-4 correspond to five levels: no wear, slight wear, moderate wear, heavy wear, and scrap. The historical data samples include the full life cycle data of PCD blades under different processing techniques, different workpiece materials, and different processing conditions. The number of samples for each wear level is kept balanced to avoid overfitting during model training.

[0068] A CNN-GRU fusion model is constructed, consisting of a CNN feature extraction layer, a GRU temporal capture layer, and an output layer. The CNN feature extraction layer employs three one-dimensional convolutional layers and pooling layers to extract local spatial features from the feature vector and perform dimensionality reduction. The calculation formula for the convolutional layer is expressed as:

[0069] In the formula, Let be the feature value at position j of the i-th sample output by the convolutional layer. It is the ReLU activation function. The kernel size is the convolution kernel size. The k-th weight value of the convolution kernel. The input feature value corresponding to the k-th weight of the convolution kernel in the i-th sample. As a bias term, the kernel size of the first convolutional layer is set to 5, and the number of kernels is 32; the kernel size of the second convolutional layer is set to 3, and the number of kernels is 64; the kernel size of the third convolutional layer is set to 3, and the number of kernels is 128. All pooling layers use max pooling with a kernel size of 2 and a stride of 2 to achieve feature reduction and local feature extraction.

[0070] The GRU temporal capture layer is used to receive local spatial features and capture the temporal dependencies of these features. The calculation formulas for the GRU update gate and reset gate are expressed as follows:

[0071]

[0072]

[0073]

[0074] In the formula, The output value of the GRU update gate at time t. The output value of the GRU reset gate at time t. It is the ReLU activation function. To update the gate weight matrix, To reset the weight matrix of the gate, Let be the weight matrix of the candidate hidden states. The hidden layer state of GRU at time t-1. Let be the input feature vector of the GRU at time t. To update the gate's bias term, To reset the door's bias, The bias term for the candidate hidden state. Let be the candidate hidden state of GRU at time t. The hyperbolic tangent activation function is used. Let t be the final hidden state of the GRU. To update the complementary values ​​of the gate outputs, the number of hidden units in the GRU layer is set to 256, employing a bidirectional GRU structure. This simultaneously captures the positive and negative dependencies of the feature time series, improving the model's ability to extract time series features. The weight matrix and bias terms are initialized using a random normal distribution with an initial mean of 0 and a variance of 0.01.

[0075] The output layer uses a fully connected layer combined with a Softmax function to output the probability of each wear state of the cutting tool, expressed by the formula:

[0076] In the formula, Let be the conditional probability that the wear state of the i-th sample belongs to the k-th class. Let i be the wear status label for the i-th sample. Let be the feature vector of the i-th sample. It is a natural exponential function. This represents the output value of the fully connected layer for the i-th sample and the k-th wear state. For the wear state category, the number of neurons in the fully connected layer is set to 128 and 5 respectively. The first fully connected layer uses the ReLU activation function, and the second fully connected layer combines the Softmax function to achieve multi-class probability output.

[0077] The cross-entropy loss function is used as the model's loss function, and the model parameters are optimized using the stochastic gradient descent algorithm. The formula is as follows:

[0078] In the formula, This is the cross-entropy loss value of the model (the smaller the loss value, the higher the model's prediction accuracy). The total number of samples, The wear state label is the one-hot encoded value (1 when the i-th sample is the k-th wear state, otherwise 0). It is the natural logarithm function. Let be the conditional probability that the wear state of the i-th sample belongs to the k-th class. The learning rate of the stochastic gradient descent algorithm is set to 0.001, the momentum coefficient to 0.9, and the weight decay coefficient to 0.0001. Batch training is used during model training, with a batch size of 64. The dataset is divided into training, validation, and test sets in a 7:2:1 ratio. The validation set is used to monitor overfitting, and the test set is used to evaluate the final performance of the model.

[0079] After the model training converges, the PCD blade wear assessment model is obtained. The convergence criterion is that the loss value of the validation set no longer decreases for 10 consecutive iterations. At this point, the optimal parameters of the model are saved.

