Fixed beam numerical control gantry machining center data processing method and system based on cloud computing

Abstract

The invention discloses a fixed-beam numerical control gantry machining center data processing method and system based on cloud computing, and the method comprises the steps: encrypting equipment machining data through an industrial Internet of Things protocol, uploading the encrypted equipment machining data to a cloud platform, designing a multi-modal feature fusion model, inputting the multi-dimensional data into the model, carrying out the self-attention modeling of time series data, and carrying out the self-attention modeling of the time series data; according to the method, features of different sensors are fused through a cross-modal attention mechanism, a PG-DRL deep reinforcement learning model based on physical guidance is established, a physical rule of a processing process is generated through FEA finite element analysis and serves as a constraint condition of a reward function, and a DQN algorithm is used to predict a health state according to the equipment, so that the processing quality of the equipment is improved. Cutting parameters are adjusted with the machining efficiency and the tool abrasion as targets, and target cutting parameters are generated; and the target cutting parameters are transmitted to a fixed beam numerical control gantry machining center, and a machining strategy is adjusted in real time. And the data analysis efficiency and accuracy of the fixed-beam numerical control gantry machining center are improved.

Description

Technical Field

[0001] This invention relates to the field of fixed-beam CNC gantry machining technology, and in particular to a data processing method and system for fixed-beam CNC gantry machining centers based on cloud computing. Background Technology

[0002] As core equipment in heavy machinery manufacturing, aerospace, and other fields, the machining accuracy, efficiency, and stability of fixed-beam CNC gantry machining centers directly impact product quality and production efficiency. Traditional methods often rely on single-sensor local monitoring, monitoring only spindle temperature or tool vibration, lacking a comprehensive understanding of the overall operating status of the equipment. Data acquisition is often limited to local storage and analysis, constrained by the equipment's own computing power, preventing in-depth multi-dimensional data mining. This leads to delayed assessments of equipment health, resulting in low data analysis efficiency and accuracy. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a data processing method and system for a fixed-beam CNC gantry machining center based on cloud computing.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned data processing method for a fixed-beam CNC gantry machining center based on cloud computing: The equipment processing data is encrypted using industrial IoT protocols and then uploaded to the cloud platform to obtain multi-dimensional data. Design a multimodal feature fusion model, input the multidimensional data into the model, use a Transformer encoder to perform self-attention modeling on the time series data, fuse features from different sensors through a cross-modal attention mechanism, and output the device's predicted health status. A physics-guided deep reinforcement learning model PG-DRL is established. Physical rules for the machining process are generated through FEA finite element analysis and used as constraints for the reward function. The DQN algorithm is used to predict the health status of the equipment and adjust the cutting parameters with machining efficiency and tool wear as objectives to generate target cutting parameters. The target cutting parameters are transmitted to the fixed-beam CNC gantry machining center, and the machining strategy is adjusted in real time.

[0005] Furthermore, in the aforementioned data processing method for a fixed-beam CNC gantry machining center based on cloud computing, the step of encrypting the equipment processing data via an Industrial Internet of Things (IIoT) protocol and uploading it to the cloud platform yields multi-dimensional data, including: Sensors are deployed at the tool position, spindle position, and guide rail position of the fixed beam CNC gantry machining center to collect vibration parameters, temperature parameters, and cutting force parameters in real time, thereby obtaining the equipment's machining data. The AES-256 symmetric encryption algorithm is used to encrypt the equipment processing data. The key is dynamically generated through an ECC-based key negotiation mechanism to obtain the encrypted equipment processing data. The encrypted device's processing data is uploaded to the cloud platform via an Industrial Internet of Things (IIoT) protocol, and then filtered and downsampled sequentially through an edge gateway to obtain multi-dimensional data.

[0006] Furthermore, in the aforementioned data processing method for a fixed-beam CNC gantry machining center based on cloud computing, the design of a multimodal feature fusion model involves inputting the multi-dimensional data into the model, using a Transformer encoder to perform self-attention modeling on the time-series data, fusing features from different sensors through a cross-modal attention mechanism, and outputting a predicted health status of the device, including: Multi-sensor data is converted into fixed-length vectors and position encoding is added to obtain time-series data; The encoder module consists of six identical processing units stacked together. Each unit contains two parts: the first part is a multi-head attention mechanism, which is used to focus on important information in the data from eight different perspectives; the second part is a feedforward neural network, which is used to further refine and transform the information processed by the attention mechanism.

[0007] Furthermore, in the aforementioned data processing method for a fixed-beam CNC gantry machining center based on cloud computing, the step of designing a multimodal feature fusion model, inputting the multi-dimensional data into the model, using a Transformer encoder to perform self-attention modeling on the time-series data, fusing features from different sensors through a cross-modal attention mechanism, and outputting a predicted health status of the device, also includes: Data from different types of sensors are converted into the same feature space, and corresponding weights are assigned according to the importance of different features. The fused features are then obtained through comprehensive calculation. Features that have a significant impact on the health status of the equipment are assigned higher weights, while features that have a smaller impact are assigned lower weights. By fusing features from different sensors through a cross-modal attention mechanism, the output device predicts the health status, including at least normal, warning, and fault.

