A comprehensive management and control system for CNC machining centers

Abstract

This invention discloses a comprehensive management and control system for CNC machining centers, relating to the field of intelligent operation and maintenance technology. It constructs a system topology graph using key components of the CNC machining center as nodes and the force flow and motion transmission relationships between components as directed edges. A graph attention network is used as the basic model, employing a multi-head attention mechanism and a multi-layer message passing mechanism to achieve end-to-end modeling of system-level degradation coupling relationships. This accurately represents the cascading degradation process where the degradation state of one component propagates to other components layer by layer through coupling relationships. Based on this, the impact of single-component degradation on the accelerated wear of coupled components is simultaneously evaluated, outputting a comprehensive health index and remaining service life range at the system assembly level. With the dual objectives of minimizing total lifecycle maintenance costs and maximizing system availability, an opportunistic maintenance strategy is introduced to optimize maintenance combinations, addressing the problems of inaccurate predictions, biased assessments, unreasonable decisions, and high cascading failure risks caused by neglecting system coupling in existing single-component independent prediction methods.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology, specifically to a comprehensive management and control system for CNC machining centers. Background Technology

[0002] CNC machining centers are core equipment in high-end manufacturing industries such as aerospace, automobile manufacturing, and precision mold making. Their reliability and availability directly determine the production efficiency, product quality, and manufacturing cost of the production line. In the full life cycle operation and maintenance management of CNC machining centers, remaining useful life (RUL) prediction and maintenance decision optimization are two core links that directly affect the overall benefits and life cycle cost of the equipment.

[0003] In existing technologies, the RUL prediction of CNC machining centers generally adopts a single-component independent modeling method, which monitors the degradation and predicts the lifespan of key components such as spindle bearings, ball screws, linear guides, and tool magazine motors. However, this single-component independent prediction method completely ignores the system-level degradation coupling relationship and has the following inherent defects: First, the prediction accuracy is insufficient, and it cannot capture the accelerated wear effect of one component degradation on other components; Second, the health assessment is one-sided, and relying solely on the health status of a single component cannot fully reflect the overall health level of the system, which may result in a situation where all components are healthy but the system performance is already substandard; Third, the maintenance decision is unreasonable, and maintenance plans based on single component lifespan thresholds will lead to frequent downtime, with only one component being maintained at a time, resulting in multiple downtimes and production capacity losses; Fourth, the risk of cascading failures is high, as the impact of cascading degradation is not considered, which may lead to a chain of maintenance phenomena where one component is maintained and its coupled components fail soon afterward, further increasing maintenance costs and downtime. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a comprehensive management and control system for CNC machining centers. This system constructs a system topology using key components of the CNC machining center as nodes and the force flow and motion transmission relationships between components as directed edges. It employs a graph attention network as the basic model and, through multi-head attention and multi-layer message passing mechanisms, achieves end-to-end modeling of system-level degradation coupling relationships. This accurately represents the cascading degradation process where the degradation state of one component propagates to other components layer by layer through coupling relationships. Based on this, it simultaneously evaluates the accelerated wear impact of single-component degradation on coupled components, outputting a comprehensive health index and remaining service life range at the system assembly level. Furthermore, with the dual objectives of minimizing total lifecycle maintenance costs and maximizing system availability, it introduces an opportunistic maintenance strategy to optimize maintenance combinations, addressing the problems of inaccurate predictions, biased assessments, unreasonable decisions, and high cascading failure risks caused by neglecting system coupling in existing single-component independent prediction methods.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a comprehensive management and control system for CNC machining centers, the system comprising: Data acquisition and preprocessing module: used to acquire multi-source operating status data of various components in the spindle system, feed system and tool changer system of CNC machining center in real time, and to preprocess the data to obtain the degradation feature vector of each component; System-level degradation modeling module: used to construct a system-level degradation model of CNC machining center based on graph neural network. It constructs a system topology graph with key components as nodes and the force flow and motion transmission relationships between components as directed edges. The system-level degradation coupling relationship is represented by the message passing mechanism of graph neural network. Assembly-level health assessment and RUL prediction module: When a component shows a degradation trend, it is used to simultaneously assess the cascading effect of the degradation on the accelerated wear of coupled components, update the degradation status and degradation rate of each coupled component, calculate the comprehensive health index at the system assembly level, and output the remaining service life range at the system assembly level. Dual-objective maintenance decision optimization module: With the dual objectives of minimizing the total lifecycle maintenance cost and maximizing system availability, when the remaining service life of a component reaches the preset maintenance trigger threshold, it evaluates whether to include coupled components in the same maintenance window in advance, optimizes the maintenance combination strategy, and outputs the optimal maintenance plan. Visualization and Interaction Module: Used to display system operating status, health assessment results, RUL prediction results and maintenance plans, and supports human-computer interaction.

