Three-dimensional cutting dynamic parameter adjusting system and method based on deep learning

Abstract

The invention discloses a three-dimensional cutting dynamic parameter adjusting system and method based on deep learning, and relates to the technical field of application of three-dimensional five-axis fiber laser cutting machines in the automobile manufacturing industry, and the three-dimensional cutting dynamic parameter adjusting system comprises a multi-mode sensor cluster, an edge calculation unit, a closed-loop control module and an AI model library, the multi-mode sensor cluster collects cutting data, the edge calculation unit receives the multi-mode sensor data and outputs dynamic adjustment parameters, and the closed-loop control module transmits the dynamic parameters to the laser generator and the five-axis motion controller through a high-speed communication bus to form a'sensing-decision-execution 'closed loop. The cutting state and risk are sensed in real time by designing a multi-mode sensor for collaborative decision, the problems that manual experience depends on and dynamic response lags are solved, the reject ratio is reduced, and the response speed is increased.

Description

Technical Field

[0001] The present invention relates to the technical field of application of three-dimensional five-axis fiber laser cutting machines in the automobile manufacturing industry, and specifically to a three-dimensional cutting dynamic parameter adjustment system and method based on deep learning. Background Art

[0002] As the automotive industry moves toward lightweighting and personalization, body-in-white (BIW) manufacturing places increasingly higher demands on laser cutting precision and efficiency. Parameter adjustment for traditional 3D five-axis fiber laser cutting machines relies heavily on manual experience, making it difficult to adapt to the dynamic demands of cutting complex curved surfaces.

[0003] Patent CN110428506B discloses a parameter-based dynamic geometric three-dimensional graphics cutting implementation method. The above patent realizes the expansion of the cutting sequence, including parallel cutting and serial cutting, for users to choose, which is more convincing and meets teaching needs.

[0004] The above patent uses multiple cutting planes to cut three-dimensional graphics, and dynamically controls the cutting rate through parameters to achieve a dynamic cutting effect. However, in the automotive manufacturing industry, the adjustment of cutting parameters mainly relies on manual experience, which is difficult to adapt to the dynamic changes required by complex processing.

[0005] To this end, this application proposes a three-dimensional cutting dynamic parameter adjustment system and method based on deep learning, which can perceive cutting status and risks in real time and solve the problems of dependence on manual experience and dynamic response lag. Summary of the Invention

[0006] The purpose of the present invention is to provide a three-dimensional cutting dynamic parameter adjustment system and method based on deep learning to solve the technical problem raised in the above background technology that relying on manual experience is difficult to adapt to the dynamic changes in the complex surface cutting process.

[0007] To achieve the above objectives, the present invention provides the following technical solutions: a deep learning-based three-dimensional cutting dynamic parameter adjustment system, comprising a multimodal sensor cluster, an edge computing unit, a closed-loop control module, and an AI model library. The multimodal sensor cluster collects cutting data, the edge computing unit receives the multimodal sensor data and outputs dynamic adjustment parameters, and the closed-loop control module transmits the dynamic parameters to the laser generator and five-axis motion controller via a high-speed communication bus, forming a "perception-decision-execution" closed loop.

[0008] The edge computing unit has a built-in deep learning model, which takes as input multimodal sensor data, workpiece 3D geometry parameters and material properties, and outputs dynamically adjusted parameters: laser power, cutting speed, nozzle height and focus position;

[0009] The AI model library stores historical cutting data, abnormal scenarios and optimization strategies, and supports online model updates and transfer learning.

[0010] Preferably, the multimodal sensor cluster includes:

[0011] Plasma detection sensor, used to monitor the laser scattering spectrum signal during the cutting process and determine missed cutting and non-penetration based on the spectral intensity threshold;

[0012] The capacitive sensing height adjustment module consists of a ceramic cutting nozzle, an amplifier board, and cables. It dynamically calculates the nozzle height deviation by detecting the change in capacitance between the cutting nozzle and the workpiece.

[0013] The magnetic anti-collision module includes upper and lower magnetic contacts and an anti-collision detection circuit. When external force causes the contacts to separate by more than 0.5mm, it triggers the device to shut down and re-route instructions.

[0014] Preferably, the deep learning model adopts a spatiotemporal dual-stream network architecture, including:

[0015] Time series network: processes sensor time series data based on LSTM units to predict cutting stability;

[0016] Spatial flow network, which analyzes workpiece geometric features based on 3D convolutional neural networks and generates spatial cutting parameter recommendations;

[0017] Feature fusion layer: Dynamically weights spatiotemporal features through the attention mechanism and outputs comprehensive parameter adjustment instructions;

[0018] The reinforcement learning optimization module constructs a multi-objective reward function based on cutting efficiency, material utilization rate, and defect rate, and optimizes the model weights online through the PPO algorithm.