[0008] Furthermore, in the aforementioned data processing method for a fixed-beam CNC gantry machining center based on cloud computing, the establishment of a PG-DRL physics-guided deep reinforcement learning model generates physical rules for the machining process through FEA finite element analysis, which serve as constraints for the reward function. The DQN algorithm is used to predict the health status of the equipment, adjusting cutting parameters with machining efficiency and tool wear as objectives, generating target cutting parameters, including: The physical rules module is used to transform the physical constraints in the processing into quantifiable decision boundaries, and the reinforcement learning network optimizes the decision strategy through dynamic interaction with the processing environment. The machining process is simulated in multiple dimensions using FEA finite element analysis technology. The simulation incorporates workpiece material properties, tool geometry parameters, and machine tool structural characteristics to analyze stress distribution, heat conduction path, and vibration propagation patterns in the cutting area.

[0009] Furthermore, in the aforementioned data processing method for a fixed-beam CNC gantry machining center based on cloud computing, the establishment of a PG-DRL physics-guided deep reinforcement learning model generates physical rules for the machining process through FEA finite element analysis, which serve as constraints for the reward function. The DQN algorithm is used to predict the health status of the equipment, adjusting cutting parameters with machining efficiency and tool wear as objectives, generating target cutting parameters, including: The basic reward function is constructed with machining efficiency and tool wear as the core optimization objectives, while physical rules are embedded as constraints into the reward mechanism. A positive reward is given when the cutting parameters are adjusted in accordance with physical rules and the machining efficiency is improved; if the parameter adjustment causes a violation of physical rules, a penalty mechanism is triggered to reduce the reward value.

[0010] Furthermore, in the aforementioned data processing method for a fixed-beam CNC gantry machining center based on cloud computing, the step of transmitting the target cutting parameters to the fixed-beam CNC gantry machining center and adjusting the machining strategy in real time includes: The target cutting parameters are transmitted to the fixed-beam CNC gantry machining center via an industrial Internet of Things (IIoT) protocol. Adjust the processing strategy according to the health status of the equipment: once every 5 minutes in normal condition, once every 1 minute in early warning condition, and immediately in fault condition.

[0011] Furthermore, in the cloud-based data processing system for a fixed-beam CNC gantry machining center, the data processing system includes the following modules: The processing data acquisition module is used to encrypt the processing data of the equipment and upload it to the cloud platform through the Industrial Internet of Things protocol to obtain multi-dimensional data. The device status prediction module is used to design a multimodal feature fusion model. The multi-dimensional data is input into the model, the Transformer encoder is used to perform self-attention modeling on the time series data, and the features of different sensors are fused through a cross-modal attention mechanism to output the device's predicted health status. The cutting parameter optimization module is used to establish a PG-DRL physics-guided deep reinforcement learning model. It generates physical rules for the machining process through FEA finite element analysis and uses them as constraints for the reward function. The DQN algorithm is used to adjust the cutting parameters based on the health status of the equipment, with machining efficiency and tool wear as objectives, to generate target cutting parameters. The machining strategy adjustment module is used to transmit the target cutting parameters to the fixed beam CNC gantry machining center and adjust the machining strategy in real time.

[0012] Furthermore, in the cloud-based data processing system for a fixed-beam CNC gantry machining center, the cutting parameter optimization module includes the following sub-modules: A submodule is set up to construct a basic reward function with machining efficiency and tool wear as the core optimization objectives, while embedding physical rules as constraints into the reward mechanism. The judgment submodule is used to give a positive reward when the cutting parameters are adjusted in accordance with the physical rules and the machining efficiency is improved; if the parameter adjustment causes a violation of the physical rules, a penalty mechanism is triggered to reduce the reward value.

[0013] Furthermore, in the cloud-based data processing system for a fixed-beam CNC gantry machining center, the cutting parameter optimization module includes the following sub-modules: The transmission submodule is used to transmit the target cutting parameters to the fixed beam CNC gantry machining center via an industrial Internet of Things protocol. The adjustment submodule is used to adjust the processing strategy according to the health status of the equipment: once every 5 minutes in normal condition, once every 1 minute in early warning condition, and immediately in fault condition.