[0080] The process involves fusing multi-dimensional monitoring data using a particle filter algorithm and constructing a correlation map of the multi-dimensional monitoring data using a graph neural network. The process includes: The multi-dimensional monitoring data after particle filtering and fusion is processed into nodes. Each feature dimension of the machining process, tool status, and machining condition is used as a graph node. The node feature is the time series data of the corresponding feature. The node processing process assigns a unique node identifier to each feature dimension. The time series data of the node feature takes the feature values ​​of 50-100 consecutive sampling times to form a fixed-length feature vector to ensure that the feature dimensions of each node are consistent.

[0081] The particle filter algorithm is used to fuse multi-dimensional monitoring data to obtain fused node feature data. The state update formula of the particle filter is expressed as:

[0082]

[0083]

[0084] In the formula, This indicates the fusion status of multi-dimensional monitoring data at time t. This represents the fusion status of multi-dimensional monitoring data at time t-1. Let be the process control parameters during the machining process at time t. The process noise at time t, It is a state transition function (used to describe the transition relationship from the state at time t-1 to the state at time t). The observation value at time t (representing the original observation value of the multi-dimensional monitoring data). The observation noise at time t, This is the observation function (used to describe the mapping relationship between the system state and the observed values). Let t be the posterior probability of the system state at time t (the state estimate probability based on all observations from the previous t times). Let be the set of all observations at time t. For the observation likelihood probability (given the system state at time t, the observed...) (probability) Let t be the prior probability of the system state at time t (the state estimation probability based on the observations at the previous time t-1). The marginal probabilities of the observations are used to normalize the posterior probabilities. The number of particles in the particle filter is set to 500-1000. Both process noise and observation noise follow a Gaussian distribution with a mean of 0. The state transition function adopts a first-order linear autoregressive model, and the observation function is a linear mapping function.

[0085] Construct the adjacency matrix of the graph neural network, and determine the connection weights between nodes based on the Pearson correlation coefficient between features. The formula is expressed as:

[0086] In the formula, For the first The correlation coefficient of the features of each node Features of the i-th node Features of the j-th node covariance, Features of the i-th node variance Features of the j-th node The variance of the Pearson correlation coefficient is calculated based on time-series data of each node's characteristics. When the absolute value of the correlation coefficient is greater than 0.3, a relationship is considered to exist between the two nodes, and the corresponding connection weight is set to the absolute value of the correlation coefficient. When the absolute value of the correlation coefficient is less than or equal to 0.3, no significant relationship is considered to exist between the two nodes, and the connection weight is set to 0.

[0087] The node features and adjacency matrix are input into a graph convolutional neural network to extract the graph features of the nodes. The calculation formula for the graph convolutional layer is expressed as:

[0088] In the formula, For the graph convolutional neural network, the first The node feature matrix of the layer, The number of layers in the graph convolutional layer. For activation function, Adjacency matrix with self-loops The degree matrix, Degree matrix The inverse square root matrix, For an adjacency matrix with self-loops ( This is the original adjacency matrix. (the identity matrix) Let be the feature matrix of the nodes in the l-th layer of the graph convolutional neural network. The weight matrix for layer l is used. The graph convolutional neural network consists of 2-3 graph convolutional layers, with 64-128 kernels per layer. The activation function is Leaky ReLU with a slope of 0.2. The inverse square root of the degree matrix effectively addresses the uneven distribution of node degree during graph convolution, improving the model's feature extraction capability.

[0089] The graph neural network is trained to learn the relationships between nodes. The trained nodes are used as vertices, and the attention weights between nodes are used as edge weights, resulting in a multi-dimensional relationship graph of the monitoring data. The graph neural network is trained using an adaptive moment estimation optimization algorithm with a batch size of 32, training continues until the loss value converges. The attention weights are calculated using a graph attention mechanism, reflecting the strength of the relationship between nodes. The final relationship graph can intuitively display the mutual influence relationships between various features of machining process, tool status, and machining conditions.

[0090] Based on the correlation graph, a multi-criteria tool-changing decision model is established, including machining accuracy, machining efficiency, and tool usage cost. The multi-criteria tool-changing decision model is solved using the particle swarm optimization algorithm to obtain the solution set of tool-changing strategies. Then, the optimal tool-changing decision strategy for the current machining condition is selected by grey relational decision theory. The establishment of the multi-criteria tool-changing decision model fully considers the technical and economic indicators in the machining process, and achieves multi-objective optimization of machining accuracy, machining efficiency, and tool usage cost. The combination of particle swarm optimization algorithm and grey relational decision theory effectively improves the scientificity and rationality of the tool-changing strategy solution.

