A pulsed laser control system integrating dynamic knowledge graph and deep learning

Abstract

This invention discloses a pulsed laser control system integrating dynamic knowledge graphs and deep learning. The invention relates to the field of pulsed laser processing control technology and comprises, in sequence, a dynamic knowledge graph construction module, a deep learning hybrid model module, and a dual-loop collaborative optimization module. This pulsed laser control system integrating dynamic knowledge graphs and deep learning achieves accurate analysis and real-time optimization of the coupling relationships of multi-domain parameters in pulsed lasers through deep collaboration between the dynamic knowledge graph and the deep learning model. By employing cross-domain coupling quantization technology of the dynamic knowledge graph, the interactive influence of time-domain, spatial-domain, and frequency-domain parameters is transformed into an online-updable coupling coefficient matrix, solving the problem of traditional methods struggling to dynamically analyze strong parameter coupling. Through the condition-triggered computational architecture of the hybrid model, based on the dynamic activation of sub-modules using an LSTM-CNN-GNN heterogeneous network, the computational load is significantly reduced while ensuring prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of pulsed laser processing control technology, specifically a pulsed laser control system that integrates dynamic knowledge graphs and deep learning. Background Technology

[0002] In the field of pulsed laser control, the core challenge of existing technologies lies in the strong coupling between time-domain, spatial-domain, and frequency-domain parameters in pulsed laser systems, and the difficulty in dynamically analyzing cross-domain effects. Specifically, parameters such as pulse width and repetition frequency in the time domain, spot size and working distance in the spatial domain, and laser wavelength in the frequency domain do not act independently but are interconnected and mutually influential, forming complex coupling relationships. This complex coupling makes it difficult for traditional control methods to comprehensively and accurately grasp the impact of each parameter on system performance, and to achieve synergistic optimization control of multi-domain parameters in pulsed laser systems. Consequently, the technical specifications of pulsed laser systems cannot meet the diverse needs of high-precision processing scenarios. For example, complex parameter coupling can lead to instability in performance indicators such as processing accuracy, ablation depth, and surface quality, making dynamic adjustment and optimization impossible based on actual processing requirements. Therefore, effectively solving the problem of complex coupling between time-domain, spatial-domain, and frequency-domain parameters in pulsed laser systems and the difficulty in dynamically analyzing cross-domain effects has become crucial for achieving multi-dimensional synergistic optimization control of pulsed laser systems and improving the system's adaptability to high-precision processing scenarios. Summary of the Invention

[0003] To achieve the above objectives, the present invention provides the following technical solution: a pulsed laser control system integrating dynamic knowledge graphs and deep learning, comprising sequentially connected components:

[0004] The module includes a dynamic knowledge graph construction module, a deep learning hybrid model module, and a dual-closed-loop collaborative optimization module.

[0005] The input of the dynamic knowledge graph construction module is connected to the time-domain sensor, spatial-domain sensor and frequency-domain sensor of the pulsed laser system, and is used to collect pulse width, repetition frequency, spot size, working distance and laser wavelength parameters in real time.

[0006] The input of the deep learning hybrid model module is connected to the output of the dynamic knowledge graph construction module, which is used to process multi-domain parameters and output predicted values ​​of processing accuracy, ablation depth and surface quality.

[0007] The input end of the dual-closed-loop collaborative optimization module is connected to the output end of the deep learning hybrid model module, its control end is connected to the actuator of the pulsed laser system, and its feedback end is connected to the output end of the sensor.

[0008] Preferably, the dynamic knowledge graph construction module includes:

[0009] The entity relationship definition unit associates the time domain parameters with the pulse width and repetition frequency entity, the spatial domain parameters with the spot size and working distance entity, and the frequency domain parameters with the laser wavelength entity.

[0010] A cross-domain coupling strength quantization unit generates dynamic coupling coefficient matrices between the time-space domain, the time-frequency domain, and the spatial-frequency domain.

[0011] The real-time update unit dynamically adjusts the relationships between entity nodes and the coupling coefficient matrix based on sensor data.

[0012] Preferably, the cross-domain coupling strength quantization unit operates in the following manner:

[0013] Initialize the coupling coefficient matrix based on the pulsed laser transmission equation;

[0014] The coupling coefficient matrix is ​​trained and updated using historical data collected by photodetectors and spot analyzers.

[0015] When an abnormality in machining accuracy is detected, the coupling coefficient matrix is ​​corrected in real time.

[0016] Preferably, the deep learning hybrid model module includes parallel connections:

[0017] The temporal processing unit employs a long short-term memory network with temporal convolutional kernels to process sequence features such as pulse width and repetition frequency.

[0018] The spatial processing unit uses a convolutional neural network with a spatial transformation layer to process the spatial distribution characteristics of spot size and working distance.

[0019] The frequency domain processing unit employs a graph neural network, embedding the feature vectors of laser wavelength nodes in the knowledge graph.

[0020] Preferably, the output of the deep learning hybrid model module is connected to a dual verification unit, which includes:

[0021] A knowledge graph rule validator verifies whether parameter combinations violate entity relationship constraints.

[0022] To combat robustness testers, perturbation samples are generated in the neighborhood of optimized parameters and performance fluctuations are detected.

[0023] Preferably, the dual verification unit performs:

[0024] If a parameter combination is deemed invalid by the knowledge graph rule validator, the combination is discarded.

[0025] If the adversarial robustness tester detects performance fluctuations exceeding a threshold, it triggers model weight fine-tuning.

[0026] Preferably, the dual-closed-loop collaborative optimization module includes:

[0027] The parameter optimizer uses an evolutionary algorithm to search for the optimal combination of parameters that satisfies the target performance.

[0028] The controller outputs the verified parameter combination to the actuator of the pulsed laser system.

[0029] The incremental learner updates the knowledge graph coupling coefficient matrix and model weights based on real-time performance data fed back from the sensors.

[0030] Preferably, the incremental learner operates as follows:

[0031] When the deviation between the actual processing accuracy and the model prediction exceeds the limit, the coupling coefficient matrix of the knowledge graph is updated first.

[0032] If the bias persists, an incremental learning algorithm is used to adjust the weights of the fully connected layers in the deep learning hybrid model.

[0033] Preferred options also include:

[0034] The fault diagnosis unit, whose input is connected to the dynamic knowledge graph construction module, locates the cross-domain coupled fault source by tracing back the entity relationship path when the processing accuracy is abnormal.