[0014] Its beneficial effects are as follows: 1. The modal feature fusion model can integrate multi-source information such as vibration, temperature, and cutting force to accurately predict the health status of equipment, provide early warning of potential faults, reduce sudden downtime, and significantly improve equipment utilization. 2. It ensures that parameter adjustments not only conform to the physical constraints of the equipment but also dynamically adapt to the equipment's health status. It achieves closed-loop control of cutting parameters from cloud-based decision-making to equipment execution. The system can automatically adjust the machining strategy according to changes in the processing material, workpiece precision requirements, and equipment status, without manual intervention. 3. It improves the efficiency and accuracy of data analysis in fixed-beam CNC gantry machining centers, enabling timely decisions based on analysis results. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1This is a schematic diagram of the first embodiment of the data processing method for a fixed-beam CNC gantry machining center based on cloud computing in this invention. Figure 2 This is a schematic diagram of the second embodiment of the data processing method for a fixed-beam CNC gantry machining center based on cloud computing in this invention. Figure 3 This is a schematic diagram of the first embodiment of the data processing system for a fixed beam CNC gantry machining center based on cloud computing in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, a data processing method for a fixed-beam CNC gantry machining center based on cloud computing is described, which includes the following steps: Step 101: Encrypt the equipment processing data using the Industrial Internet of Things (IIoT) protocol and upload it to the cloud platform to obtain multi-dimensional data; Specifically, in this embodiment, sensors are deployed at the tool position, spindle position, and guide rail position of the fixed beam CNC gantry machining center to collect vibration parameters, temperature parameters, and cutting force parameters in real time, thereby obtaining the equipment's machining data; The AES-256 symmetric encryption algorithm is used to encrypt the equipment processing data. The key is dynamically generated through an ECC-based key negotiation mechanism to obtain the encrypted equipment processing data. The encrypted processing data of the equipment is uploaded to the cloud platform through the Industrial Internet of Things (IIoT) protocol, and then the data is filtered and downsampled sequentially through the edge gateway to obtain multi-dimensional data.

[0020] Multi-sensor data is converted into fixed-length vectors and position encoding is added to obtain time-series data; The encoder module consists of six identical processing units stacked together. Each unit contains two parts: the first part is a multi-head attention mechanism, which is used to focus on important information in the data from eight different perspectives; the second part is a feedforward neural network, which is used to further refine and transform the information processed by the attention mechanism.

[0021] Specifically, (I) Sensor Deployment Scheme Tool position: A triaxial acceleration vibration sensor (model: PCB356A16) is used, with a measurement range of ±50g, a frequency response of 0.5-10kHz, and a sampling rate of 20kHz. It is fixed to the end of the tool holder by a magnetic mounting bracket to ensure that the distance from the cutting point is ≤10cm.

[0022] Spindle unit: Temperature sensor: PT1000 platinum resistance sensor is selected, with a measurement range of -50-200℃ and an accuracy of ±0.1℃. It is embedded in the reserved hole of the spindle bearing housing and thermally conductive silicone is used to ensure that the thermal response time is ≤1s.

[0023] Vibration sensor: Install a non-contact eddy current displacement sensor (model: KD2306), with a measurement gap of 0.5-2mm and a resolution of 0.1μm, to monitor the radial runout of the spindle.

[0024] Guide rail position: Cutting force sensor: A piezoelectric triaxial force sensor (model: Kistler9257B) is used, with a measurement range of 0-20kN in each of the X / Y / Z axes and an accuracy of ±0.5%FS. It is integrated into the connection surface between the worktable and the guide rail slider.

[0025] Temperature sensor: Distributed fiber optic temperature sensor (model: SensornetDTS8000), spatial resolution 1m, temperature measurement accuracy ±0.5℃, laid along the entire length of the guide rail.

[0026] (ii) Data encryption and transmission mechanism Data encryption: The raw sensor data is encrypted using the AES-256 symmetric encryption algorithm. The key is dynamically generated through a key negotiation mechanism based on ECC (elliptic curve cryptography) and is updated every hour.

[0027] Transmission Protocol: The OPCUA over TSN industrial IoT protocol is adopted, which supports precise time synchronization (±1μs) of Time Sensitive Network (TSN). The data transmission frame structure includes device ID, sensor ID, timestamp, encrypted data segment and CRC check code. The transmission rate is 100Mbps and the packet loss rate is controlled below 0.1%.

[0028] Cloud access: Data is preprocessed (filtered and downsampled) through an edge gateway (equipped with an Intel Xeon D-1541 processor), and then uploaded to the Alibaba Cloud IoT platform via a 5G industrial module. Data storage uses the time-series database InfluxDB with a retention period of one year.

[0029] Step 102: Design a multimodal feature fusion model, input multi-dimensional data into the model, use a Transformer encoder to perform self-attention modeling on time series data, fuse features from different sensors through a cross-modal attention mechanism, and output the device's predicted health status. Specifically, in this embodiment, multi-sensor data is converted into a fixed-length vector form and position encoding is added to obtain time-series data; The encoder module consists of six identical processing units stacked together. Each unit contains two parts: the first part is a multi-head attention mechanism, which is used to focus on important information in the data from eight different perspectives; the second part is a feedforward neural network, which is used to further refine and transform the information processed by the attention mechanism.