[0006] Furthermore, the multi-source operating status data includes: vibration acceleration, temperature, current, torque, speed, displacement sensor data, cutting depth, cutting width, feed rate machining process parameters, as well as historical maintenance records, fault records, and spare parts replacement records.

[0007] Furthermore, the system-level degradation modeling module uses key components of the CNC machining center as nodes and the force flow and motion transmission relationships between components as directed edges to construct a topology graph of the CNC machining center system. The node attributes are the multi-dimensional degradation feature vectors of the corresponding components, and the edge attributes are the initial strength coefficients of the corresponding coupling relationships. A graph attention network is used as the basic model, automatically learning the dynamic weights of different coupling edges through a multi-head attention mechanism. Supervised training is performed using historical operating data and fault data from the entire lifecycle of the CNC machining center. The loss function is a weighted sum of mean squared error and cascaded degradation loss. Through the multi-layer message passing mechanism of the graph neural network, the process of the degradation state of a component being transmitted layer by layer to other components through coupling edges is accurately represented, achieving end-to-end modeling of system-level degradation coupling relationships.

[0008] Furthermore, the system-level degradation modeling module uses a graph attention network as the basic model to construct a network structure that includes an input layer, three hidden layers, and an output layer. Each hidden layer has eight attention heads and uses LeakyReLU as the activation function. The attention weight of each node to its neighboring nodes is calculated through a multi-head attention mechanism, and the dynamic strength changes of different coupling edges under different working conditions are automatically learned. Supervised training was performed using historical operating data and fault data from the entire lifecycle of the CNC machining center. The training process used the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 200 training rounds. The loss function is a weighted sum of mean squared error and cascaded degradation loss, specifically expressed as: ,in, This is the total training loss value of the graph attention network. This represents the mean square error of the prediction of the degradation state of a single component. The mean square error is the prediction of the effects of cascade degradation. This is the weight for cascaded degradation loss.

[0009] Furthermore, the assembly-level health assessment and RUL prediction module inputs the pre-processed degradation feature vectors of each component into the system-level degradation model in real time to obtain the real-time degradation status value of each component. When the degradation status value of a component exceeds the preset degradation trend detection threshold, it is determined that the component has an early degradation trend. When a degradation trend is detected in a component, the degradation increment of the component per unit time is calculated. Through the message passing mechanism of the system-level degradation model, the accelerated wear impact of the degradation increment on all coupled components is calculated synchronously, and the degradation status and degradation rate of each coupled component are updated. Based on the real-time degradation status and coupling impact of all components, the weighted summation method is used to calculate the comprehensive health index at the assembly level of the CNC machining center system. The weights are determined according to the importance of the components and the degree of failure impact. The Monte Carlo simulation method is used to combine the degradation evolution law of each component, the cascade degradation impact, and the randomness of the degradation process to output the remaining service life range and corresponding confidence level at the system assembly level.

[0010] Furthermore, the accelerated wear effect of the degradation increment on all coupled components is expressed as follows: ,in, For the degradation increment of component i, Let be the dynamic coupling strength coefficient between component i and component j at time t. The coupling component j is affected by the degradation of component i.

[0011] Furthermore, the dual-objective maintenance decision optimization module establishes a full lifecycle maintenance cost model and a system availability model. The full lifecycle maintenance cost includes preventive maintenance cost, post-failure repair cost, downtime loss cost, and spare parts cost. System availability quantifies the impact of different maintenance combinations on system downtime and availability. A maintenance trigger threshold is set, and when the predicted remaining service life of any component is less than or equal to the threshold, the opportunistic maintenance assessment process is automatically triggered. A non-dominated sorting genetic algorithm is used to generate all feasible maintenance combination schemes, evaluate the full lifecycle maintenance cost and system availability of each scheme, and output a Pareto optimal solution set. Based on the Pareto optimal solution set, combined with the current production plan, spare parts inventory, and maintenance resource constraints, the optimal maintenance combination strategy is selected to determine the list of components to be replaced or repaired, maintenance time, and resource requirements in this maintenance window.