[0035] A pulsed laser control method based on a pulsed laser control system fusion of dynamic knowledge graph and deep learning includes the following steps:

[0036] Step a: Collect time-domain, spatial-domain, and frequency-domain parameters in real time using sensors;

[0037] Step b: Dynamically update the entity relationships and coupling coefficient matrix of the knowledge graph;

[0038] Step c: Predict the system performance of parameter combinations using a deep learning hybrid model;

[0039] Step d: After double verification, output the optimal parameters to the pulsed laser system;

[0040] Step e: Update the knowledge graph and model weights based on real-time performance feedback.

[0041] This invention provides a pulsed laser control system that integrates dynamic knowledge graphs and deep learning. It offers the following advantages:

[0042] This pulsed laser control system, which integrates a dynamic knowledge graph with deep learning, achieves precise analysis and real-time optimization of the coupling relationships between multi-domain parameters of pulsed lasers through deep collaboration between the dynamic knowledge graph and the deep learning model. By employing cross-domain coupling quantization technology based on the dynamic knowledge graph, the interactive influence of time-domain, spatial-domain, and frequency-domain parameters is transformed into an online-updable coupling coefficient matrix, solving the problem of traditional methods struggling to dynamically analyze strong parameter coupling. Through a hybrid model's condition-triggered computational architecture, dynamically activating submodules based on a heterogeneous LSTM-CNN-GNN network, the system significantly reduces computational load while maintaining prediction accuracy. Based on a dual-loop verification and incremental learning system, physical conflict parameters are intercepted through knowledge graph rule verification and adversarial robustness testing, combined with an error-driven hierarchical update strategy to ensure continuous system optimization. Attached Figure Description

[0043] Figure 1 This is a data flow diagram between modules of the pulsed laser control system of the present invention;

[0044] Figure 2 This is a schematic diagram illustrating the construction of a cross-entity relationship tracing tree in this invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 and Figure 2 This invention provides a technical solution: a pulsed laser control system that integrates dynamic knowledge graphs and deep learning, comprising the following sequentially connected components:

[0047] The module includes a dynamic knowledge graph construction module, a deep learning hybrid model module, and a dual-closed-loop collaborative optimization module.

[0048] The input of the dynamic knowledge graph construction module is connected to the time-domain sensor, spatial-domain sensor and frequency-domain sensor of the pulsed laser system, which is used to collect pulse width, repetition frequency, spot size, working distance and laser wavelength parameters in real time.

[0049] The input of the deep learning hybrid model module is connected to the output of the dynamic knowledge graph construction module, which is used to process multi-domain parameters and output predicted values ​​of processing accuracy, ablation depth and surface quality.

[0050] The input of the dual-loop collaborative optimization module is connected to the output of the deep learning hybrid model module, its control end is connected to the actuator of the pulsed laser system, and its feedback end is connected to the output of the sensor.

[0051] It should be further explained that, during the specific implementation process, upon system startup, the pulse width and repetition frequency are collected in real time by a time-domain sensor, the spot size and working distance are collected by a spatial-domain sensor, and the laser wavelength parameters are collected by a frequency-domain sensor. The time-domain sensor can be a high-speed photodetector, and the spatial-domain sensor can be a spot analyzer. After receiving the above parameters, the dynamic knowledge graph construction module performs the following operations:

[0052] S1: Dynamic binding of entity relationships: bind pulse width and repetition frequency as time-domain entities, spot size and working distance as spatial entities, and laser wavelength as frequency-domain entities;

[0053] Establish a cross-domain coupling relationship library. For example, if an increase in pulse width and a decrease in spot size are detected, an energy density anomaly warning is triggered; if the laser wavelength deviates from the material absorption peak range, a material suitability conflict is marked.

[0054] S2: Deep learning hybrid model collaborative processing: The time series processing unit analyzes the pulse width sequence through the LSTM network. If the fluctuation of the continuous pulse interval exceeds the preset tolerance, the pulse cluster effect compensation submodule is activated.

[0055] The spatial processing unit identifies light spot deformation features through CNN. If the working distance exceeds the depth of field range, causing the light spot ellipticity to exceed the standard, the spatial transformation layer is called to correct the energy distribution.

[0056] The frequency domain processing unit inputs the feature vectors of wavelength nodes in the knowledge graph into the GNN. When the wavelength is in the high-risk range of the material's reflection spectrum, a frequency domain risk identifier is generated.

[0057] S3: Dual-loop optimization and verification execution: After the parameter optimizer generates candidate parameter combinations, it enters a dual verification channel, including the following channels:

[0058] Rule verification channel: If the wavelength in the combination is equal to 1064nm and the processing object is a polymer material, and the "Metal Processing Applicability" rule of the knowledge graph is violated, it will be discarded directly;

[0059] Robustness test channel: Add ±5% perturbation to the neighborhood of candidate parameters to generate test samples. If the fluctuation range of the surface quality prediction value exceeds the safety threshold, model fine-tuning will be initiated.

[0060] The verified parameters are output to the laser system actuator, while the photodetector monitors the actual ablation depth in real time. If the deviation between the actual value and the predicted value exceeds the limit for three consecutive times, the pulse width-spot size coupling coefficient in the knowledge graph is updated first. If the deviation still does not converge after the update, the incremental learner is triggered to adjust the weights of the LSTM-CNN-GNN fusion layer.

[0061] S4: Fault tracing linkage mechanism: When the machining accuracy is abnormal, the fault diagnosis unit traces back along the knowledge graph relationship edge: If the surface quality decline is associated with the spatial domain node, and the confidence of the frequency domain wavelength node is less than 0.9 at the same time, it is determined to be a cross-domain coupling mismatch fault; if the energy density is abnormal, it is only associated with the time domain-spatial domain edge, and it is located as a single-domain parameter actuator drift fault.

[0062] The dynamic knowledge graph construction module includes:

[0063] The entity relationship definition unit associates the time domain parameters with the pulse width and repetition frequency entity, the spatial domain parameters with the spot size and working distance entity, and the frequency domain parameters with the laser wavelength entity.

[0064] A cross-domain coupling strength quantization unit generates dynamic coupling coefficient matrices between the time-space domain, the time-frequency domain, and the spatial-frequency domain.

[0065] The real-time update unit dynamically adjusts the relationships between entity nodes and the coupling coefficient matrix based on sensor data.