[0030] Data from different types of sensors are converted into the same feature space, and corresponding weights are assigned according to the importance of different features. The fused features are then obtained through comprehensive calculation. Features that have a significant impact on the health status of the equipment are assigned higher weights, while features that have a smaller impact are assigned lower weights. By fusing features from different sensors through a cross-modal attention mechanism, the output device predicts the health status, including at least normal, warning, and fault.

[0031] Specifically, 1) Data Preprocessing Module Time series data segmentation: The sliding window method is used with a window size of 512ms and a step size of 128ms to convert vibration and temperature signals into two-dimensional tensors (time × feature). Feature standardization: Z-score standardization is performed on the data of each sensor channel to eliminate dimensional differences. Outlier handling: Outlier data points are identified using the 3σ criterion and repaired using linear interpolation. (II) Transformer Encoder Structure Input layer: Converts multi-sensor data into 512-dimensional embedding vectors and adds position encoding (sine and cosine functions). The encoder module contains a 6-layer Transformer encoder, each consisting of an 8-head self-attention mechanism (head dimension 64) and a feedforward neural network (hidden layer dimension 2048), using the GELU activation function. Temporal feature extraction: The dependencies between different time steps are captured by the self-attention weight matrix, and the temporal feature vector (dimension 512) is output. (III) Cross-modal attention fusion mechanism Modal feature mapping: Vibration (3 channels), temperature (2 channels), and cutting force (3 channels) features are mapped to 256-dimensional subspaces respectively. Cross-modal attention calculation: Using vibration features as the query vector and temperature and cutting force features as the key and value vectors, respectively, the attention weights between modes are calculated. Feature fusion: Multimodal features are fused using a weighted summation method, outputting a 1024-dimensional fused feature vector. (iv) Health Status Prediction Output Prediction Head: The fused features are output through a 2-layer fully connected network (512→128→3) to determine the device health status, which is divided into three categories: "normal", "early warning" and "fault". Model training: The cross-entropy loss function was used, the optimizer was AdamW (β1=0.9, β2=0.98), the initial learning rate was 5e-4, the number of training epochs was 200, the batch size was 32, and an early stopping strategy (patience=10) was adopted.

[0032] Specifically, I. Data Preprocessing (a) Feature standardization Data collected from different sensors is processed uniformly to eliminate the influence of different units and measurement ranges. Specifically, the average value and fluctuation range of each sensor's data are calculated, and then all data are converted to the same numerical range. For example, temperature and vibration data are both adjusted to the range of 0-1, allowing different types of data to be compared and calculated within the same model.

[0033] (ii) Outlier handling Statistical analysis is used to determine the normal range of data. When abnormal data exceeding this range is found, appropriate methods are used to correct it. For example, if the temperature value at a certain moment suddenly exceeds the normal processing temperature and deviates significantly from adjacent data, it is judged as an outlier and replaced with the average of the normal data before and after that moment to ensure the continuity of the data sequence.

[0034] II. Transformer Encoder Structure (a) Input layer processing The preprocessed multi-sensor data is converted into fixed-length vectors, each containing 512 data points, and labeled with information indicating the order of the data. This labeling acts like a timestamp, allowing the model to identify the temporal characteristics of the data, such as the temporal relationship between vibration data collected earlier and temperature data collected later.

[0035] (II) Encoder Module Composition It consists of six identical processing units stacked together, each containing two key parts. The first part is a multi-head attention mechanism, which can simultaneously focus on important information in the data from eight different angles. For example, in vibration data, it can focus on both the parts with large amplitude and the parts with significant frequency changes. The second part is a feedforward neural network, which further refines and transforms the information processed by the attention mechanism, enhancing the model's ability to express data features.

[0036] (III) Temporal Feature Extraction Through the above processing, the model can automatically identify key features in the data that change over time. For example, during continuous machining, it can capture the changing pattern of vibration signals as tool wear intensifies, and the rising trend of temperature signals with spindle running time. These features will provide a basis for subsequent assessment of equipment health status.

[0037] III. Cross-modal attention fusion mechanism (a) Modal Feature Mapping Transforming data from different types of sensors into the same feature space makes data from different modes, such as vibration, temperature, and cutting force, comparable. For example, converting the features of vibration signals and temperature signals into 256-dimensional vectors allows them to be fused and calculated on the same dimension.

[0038] (ii) Cross-modal attention computation Using vibration data characteristics as a reference, the model seeks to identify the components in temperature and cutting force data that are correlated with the vibration characteristics. For example, when an abnormal vibration signal occurs, the model automatically focuses on the temperature and cutting force data at that time, analyzes the interaction between them, and determines whether the abnormal vibration is caused by excessive temperature or excessive cutting force.

[0039] (III) Feature Fusion Based on the importance of different features, appropriate weights are assigned, and then a comprehensive calculation is performed to obtain the fused features. Features that have a greater impact on the health status of the equipment, such as cutting force, are assigned higher weights; features with a smaller impact, such as ambient temperature, are assigned lower weights. In this way, multi-dimensional data is integrated into a comprehensive vector containing 1024 feature points, comprehensively reflecting the operating status of the equipment.