[0012] Furthermore, the full lifecycle maintenance cost model is expressed as follows: ,in, It refers to the overall maintenance cost incurred throughout the entire lifecycle of a CNC machining center. To cover preventative maintenance costs, For repair costs after a failure, To avoid downtime losses, For spare parts costs; The system availability model is represented as follows: ,in, It refers to the overall availability of CNC machining centers. For the total running time, for the first... Downtime for maintenance or malfunction. Total number of maintenance or failures.

[0013] Furthermore, the dual-objective maintenance decision optimization module uses a non-dominated sorting genetic algorithm to generate all feasible maintenance combinations. Specifically, this includes: encoding the maintenance combinations using binary encoding, where each gene bit corresponds to a component, with a gene value of 1 indicating that the component is maintained in the current maintenance window and a gene value of 0 indicating that it is not maintained; initializing the population and setting the number of iterations; performing genetic operations using simulated binary crossover operators and polynomial mutation operators; introducing an elite retention strategy, directly retaining the best individuals from each generation to the next generation; performing non-dominated sorting and crowding calculations on each generation of the population to generate a Pareto optimal front; evaluating the full lifecycle maintenance cost and system availability corresponding to each Pareto optimal solution, and outputting the final Pareto optimal solution set.

[0014] Furthermore, the dual-objective maintenance decision optimization module, based on the Pareto optimal solution set and considering the current production plan, spare parts inventory, and maintenance resource constraints, employs a combined weighting method of analytic hierarchy process (AHP) and entropy weighting to select the optimal maintenance combination strategy. Specifically, this includes: establishing an evaluation index system, including total lifecycle maintenance cost, system availability, spare parts inventory fulfillment rate, and maintenance personnel availability; determining subjective weights using AHP and objective weights using entropy weighting, and obtaining a comprehensive weight through linear weighting; calculating the Euclidean distance between each Pareto optimal solution and the positive and negative ideal solutions to obtain the relative proximity of each solution; selecting the solution with the highest relative proximity as the optimal maintenance combination strategy, and determining the list of parts to be replaced or repaired, maintenance time, and resource requirements within the current maintenance window.

[0015] Compared with existing technologies, this integrated management and control system for CNC machining centers has the following advantages: This invention employs a graph neural network degradation modeling method using key components of a CNC machining center as nodes and force flow and motion transmission relationships as directed edges. Through the multi-head attention mechanism and multi-layer message passing mechanism of the graph attention network, it can adaptively learn the dynamic changes in coupling strength under different operating conditions, characterizing the cascading degradation process of one component's degradation state propagating to other components layer by layer through coupling edges. This solves the problem of neglecting system coupling in independent modeling of single components. Simultaneously, when an early degradation trend appears in a component, this invention can simultaneously assess the accelerated wear impact of that degradation on all coupled components, identifying potential cascading failure risks in advance, making system health assessment more comprehensive and accurate. In terms of maintenance decisions, this invention aims to minimize the total lifecycle maintenance cost and maximize system availability, introducing an opportunistic maintenance strategy. This concentrates the maintenance of multiple coupled components within the same window, effectively reducing downtime.

[0016] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 A flowchart of a comprehensive management and control system for a CNC machining center; Figure 2 A flowchart of a health assessment and RUL prediction module for an assembly-level integrated management and control system for a CNC machining center; Figure 3 This is a flowchart of a dual-objective maintenance decision optimization module in a comprehensive management and control system for CNC machining centers. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0020] Reference Figure 1 This invention proposes a comprehensive management and control system for CNC machining centers. It adopts a system-level degradation coupling model based on graph attention networks and a dual-objective opportunistic maintenance decision-making mechanism that considers cascading effects. This overcomes the shortcomings of traditional single-component independent health assessments that ignore system-level coupling effects and the disconnect between cost and availability in maintenance strategies. It achieves accurate separation between component-level degradation transmission patterns and system-level assembly degradation trends, and performs high-confidence online health assessments, remaining service life predictions, and optimal maintenance decisions for the entire life cycle of CNC machining centers.

[0021] Specifically, this invention achieves standardized extraction of degradation features through unified preprocessing of multi-source heterogeneous data, accurately characterizes the degradation cascade diffusion effect on force flow and motion transmission paths through the message passing mechanism of graph neural networks, realizes an upgraded assessment from the health status of a single component to the overall health status of the system through assembly-level fusion calculation, seeks the optimal balance between full life cycle cost and system availability through a dual-objective optimization algorithm, and finally realizes intelligent operation and maintenance management through a visual interaction module. The modules achieve real-time information interaction through standardized data interfaces, forming a complete closed loop from data collection to decision execution.