[0066] It should be further explained that, in the specific implementation process, after the sensor data is input into the dynamic knowledge graph construction module, the entity relationship definition unit first performs parameter entityization processing: if the variance of the pulse width sequence is detected to exceed the time domain stability threshold, the current pulse width entity is marked as a high fluctuation state, and its cooperative constraint relationship with the repetition frequency entity is activated; when the spot ellipticity returned by the spot analyzer is greater than the deformation tolerance, the spot size entity is automatically associated with the spatial distortion correction category, and its topological association weight with the working distance entity is strengthened.

[0067] If the laser wavelength falls into the high-risk range of the material's reflection spectrum, such as the high reflection band of polymer materials around 1064nm, it will trigger the absorption risk indicator of the frequency domain entity and be associated with the material's applicability performance node.

[0068] The cross-domain coupling strength quantization unit synchronously initiates a dynamic coupling mechanism: In the initialization phase, pulse transmission characteristics are derived based on the nonlinear Schrödinger equation to generate a time-space basic coupling matrix. For example, when the working distance increases, causing the spot size to expand, the energy density attenuation compensation coefficient is automatically calculated; if the pulse width shortens and the wavelength blue shifts, a frequency domain dispersion compensation factor is generated; In the operation phase, the matrix is ​​corrected in real time by training with historical data: when the actual ablation depth deviates from the predicted value three times consecutively and the error increases in the same direction, the enhanced training of the pulse width-ablation depth coupling weight is initiated; if the surface quality anomaly only occurs under specific wavelength conditions, the confidence score of the wavelength-surface quality relationship edge is updated accordingly.

[0069] The real-time update unit dynamically adjusts the map based on sensor stream data, including the following modes:

[0070] Regular update mode: After each processing task is completed, the entity node attributes are refreshed using the average parameter value;

[0071] Abnormal triggering mode: If the processing accuracy suddenly drops beyond the safety threshold, immediately backtrack the last 10 sets of parameter combinations: when the change in spot size is detected as the main cause, reset the weight of the relationship between spot size and processing accuracy; if wavelength shift also exists, then synchronously correct the time-frequency cross-coupling matrix.

[0072] When the signal input is changed to a different material: the original time-space coupling relationship is preserved, and the mapping relationship library between the wavelength entity and the absorption spectrum of the new material is reconstructed.

[0073] Accurate relationship binding is achieved through a dynamic classification mechanism of entity states with high volatility, distortion, and risk indicators; the limitations of empirical formulas are overcome through a dual-source evolution strategy of coupling matrix with physical basis and data-driven correction; and efficiency and real-time requirements are balanced through a three-level update response system of regular, abnormal, and emergency updates, which is significantly different from the construction method of static knowledge graphs.

[0074] The cross-domain coupling strength quantization unit operates in the following manner:

[0075] Initialize the coupling coefficient matrix based on the pulsed laser transmission equation;

[0076] The coupling coefficient matrix is ​​trained and updated using historical data collected by photodetectors and spot analyzers.

[0077] When an abnormality in machining accuracy is detected, the coupling coefficient matrix is ​​corrected in real time.

[0078] It should be further explained that, in the specific implementation process, during the system initialization phase, the cross-domain coupling strength quantization unit extracts the fundamental physical constraints from the pulsed laser transmission equation:

[0079] When the increase in working distance is detected to cause the spot size to expand, an energy density attenuation compensation coefficient is automatically generated based on the beam propagation model. This coefficient is non-linearly positively correlated with the distance increment.

[0080] If the pulse width is shortened to the femtosecond level and accompanied by a blue shift in wavelength, the calculation of the frequency domain dispersion compensation factor is triggered. The compensation factor value is determined by the wavelength shift and the material group velocity dispersion characteristics.

[0081] For specific materials, such as ceramics, when the repetition frequency exceeds the thermal diffusion critical value, a thermal accumulation inhibition coefficient is generated and bound to the time-domain-surface quality relation edge.

[0082] During system operation, the coupling coefficient matrix is ​​dynamically corrected through real-time data streams, as follows:

[0083] Continuous error triggering mechanism: If the actual ablation depth is greater than the predicted value three times in a row and the error increases with each time, it is determined that the pulse width-ablation depth coupling weight is too weak, and the reinforcement training of the weight is initiated: collect nearly 50 sets of high energy density operating condition data; refit the coupling curve under the premise of maintaining wavelength-material constraints;

[0084] If surface quality anomalies only occur when processing polymers at a wavelength of 1064nm, then the wavelength-surface quality relationship edge is locked as the correction target: the initial confidence level of this relationship edge is reduced to below 0.7; and a polymer material reflectivity compensation parameter is added.

[0085] Multi-condition adaptive strategy: In the micro-hole processing scenario, when the hole diameter-depth-width ratio exceeds 5:1, the pulse width-ablation depth coupling weight is strengthened, and the influence of the working distance-processing efficiency relationship is weakened; if the system switches to the brittle material etching mode, the thermal accumulation suppression coefficient is disabled and the wavelength-thermal stress coupling channel is activated.

[0086] Emergency response under abnormal operating conditions includes: when irreversible ablation damage occurs at the processing location: backtrack the spatial parameters in the last 5 parameter combinations; if the spot size fluctuation exceeds the safety tolerance and coincides with the damage location in time and space, reset the spot size-processing accuracy relationship edge to the initial weight and trigger the spatial sensor recalibration; if the damage is accompanied by a sudden wavelength jump, such as a ±10nm jump, freeze the time-frequency cross-coupling matrix and initiate the frequency stabilization device intervention process.

[0087] The basic reliability is ensured by prioritizing physical constraints and data corrections. Precise control is achieved by adaptive coupling channel switching in processing scenarios with differentiated treatments for deep pores, brittle materials, and polymers. An anomaly hierarchical response mechanism from parameter training to equipment intervention constructs a fault defense system, solving the static nature and scenario blind spots of cross-domain coupling analysis in traditional methods.

[0088] The deep learning hybrid model module contains parallel connections:

[0089] The temporal processing unit employs a long short-term memory network with temporal convolutional kernels to process sequence features such as pulse width and repetition frequency.

[0090] The spatial processing unit uses a convolutional neural network with a spatial transformation layer to process the spatial distribution characteristics of spot size and working distance.

[0091] The frequency domain processing unit employs a graph neural network, embedding the feature vectors of laser wavelength nodes in the knowledge graph.

[0092] It should be further explained that, in the specific implementation process, after the sensor data is input into the deep learning hybrid model module, the timing processing unit first analyzes the pulse width sequence: when the coefficient of variation of the continuous pulse interval is detected to exceed the dynamic stability threshold, the temporal convolution kernel is automatically activated to capture the pulse cluster effect. If the width range within the same pulse cluster exceeds the preset tolerance, the pulse energy equalization compensation submodule is triggered. For repetitive frequency change conditions, such as switching from 1kHz to 100kHz, the LSTM gated memory reset function is enabled to clear the historical state to avoid frequency aliasing error.