[0040] Step 103: Establish a PG-DRL physics-guided deep reinforcement learning model. Generate physical rules for the machining process through FEA finite element analysis and use them as constraints for the reward function. Use the DQN algorithm to adjust the cutting parameters based on the equipment's predicted health status, with machining efficiency and tool wear as objectives, and generate target cutting parameters. Specifically, in this embodiment, the physical rule module is used to transform the physical constraints in the processing into quantifiable decision boundaries, and the reinforcement learning network optimizes the decision strategy through dynamic interaction with the processing environment. The machining process is simulated in multiple dimensions using FEA finite element analysis technology. The simulation incorporates workpiece material properties, tool geometry parameters, and machine tool structural characteristics to analyze stress distribution, heat conduction path, and vibration propagation patterns in the cutting area.

[0041] The basic reward function is constructed with machining efficiency and tool wear as the core optimization objectives, while physical rules are embedded as constraints into the reward mechanism. A positive reward is given when the cutting parameters are adjusted in accordance with physical rules and the machining efficiency is improved; if the parameter adjustment causes a violation of physical rules, a penalty mechanism is triggered to reduce the reward value.

[0042] Specifically, (I) Generation of FEA Physics Rules Finite element modeling: A complete model of the machining center was built based on ANSYS Workbench, including components such as the spindle (45 steel), guide rail (HT300), and cutting tools (carbide). The mesh generation accuracy was 1mm, and the number of nodes was approximately 500,000. Physical field analysis: Thermal analysis: Simulate the temperature field distribution at spindle speeds of 1000-6000 rpm to obtain the critical threshold for thermal deformation (≤0.02 mm). Structural analysis: Calculate the guide rail deflection (≤0.01mm) under cutting forces of 5000-20000N. Vibration analysis: The first-order resonant frequency (250Hz) of the main shaft is obtained through modal analysis and used as the upper limit of the vibration constraint. Rule quantification: Transforming FEA results into mathematical constraints. State space: includes equipment health status (3 categories), current cutting parameters (spindle speed, feed rate, depth of cut), and real-time monitoring data (temperature, vibration, cutting force), with 20 dimensions. Motion space: Discretized cutting parameter adjustment (speed ±500rpm, feed ±50mm / min, depth ±0.1mm), a total of 3×3×3=27 motions. Reward function:

[0043] Processing efficiency: Metal removal rate per unit time Tool wear: Estimated using an empirical model based on cutting force and time Constraint Penalty: Triggered when physical parameters exceed the FEA threshold; the value is the square of the constraint deviation. Weighting coefficients: α=0.6, β=0.3, γ=0.1 DQN algorithm implementation Network structure: A dual DQN architecture is adopted. Both the evaluation network and the target network are 3-layer fully connected networks (input 20 → hidden layer 128 → hidden layer 64 → output 27), and the activation function is ReLU.

[0044] Training parameters: Experience replay pool capacity 10^6, batch size 64, discount factor γ=0.95, ε-greedy policy (ε decays linearly from 1.0 to 0.1), learning rate 0.0001, target network updated every 1000 steps. Convergence condition: Training stops when the average reward fluctuation is ≤5% over 100 consecutive rounds.

[0045] Specifically, Model Construction: A dual-driven framework integrating physical constraints and deep reinforcement learning is constructed, with a reinforcement learning network as the decision-making core and a physical rule module embedded as the decision guide. The physical rule module is responsible for transforming the physical constraints in the machining process into quantifiable decision boundaries, while the reinforcement learning network optimizes the decision strategy through dynamic interaction with the machining environment. The two communicate in real time through a data interface, ensuring that the model always follows physical laws when exploring optimal cutting parameters.

[0046] Physical rule generation: FEA (Finite Element Analysis) technology is used to perform multi-dimensional physical simulation of the machining process. Key variables such as workpiece material properties, tool geometry parameters, and machine tool structural characteristics are incorporated into the simulation. The analysis focuses on stress distribution, heat conduction paths, and vibration propagation patterns in the cutting zone. Key physical thresholds are extracted from the simulation results, such as the maximum torque the spindle can withstand, the maximum allowable deformation of the guide rail, and the critical temperature of the tool edge. These thresholds are then transformed into specific physical rules, such as "the feed rate should be reduced when the cutting force exceeds 1.2 times the material's yield strength" and "the cooling enhancement mechanism should be activated when the spindle temperature approaches the critical value of 3°C."

[0047] Reward Function Design: A basic reward function is constructed with machining efficiency and tool wear as the core optimization objectives, while physical rules are embedded as constraints in the reward mechanism. When the cutting parameters are adjusted in accordance with the physical rules and machining efficiency is improved, a positive reward is given; if the parameter adjustment leads to approaching or violating the physical rules (vibration amplitude exceeds the safety threshold), a penalty mechanism is triggered to reduce the reward value; when the tool wear rate is lower than the expected threshold, an additional reward weight is added, and the dynamic change of the reward value guides the model to balance efficiency and wear within a safe range.