[0022] Optionally, the data acquisition and preprocessing module is used to acquire multi-source operating status data of each component in the spindle system, feed system and tool changer system of the CNC machining center in real time, and perform preprocessing to obtain the degradation feature vector of each component.

[0023] Specifically, vibration acceleration sensors, temperature sensors, current sensors, torque sensors, speed sensors, and displacement sensors are deployed at key components such as the front and rear bearings of the spindle, the spindle motor, the X / Y / Z axis ball screws, servo motors, the tool magazine robot, and the tool changer motor. The sampling frequency of the vibration acceleration sensor is set to 10kHz, the sampling frequency of the current, torque, and speed sensors is set to 1kHz, and the sampling frequency of the temperature sensor is set to 1Hz. The machining process parameters such as the current cutting depth, cutting width, feed rate, and spindle speed are read in real time through the OPCUA interface of the CNC system. Historical maintenance records, fault records, and spare parts replacement records are synchronized through the API interface of the enterprise equipment management system. The collected raw data were subjected to outlier removal (using the 3σ criterion), wavelet denoising (using the db4 wavelet basis with a decomposition level of 5), timestamp alignment (with a 1-second time interval as the baseline), and min-max normalization. Time-domain features (mean, variance, kurtosis, margin, peak factor), frequency-domain features (center frequency, spectral entropy, peak frequency, mean square frequency), and time-frequency-domain features (wavelet packet energy entropy) were extracted to construct a 24-dimensional standardized degradation feature vector for each key component.

[0024] For example, a piezoelectric vibration acceleration sensor is deployed at the front bearing of the spindle of a certain type of vertical CNC machining center to continuously collect raw vibration data for 10 seconds. After outlier removal and wavelet denoising, 16-dimensional vibration features are extracted. At the same time, the three-phase current data of the spindle motor is collected, and 8-dimensional current features are extracted. Finally, a 24-dimensional degradation feature vector of the front bearing of the spindle is generated by splicing. All feature values ​​are normalized to the [0,1] interval for subsequent system-level degradation modeling.

[0025] Optionally, the system-level degradation modeling module is used to construct a system-level degradation model of a CNC machining center based on a graph neural network. It constructs a system topology graph with key components as nodes and the force flow transmission relationship and motion transmission relationship between components as directed edges. The system-level degradation coupling relationship is represented by the message passing mechanism of the graph neural network.

[0026] Specifically, firstly, 12 core key components of the CNC machining center are identified as nodes in the topology graph, with node attributes being the corresponding component's 24-dimensional degenerate feature vector; then, 18 directed edges are constructed based on the mechanical transmission principle and force flow transmission path, for example: The spindle motor to the spindle bearing; the spindle bearing to the worktable; the X-axis servo motor to the X-axis ball screw, the X-axis ball screw to the worktable, etc., the edge attribute is the initial strength coefficient of the corresponding coupling relationship, and the initial value is set in the range of 0.1-0.3 according to the mechanical design manual and engineering experience.

[0027] A graph attention network architecture consisting of an input layer, three hidden layers, and an output layer is constructed. Each hidden layer has eight attention heads, and LeakyReLU is used as the activation function with a negative slope of 0.2. A multi-head attention mechanism is used to calculate the attention weights of each node to its neighbors, automatically learning the dynamic intensity changes of different coupling edges under various operating conditions. A linear transformation is performed on the 24-dimensional degenerate feature vector of each node, and it is mapped to a higher-dimensional feature space through a shared learnable weight matrix to obtain the transformed feature vector of each node. This transformation process is used to enhance the expressive power of the features. For each node and all its neighboring nodes, the transformed feature vectors of the two nodes are concatenated and input into a single-layer feedforward neural network to calculate the original attention coefficient between the two nodes. This coefficient reflects the importance of the neighboring nodes to the current node. The original attention coefficients of all neighboring nodes of each node are normalized by softmax so that the sum of the attention coefficients of all neighboring nodes is 1, thus obtaining the normalized single-head attention weight. Repeat the above steps 8 times (corresponding to 8 attention heads) to obtain 8 different attention weights; concatenate the node features corresponding to the 8 attention weights to obtain the final output features of the node in the current hidden layer. By stacking three hidden layers, the degradation state is propagated in multiple layers in the system topology graph. The output features of each layer serve as the input features of the next layer. Finally, the degradation state prediction value of each node and the dynamic coupling strength coefficient between components are obtained in the output layer.

[0028] Supervised training was conducted using historical operating data from 10 CNC machining centers of this model throughout their entire lifecycle and 32 fault data points. The Adam optimizer was used during the training process, with a learning rate of 0.001, a batch size of 32, and 200 training rounds. An early stopping mechanism was employed to prevent overfitting.