[0019] Preferably, the magnetic anti-collision module further includes:

[0020] Collision prediction algorithm: Based on the five-axis motion trajectory and the workpiece's 3D point cloud data, the Kalman filter predicts the cutting head's position offset within the next 5ms. If the offset exceeds the safety threshold, path replanning is triggered in advance.

[0021] Multi-level response mechanism:

[0022] Level 1 response: In case of minor collision, only adjust the cutting speed to the safety level;

[0023] Secondary response: In the event of a serious collision, the vehicle will be shut down and the A* algorithm will be activated to dynamically generate an obstacle avoidance path.

[0024] Collision scenario learning module: uploads abnormal data to the AI model library to generate an anti-collision rule library for active avoidance when cutting similar workpieces.

[0025] Preferably, the adjustment logic of the capacitive sensing height adjustment module includes:

[0026] Dynamic compensation algorithm: Dynamically adjust the capacitance signal amplification factor according to the workpiece surface roughness and material conductivity to ensure that the height detection error is ≤0.03mm;

[0027] Adaptive filtering technology: uses wavelet transform to filter out high-frequency noise and retain effective capacitance signals;

[0028] Double closed-loop control:

[0029] Inner loop: real-time adjustment of nozzle height based on PID controller;

[0030] Outer loop: Predict the thermal deformation of the workpiece through a deep learning model and compensate for height offset in advance.

[0031] Preferably, the edge computing unit integrates the following model update mechanism:

[0032] Incremental learning module: After each cutting task is completed, key data including cutting quality score and sensor abnormality mark are automatically extracted, and model parameters are updated through online gradient descent method with an update cycle of ≤3 minutes;

[0033] Model version management: retain historical model versions and automatically roll back to the optimal version when new data causes performance, or the failure rate, to increase by more than 0.1%.

[0034] Federated Learning Interface: supports encrypted sharing of local model parameters among multiple devices to build a global optimization model.

[0035] Preferably, the parameter adjustment system further integrates a visual detection subsystem, including:

[0036] High-speed industrial camera to capture infrared thermal imaging and visible light images of the cutting area;

[0037] Image fusion algorithm: superimposes thermal imaging data with visible light images and uses the U-Net network to segment and cut defect areas;

[0038] Multi-sensor collaborative decision-making:

[0039] When the magnetic collision avoidance module detects a collision risk and the visual detection subsystem simultaneously identifies an obstacle in the path, an emergency stop is triggered;

[0040] When the visual inspection subsystem detects local overheating, it automatically reduces the laser power by 10%-20%.

[0041] Preferably, the deep learning model is embedded in an energy consumption optimization module, specifically including:

[0042] Material thermal characteristics analysis unit: Calculates the minimum energy threshold based on the thermal conductivity and specific heat capacity of the material to ensure that the cutting depth meets the standard;

[0043] Laser pulse optimization strategy: Changing the continuous laser to an adaptive pulse mode, with the pulse frequency and duty cycle dynamically adjusted by the model, reduces energy consumption by ≥15%;

[0044] Scrap recycling feedback: Real-time monitoring of scrap quality through weighing sensors, optimizing cutting paths to reduce scrap, and improving material utilization by ≥8%.

[0045] Preferably, the AI model library supports the following functions:

[0046] Knowledge distillation technology: distills expert experience and historical data into lightweight rule models;

[0047] Geometric feature matching engine: uses point cloud registration algorithm to compare the curvature distribution of new workpieces with those in the model library, automatically recommends similar process parameters, and the adaptation time is ≤ 2 minutes;

[0048] Cross-material transfer learning: If the new workpiece material is not in the library, it is mapped to the existing parameter combination based on the similarity of thermophysical properties.

[0049] Preferably, the adjustment method comprises the following steps:

[0050] S1. Data acquisition stage: The multimodal sensor cluster collects plasma spectrum, capacitance value, magnetic contact status, workpiece geometry data and ambient temperature in real time, and the visual inspection subsystem obtains thermal imaging and topographic images of the cutting area;

[0051] S2, model reasoning stage: the collected data is input into the spatiotemporal dual-stream network, and the dynamic parameter combination of laser power, cutting speed, nozzle height, and focus position is output, and the pulse laser parameters are generated through the energy consumption optimization module;

[0052] S3, execution and feedback stage: The closed-loop control module drives the laser generator and the five-axis motion mechanism to perform parameter adjustment, and updates the model weights through the incremental learning module based on the cutting quality and energy consumption data;

[0053] S4, exception handling stage: If a collision risk or parameter out-of-limit is detected, a multi-level response mechanism is triggered and the abnormal scenario is recorded in the AI model library;

[0054] S5: Cloud-based collaborative optimization: Local model parameters and abnormal data are encrypted and uploaded to the cloud. A global optimization model is generated through federated learning and distributed to all devices.