[0091] The process of establishing a multi-criteria tool change decision model based on the correlation graph, including machining accuracy, machining efficiency, and tool usage cost, includes: Machining accuracy is defined as the deviation rate between the actual machined dimension and the design dimension. Machining accuracy is calculated based on the dimensional deviation value and the design dimension, expressed by the following formula:

[0092] In the formula, For machining accuracy, These are the actual machining dimensions. To achieve the designed dimensions, the actual machining dimensions are detected in real time by online inspection equipment on the workpiece. The detection accuracy matches the machining process requirements, and the dimensional deviation value is taken as the average of multiple inspections to avoid the random error of a single inspection affecting the calculation results of machining accuracy.

[0093] Processing efficiency is defined as the number of workpieces processed effectively per unit time. The processing efficiency is calculated based on the number of workpieces processed and the total processing time, expressed by the following formula:

[0094] In the formula, For processing efficiency, To determine the effective number of workpieces processed, The total processing time is the number of workpieces that can be processed effectively to meet the processing accuracy requirements. The total processing time includes the actual cutting time, tool change time, workpiece clamping time, and other processing time. The unit is uniformly hours. The unit of processing efficiency is pieces / hour.

[0095] Tool usage cost is defined as the comprehensive tool cost per unit of workpiece processed, including tool purchase cost, tool change time cost, and cost of scrap. The formula is as follows:

[0096] In the formula, For the cost of using cutting tools, For PCD blade procurement costs, This refers to the labor cost corresponding to the tool changer. Costs of machining scrap due to blade wear, To determine the effective number of workpieces to be processed, the PCD insert purchase cost is the price of a single insert. The tool change time cost is calculated based on the time required for tool change and the hourly wage of the machine tool operator. The cost of processing scrap is the sum of the raw material cost of the scrap workpiece and the labor and equipment costs corresponding to the processing time.

[0097] A comprehensive optimization function for tool change decision based on multiple criteria is established, taking into account machining accuracy, machining efficiency, and tool usage cost. The formula is expressed as follows:

[0098] In the formula, To optimize the function, For the decision variable vector, For cutting speed, For feed rate, For cutting depth, This is the tool change time threshold.

[0099] The process of solving the multi-criteria tool-changing decision model using the particle swarm optimization algorithm to obtain the solution set of the tool-changing strategy includes: Based on the process constraints of PCD cutting tool machining, randomly generate A vector of decision variables is used as the initial particle swarm, with each particle corresponding to a tool-changing strategy. The maximum number of iterations for the particle swarm is set. Initialize the position of the particles and speed The number of particles in the initial particle swarm The number of particles is set to 100-200, and the maximum number of particle swarm iterations is set to 200-500. The position and velocity of the particles are randomly initialized within the range of the decision variables, and the initial value of the velocity is set to 10%-20% of the position value range.

[0100] For each particle in the particle swarm, based on the correlation graph and the machining process model, the corresponding machining accuracy, machining efficiency, and tool usage cost are calculated to obtain the comprehensive optimization function value.

[0101] Determine the individual optimal position of each particle. and global optimal position The individual optimal position is the position with the best overall optimization effect during the particle's own iteration process, while the global optimal position is the position with the best overall optimization effect in the entire particle swarm.

[0102] Non-dominated solutions from the iteration process are introduced into the external archive set. Redundant solutions with a crowding degree less than a preset threshold are removed using a crowding degree calculation method. The crowding degree calculation formula is expressed as:

[0103] In the formula, For the first The crowding of individual particles For the first The function values ​​of neighboring particles under each optimization objective. For the first The maximum and minimum values ​​of the optimization objectives, the preset threshold for congestion is set to 0.1-0.3, and the maximum capacity of the external archive set is set to 50-100.

[0104] The position and velocity of particles are updated based on individual optimality and global optimality. The update formula is expressed as follows:

[0105]

[0106] In the formula, Let be the velocity of the i-th particle at time t+1. For inertial weights, Let be the velocity of the i-th particle at time t. As a learning factor, A random number between 0 and 1 Let be the optimal position for the i-th particle. Let be the position of the i-th particle at time t. This represents the globally optimal position of the particle swarm. Let be the position of the i-th particle at time t+1.