[0029] The loss function of this module can be expressed as: In the formula, The mean square error is the prediction of the degradation state of a single component. The mean square error is used to predict the effects of cascade degradation. The value is the weight for cascaded degradation loss, ranging from 0.2 to 0.5, and is 0.3 in this embodiment.

[0030] For example, in the constructed system topology diagram, the initial coupling strength coefficient from the spindle front bearing to the X-axis ball screw is 0.25. After 200 rounds of supervised training, the dynamic attention weight of this coupling edge increases to 0.42 under heavy cutting conditions (cutting depth 5mm, feed rate 0.2mm / r) and to 0.18 under light cutting conditions (cutting depth 1mm, feed rate 0.1mm / r), accurately reflecting the dynamic change law of coupling strength between components under different machining conditions.

[0031] Optionally, such as Figure 2 As shown, the assembly-level health assessment and RUL prediction module is used to simultaneously assess the cascading effect of degradation on the accelerated wear of coupled components when a component shows a degradation trend, update the degradation status and degradation rate of each coupled component, calculate the comprehensive health index at the system assembly level, and output the remaining service life range at the system assembly level.

[0032] Specifically, the preprocessed real-time degradation feature vectors of each component are input into the trained system-level degradation model to obtain the real-time degradation state value of each component. The degradation state value ranges from 0 to 1, where 0 represents the component in a brand-new state and 1 represents the component in a completely failed state.

[0033] The degradation trend detection threshold is set to 0.15. When the degradation state value of a component exceeds this threshold for three consecutive sampling periods (sampling period is 1 minute), the component is determined to have an early degradation trend. When a degradation trend is detected in a component, the degradation increment of the component per unit time is calculated. Through the multi-layer message passing mechanism of the system-level degradation model, the accelerated wear impact of the degradation increment on all coupled components is calculated synchronously, and the degradation state and degradation rate of each coupled component are updated in real time.

[0034] Based on the real-time degradation status of all key components, a weighted summation method is used to calculate the comprehensive health index at the system assembly level of the CNC machining center. The weights are pre-determined using the analytic hierarchy process (AHP) based on the importance of the components and the degree of failure impact. A Monte Carlo simulation method is employed to conduct 1000 random simulations. Combining the degradation evolution patterns of each component, the cascading degradation effects, and the randomness of the degradation process, the remaining service life range at the system assembly level, along with the corresponding 95% confidence level, is output.

[0035] The calculation process for the cascading degradation effect in this module can be expressed as follows: ,in, For the degradation increment of component i, Let be the dynamic coupling strength coefficient between component i and component j at time t. The coupling component j is affected by the degradation of component i.

[0036] For example, the system detects that the degradation status value of the spindle front bearing is 0.18, and exceeds the threshold of 0.15 for three consecutive sampling periods, indicating an early degradation trend. The degradation increment per unit time is calculated to be 0.002 / hour. The dynamic coupling strength coefficient of the bearing to the X-axis ball screw is obtained as 0.35 through the system-level degradation model. Therefore, the degradation rate of the X-axis ball screw is updated to the original rate plus 0.0007 / hour. Based on the real-time degradation status of 12 key components, the overall system health index is calculated to be 0.87. After 1000 Monte Carlo simulations, the remaining service life of the system assembly is output as [820, 960] hours with a confidence level of 95%.

[0037] Optionally, such as Figure 3 As shown, the dual-objective maintenance decision optimization module is used to minimize the total life cycle maintenance cost and maximize system availability as dual objectives. When the remaining service life of a certain component reaches the preset maintenance trigger threshold, it evaluates whether to include the coupled components in the same maintenance window in advance, optimizes the maintenance combination strategy, and outputs the optimal maintenance plan.

[0038] Specifically, a full lifecycle maintenance cost model and a system availability model are pre-established. The full lifecycle maintenance cost includes preventative maintenance costs, post-failure repair costs, downtime loss costs, and spare parts costs. System availability quantifies the impact of different maintenance combinations on system downtime and availability. A maintenance trigger threshold is set at 100 hours. When the predicted remaining useful life of any component is less than or equal to this threshold, the opportunistic maintenance assessment process is automatically triggered.