[0055] S6. Process chain traceability: Generate a unique ID for each cutting task, associate parameter settings, sensor data and quality reports, and support full life cycle quality analysis.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. This invention uses multi-modal sensors to collaboratively make decisions and perceive cutting status and risks in real time, solving the problems of manual experience dependence and dynamic response lag, reducing defect rates and improving response speed.

[0058] 2. This invention achieves accurate prediction and adaptive adjustment of dynamic parameters by designing a time-space dual-flow network architecture, solving the problem of difficult parameter matching in complex working conditions, improving cutting efficiency and reducing energy consumption;

[0059] 3. The present invention is designed with a liberalized process parameter library to achieve the function of rapid adaptation to new workpieces and materials, solving the problems of long new process development cycle and high debugging costs, reducing debugging time and lowering material waste rate;

[0060] 4. The present invention is designed with a green energy consumption optimization closed loop to achieve synergy between energy saving and consumption reduction and waste recycling, thereby solving the problem of high energy consumption and low material utilization rate of traditional continuous laser, improving material utilization rate and reducing energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Schematic diagram of the parameter adjustment system framework of the present invention;

[0062] Figure 2 Schematic diagram of the parameter adjustment method of the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] See also Figure 1 and Figure 2 The present invention provides an embodiment of a deep learning-based three-dimensional cutting dynamic parameter adjustment system, comprising a multimodal sensor cluster, an edge computing unit, a closed-loop control module, and an AI model library. The multimodal sensor cluster collects cutting data, the edge computing unit receives the multimodal sensor data and outputs dynamic adjustment parameters, and the closed-loop control module transmits the dynamic parameters to a laser generator and a five-axis motion controller via a high-speed communication bus, forming a "perception-decision-execution" closed loop.

[0065] The edge computing unit has a built-in deep learning model, which takes as input multimodal sensor data, workpiece 3D geometry parameters and material properties, and outputs dynamically adjusted parameters: laser power, cutting speed, nozzle height and focus position;

[0066] The AI model library stores historical cutting data, abnormal scenarios and optimization strategies, and supports online model updates and transfer learning;

[0067] The edge computing unit integrates the following model update mechanisms:

[0068] Incremental learning module: After each cutting task is completed, key data including cutting quality score and sensor abnormality mark are automatically extracted, and model parameters are updated through online gradient descent method with an update cycle of ≤3 minutes;

[0069] Model version management: retain historical model versions and automatically roll back to the optimal version when new data causes performance, or the failure rate, to increase by more than 0.1%.

[0070] Federated learning interface: supports encrypted sharing of local model parameters among multiple devices to build a global optimization model;

[0071] Furthermore, data acquisition and input: sensor data: plasma spectrum intensity: real-time monitoring value 5200 lux; capacitance value: the current nozzle height deviation is detected to be +0.08mm; magnetic contact state: closed; working parameters: three-dimensional geometric data: STL file analysis of surface curvature; material properties: aluminum alloy 5052; deep learning model reasoning: spatiotemporal dual-stream network processing: time series flow: analyze 10 consecutive frames of capacitance value data and predict the height deviation trend to be +0.12mm within the next 0.5s; spatial flow: identify high curvature areas of the surface and recommend reducing the cutting speed by 15% to avoid heat accumulation; special Feature fusion: The attention mechanism weights spatiotemporal features to generate parameter instructions: laser power is increased from 500W to 530W, cutting speed is reduced from 4.0m / min to 3.4m / min, and nozzle height is adjusted from 2.0mm to 1.92mm. Closed-loop execution and feedback: After receiving the instruction, the laser generator switches to 530W mode within 0.5ms. The cutting head moves along the optimized path, the speed is reduced to 3.4m / min, and the follow-up height adjustment device dynamically maintains the nozzle height at 1.92mm. After the cutting is completed, the visual inspection subsystem detects the cross-section roughness Ra = 6.3um and judges it as qualified.

[0072] Incremental learning module: Key data extraction: The cutting quality score in this task was 92 points, and the sensor had no abnormal markings; Online gradient descent: The model updates the weights of the fully connected layer based on the score, and the update takes 2 minutes; Model version management: Version rollback logic: If the cutting defect rate is greater than 0.5% for three consecutive times after the update, it will automatically switch to the historical optimal version; Federated learning interface: Multi-device collaboration: Share local model parameters with 10-state devices in the factory, generate a global model after aggregation in the cloud, and send it to each device. The global model reduces the adaptation time for new workpieces from 15 minutes to 7 minutes.