[0107] Boundary constraints are applied to the updated particle positions to ensure that the decision variables meet the process requirements of PCD cutting tool machining. The boundary constraints adopt the truncation method. When the position of a particle exceeds the value range of the decision variable, its position is corrected to the boundary value of the value range, and the particle velocity is set to 0 to prevent the particle from repeatedly exceeding the boundary range and to ensure that the particle swarm search is always carried out within the effective process range.

[0108] When the particle swarm iteration count reaches Alternatively, when the optimization function converges, all non-dominated solutions in the external archive are taken as the tool-changing strategy solution set. The convergence criterion for the optimization function is that the non-dominated solutions in the external archive have not been updated for 20 consecutive iterations. At this point, the particle swarm is considered to have found the optimal tool-changing strategy solution set. The tool-changing strategy solution set contains multiple different combinations of decision variables, each combination corresponding to a tool-changing scheme that balances machining accuracy, machining efficiency, and tool usage cost.

[0109] The process of selecting the optimal tool-changing strategy for the current machining condition using grey relational decision theory includes: The set of factors and the set of alternatives for determining the tool-changing decision; factor set ,in These correspond to machining accuracy, machining efficiency, and tooling cost, respectively. Solution set , For the solution set of the tool-changing strategy, the first A tool-changing strategy.

[0110] Based on the requirements of the processing conditions, the weight vector of each factor is determined using the analytic hierarchy process (AHP). ,satisfy and The Analytic Hierarchy Process (AHP) constructs pairwise comparison judgment matrices and calculates the weights of each factor using the root mean square method. If the consistency ratio (CR) of the judgment matrix is ​​less than 0.1, the consistency of the judgment matrix is ​​considered to meet the requirements; otherwise, the judgment matrix needs to be readjusted. The weight allocation can be dynamically adjusted according to the needs of actual machining conditions, such as increasing machining accuracy in precision machining and increasing machining efficiency and tool usage costs in batch machining.

[0111] Construct the grey relational coefficient matrix of each scheme with respect to the ideal scheme. The formula for calculating the grey relational coefficient is expressed as:

[0112] In the formula, Let be the grey relational coefficient between the j-th scheme and the ideal scheme under the i-th factor. The ideal value of the i-th factor (the optimal target value of this factor, taking the maximum value of machining accuracy and machining efficiency, and the minimum value of tool usage cost). Let j be the actual value of the j-th scheme under the i-th factor. This represents the minimum deviation between the actual and ideal values ​​under all possible scenarios and all factors. This represents the maximum deviation between the actual and ideal values ​​under all scenarios and all factors. The discrimination coefficient is set to 0.5. The ideal solution is determined based on the theoretical optimal value of each factor. If the theoretical optimal value cannot be achieved, the optimal actual value of each factor in the solution set of the tool-changing strategy is taken as the ideal value. The grey relational coefficient ranges from (0,1). The closer the value is to 1, the higher the correlation between the solution and the ideal solution under the corresponding factor.

[0113] The grey relational degree of each tool-changing strategy is calculated using the following formula:

[0114] In the formula, For the first The overall grey relational degree of each tool-switching strategy Let i be the weight of the i-th factor. Let be the grey relational coefficient of the j-th scheme under the i-th factor. The comprehensive grey relational degree is the weighted sum of the grey relational coefficients of each factor and their corresponding weights, with a value range of (0,1]. The magnitude of the comprehensive grey relational degree directly reflects the overall closeness of the tool-changing strategy to the ideal scheme.

[0115] The tool-changing strategy with the highest grey correlation degree is selected as the optimal tool-changing decision strategy for the current machining condition. If there are multiple tool-changing strategies with the same grey correlation degree and the highest value, they are further filtered according to the priority of the actual machining condition, such as prioritizing the solution with higher machining efficiency or lower tool usage cost.

[0116] The optimal tool change decision strategy is translated into execution instructions, and PCD insert tool change decisions are made in conjunction with real-time feedback from the machining process. The process includes: The execution instructions for the tool changing decision strategy are encapsulated using a communication protocol compatible with the CNC machining system and transmitted to the PLC controller via industrial Ethernet or fieldbus. After receiving the instructions, the PLC controller stores and parses the specific parameters of the decision variables in real time.