[0039] A non-dominated sorting genetic algorithm is used to generate all feasible maintenance combinations: the maintenance combinations are encoded using binary encoding, with each gene bit corresponding to a key component. A gene value of 1 indicates that the component will be maintained in the current maintenance window, and a gene value of 0 indicates that it will not be maintained. The population size is initialized to 100, and the number of iterations is set to 200. Genetic operations are performed using a simulated binary crossover operator (crossover probability 0.9) and a polynomial mutation operator (mutation probability 0.1). An elite retention strategy is introduced, directly preserving the 10 best individuals from each generation to the next. Non-dominated ranking and crowding calculations are performed on each generation to generate a Pareto optimal front. Based on the Pareto optimal solution set, an evaluation index system is established, encompassing full lifecycle maintenance cost, system availability, spare parts inventory fulfillment rate, and maintenance personnel availability. The analytic hierarchy process (AHP) is used to determine subjective weights, and the entropy weight method is used to determine objective weights. A comprehensive weight is obtained through linear weighting. The Euclidean distance between each Pareto optimal solution and the positive and negative ideal solutions is calculated to obtain the relative proximity of each solution. The solution with the highest relative proximity is selected as the optimal maintenance combination strategy, determining the list of parts to be replaced or repaired, maintenance time, and resource requirements within the current maintenance window.

[0040] The calculation process for total lifecycle maintenance cost can be expressed as follows: ,in, To cover preventative maintenance costs, For repair costs after a failure, To avoid downtime losses, For spare parts costs.

[0041] The calculation process for system availability can be expressed as follows: ,in, Total running time, For the first Downtime for maintenance or malfunction. Total number of maintenance or failures.

[0042] For example, when the predicted remaining service life of the spindle front bearing is 92 hours, the opportunistic maintenance assessment process is triggered. A non-dominated sorting genetic algorithm generates 28 Pareto optimal solutions, one of which has a total lifecycle maintenance cost of 12,800 yuan and a system availability of 98.7%. Using a combination of analytic hierarchy process (AHP) and entropy weighting, the solution with the highest relative proximity is determined. The optimal maintenance strategy is to simultaneously replace the spindle front bearing, X-axis ball screw, and tool magazine robot seal during the non-production period this Saturday. This is expected to result in a 4-hour downtime, requiring 2 intermediate-level maintenance personnel and corresponding spare parts.

[0043] Optionally, the visualization and interaction module is used to display the system's operating status, health assessment results, RUL prediction results, and maintenance plans, and supports human-computer interaction.

[0044] Specifically, a visual interactive interface based on a B / S architecture built using Web technology is included, comprising four core functional modules: a real-time operation status monitoring panel, a health assessment dashboard, a remaining service life prediction curve, and a maintenance plan Gantt chart. The real-time operation status monitoring panel displays real-time data from each sensor and current processing parameters in card format; abnormal data is highlighted in red and triggers an audible alarm. The health assessment dashboard uses a pie chart to display the degradation status values ​​of each component and the overall system health index, using four colors to indicate different health levels: green (healthy, 0≤h<0.1), yellow (sub-healthy, 0.1≤h<0.15), orange (early degradation, 0.15≤h<0.5), and red (severe degradation, h≥0.5). The remaining service life prediction curve displays the remaining service life trend of each component and system assembly in line graph form, with 95% confidence intervals marked by shaded areas. The maintenance plan Gantt chart displays all maintenance tasks for the next three months, including maintenance content, maintenance time, required resources, and responsible personnel. It supports users to manually adjust maintenance trigger thresholds, model training parameters, and maintenance plans, and provides functions for historical data query, data export, and automatic generation of operation and maintenance reports.

[0045] For example, when the system detects an early degradation trend in the front bearing of the spindle, the health assessment dashboard automatically marks the component as orange, the remaining service life prediction curve shows that its remaining service life is 92 hours, the maintenance plan Gantt chart automatically generates the corresponding maintenance task, and reminds the operation and maintenance management personnel through pop-up windows and SMS.

[0046] Optionally, the integrated management and control system of the CNC machining center also includes a sensor fault compensation and data completion submodule: acquiring historical operating data of each sensor, constructing a benchmark model of the normal operating state of the sensor using a Long Short-Term Memory (LSTM) network, inputting historical data of the previous 10 sampling periods of the sensor, and predicting the measurement value at the current moment; calculating the residual between the sensor measurement value and the predicted value of the benchmark model in real time, setting the residual threshold to 3 times the standard deviation, and determining that the sensor has failed when the residual exceeds the threshold for 5 consecutive sampling periods; and using a time series-based linear interpolation method and correlation data of adjacent sensors to complete the missing data of the faulty sensor, ensuring continuous and stable operation of the system.