[0073] See also Figure 1 and Figure 2 The present invention provides an embodiment of a three-dimensional cutting dynamic parameter adjustment system based on deep learning, wherein the multimodal sensor cluster includes:

[0074] Plasma detection sensor, used to monitor the laser scattering spectrum signal during the cutting process and determine missed cutting and non-penetration based on the spectral intensity threshold;

[0075] The capacitive sensing height adjustment module consists of a ceramic cutting nozzle, an amplifier board, and cables. It dynamically calculates the nozzle height deviation by detecting the change in capacitance between the cutting nozzle and the workpiece.

[0076] The magnetic anti-collision module includes upper and lower magnetic contacts and an anti-collision detection circuit. When external force causes the contacts to separate by more than 0.5mm, it triggers a device shutdown and path replanning command.

[0077] The magnetic anti-collision module further includes:

[0078] Collision prediction algorithm: Based on the five-axis motion trajectory and the workpiece's 3D point cloud data, the Kalman filter predicts the cutting head's position offset within the next 5ms. If the offset exceeds the safety threshold, path replanning is triggered in advance.

[0079] Multi-level response mechanism:

[0080] Level 1 response: In case of minor collision, only adjust the cutting speed to the safety level;

[0081] Secondary response: In the event of a serious collision, the vehicle will be shut down and the A* algorithm will be activated to dynamically generate an obstacle avoidance path.

[0082] Collision scenario learning module: uploads abnormal data to the AI model library to generate a collision avoidance rule library for active avoidance when cutting similar workpieces;

[0083] The adjustment logic of the capacitive sensing height adjustment module includes:

[0084] Dynamic compensation algorithm: Dynamically adjust the capacitance signal amplification factor according to the workpiece surface roughness and material conductivity to ensure that the height detection error is ≤0.03mm;

[0085] Adaptive filtering technology: uses wavelet transform to filter out high-frequency noise and retain effective capacitance signals;

[0086] Double closed-loop control:

[0087] Inner loop: real-time adjustment of nozzle height based on PID controller;

[0088] Outer loop: Predicts the thermal deformation of the workpiece through a deep learning model and compensates for height offset in advance;

[0089] Furthermore, the plasma detection sensor monitors the laser scattering spectral signal during the cutting process in real time to judge the cutting quality. During the cutting process, the laser interacts with the material to generate plasma. The plasma detection sensor collects the spectral signal at a sampling frequency of 1kHz and sets a spectral intensity threshold. When the detected signal is lower than the threshold, it is judged as "not penetrated". If the signal continues to fluctuate beyond the limit, it is marked as not "missed cut". If an abnormality is found, the system immediately reduces the cutting speed to a safe level and starts process parameter optimization through the edge computing unit.

[0090] The capacitive sensing height adjustment module dynamically adjusts the nozzle height to avoid scraping or distance deviation caused by workpiece deformation or surface unevenness. The ceramic cutting nozzle forms a capacitor with the workpiece surface. The amplifier board detects changes in capacitance with a resolution of 0.01pF. If the workpiece surface roughness Ra value is high, the amplification factor is automatically increased to 50 times to enhance signal sensitivity. For non-conductive materials such as composite materials, a low-frequency excitation signal of 1kHz is used to reduce interference. PID control adjusts the nozzle height in real time with a response time of ≤5ms, ensuring a height error of ≤0.03mm. A deep learning model analyzes the workpiece's thermal expansion coefficient, predicts height offset within the next 2s, and compensates in advance.

[0091] The magnetic anti-collision module detects collision risks and triggers avoidance strategies to ensure equipment safety. Based on the five-axis motion trajectory and the three-dimensional point cloud of the workpiece, the Kalman filter predicts the position offset of the cutting head within the next 5ms. If the predicted offset exceeds ±0.3mm, the path optimization is triggered in advance; the first-level response: when the contact wind force is ≤0.5mm, the cutting speed is reduced to a safe value, and the laser power is simultaneously reduced by 20%; the second-level response: when the contact separation is greater than 0.5mm, the machine is shut down and an obstacle avoidance path is generated within 10ms through the A* algorithm; abnormal data such as collision position and external force direction are uploaded to the AI model library to build an anti-collision rule library.

[0092] See also Figure 1 and Figure 2 The present invention provides an embodiment of a three-dimensional cutting dynamic parameter adjustment system based on deep learning, wherein the deep learning model adopts a spatiotemporal dual-stream network architecture, including:

[0093] Time series network: processes sensor time series data based on LSTM units to predict cutting stability;

[0094] Spatial flow network, which analyzes workpiece geometric features based on 3D convolutional neural networks and generates spatial cutting parameter recommendations;

[0095] Feature fusion layer: Dynamically weights spatiotemporal features through the attention mechanism and outputs comprehensive parameter adjustment instructions;

[0096] The reinforcement learning optimization module constructs a multi-objective reward function based on cutting efficiency, material utilization, and defect rate, and optimizes the model weights online through the PPO algorithm;

[0097] Furthermore, the temporal flow network processes sensor time series data such as capacitance values and plasma signals to predict the stability of the cutting process. The spatial flow network analyzes the three-dimensional geometric features of the workpiece, such as surface curvature and thickness distribution, and generates spatial cutting parameter recommendations. The feature fusion layer dynamically weights the spatiotemporal features and outputs comprehensive adjustment instructions. The reinforcement learning module optimizes the model weights online to achieve a balance among multiple objectives, namely efficiency, material utilization rate, and defect rate.