[0117] A real-time feedback model for the machining process is established, with tool wear state, machining accuracy deviation, and cutting force variation as feedback variables. The formula is as follows:

[0118] In the formula, Let be the adjustment amount of the decision variable at time t. For the feedback gain matrix, Let be the deviation value of the feedback variable at time t, and be the feedback gain matrix. Based on the dynamic characteristics of the machining process, the pole placement method is used for design. The deviation value of the feedback variable is the difference between the real-time detection value and the set threshold. The deviation value of the tool wear state is calculated based on the output results of the wear assessment model. The machining accuracy deviation is the difference between the actual machined size and the design size. The cutting force change is the difference between the real-time cutting force and the normal cutting force.

[0119] The PLC controller receives real-time feedback data from the machining process. When the feedback variable reaches a preset warning threshold, it issues a tool change warning signal. When the feedback variable reaches the tool change execution threshold, it triggers the automatic tool changer of the CNC machining system or issues a manual tool change command. The tool change warning threshold and execution threshold are set according to the wear characteristics of the PCD cutting tool, machining process requirements, and equipment performance. The warning signal is issued in the form of an audible and visual alarm and a system pop-up window, reminding the operator to prepare for tool change. If the CNC machining system is equipped with an automatic tool changer, the PLC controller will directly send a tool change command to complete the automatic replacement of the cutting tool. If there is no automatic tool changer, a manual tool change command will be issued, and the operator will perform the tool change operation.

[0120] In summary, this embodiment provides a deep learning-based PCD insert replacement decision method. By denoising and extracting features from multi-dimensional monitoring data and combining it with a CNN-GRU fusion model, it achieves real-time and accurate assessment of tool wear status, avoiding the subjectivity of human judgment, improving the accuracy and reliability of wear assessment, and providing a scientific basis for tool replacement decisions.

[0121] By fusing multi-dimensional data using particle filtering algorithms and constructing a relational graph using graph neural networks, the intrinsic correlations between various monitoring parameters can be fully explored, laying the foundation for establishing a multi-criteria tool-changing decision model. Combining multi-objective optimization algorithms and grey relational decision theory, a multi-objective balance between machining accuracy, machining efficiency, and tool usage costs is achieved, yielding the optimal tool-changing strategy adapted to the current machining conditions. Simultaneously, a real-time feedback mechanism dynamically fine-tunes the decision parameters, further improving the accuracy and adaptability of tool-changing decisions, effectively reducing scrap generation, lowering tool usage costs, and increasing production efficiency. This aligns with the intelligent and efficient demands of modern precision machining and has broad application prospects.

[0122] Based on the same general inventive concept, this invention also protects a deep learning-based PCD tool changing decision system. The deep learning-based PCD tool changing decision system provided by this invention will be described below. The deep learning-based PCD tool changing decision system described below can be referred to in correspondence with the deep learning-based PCD tool changing decision method and system described above.

[0123] The deep learning-based PCD blade changing decision system includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The processor includes a data acquisition module, a feature extraction module, a model building module, a data association module, a decision selection module, and a decision execution module.

[0124] The data acquisition module is used to collect multi-dimensional monitoring data in real time during the PCD cutting tool machining process. The multi-dimensional monitoring data includes machining process data, tool status data, and machining condition data.

[0125] The feature extraction module is used to denoise multi-dimensional monitoring data using empirical mode decomposition methods and to extract features using the t-distribution random neighborhood embedding algorithm to obtain process features, tool features, and working condition features.

[0126] The model building module is used to build a wear assessment model based on the fusion of convolutional neural networks and gated loop units. It evaluates the wear status of PCD inserts in real time according to process characteristics, tool characteristics and working condition characteristics, and makes preliminary tool change warning judgments based on the wear status and preset thresholds.

[0127] The data association module is used to fuse multi-dimensional monitoring data using a particle filter algorithm and to construct a correlation map of the multi-dimensional monitoring data using a graph neural network.

[0128] The decision selection module is used to establish a multi-criteria tool changing decision model based on the correlation graph, including machining accuracy, machining efficiency and tool usage cost. The multi-criteria tool changing decision model is solved by particle swarm optimization algorithm to obtain the solution set of tool changing strategies. Finally, the optimal tool changing decision strategy for the current machining condition is selected by grey relational decision theory.