[0093] The spatial processing unit synchronously processes spot features: if the ellipticity index returned by the spot analyzer exceeds the deformation threshold of 0.25, the spatial transformation layer is activated for real-time correction: when the working distance exceeds the depth of field range, resulting in a ring-shaped spot, an ellipse-to-circle transformation matrix is ​​generated; if the position of the reading head shifts, causing the spot center to drift, a translation compensation vector is calculated; in the processing of microstructures, when the feature size is smaller than the spot diameter, a multi-scale convolutional kernel group is activated to synchronously extract local energy distribution and global morphological features.

[0094] The frequency domain processing unit performs deep coupling analysis: the feature vectors of wavelength nodes in the knowledge graph are input into the GNN. When the wavelength falls into the high-risk range of material reflection, such as 1064nm processed polycarbonate, an absorption risk identifier is generated and passed to the output layer of the hybrid model. If the material is marked as a high reflectivity category in the knowledge graph, an additional reflection compensation weight factor is added.

[0095] When the system switches to a new wavelength, such as from 1064nm to 532nm: freeze the historical characteristics of the original wavelength nodes; reconstruct the graph relationship between the frequency domain entity and the absorption spectrum of the new material.

[0096] The feature fusion stage employs a hierarchical stitching strategy, including the following:

[0097] Primary fusion: The pulse cluster feature vector output by LSTM is concatenated with the spot shape vector extracted by CNN along the channel dimension;

[0098] Cross-domain enhancement: When a high-frequency pulse and a small light spot are detected simultaneously, the spatiotemporal feature cross-attention layer is activated;

[0099] Frequency domain intervention: If the confidence level of the absorption risk identifier is >0.9, then apply reflection compensation scaling to the fused features.

[0100] It achieves efficient processing through a dynamic allocation mechanism of computational resources, overcomes the problem of micromachining precision through spatiotemporal feature cross-enhancement technology, and prevents the risk of processing failure through material reflectivity feedforward compensation, effectively distinguishing it from neural network models with fixed structures.

[0101] The output of the deep learning hybrid model module is connected to a dual verification unit, which includes:

[0102] A knowledge graph rule validator verifies whether parameter combinations violate entity relationship constraints.

[0103] To combat robustness testers, perturbation samples are generated in the neighborhood of optimized parameters and performance fluctuations are detected.

[0104] It should be further explained that, in the specific implementation process, when the deep learning hybrid model outputs candidate parameter combinations, the knowledge graph rule validator first performs a physical compliance review: if it detects that the wavelength parameter is at 1064nm and the knowledge graph node of the processing object is marked as "polymer material", then the material absorption conflict rule is triggered, the combination is automatically discarded and a wavelength tuning suggestion is generated;

[0105] When the pulse width is less than 100 fs and the working distance exceeds the depth of field limit, the beam transmission stability rule is activated, and it is determined to be a combination of spatial-temporal coupling failure. For micro-hole processing scenarios, if the spot diameter is larger than the target aperture and the multi-beam strategy is not enabled, the spatial resolution constraint rule is violated, and the parameter flow to the execution end is forcibly terminated.

[0106] The parameters that pass the rule validation are fed into the adversarial robustness tester to perform stability evaluation, including:

[0107] Perturbation sample generation stage: Generate ±5% random perturbations in the neighborhood of candidate parameters. If the pulse width in the original combination is in the femtosecond range, then shrink the perturbation range to ±2%.

[0108] When the spot size is close to the diffraction limit, disable lateral position perturbation and adjust only the energy intensity;

[0109] Fluctuation detection phase: Input the disturbance sample into the hybrid model to obtain the performance prediction value. If the range of the surface quality index exceeds the upper limit of the material roughness tolerance, it is marked as a high-sensitivity combination. When the ablation depth prediction value shows a non-monotonic change, such as fluctuation exceeding 3 oscillation cycles, it is determined that there is a hidden risk of instability.

[0110] Response decision-making phase: For highly sensitive combination start-up parameter solidification protocol: keep the core parameters unchanged, only optimize the auxiliary variables, where the core parameters include wavelength and pulse width, and the auxiliary variables include repetition frequency and working distance;

[0111] If latent instability is detected, the adversarial training mode of the hybrid model is immediately triggered: a noise-tolerant regularization term is injected into the hidden layer, and the iteration continues until the fluctuations converge.

[0112] The anomaly handling linkage mechanism includes: when a single batch of parameters fails rule verification three times in a row: backtrack the confidence of the relevant entity relationship edges in the knowledge graph; if the confidence of the "wavelength-material suitability" edge is less than 0.8, start the frequency domain entity library reconstruction process;

[0113] When adversarial testing reveals systemic sensitivity: freeze the current optimizer output channel and activate prior knowledge to guide optimization: match similar operating condition parameter templates from historical successful cases.

[0114] By combining physical rules with random disturbances to eliminate the risk of parameter failure, the specific protection of ultra-precision working conditions ensures the reliability of extreme scenarios, and the multi-level emergency degradation mechanism ensures the continuous availability of the system, thus solving the problem of processing accidents caused by the lack of verification in the optimization scheme.

[0115] Dual verification unit execution:

[0116] If a parameter combination is deemed invalid by the knowledge graph rule validator, the combination is discarded.

[0117] If the adversarial robustness tester detects performance fluctuations exceeding a threshold, it triggers model weight fine-tuning.

[0118] It should be further explained that, in the specific implementation process, when candidate parameter combinations enter the dual verification unit, the system first performs a deep review by the knowledge graph rule verifier: if the wavelength parameter is detected to be at 1064nm and the processing object is marked as "polymer material" by the knowledge graph, the material absorption conflict rule is immediately triggered, the combination is discarded and the wavelength tuning suggestion range is fed back to the optimizer; when the pulse width is lower than the critical femtosecond value and the working distance exceeds the depth of field limit of the optical system, the beam transmission instability rule is activated, the combination is determined to have a spatial-temporal coupling failure risk and an alarm code is generated;

[0119] For combinations where the spot diameter is larger than the current processing feature size, if no multi-beam strategy node is associated, the spatial resolution constraint rule is violated, the execution link is forcibly interrupted, and the self-test process of the microscopic optics extension module is activated.