[0098] The time series flow network inputs the time series sensor signal and time window, normalizes the capacitance value to the range of 0-1, removes transient noise, and processes it through the LSTM unit. The input is three-dimensional, including capacitance, plasma, and contact status. The hidden layer has 64 neurons, memorizes the cutting stability trend, and outputs the cutting stability score. A score of <80 triggers parameter correction.

[0099] Time-flow network input: 3D workpiece geometry data: STL files parsed into voxel grids; material properties: thermal conductivity, reflectivity, and melting point; processing logic: 3D convolution operation: First layer: 3×3×3 convolution kernel extracts local geometric features such as curvature changes and thickness mutations; second layer: hole convolution expands the receptive field and identifies large-scale structural features such as rib locations; feature mapping: outputs recommendations for reducing cutting speeds in high-curvature areas to avoid heat accumulation, and outputs recommendations for reducing laser power in thin-walled areas to prevent burn-through; parameter recommendation generation: spatial cutting speed distribution map and focus position offset compensate for surface focal length deviation;

[0100] The feature fusion layer inputs temporal stream features and spatial stream features. The fusion logic is as follows: Feature alignment: upsample the temporal features to the same temporal resolution as the spatial features; Attention weight calculation: generate attention scores through the fully connected layer and dynamically assign weights; Weighted output: comprehensive parameter instruction = 0.7 × spatial suggestion + 0.3 × temporal suggestion;

[0101] Optimization objectives of the reinforcement learning optimization module: cutting efficiency: cutting length per unit time, material utilization rate and defect rate; training process: state space: sensor data + workpiece geometric features; action space: continuous adjustment of laser power, cutting speed and nozzle height; reward function design: basic reward = efficiency × 0.4 + material utilization rate × 0.3 + (1-defect rate) × 0.3; penalty item: collision trigger deduction of 50 points, parameter exceeding the limit deduction of 30 points; PPO strategy update: 1000 cutting trajectory data are collected online, and the strategy network is updated through importance sampling.

[0102] See also Figure 1 and Figure 2 The present invention provides an embodiment of a three-dimensional cutting dynamic parameter adjustment system based on deep learning, wherein the parameter adjustment system further integrates a visual detection subsystem, including:

[0103] High-speed industrial camera to capture infrared thermal imaging and visible light images of the cutting area;

[0104] Image fusion algorithm: superimposes thermal imaging data with visible light images and uses the U-Net network to segment and cut defect areas;

[0105] Multi-sensor collaborative decision-making:

[0106] When the magnetic collision avoidance module detects a collision risk and the visual detection subsystem simultaneously identifies an obstacle in the path, an emergency stop is triggered;

[0107] When only the visual inspection subsystem detects local overheating, it automatically reduces the laser power by 10%-20%;

[0108] Furthermore, during the cutting process, visible light cameras and infrared thermal imaging cameras synchronously capture images. The image data is transmitted to the image processing unit via optical fiber, and the processing results are uploaded to the edge computing unit to participate in multi-sensor decision-making. Image alignment and registration: Calibration: The spatial coordinate systems of the visible light and infrared cameras are aligned using the checkerboard calibration method, with a registration error of ≤0.1 pixel. Real-time synchronization: Dynamically compensate for image offset based on the motion trajectory of the cutting head.

[0109] Image fusion and U-Net segmentation: Fusion algorithm: Visible light images: Extract texture details such as the sharpness of the cut edge; Infrared images: Map the temperature field (high temperature areas > material melting point + 50°C are marked in red); Fusion output: A 4-channel image superimposed with RGB and heat maps; U-Net network segmentation: Input: Fusion image; Output: Defect area mask; Network structure: 5-layer encoder-decoder with skip connections to preserve details. The training set contains more than 100,000 material cutting images.