[0047] For example, when a circuit fault occurs in the X-axis vibration sensor, causing data interruption, the system automatically triggers a data completion process, using historical data from the previous 10 minutes and correlation data from the Y-axis and Z-axis vibration sensors to complete the data. The average absolute error of the completed data is less than 0.05g, which meets the accuracy requirements for subsequent degradation feature extraction and health assessment.

[0048] Optionally, the system also includes an online incremental update submodule for the degradation model: automatically collecting the latest operating and maintenance data of the CNC machining center every 7 days, and using an incremental learning method to update the system-level degradation model, retaining the coupling relationship knowledge already learned by the model, while incorporating new working condition data and degradation data, avoiding catastrophic forgetting of the model, and improving the model's adaptability and prediction accuracy under different machining conditions.

[0049] For example, when a CNC machining center changes to a new workpiece and the cutting parameters are switched from light cutting to heavy cutting, the system completes the adjustment of model parameters within 2 hours through online incremental updates, reducing the average absolute error of the remaining service life prediction from 12% to 5%.

[0050] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A comprehensive management and control system for a CNC machining center, characterized in that, The system includes: Data acquisition and preprocessing module: used to acquire multi-source operating status data of various components in the spindle system, feed system and tool changer system of CNC machining center in real time, and to preprocess the data to obtain the degradation feature vector of each component; System-level degradation modeling module: used to construct a system-level degradation model of CNC machining center based on graph neural network. It constructs a system topology graph with key components as nodes and the force flow and motion transmission relationships between components as directed edges. The system-level degradation coupling relationship is represented by the message passing mechanism of graph neural network. Assembly-level health assessment and RUL prediction module: When a component shows a degradation trend, it is used to simultaneously assess the cascading effect of the degradation on the accelerated wear of coupled components, update the degradation status and degradation rate of each coupled component, calculate the comprehensive health index at the system assembly level, and output the remaining service life range at the system assembly level. Dual-objective maintenance decision optimization module: With the dual objectives of minimizing the total lifecycle maintenance cost and maximizing system availability, when the remaining service life of a component reaches the preset maintenance trigger threshold, it evaluates whether to include coupled components in the same maintenance window in advance, optimizes the maintenance combination strategy, and outputs the optimal maintenance plan. Visualization and Interaction Module: Used to display system operating status, health assessment results, RUL prediction results and maintenance plans, and supports human-computer interaction.

2. The integrated management and control system for a CNC machining center according to claim 1, characterized in that, The multi-source operating status data includes: vibration acceleration, temperature, current, torque, speed, displacement sensor data, cutting depth, cutting width, feed rate machining parameters, as well as historical maintenance records, fault records, and spare parts replacement records.

3. The integrated management and control system for a CNC machining center according to claim 2, characterized in that, The system-level degradation modeling module uses key components of the CNC machining center as nodes and the force flow and motion transmission relationships between components as directed edges to construct a topology graph of the CNC machining center system. Node attributes are the multi-dimensional degradation feature vectors of the corresponding components, and edge attributes are the initial strength coefficients of the corresponding coupling relationships. A graph attention network is used as the basic model, automatically learning the dynamic weights of different coupling edges through a multi-head attention mechanism. Supervised training is performed using historical operating data and fault data from the entire lifecycle of the CNC machining center. The loss function is a weighted sum of mean squared error and cascaded degradation loss. Through the multi-layer message passing mechanism of the graph neural network, the process of the degradation state of one component being transmitted layer by layer to other components through coupling edges is accurately represented, achieving end-to-end modeling of system-level degradation coupling relationships.

4. The integrated management and control system for a CNC machining center according to claim 3, characterized in that, The system-level degradation modeling module uses a graph attention network as the basic model and constructs a network structure that includes an input layer, three hidden layers and an output layer. Each hidden layer has eight attention heads and uses LeakyReLU as the activation function. The attention weight of each node to its neighboring nodes is calculated through a multi-head attention mechanism, and the dynamic strength changes of different coupling edges under different working conditions are automatically learned. Supervised training was performed using historical operating data and fault data from the entire lifecycle of the CNC machining center. The training process used the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 200 training rounds. The loss function is a weighted sum of mean squared error and cascaded degradation loss, specifically expressed as: ,in, This is the total training loss value of the graph attention network. This represents the mean square error of the prediction of the degradation state of a single component. The mean square error is the prediction of the effects of cascade degradation. This is the weight for cascaded degradation loss.