[0129] The decision execution module is used to convert the optimal tool change decision strategy into execution instructions, and realize PCD tool change decision by combining real-time feedback from the machining process.

[0130] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based PCD insert changing decision method, characterized in that, include: Real-time acquisition of multi-dimensional monitoring data during the PCD insert machining process, including machining process data, tool status data, and machining condition data; The multi-dimensional monitoring data is denoised using the empirical mode decomposition method, and the t-distributed random neighborhood embedding algorithm is used for feature extraction to obtain process features, tool features and working condition features. A wear assessment model based on CNN-GRU fusion is constructed to evaluate the wear status of PCD inserts in real time according to the process characteristics, tool characteristics and working condition characteristics, and to make a preliminary tool replacement warning judgment based on the wear status and a preset threshold. The multi-dimensional monitoring data is fused using a particle filter algorithm, and a graph neural network is used to construct a correlation map of the multi-dimensional monitoring data. Based on the correlation graph, a multi-criteria tool changing decision model including machining accuracy, machining efficiency and tool usage cost is established. The multi-criteria tool changing decision model is solved by particle swarm optimization algorithm to obtain the tool changing strategy solution set. The optimal tool changing decision strategy for the current machining condition is selected by grey relational decision theory. The optimal tool change decision strategy is transformed into execution instructions, and PCD tool change decision is realized by combining real-time feedback from the machining process.

2. The PCD insert tool changing decision method based on deep learning according to claim 1, characterized in that, The machining process data includes cutting speed, feed rate, depth of cut, and cutting fluid flow rate; the tool condition data includes insert tip wear, flank wear width, edge chipping size, and tool vibration amplitude; the machining operating condition data includes workpiece material hardness, machining environment temperature, spindle speed fluctuation, and cutting force.

3. The PCD insert tool changing decision method based on deep learning according to claim 1, characterized in that, The process of feature extraction using the t-distributed random neighborhood embedding algorithm includes: Construct a sample feature matrix of multi-dimensional monitoring data, and normalize each feature dimension in the sample feature matrix; Calculate the Gaussian conditional probability between samples and construct the conditional probability of the corresponding samples in the low-dimensional space. Use the Student's t-distribution to characterize the similarity between samples in the low-dimensional space. The optimal solution for the feature map is obtained by minimizing the KL divergence between the conditional probabilities in the low-dimensional space and the high-dimensional space. The features mapped from the low-dimensional space are classified according to the dimensions of process, tool, and working condition to obtain process features, tool features, and working condition features.

4. The PCD insert tool changing decision method based on deep learning according to claim 1, characterized in that, The process of constructing a wear assessment model based on the fusion of convolutional neural networks and gated recurrent units includes: Collect multi-dimensional monitoring data samples of PCD cutting tools throughout their entire life cycle, and label the cutting tool wear status label corresponding to the feature vector of each sample. The cutting tool wear status label includes no wear, slight wear, moderate wear, heavy wear, and scrap. A CNN-GRU fusion model is constructed, comprising a CNN feature extraction layer, a GRU temporal capture layer, and an output layer. The CNN feature extraction layer uses three one-dimensional convolutional layers and pooling layers to extract local spatial features from the feature vector and perform dimensionality reduction. The GRU temporal capture layer is used to receive the local spatial features and capture the temporal dependencies of the features. The output layer uses a fully connected layer combined with a Softmax function to output the probability of each wear state of the blade. The cross-entropy loss function is used as the loss function of the model, and the model parameters are optimized by combining the stochastic gradient descent algorithm. The model parameters that meet the preset accuracy are retained to obtain the wear assessment model of PCD blade.

5. The deep learning-based PCD insert tool changing decision method according to claim 1, characterized in that, The process of constructing the correlation graph of the multi-dimensional monitoring data using graph neural networks includes: The multi-dimensional monitoring data after particle filtering and fusion is processed into nodes, with each feature dimension of machining process, tool status and machining condition as the map nodes, and the node features as the time series data of the corresponding features. The particle filter algorithm is used to fuse multi-dimensional monitoring data to obtain fused node feature data; Construct the adjacency matrix of the graph neural network and determine the connection weights between nodes based on the Pearson correlation coefficient between features; Input the node features and adjacency matrix into the graph convolutional neural network to extract the graph features of the nodes; By training a graph neural network to learn the relationships between nodes, the trained graph nodes are used as vertices, and the attention weights between nodes are used as edge weights to obtain a multi-dimensional monitoring data relationship graph.