[0110] Defect classification and feedback: Classification rules: Burr: edge protrusion height > 0.1mm, trigger nozzle height adjustment; Slag: temperature residue > material melting point + 100℃, it is recommended to increase the auxiliary gas pressure; No penetration: abnormal temperature gradient, trigger laser power compensation;

[0111] Multi-sensor collaborative decision-making mechanism: Double confirmation of collision risk emergency stop: Trigger condition: The separation of the detection points of the magnetic anti-collision module is greater than 0.5mm, and the visual detection subsystem identifies the path obstacle; Decision process: The binyuan computing unit receives the dual sensor alarm information; Initiate the emergency stop protocol: The laser power is reset to zero and the five-axis motion is paused; Trigger the A* algorithm to replan the path to avoid the obstacle area; Record the obstacle position and image to the AI model library to generate the "welding slag pre-scan" rule;

[0112] Adaptive power adjustment for local overheating: Trigger condition: Only the visual inspection subsystem detects local overheating (temperature > material melting point + 150°C), and there is no magnetic collision signal; Adjustment strategy: If the overheating area is located in the center of the cutting seam: reduce the power by 10%; if the overheating area spreads to the edge: reduce the power by 20% and increase the cutting speed by 15%.

[0113] See also Figure 1 and Figure 2 The present invention provides an embodiment of a three-dimensional cutting dynamic parameter adjustment system based on deep learning, wherein the deep learning model is embedded in an energy consumption optimization module, specifically comprising:

[0114] Material thermal characteristics analysis unit: Calculates the minimum energy threshold based on the thermal conductivity and specific heat capacity of the material to ensure that the cutting depth meets the standard;

[0115] Laser pulse optimization strategy: Changing the continuous laser to an adaptive pulse mode, with the pulse frequency and duty cycle dynamically adjusted by the model, reduces energy consumption by ≥15%;

[0116] Scrap recycling feedback: Real-time monitoring of scrap quality through weighing sensors, optimizing cutting paths to reduce scrap, and improving material utilization by ≥8%;

[0117] Furthermore, material database matching: input material type, call pre-stored thermal conductivity, specific heat capacity and melting point. If the material is not entered into the database, the thermal characteristics are estimated by real-time measurement of laser reflectivity; energy threshold calculation formula: Where ρ is the material density, Cp is the material specific heat capacity, Tmelt is the material melting point temperature, Tambient is the ambient temperature, Lfusion is the material melting latent heat, η is the laser absorptivity, and α is the spot diameter correction factor. If the real-time monitoring shows that the cutting depth is insufficient, the energy is increased in steps of 10% until it reaches the standard.

[0118] Laser pulse optimization strategy: Dynamic adjustment of pulse parameters: Frequency adjustment: Thin materials <3mm use a high frequency of 5kHz to reduce heat accumulation, and thick materials >6mm use a low frequency of 2kHz to enhance single pulse energy; Duty cycle adjustment: The duty cycle in high curvature areas is reduced to 30% to avoid local overheating; The duty cycle of straight-line cutting is increased to 70% to speed up the process; Real-time energy consumption monitoring: Power calculation of actual energy consumption is compared with the continuous laser baseline value; Adaptive learning: The reinforcement learning module optimizes the pulse parameter combination based on historical data, and the reward function weights energy consumption and cutting quality;

[0119] Waste recycling feedback and path optimization: Real-time monitoring of waste: Weighing sensors are installed in the waste collection trough to monitor the quality of waste in real time; Path optimization algorithm: Nested algorithm: Based on the geometric layout of the workpiece, optimize the sheet material utilization rate, such as from 82% to 90%; Adaptive compensation: If it is detected that the proportion of scrap exceeds the limit, such as >12%, the path re-planning is triggered: reduce the idle movement speed to 120%, increase the common edge cutting and shared cutting seam to reduce waste; the waste data is uploaded to the AI model library to generate a "high-utilization path template".

[0120] See also Figure 1 and Figure 2 The present invention provides an embodiment of a three-dimensional cutting dynamic parameter adjustment system based on deep learning, wherein the AI model library supports the following functions:

[0121] Knowledge distillation technology: distills expert experience and historical data into lightweight rule models;

[0122] Geometric feature matching engine: uses point cloud registration algorithm to compare the curvature distribution of new workpieces with those in the model library, automatically recommends similar process parameters, and the adaptation time is ≤ 2 minutes;

[0123] Cross-material transfer learning: If the new workpiece material is not in the library, it is mapped to the existing parameter combination based on the similarity of thermophysical properties;

[0124] Furthermore, data input includes: expert experience: manually debugging parameter tables such as laser power and speed combinations for different curvatures; historical data: 100,000 cutting records containing sensor data, parameter settings, and quality scores; key rule mining: extracting high-frequency parameter combinations through a decision tree algorithm, for example: when the curvature is greater than 0.1 / mm, the power is increased by 10%; model compression: distilling the original 1.2GB deep learning model into a 50MB rule model while retaining 95% accuracy; embedding the rule model into the edge computing unit, reducing the inference speed from 15ms to 3ms;

[0125] New workpiece scanning: 3D laser scanning generates a point cloud with a resolution of 0.1mm. Feature extraction: Calculates the Gaussian curvature, average curvature, thickness gradient, and edge sharpness of the curvature distribution. ICP registration algorithm: Aligns the new workpiece point cloud with 500 workpieces in the model library, calculating the mean square error of the curvature difference to less than 0.05. Top-3 recommendations: Output the three most similar sets of process parameters, such as power, speed, and focus position. If the recommended parameters do not meet the cutting quality standards, such as burr height greater than 0.1mm, the online learning module triggers dynamic optimization.