[0048] Dynamic parameter adjustment: The DQN algorithm receives predicted health status data of the equipment (including tool wear, real-time spindle temperature, guideway vibration frequency, etc.) and uses it as state input to construct the decision space. The algorithm first assesses machining capability based on the current health status. For example, when tool wear reaches 30% of its initial value, a wear suppression mode is activated; if the spindle temperature is within the normal range and the vibration amplitude is low, an efficiency-first mode is activated. Within this mode framework, the algorithm iteratively learns to generate candidate cutting parameter combinations (feed rate, depth of cut, and spindle speed combinations), selects feasible solutions based on physical rule constraints, and chooses the optimal adjustment direction based on the feedback value of the reward function.

[0049] Target-based balancing optimization: A dual-objective dynamic balancing mechanism is established. When improving machining efficiency may lead to increased tool wear, the algorithm automatically calculates the cost-effectiveness of efficiency gain versus tool wear. For example, if increasing the feed rate by 10% would shorten tool life by 20%, then the second-highest efficiency parameter is prioritized. If the equipment is in good health (tool recently replaced, spindle cooling system running at full load), then wear limits are appropriately relaxed to pursue higher efficiency. By comparing the efficiency-wear curves under current parameters with historical best curves in real time, the parameter combination is continuously optimized to ensure maximum overall benefit during long-term machining.

[0050] Parameter generation and verification: The initially generated target cutting parameters are verified for compliance with physical rules, checking whether they meet the constraints such as stress, temperature, and vibration derived from the FEA analysis. If potential risks exist (a certain parameter combination may cause the guide rail deformation to exceed the allowable range), the parameters are backtracked and adjusted until they meet all physical rules, ultimately outputting target cutting parameters that combine safety, efficiency, and low wear characteristics.

[0051] Step 104: Transmit the target cutting parameters to the fixed beam CNC gantry machining center and adjust the machining strategy in real time.

[0052] Specifically, in this embodiment, the target cutting parameters are transmitted to the fixed beam CNC gantry machining center via an industrial Internet of Things (IIoT) protocol. Adjust the processing strategy according to the health status of the equipment: once every 5 minutes in normal condition, once every 1 minute in early warning condition, and immediately in fault condition.

[0053] Specifically, (a) Parameter transmission mechanism Cloud-to-device communication: The MQTT protocol is used, with the cloud acting as the publisher and the PLC in the processing center acting as the subscriber, with QoS level 2 (ensuring that messages are delivered only once).

[0054] Real-time performance guarantee: 10 sets of alternative parameter schemes are pre-stored at the edge node. When the network latency is greater than 100ms, the local scheme is automatically activated and updated synchronously after the network is restored.

[0055] (ii) Dynamic adjustment of processing strategy Adjustment cycle: dynamically set according to the health status of the equipment: 5 minutes / time in normal state, 1 minute / time in warning state, and immediate adjustment in fault state.

[0056] Feedback loop: The machining center transmits the adjusted parameter execution results (actual cutting efficiency, sensor data) back in real time, the cloud updates the experience playback pool of the DQN model every hour, and performs incremental training once a week.

[0057] Emergency mechanism: If the constraint penalty is still triggered after 3 consecutive parameter adjustments, the system will automatically switch to safe mode (reduce the speed by 30% and the feed by 20%) and push alarm information to the operation and maintenance terminal.

[0058] (III) Visual Monitoring Platform Data dashboard: Real-time display of equipment status (health score 0-100), cutting parameter curves, processing efficiency trends, etc., with a refresh rate of 1Hz.

[0059] Historical tracking: Supports querying parameter adjustment records, health status changes, and fault diagnosis reports for any time period, with data granularity accurate to the second.

[0060] Predictive maintenance: Based on health status prediction results, it pushes maintenance suggestions such as tool replacement and spindle maintenance 24 hours in advance.

[0061] (iv) System performance indicators Data processing latency: End-to-end latency ≤ 500ms (from sensor acquisition to parameter adjustment completion).

[0062] Prediction accuracy: Equipment health status prediction accuracy ≥95%, fault early warning time ≥30min.

[0063] Optimization results: Processing efficiency increased by ≥15%, tool life extended by ≥20%, and equipment failure rate reduced by ≥30%.

[0064] Its beneficial effects are as follows: 1. The modal feature fusion model can integrate multi-source information such as vibration, temperature, and cutting force to accurately predict the health status of equipment, provide early warning of potential faults, reduce sudden downtime, and significantly improve equipment utilization. 2. It ensures that parameter adjustments not only conform to the physical constraints of the equipment but also dynamically adapt to the equipment's health status. It achieves closed-loop control of cutting parameters from cloud-based decision-making to equipment execution. The system can automatically adjust the machining strategy according to changes in the processing material, workpiece precision requirements, and equipment status, without manual intervention. 3. It improves the efficiency and accuracy of data analysis in fixed-beam CNC gantry machining centers, enabling timely decisions based on analysis results.