6. The PCD insert tool changing decision method based on deep learning according to claim 1, characterized in that, The process of establishing a multi-criteria tool change decision model based on the aforementioned correlation graph, including machining accuracy, machining efficiency, and tool usage cost, includes: Machining accuracy is defined as the deviation rate between the actual machined dimension and the design dimension. Machining accuracy is calculated based on the dimensional deviation value and the design dimension, expressed by the following formula: In the formula, For machining accuracy, These are the actual machining dimensions. For design dimensions; Processing efficiency is defined as the number of workpieces processed effectively per unit time. The processing efficiency is calculated based on the number of workpieces processed and the total processing time, expressed by the following formula: In the formula, For processing efficiency, To determine the effective number of workpieces processed, Total processing time; Tool usage cost is defined as the comprehensive tool cost per unit of workpiece processed, including tool purchase cost, tool change time cost, and cost of scrap. The formula is as follows: In the formula, For the cost of using cutting tools, For PCD blade procurement costs, This refers to the labor cost corresponding to the tool changer. Costs of machining scrap due to blade wear, The number of workpieces that can be processed effectively; A comprehensive optimization function for tool change decision based on multiple criteria is established, taking into account machining accuracy, machining efficiency, and tool usage cost. The formula is expressed as follows: In the formula, To optimize the function, For the decision variable vector, For cutting speed, For feed rate, For cutting depth, This is the tool change time threshold.

7. The PCD insert tool changing decision method based on deep learning according to claim 1, characterized in that, The process of solving the multi-criteria tool-changing decision model using the particle swarm optimization algorithm to obtain the solution set of the tool-changing strategy includes: Based on the process constraints of PCD cutting tool machining, randomly generate A vector of decision variables is used as the initial particle swarm, each particle corresponds to a tool-changing strategy, the maximum number of iterations of the particle swarm is set, and the position and velocity of the particles are initialized. For each particle in the particle swarm, based on the correlation graph and the machining process model, its corresponding machining accuracy, machining efficiency and tool usage cost are calculated to obtain the comprehensive optimization function value; Determine the individual optimal position and the global optimal position for each particle. The individual optimal position is the position with the best overall optimization effect during the particle's own iteration process, and the global optimal position is the position with the best overall optimization effect in the entire particle swarm. An external archive set is introduced to store non-dominated solutions during the iteration process, and a congestion calculation method is used to remove redundant solutions with a congestion level less than a preset threshold. The particle's position and velocity are updated based on individual optimality and global optimality, and boundary constraints are applied to the updated particle position. When the number of particle swarm iterations reaches the maximum number of particle swarm iterations, all non-dominated solutions in the external archive set are used as the solution set for the tool-changing strategy.

8. The PCD insert tool changing decision method based on deep learning according to claim 1, characterized in that, The process of selecting the optimal tool-changing strategy for the current machining condition using grey relational decision theory includes: The set of factors and the set of solutions for tool changing decisions are determined. The set of factors corresponds to machining accuracy, machining efficiency, and tool usage cost; the set of solutions is the solution set for tool changing strategies. Based on the requirements of the processing conditions, the weight vector of each factor is determined using the analytic hierarchy process (AHP). Construct the grey relational coefficient matrix of each scheme with respect to the ideal scheme, and calculate the grey relational degree of each tool-changing strategy; The tool-changing strategy with the highest grey relational degree is selected as the optimal tool-changing decision strategy for the current machining condition.

9. The deep learning-based PCD insert tool changing decision method according to claim 1, characterized in that, The process of making PCD insert changing decisions based on real-time feedback from the machining process includes: The optimal tool changing decision strategy is transmitted to the PLC controller of the CNC machining system. A real-time feedback model for the machining process is established using the blade wear state, machining accuracy deviation, and cutting force variation as feedback variables. The PLC controller receives feedback data from the machining process in real time. When the feedback variable reaches the preset warning threshold, it issues a tool change warning signal. When the feedback variable reaches the tool change execution threshold, it triggers the automatic tool change mechanism of the CNC machining system or issues a manual tool change command.

10. A deep learning-based PCD blade changing decision system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the deep learning-based PCD blade changing decision method as described in any one of claims 1 to 9.

Images