5. The integrated management and control system for a CNC machining center according to claim 1, characterized in that, The assembly-level health assessment and RUL prediction module inputs the preprocessed degradation feature vectors of each component into the system-level degradation model in real time to obtain the real-time degradation status value of each component. When the degradation status value of a component exceeds the preset degradation trend detection threshold, it is determined that the component has an early degradation trend. When a degradation trend is detected in a component, the degradation increment of that component per unit time is calculated. Through the message passing mechanism of the system-level degradation model, the accelerated wear impact of this degradation increment on all coupled components is calculated synchronously, and the degradation status and degradation rate of each coupled component are updated. Based on the real-time degradation status and coupling impact of all components, a weighted summation method is used to calculate the comprehensive health index at the system assembly level of the CNC machining center. The weights are determined according to the importance of the components and the degree of failure impact. Using the Monte Carlo simulation method, combined with the degradation evolution law of each component, cascade degradation impact, and the randomness of the degradation process, the remaining service life range and corresponding confidence level at the system assembly level are output.

6. The integrated management and control system for a CNC machining center according to claim 5, characterized in that, The effect of the degradation increment on the accelerated wear of all coupled components is expressed as follows: ,in, For the degradation increment of component i, Let be the dynamic coupling strength coefficient between component i and component j at time t. The coupling component j is affected by the degradation of component i.

7. The integrated management and control system for a CNC machining center according to claim 1, characterized in that, The dual-objective maintenance decision optimization module establishes a full lifecycle maintenance cost model and a system availability model. The full lifecycle maintenance cost includes preventive maintenance cost, post-failure repair cost, downtime loss cost, and spare parts cost. System availability quantifies the impact of different maintenance combinations on system downtime and availability. A maintenance trigger threshold is set; when the predicted remaining service life of any component is less than or equal to this threshold, the opportunistic maintenance assessment process is automatically triggered. A non-dominated sorting genetic algorithm is used to generate all feasible maintenance combination schemes, evaluate the full lifecycle maintenance cost and system availability of each scheme, and output a Pareto optimal solution set. Based on the Pareto optimal solution set, combined with the current production plan, spare parts inventory, and maintenance resource constraints, the optimal maintenance combination strategy is selected, and the list of components to be replaced or repaired, maintenance time, and resource requirements in this maintenance window are determined.

8. The integrated management and control system for a CNC machining center according to claim 7, characterized in that, The full lifecycle maintenance cost model is expressed as follows: ,in, It refers to the overall maintenance cost incurred throughout the entire life cycle of a CNC machining center. To cover preventative maintenance costs, For repair costs after a failure, To avoid downtime losses, For spare parts costs; The system availability model is represented as follows: ,in, It refers to the overall availability of CNC machining centers. Total running time, For the first Downtime for maintenance or malfunction. Total number of maintenance or failures.

9. A comprehensive management and control system for a CNC machining center according to claim 7, characterized in that, The dual-objective maintenance decision optimization module uses a non-dominated sorting genetic algorithm to generate all feasible maintenance combinations. Specifically, it includes: encoding the maintenance combinations using binary encoding, with each gene bit corresponding to a component; a gene value of 1 indicates that the component will be maintained in the current maintenance window, and a gene value of 0 indicates that it will not be maintained; initializing the population and setting the number of iterations; performing genetic operations using simulated binary crossover operators and polynomial mutation operators; introducing an elite retention strategy, directly retaining the best individuals from each generation to the next generation; performing non-dominated sorting and crowding calculation on each generation of the population to generate a Pareto optimal front; evaluating the full lifecycle maintenance cost and system availability corresponding to each Pareto optimal solution, and outputting the final Pareto optimal solution set.

10. A comprehensive management and control system for a CNC machining center according to claim 7, characterized in that, The dual-objective maintenance decision optimization module, based on the Pareto optimal solution set, combines the current production plan, spare parts inventory status, and maintenance resource constraints. It employs a combined weighting method of analytic hierarchy process (AHP) and entropy weighting to select the optimal maintenance combination strategy. Specifically, this includes: establishing an evaluation index system, including total lifecycle maintenance cost, system availability, spare parts inventory fulfillment rate, and maintenance personnel availability; determining subjective weights using AHP and objective weights using entropy weighting, and obtaining a comprehensive weight through linear weighting; calculating the Euclidean distance between each Pareto optimal solution and the positive and negative ideal solutions to obtain the relative proximity of each solution; selecting the solution with the highest relative proximity as the optimal maintenance combination strategy, and determining the list of parts to be replaced or repaired, maintenance time, and resource requirements within the current maintenance window.

Images