[0065] Please see Figure 2 In the data processing method of a cloud-based CNC gantry machining center with a fixed beam, a multimodal feature fusion model is designed. Multi-dimensional data is input into the model, and a Transformer encoder is used to perform self-attention modeling on the time-series data. Features from different sensors are fused through a cross-modal attention mechanism, and the output device prediction of health status includes the following steps: Step 201: Deploy sensors at the tool position, spindle position, and guide rail position of the fixed beam CNC gantry machining center to collect vibration parameters, temperature parameters, and cutting force parameters in real time to obtain equipment machining data; Step 202: Encrypt the equipment processing data using the AES-256 symmetric encryption algorithm. The key is dynamically generated through an ECC-based key negotiation mechanism to obtain the encrypted equipment processing data. Step 203: Upload the encrypted equipment processing data to the cloud platform through the Industrial Internet of Things protocol, and then perform data filtering and downsampling through the edge gateway to obtain multi-dimensional data.

[0066] The above describes an embodiment of the data processing method for a fixed-beam CNC gantry machining center based on cloud computing according to the present invention. Please refer to [link / reference]. Figure 3 The cloud-based data processing system for a fixed-beam CNC gantry machining center includes the following modules: The processing data acquisition module is used to encrypt the processing data of the equipment and upload it to the cloud platform through the Industrial Internet of Things protocol to obtain multi-dimensional data. The device status prediction module is used to design a multimodal feature fusion model. It inputs multi-dimensional data into the model, uses a Transformer encoder to perform self-attention modeling on time-series data, and fuses features from different sensors through a cross-modal attention mechanism to output the device's predicted health status. The cutting parameter optimization module is used to establish a PG-DRL physics-guided deep reinforcement learning model. It generates physical rules for the machining process through FEA finite element analysis and uses them as constraints for the reward function. The DQN algorithm is used to adjust the cutting parameters based on the equipment's predicted health status, with machining efficiency and tool wear as objectives, to generate target cutting parameters. The machining strategy adjustment module is used to transmit the target cutting parameters to the fixed beam CNC gantry machining center and adjust the machining strategy in real time.

[0067] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A data processing method for a fixed-beam CNC gantry machining center based on cloud computing, characterized in that, The data processing method for the fixed-beam CNC gantry machining center includes the following steps: The equipment processing data is encrypted using industrial IoT protocols and then uploaded to the cloud platform to obtain multi-dimensional data. Design a multimodal feature fusion model, input the multidimensional data into the model, use a Transformer encoder to perform self-attention modeling on the time series data, fuse features from different sensors through a cross-modal attention mechanism, and output the device's predicted health status. A physics-guided deep reinforcement learning model PG-DRL is established. Physical rules for the machining process are generated through FEA finite element analysis and used as constraints for the reward function. The DQN algorithm is used to predict the health status of the equipment and adjust the cutting parameters with machining efficiency and tool wear as objectives to generate target cutting parameters. The target cutting parameters are transmitted to the fixed-beam CNC gantry machining center, and the machining strategy is adjusted in real time.

2. The data processing method for a fixed-beam CNC gantry machining center based on cloud computing as described in claim 1, characterized in that, The process involves encrypting the equipment processing data using an Industrial Internet of Things (IIoT) protocol and uploading it to a cloud platform to obtain multi-dimensional data, including: Sensors are deployed at the tool position, spindle position, and guide rail position of the fixed beam CNC gantry machining center to collect vibration parameters, temperature parameters, and cutting force parameters in real time, thereby obtaining the equipment's machining data. The AES-256 symmetric encryption algorithm is used to encrypt the equipment processing data. The key is dynamically generated through an ECC-based key negotiation mechanism to obtain the encrypted equipment processing data. The encrypted device's processing data is uploaded to the cloud platform via an Industrial Internet of Things (IIoT) protocol, and then filtered and downsampled sequentially through an edge gateway to obtain multi-dimensional data.

3. The data processing method for a fixed-beam CNC gantry machining center based on cloud computing as described in claim 1, characterized in that, The aforementioned multimodal feature fusion model inputs the multidimensional data into the model, uses a Transformer encoder to perform self-attention modeling on the time-series data, fuses features from different sensors through a cross-modal attention mechanism, and outputs a device prediction of health status, including: Multi-sensor data is converted into fixed-length vectors and position encoding is added to obtain time-series data; The encoder module consists of six identical processing units stacked together. Each unit contains two parts: the first part is a multi-head attention mechanism, which is used to focus on important information in the data from eight different perspectives; the second part is a feedforward neural network, which is used to further refine and transform the information processed by the attention mechanism.