[0120] The parameters that pass the rule validation are fed into the adversarial robustness tester to perform multi-level stability evaluation, including the following stages:

[0121] Intelligent perturbation injection stage: Automatically compress the perturbation amplitude under femtosecond pulse conditions to avoid distortion of ultrashort pulse characteristics; lock the spatial coordinate perturbation channel when the light spot approaches the diffraction limit, and only open the energy intensity adjustment dimension; disable random perturbation of heat-sensitive parameters, including repetition frequency, in brittle material processing scenarios;

[0122] Dynamic fluctuation monitoring phase: Real-time tracking of the range change of the predicted surface quality value in the disturbed sample; if it exceeds the material roughness tolerance boundary, it is marked as a high-sensitivity combination; when the ablation depth prediction curve shows non-monotonic oscillation characteristics, the hidden instability analysis algorithm is activated to identify potential risk points.

[0123] Hierarchical decision execution phase: For highly sensitive combinations, a parameter solidification protocol is enabled: Keeping the core parameters of wavelength and pulse width unchanged, the repetition frequency and working distance are locally optimized through evolutionary algorithms; if oscillatory instability is detected, noise-tolerant regularization constraints are immediately injected into the fully connected layer of the hybrid model, and adversarial training is initiated until the output fluctuations converge to within the safe threshold.

[0124] Anomaly handling employs a multi-level linkage response, including: when consecutive rule verification failures occur, the confidence scores of the related edges in the knowledge graph are traced back in reverse; when the score of the "wavelength-material applicability" edge continues to be lower than the dynamic threshold, the emergency reconstruction process of the frequency domain entity library is triggered.

[0125] When adversarial testing reveals systemic sensitivity, the real-time optimizer output channel is paused, the historical case matching engine is activated, and the optimal parameter template is retrieved based on the processing material and precision requirements to implement degradation control.

[0126] Precise protection is achieved through a strong binding mechanism between physical rules and operating conditions; multi-channel disturbance constraint technology prevents failure in extreme scenarios; and an instability risk classification and handling system ensures continuous and reliable operation, fundamentally breaking through the limitations of single threshold judgment.

[0127] The dual-closed-loop collaborative optimization module includes:

[0128] The parameter optimizer uses an evolutionary algorithm to search for the optimal combination of parameters that satisfies the target performance.

[0129] The controller outputs the verified parameter combination to the actuator of the pulsed laser system.

[0130] The incremental learner updates the knowledge graph coupling coefficient matrix and model weights based on real-time performance data fed back from the sensors.

[0131] It should be further explained that, in the specific implementation process, after the parameter combination that has passed dual verification is input into the dual closed-loop collaborative optimization module, the parameter optimizer starts the intelligent search process: In the deep hole machining scenario, if the target ablation depth is detected to exceed the material thermal diffusion critical value, parameter sampling is preferentially performed in the low repetition frequency region to avoid thermal accumulation effect causing hole wall carbonization; for brittle material etching tasks, when the surface quality requirement reaches the submicron level, the pulse width search range is automatically constrained to the picosecond level or above to suppress the risk of microcracks caused by stress waves; if the current material is marked as a high reflectivity category in the knowledge graph, a wavelength-reflectivity compensation weight factor is injected into the evolutionary algorithm to guide the population to evolve towards the near-infrared band.

[0132] When the execution controller outputs optimized parameters to the laser system, it simultaneously starts monitoring: the actual ablation depth is captured in real time by a high-speed photodetector. When there is a deviation in the same direction for three consecutive processing cycles and the deviation increases: if the deviation source is associated with the time domain parameters, the repetition frequency is dynamically adjusted to compensate for the energy gap; when the spatial sensor reports uneven spot energy distribution, the working distance fine-tuning mechanism is immediately intervened to refocus.

[0133] In femtosecond-level micromachining, if the actual surface roughness suddenly increases but the model prediction value remains stable, the current parameter execution channel is frozen, and an online spectrometer is activated to scan the material phase transformation characteristics to verify whether amorphization damage has occurred.

[0134] The incremental learner initiates hierarchical updates based on the monitored error, including the following responses:

[0135] Primary Response: When the actual ablation depth deviates from the predicted value three times consecutively and the absolute value of the error exceeds the material single-pulse removal threshold, online training of the coupling coefficient of the pulse width-ablation depth relationship edge in the knowledge graph is triggered first, and the fitting curve of the most recent 50 sets of data is updated using a sliding window; if the error reduction in the next cycle after the update is less than 50%, it is marked as a risk of model failure.

[0136] Advanced Response: For scenarios marked as model failure, extract the current sensor data stream to generate an incremental dataset, keep the CNN convolutional kernels and LSTM gate weights unchanged, and only fine-tune the cross parameters of the fully connected layers. When high-frequency parameters switch scenarios, enable the historical state protection mechanism to prevent feature drift.

[0137] By strongly linking material properties with optimized search, precise avoidance is achieved; multi-dimensional diagnosis during execution blocks hidden failures; and intelligent sorting through incremental updates ensures learning efficiency, thus solving the technical defect of the separation between optimization, execution, and learning.

[0138] The incremental learner operates as follows:

[0139] When the deviation between the actual processing accuracy and the model prediction exceeds the limit, the coupling coefficient matrix of the knowledge graph is updated first.

[0140] If the bias persists, an incremental learning algorithm is used to adjust the weights of the fully connected layers in the deep learning hybrid model.

[0141] It should be further explained that, in the specific implementation process, when the actual processing accuracy fed back by the sensor deviates from the model prediction value, the system activates the incremental learner's hierarchical response protocol, which includes the following two stages:

[0142] Initial response phase: If the ablation depth deviates in the same direction for three consecutive processing cycles and the absolute value of the deviation exceeds the material single-pulse removal threshold, the dynamic update of the pulse width-ablation depth relationship edge in the knowledge graph is triggered first: collect the most recent 50 sets of high energy density operating condition data, and refit the coupling curve while maintaining the wavelength-material constraint; when the spot size fluctuation is detected as the main cause, the weight coefficient of the spot size-energy density relationship edge is simultaneously corrected; if the deviation reduction in the next processing cycle does not reach the expected improvement level, it is marked as a model failure risk condition.