[0126] New material parameters: thermal conductivity 160 W / m·K, specific heat capacity 0.9 J / g·K, reflectivity 80%. Euclidean distance calculation: Compare the thermophysical properties of the materials in the library to find the closest material, such as aluminum alloy 6061, with a thermal conductivity of 167 W / m·K and an error of 4.3%. Weight adjustment: Based on the difference in thermal conductivity, the power is proportionally corrected, such as +5%. Output mapping parameters: Laser power = aluminum alloy 6061 baseline value × 1.05, cutting speed = baseline value × 0.95.

[0127] Working Principle: A plasma sensor monitors the laser scattering spectrum to determine cutting depth and quality. A capacitive height adjustment module detects the nozzle-workpiece distance in real time. A magnetic collision avoidance module predicts collision risks and triggers a multi-level response mechanism. Sensor timing data and workpiece geometric features are weightedly fused through an attention mechanism to generate dynamic parameter recommendations.

[0128] The temporal flow predicts the cutting stability trend, the spatial flow analyzes the geometric complexity of the workpiece, and generates a spatial parameter distribution map. The multi-objective reward function is based on cutting efficiency, material utilization rate, and defect rate. The PPO algorithm is used to optimize the model weights online to achieve parameter self-exploration.

[0129] Dynamic parameters are transmitted to the laser generator and five-axis controller via the edge computing unit to adjust the power, speed and focus position in real time. Cutting quality data is fed back to the AI model library, triggering incremental learning and federated learning, and continuously improving the model's generalization capabilities.

[0130] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A deep learning-based 3D cutting dynamic parameter adjustment system, comprising a multimodal sensor cluster, an edge computing unit, a closed-loop control module, and an AI model library, characterized by: The multimodal sensor cluster collects cutting data, the edge computing unit receives the multimodal sensor data and outputs dynamic adjustment parameters, and the closed-loop control module transmits the dynamic parameters to the laser generator and five-axis motion controller via a high-speed communication bus, forming a "perception-decision-execution" closed loop; The edge computing unit has a built-in deep learning model, which takes as input multimodal sensor data, workpiece 3D geometry parameters and material properties, and outputs dynamically adjusted parameters: laser power, cutting speed, nozzle height and focus position; The AI model library stores historical cutting data, abnormal scenarios and optimization strategies, and supports online model updates and transfer learning.

2. The deep learning-based 3D cutting dynamic parameter adjustment system according to claim 1, characterized in that: The multimodal sensor cluster comprises: Plasma detection sensor, used to monitor the laser scattering spectrum signal during the cutting process and determine missed cutting and non-penetration based on the spectral intensity threshold; The capacitive sensing height adjustment module consists of a ceramic cutting nozzle, an amplifier board, and cables. It dynamically calculates the nozzle height deviation by detecting the change in capacitance between the cutting nozzle and the workpiece. The magnetic anti-collision module includes upper and lower magnetic contacts and an anti-collision detection circuit. When external force causes the contacts to separate by more than 0.5mm, it triggers the device to shut down and re-route instructions.

3. The deep learning-based 3D cutting dynamic parameter adjustment system according to claim 1, characterized in that: The deep learning model adopts a spatiotemporal dual-stream network architecture, including: Time series network: processes sensor time series data based on LSTM units to predict cutting stability; Spatial flow network, which analyzes workpiece geometric features based on 3D convolutional neural networks and generates spatial cutting parameter recommendations; Feature fusion layer: Dynamically weights spatiotemporal features through the attention mechanism and outputs comprehensive parameter adjustment instructions; The reinforcement learning optimization module constructs a multi-objective reward function based on cutting efficiency, material utilization rate, and defect rate, and optimizes the model weights online through the PPO algorithm.

4. The deep learning-based 3D cutting dynamic parameter adjustment system according to claim 2, characterized in that: The magnetic anti-collision module further includes: Collision prediction algorithm: Based on the five-axis motion trajectory and the workpiece's 3D point cloud data, the Kalman filter predicts the cutting head's position offset within the next 5ms. If the offset exceeds the safety threshold, path replanning is triggered in advance. Multi-level response mechanism: Level 1 response: In case of minor collision, only adjust the cutting speed to the safety level; Secondary response: In the event of a serious collision, the vehicle will be shut down and the A* algorithm will be activated to dynamically generate an obstacle avoidance path. Collision scenario learning module: uploads abnormal data to the AI model library to generate an anti-collision rule library for active avoidance when cutting similar workpieces.