4. The data processing method for a fixed-beam CNC gantry machining center based on cloud computing as described in claim 1, characterized in that, The multimodal feature fusion model is designed by inputting the multidimensional data into the model, using a Transformer encoder to perform self-attention modeling on the time-series data, fusing features from different sensors through a cross-modal attention mechanism, and outputting a device prediction of health status. It also includes: Data from different types of sensors are converted into the same feature space, and corresponding weights are assigned according to the importance of different features. The fused features are then obtained through comprehensive calculation. Features that have a significant impact on the health status of the equipment are assigned higher weights, while features that have a smaller impact are assigned lower weights. By fusing features from different sensors through a cross-modal attention mechanism, the output device predicts the health status, including at least normal, warning, and fault.

5. The data processing method for a fixed-beam CNC gantry machining center based on cloud computing as described in claim 1, characterized in that, The PG-DRL-based physics-guided deep reinforcement learning model is established by generating physical rules for the machining process through FEA finite element analysis, which serve as constraints for the reward function. The DQN algorithm is used to predict the health status of the equipment, and cutting parameters are adjusted with machining efficiency and tool wear as objectives, generating target cutting parameters, including: The physical rules module is used to transform the physical constraints in the processing into quantifiable decision boundaries, and the reinforcement learning network optimizes the decision strategy through dynamic interaction with the processing environment. The machining process is simulated in multiple dimensions using FEA finite element analysis technology. The simulation incorporates workpiece material properties, tool geometry parameters, and machine tool structural characteristics to analyze stress distribution, heat conduction path, and vibration propagation patterns in the cutting area.

6. The data processing method for a fixed-beam CNC gantry machining center based on cloud computing as described in claim 1, characterized in that, The PG-DRL-based physics-guided deep reinforcement learning model is established by generating physical rules for the machining process through FEA finite element analysis, which serve as constraints for the reward function. The DQN algorithm is used to predict the health status of the equipment, and cutting parameters are adjusted with machining efficiency and tool wear as objectives, generating target cutting parameters, including: The basic reward function is constructed with machining efficiency and tool wear as the core optimization objectives, while physical rules are embedded as constraints into the reward mechanism. A positive reward is given when the cutting parameters are adjusted in accordance with physical rules and the machining efficiency is improved; if the parameter adjustment causes a violation of physical rules, a penalty mechanism is triggered to reduce the reward value.

7. The data processing method for a fixed-beam CNC gantry machining center based on cloud computing as described in claim 1, characterized in that, The process of transmitting the target cutting parameters to a fixed-beam CNC gantry machining center and adjusting the machining strategy in real time includes: The target cutting parameters are transmitted to the fixed-beam CNC gantry machining center via an industrial Internet of Things (IIoT) protocol. Adjust the processing strategy according to the health status of the equipment: once every 5 minutes in normal condition, once every 1 minute in early warning condition, and immediately in fault condition.

8. A data processing system for a cloud-based fixed-beam CNC gantry machining center, characterized in that: The data processing method for the fixed-beam CNC gantry machining center includes the following modules: The processing data acquisition module is used to encrypt the processing data of the equipment and upload it to the cloud platform through the Industrial Internet of Things protocol to obtain multi-dimensional data. The device status prediction module is used to design a multimodal feature fusion model. The multi-dimensional data is input into the model, the Transformer encoder is used to perform self-attention modeling on the time series data, and the features of different sensors are fused through a cross-modal attention mechanism to output the device's predicted health status. The cutting parameter optimization module is used to establish a PG-DRL physics-guided deep reinforcement learning model. It generates physical rules for the machining process through FEA finite element analysis and uses them as constraints for the reward function. The DQN algorithm is used to adjust the cutting parameters based on the health status of the equipment, with machining efficiency and tool wear as objectives, to generate target cutting parameters. The machining strategy adjustment module is used to transmit the target cutting parameters to the fixed beam CNC gantry machining center and adjust the machining strategy in real time.

9. The data processing system for a fixed-beam CNC gantry machining center based on cloud computing as described in claim 8, characterized in that, The cutting parameter optimization module Includes the following sub-modules: A submodule is set up to construct a basic reward function with machining efficiency and tool wear as the core optimization objectives, while embedding physical rules as constraints into the reward mechanism. The judgment submodule is used to give a positive reward when the cutting parameters are adjusted in accordance with the physical rules and the machining efficiency is improved; if the parameter adjustment causes a violation of the physical rules, a penalty mechanism is triggered to reduce the reward value.

10. The data processing system for a fixed-beam CNC gantry machining center based on cloud computing as described in claim 8, characterized in that, The cutting parameter optimization module includes the following sub-modules: The transmission submodule is used to transmit the target cutting parameters to the fixed beam CNC gantry machining center via an industrial Internet of Things protocol. The adjustment submodule is used to adjust the processing strategy according to the health status of the equipment: once every 5 minutes in normal condition, once every 1 minute in early warning condition, and immediately in fault condition.

Images