[0143] Advanced Response Phase: For conditions marked as model failure, initiate neural network fine-tuning: keep the CNN convolutional kernels and LSTM gate weights constant, and only open the cross parameters of the fully connected layers for incremental learning; use a few-sample sliding window mechanism to extract the current sensor data stream to generate an incremental training set;

[0144] When encountering high-frequency parameter switching scenarios, historical state snapshot protection is enabled to save the average value of pulse cluster characteristics before the switching as a benchmark reference; if the pulse width distribution deviation under the new frequency exceeds the stability tolerance, the snapshot data is injected to constrain the weight update direction.

[0145] Emergency avoidance mechanism: When a systematic instability is detected during the incremental learning process, if the predicted surface roughness value fluctuates in the opposite direction twice in a row, such as first rising, then falling, and then rising again, the learning process will be immediately suspended.

[0146] Activate cross-domain coupling backtracking diagnosis: Check whether the time-space cross matrix in the knowledge graph is outdated; if the spot analyzer shows ring distortion but the graph is not marked, reset the spatial entity association rules; after confirming that the fault is resolved, continue incremental learning from the breakpoint and compress the learning rate to a safe level.

[0147] Accurate response is achieved through error feature-driven update priority decision-making, and state anchoring technology for high-frequency switching scenarios ensures system stability; the incremental learning full-process monitoring mechanism eliminates the risk of learning out of control; and it overcomes the problems of "blind full-weight update" and "lack of scenario adaptability" in incremental learning.

[0148] It also includes a fault diagnosis unit, whose input is connected to the dynamic knowledge graph construction module. When machining accuracy is abnormal, it locates the cross-domain coupled fault source by tracing back the entity relationship path. It should be further noted that in the specific implementation process, when abnormal machining accuracy triggers the fault diagnosis unit, the system starts a multi-level tracing engine to accurately locate the fault source, including the following:

[0149] Primary diagnostic layer: If the surface roughness suddenly increases and the spatial sensor detects that the ellipticity of the light spot exceeds the standard, immediately trace back the association path of the light spot size node in the knowledge graph: when there is a recent jump record in the working distance, it is determined to be a light spot distortion fault caused by defocusing, and the automatic focusing mechanism is triggered to intervene; if the confidence of the reading head position is continuously lower than 0.8, it is located as a mechanical displacement deviation fault, and the optical calibration protocol is activated.

[0150] Intermediate diagnostic layer: When ablation depth is abnormal and accompanied by frequency domain wavelength shift: Search along the knowledge graph for wavelength-material absorption relationship edges. If the current wavelength is in the material reflection peak range, generate an insufficient absorption warning and suggest switching to the ultraviolet band; Simultaneously check the time-domain-frequency domain cross-coupling matrix. If the pulse width is compressed to the femtosecond level but dispersion compensation is not activated, mark it as a group velocity dispersion mismatch fault and inject compensation parameters into the actuator.

[0151] Advanced diagnostic layer: For multi-domain coupling anomalies, such as a sudden drop in processing efficiency accompanied by an expansion of the heat-affected zone: construct a cross-entity relationship tracing tree and prioritize the detection of high-frequency associated paths; if the end node of the tracing tree contains "material thermal diffusivity exceeds the threshold", then activate the cooling system enhancement command.

[0152] A dynamic confidence adjustment mechanism is implemented throughout the entire diagnostic process: when a certain relation edge fails to detect the real fault three times in a row, its confidence score is automatically reduced; if the score of the "pulse width-processing efficiency" edge falls below 0.7, data-driven retraining of that edge is triggered; diagnostic paths are downweighted for edges with scores below 0.5 to avoid misleading subsequent analysis; after successfully locating the fault, the confidence of the relevant relation edge is increased: for example, when defocus distortion is correctly identified, the score of the "working distance-spot deformation" edge increases by 0.15; if the cross-domain coupling diagnosis takes less time than the preset standard, an additional 0.1 score bonus is added.

[0153] Material compatibility verification serves as the ultimate guarantee: when the fault cannot be resolved by parameter adjustment, extract the current material surface micro-area spectral data and compare it with the standard absorption spectrum characteristics in the knowledge graph: if an unknown absorption peak is detected with a surge in reflectance, create a new material sub-graph; activate the emergency processing parameter template: for metallic impurities, activate the high-energy pulse burst mode, and for organic pollutants, switch to low-thermal-effect long pulse.

[0154] Accurate fault location is achieved through a three-level penetrating diagnostic architecture. The dynamic evolution system of confidence, which includes misdiagnosis attenuation, success reward, and low score reduction, ensures continuous optimization of the map. The emergency response to unknown materials solves the problem of sudden processing failure and addresses the shortcomings of single-level analysis blind spots and static knowledge base rigidity.

[0155] A pulsed laser control method based on a pulsed laser control system fusion of dynamic knowledge graph and deep learning includes the following steps:

[0156] Step a: Collect time-domain, spatial-domain, and frequency-domain parameters in real time using sensors;

[0157] Step b: Dynamically update the entity relationships and coupling coefficient matrix of the knowledge graph;

[0158] Step c: Predict the system performance of parameter combinations using a deep learning hybrid model;

[0159] Step d: After double verification, output the optimal parameters to the pulsed laser system;

[0160] Step e: Update the knowledge graph and model weights based on real-time performance feedback.

[0161] It should be further explained that, in the specific implementation process, after the system is started, the time-domain sensor captures the pulse width sequence in real time. When the coefficient of variation of the continuous pulse interval is detected to exceed the dynamic stability threshold, the time convolution kernel is automatically activated to capture the pulse cluster effect. The spatial-domain sensor synchronously collects the spot distribution data. If the ellipticity index exceeds the deformation tolerance, the spatial transformation layer is triggered to correct the energy distribution. The wavelength parameters obtained by the frequency-domain sensor are input into the knowledge graph. When the wavelength falls into the high-risk range of material reflection, an absorption risk identifier is generated.

[0162] The dynamic knowledge graph construction module receives multi-domain parameters and performs intelligent updates: if the variance of the pulse width sequence continues to exceed the standard, the time-domain entity is marked as a high-fluctuation state and its cooperative constraint with the repetition frequency is strengthened; when the working distance jump causes spot distortion, the correlation weight between the spot size node and the processing accuracy is corrected; when a material replacement signal is detected, the mapping relationship library between wavelength entities and the absorption spectrum of the new material is reconstructed.

[0163] When processing fused features, the deep learning hybrid model implements dynamic computation optimization: the CNN spatial transformation layer is activated only when the ellipticity of the light spot is greater than 0.25 to avoid consuming resources for full-time computation; in the scenario where high-frequency pulses and micro-spots coexist, the spatiotemporal feature cross-attention layer is activated to strengthen the weight of key energy density regions; when the confidence of the frequency domain risk identifier is greater than 0.9, reflection compensation scaling is applied to the output prediction value.