5. The deep learning-based 3D cutting dynamic parameter adjustment system according to claim 2, characterized in that: The adjustment logic of the capacitive sensing height adjustment module includes: Dynamic compensation algorithm: Dynamically adjust the capacitance signal amplification factor according to the workpiece surface roughness and material conductivity to ensure that the height detection error is ≤0.03mm; Adaptive filtering technology: uses wavelet transform to filter out high-frequency noise and retain effective capacitance signals; Double closed-loop control: Inner loop: real-time adjustment of nozzle height based on PID controller; Outer loop: Predict the thermal deformation of the workpiece through a deep learning model and compensate for height offset in advance.

6. The deep learning-based 3D cutting dynamic parameter adjustment system according to claim 1, characterized in that: The edge computing unit integrates the following model update mechanisms: Incremental learning module: After each cutting task is completed, key data including cutting quality score and sensor abnormality mark are automatically extracted, and model parameters are updated through online gradient descent method with an update cycle of ≤3 minutes; Model version management: retain historical model versions and automatically roll back to the optimal version when new data causes performance, or the failure rate, to increase by more than 0.1%. Federated Learning Interface: supports encrypted sharing of local model parameters among multiple devices to build a global optimization model.

7. The deep learning-based 3D cutting dynamic parameter adjustment system according to claim 1, characterized in that: The parameter adjustment system further integrates a visual detection subsystem, including: High-speed industrial camera to capture infrared thermal imaging and visible light images of the cutting area; Image fusion algorithm: superimposes thermal imaging data with visible light images and uses the U-Net network to segment and cut defect areas; Multi-sensor collaborative decision-making: When the magnetic collision avoidance module detects a collision risk and the visual detection subsystem simultaneously identifies an obstacle in the path, an emergency stop is triggered; When the visual inspection subsystem detects local overheating, it automatically reduces the laser power by 10%-20%.

8. The deep learning-based 3D cutting dynamic parameter adjustment system according to claim 1, characterized in that: The deep learning model is embedded in the energy consumption optimization module, which specifically includes: Material thermal characteristics analysis unit: Calculates the minimum energy threshold based on the thermal conductivity and specific heat capacity of the material to ensure that the cutting depth meets the standard; Laser pulse optimization strategy: Changing the continuous laser to an adaptive pulse mode, with the pulse frequency and duty cycle dynamically adjusted by the model, reduces energy consumption by ≥15%; Scrap recycling feedback: Real-time monitoring of scrap quality through weighing sensors, optimizing cutting paths to reduce scrap, and improving material utilization by ≥8%.

9. The deep learning-based 3D cutting dynamic parameter adjustment system according to claim 1, characterized in that: The AI model library supports the following functions: Knowledge distillation technology: distills expert experience and historical data into lightweight rule models; Geometric feature matching engine: uses point cloud registration algorithm to compare the curvature distribution of new workpieces with those in the model library, automatically recommends similar process parameters, and the adaptation time is ≤ 2 minutes; Cross-material transfer learning: If the new workpiece material is not in the library, it is mapped to the existing parameter combination based on the similarity of thermophysical properties.

10. A method for adjusting dynamic parameters of 3D cutting based on deep learning, applicable to a system for adjusting dynamic parameters of 3D cutting based on deep learning according to any one of claims 1 to 9, characterized in that: The adjustment method comprises the following steps: S1. Data acquisition stage: The multimodal sensor cluster collects plasma spectrum, capacitance value, magnetic contact status, workpiece geometry data and ambient temperature in real time, and the visual inspection subsystem obtains thermal imaging and topographic images of the cutting area; S2, model reasoning stage: the collected data is input into the spatiotemporal dual-stream network, and the dynamic parameter combination of laser power, cutting speed, nozzle height, and focus position is output, and the pulse laser parameters are generated through the energy consumption optimization module; S3, execution and feedback stage: The closed-loop control module drives the laser generator and the five-axis motion mechanism to perform parameter adjustment, and updates the model weights through the incremental learning module based on the cutting quality and energy consumption data; S4, exception handling stage: If a collision risk or parameter out-of-limit is detected, a multi-level response mechanism is triggered and the abnormal scenario is recorded in the AI model library; S5: Cloud-based collaborative optimization: Local model parameters and abnormal data are encrypted and uploaded to the cloud. A global optimization model is generated through federated learning and distributed to all devices. S6. Process chain traceability: Generate a unique ID for each cutting task, associate parameter settings, sensor data and quality reports, and support full life cycle quality analysis.

Images