[0164] Parameter combinations undergo rigorous screening via a dual verification channel: combinations violating wavelength-material suitability rules are immediately blocked; parameters that pass the rule verification undergo intelligent perturbation testing, namely: perturbation amplitude is compressed to a safe range in femtosecond pulse conditions, and repetition frequency perturbation is disabled in brittle material scenarios. The optimal parameters that pass verification are output to the laser system actuator.

[0165] The execution process is monitored in multiple dimensions: when the actual ablation depth deviates from the predicted value in the same direction for three consecutive cycles, the repetition frequency is dynamically adjusted to compensate for the energy gap; when the surface roughness suddenly increases but the model prediction remains stable, online spectral analysis is initiated to verify the material phase transformation damage. Real-time errors trigger incremental learning hierarchical responses: the primary response updates the knowledge graph coupling coefficient; the advanced response fine-tunes the weights of the fully connected layers of the model; and historical state snapshots are enabled during high-frequency parameter switching to prevent feature drift.

[0166] When machining accuracy is abnormal, a multi-level fault tracing is initiated: spot distortion accompanied by a jump in working distance is identified as a defocusing fault; when ablation anomalies coexist with wavelength shifts, material absorption spectrum conflicts are detected; for systemic failure scenarios, a cross-entity tracing tree is constructed to locate the root cause of thermal accumulation. Each successful diagnosis increases the confidence of the relevant edge, while continuous misdiagnosis triggers data-driven retraining. In the final material verification stage, micro-area spectral characteristics are compared, and after identifying unknown impurities, an emergency template for high-energy blasting or cryogenic removal is switched.

[0167] System startup and real-time sensing: The time-domain sensor continuously monitors changes in the pulse width sequence. When the fluctuation amplitude of the continuous pulse interval exceeds the dynamic stability set limit, the system automatically activates the time convolution kernel function to capture pulse cluster effects. The spatial-domain sensor synchronously acquires spot energy distribution data. If the analysis results show that the spot ellipticity exceeds the preset deformation tolerance, the spatial transformation layer is immediately triggered to correct the energy distribution in real time. After the laser wavelength parameters acquired by the frequency-domain sensor are input into the knowledge graph, the system automatically compares them with the material absorption characteristic database. When it is identified that the current wavelength is in the high reflectivity danger range of the material, a frequency-domain absorption risk identifier is generated.

[0168] Dynamic knowledge graph construction: The dynamic knowledge graph construction module receives multi-domain sensor data and performs intelligent updates: If the fluctuation variance of the pulse width sequence continues to exceed the time domain stability threshold, the system marks the time domain entity as a high fluctuation state and strengthens its cooperative constraint relationship with the repetition frequency entity; when the working distance parameter changes abruptly and causes spot distortion, the system automatically corrects the correlation weight coefficient between the spot size node and the processing accuracy performance; when a material replacement command signal is received, the system immediately reconstructs the wavelength entity and the absorption spectrum mapping relationship library of the new material, retaining the original entity topology.

[0169] Deep learning hybrid model optimization: The hybrid model implements dynamic resource allocation when processing features: the spatial transformation layer of the convolutional neural network is activated only when the spot ellipticity exceeds the deformation threshold to avoid the resource consumption of full-time operation; when high-frequency pulse parameters and small spot size parameters coexist, the temporal and spatial feature cross-attention layer is automatically activated to enhance the weight focus on key energy density regions; if the confidence score of the frequency domain absorption risk identifier exceeds the dynamic threshold, the system applies a reflection compensation scaling factor to the output prediction value.

[0170] A dual parameter verification mechanism is implemented: Candidate parameter combinations are first verified using knowledge graph rules. Combinations that violate wavelength-appropriate material suitability rules, such as processing polymer materials at near-infrared wavelengths, are immediately intercepted, and tuning suggestions are provided. Parameters that pass verification proceed to the robustness test phase, where the system intelligently adjusts the perturbation strategy based on the operating conditions: under femtosecond pulse conditions, the parameter perturbation amplitude is compressed to a safe range; in brittle material processing scenarios, random perturbations to heat-sensitive parameters are prohibited. The optimal parameters that pass verification are output to the laser actuator.

[0171] Multi-dimensional monitoring during execution: In actual processing, if the ablation depth deviates from the predicted value in the same direction for three consecutive processing cycles, the system dynamically adjusts the repetition frequency parameter to compensate for the energy gap; when the measured surface roughness suddenly deteriorates but the model prediction remains stable, the online spectral analysis module is immediately activated to scan the phase transformation characteristics of the material surface to verify whether amorphization damage has occurred. Real-time error data triggers an incremental learning hierarchical response mechanism: The primary response prioritizes updating the coupling coefficients of key relationship edges in the knowledge graph, such as the correlation function between pulse width and ablation depth; the advanced response selectively fine-tunes the weight parameters of the fully connected layer of the model, and in high-frequency parameter switching scenarios, historical state snapshot anchoring technology is enabled to prevent feature drift.

[0172] Multi-level fault diagnosis and tracing: When machining accuracy is abnormal, a hierarchical diagnostic engine is activated: when spot distortion is accompanied by a jump in working distance, the system determines it as an optical defocusing fault and triggers the autofocus mechanism; if the ablation depth is abnormal and wavelength shift is detected simultaneously, conflicting nodes in the material absorption spectrum are retrieved from the knowledge graph; a cross-entity relationship tracing tree is constructed for systemic failure scenarios, and the root cause of heat accumulation is located by analyzing terminal nodes such as the thermal diffusivity coefficient. The confidence score of the relevant relationship edges is increased after each successful diagnosis, and a data-driven retraining mechanism is triggered if diagnoses fail consecutively. The final assurance stage identifies unknown impurity components through micro-area spectral feature comparison and switches between high-energy pulse blasting or low-temperature long-pulse removal emergency templates based on the impurity type.

[0173] Dynamic Confidence Evolution System: The confidence scores of knowledge graph relation edges are dynamically managed. When a specific relation edge fails to locate the true fault in diagnosis three consecutive times, the system automatically lowers its confidence score. If the score falls below a set threshold, the diagnostic path for that edge is downweighted to avoid misleading subsequent analyses. Successfully locating the fault significantly increases the score of the relevant edge, and additional score rewards are given when the cross-domain coupled diagnostic efficiency exceeds the baseline. Confidence changes are synchronously fed back to the frequency domain processing unit of the hybrid model, enabling the co-evolution of the knowledge base and the learning model.

[0174] Emergency Response and System Recovery: When continuous reverse fluctuations in the output predicted value are detected during incremental learning, the learning process is immediately paused and cross-domain coupling backtracking diagnosis is initiated: the timeliness of the temporal and spatial cross-matrix in the knowledge graph is checked. When the spot analyzer displays ring distortion but the graph label is not updated, the spatial entity association rules are reset. After confirming that the fault has been resolved, the learning process is resumed with a compressed learning rate. When an unknown absorption peak is found in the material compatibility verification stage, a temporary material sub-graph is created and the impurity characteristic spectral fingerprint is recorded. In subsequent processing tasks, a high-energy blasting or low-temperature removal template is automatically matched.

[0175] It should be further explained that, in the specific implementation process:

[0176] Knowledge graph update priority: After routine tasks, use sliding window mean update; when accuracy drops suddenly, backtrack multiple parameter combinations to implement cross-validation; when materials are changed, retain the basic temporal and spatial relationships to reconstruct the frequency domain mapping.

[0177] Hybrid model feature fusion: temporal features and spatial features are concatenated along the channel dimension; when high-frequency micromachining conditions are detected, the cross-attention layer is activated to allocate regional weights; the frequency domain risk factor is used as a scaling factor in the final output.

[0178] Incremental learning safety mechanisms: verify feature distribution offset before updating model weights; inject regularization constraints when oscillation characteristics are detected; and call historical state snapshots to stabilize the feature space during high-frequency switching scenarios.

[0179] Fault tracing tree construction: Starting from the abnormal performance node, the search proceeds in descending order of confidence along the relation edges; priority is given to detecting paths with recently changed parameters; thermodynamic nodes serve as the final filter for systematic failures.

[0180] A pulsed laser modulation method integrating dynamic knowledge graphs and deep learning includes the following steps:

[0181] Step S1: The time-domain sensor acquires the pulse width sequence in real time. When the fluctuation of the continuous pulse interval exceeds the dynamic stability threshold, the time convolution kernel is automatically activated to capture the pulse cluster effect. The spatial-domain sensor monitors the spot distribution in real time. If the ellipticity exceeds the deformation tolerance, the spatial transformation layer is triggered to correct the energy distribution. The frequency-domain sensor acquires the wavelength parameters and inputs them into the knowledge graph. When the material is identified as having a high reflection risk, an absorption risk identifier is generated.

[0182] Step S2: The dynamic knowledge graph construction module performs intelligent updates: when the pulse width fluctuation continues to exceed the standard, the time domain entity is marked as a high fluctuation state and the cooperative constraint with the repetition frequency is strengthened; when the working distance jump causes the spot distortion, the correlation weight between the spot size node and the processing accuracy is corrected; when the material is changed, the wavelength entity and the absorption spectrum mapping library of the new material are reconstructed.

[0183] Step S3: Deep learning hybrid model initiates dynamic computation optimization: Spatial transformation layer is activated only when the spot ellipticity exceeds the limit; Temporal and spatial feature cross-attention layer is activated in the scenario where high-frequency pulses and micro-spots coexist, enhancing the focusing of key energy density regions; When the confidence of the frequency domain risk identifier exceeds the threshold, reflection compensation scaling is applied to the output prediction value.

[0184] Step S4: Candidate parameter combinations enter the dual verification channel: First, combinations that violate the wavelength material suitability rules are intercepted; parameters that pass the verification undergo intelligent perturbation testing, namely: the perturbation amplitude is compressed in femtosecond pulse conditions, and thermally sensitive parameter perturbations are disabled in brittle material scenarios.

[0185] Step S5: The verified optimal parameters are output to the laser actuator, and multi-dimensional monitoring is initiated: when the ablation depth continuously deviates from the predicted value in the same direction, the repetition frequency is dynamically adjusted to compensate for the energy; when the surface roughness suddenly increases but the model prediction remains stable, online spectral analysis is initiated to verify the material phase transformation damage.

[0186] Step S6: Real-time error triggers incremental learning hierarchical response: primary response updates key coupling coefficients of the knowledge graph; advanced response fine-tunes the weights of the fully connected layers of the model, and enables historical state snapshots to prevent feature drift when switching high-frequency parameters.

[0187] Step S7: Initiate multi-level diagnosis when processing abnormalities occur: spot distortion accompanied by working distance jumps is identified as defocusing fault; when ablation abnormalities coexist with wavelength shifts, material absorption spectrum conflicts are detected; for systematic failures, a cross-entity tracing tree is constructed to locate the root cause of thermal accumulation. Successful diagnosis increases the confidence of relational edges, while continuous misdiagnosis triggers retraining.

[0188] Step S8: Final material verification process: Compare micro-area spectral characteristics. After identifying unknown impurities, switch to high-energy pulse explosion template for metals and use low-temperature long pulse removal template for organic pollutants.

[0189] Step S2 also includes:

[0190] High volatility state marking mechanism: When the pulse width fluctuation variance exceeds the threshold three times consecutively, the temporal entity collaborative constraint reinforcement is activated;

[0191] Spot distortion weight correction: The weight coefficient is dynamically adjusted based on the product of the working distance jump amplitude and the spot ellipticity increment;

[0192] Material mapping library reconstruction: Preserve the basic topology in the time and spatial domains, and only replace the eigenvectors of the frequency domain absorption spectrum.

[0193] Step S3 also includes:

[0194] Spatial transformation layer condition activation: Spot ellipticity is used as the sole criterion for start / stop judgment, avoiding full-time calculation;

[0195] Cross-attention dynamic focusing: When the pulse frequency > critical value and the spot diameter < feature size, energy density region weights are automatically assigned;

[0196] Reflection compensation scaling: The scaling factor is proportional to the wavelength reflectance confidence score.

[0197] Step S4 also includes:

[0198] Rule-based hard interception: Establish a wavelength material conflict rule library, and polymer materials trigger absolute interception in the near-infrared band;

[0199] Disturbance channel management: Femtosecond-level pulse activation of disturbance amplitude compression algorithm, locking the repetition frequency disturbance dimension in brittle material scenarios.

[0200] Step S5 also includes:

[0201] Same-direction deviation compensation: Step-by-step incremental compensation for repetition frequency triggered by deviation in three consecutive cycles;

[0202] Spectral verification mechanism: When the measured surface roughness exceeds the tolerance band predicted by the model, laser-induced breakdown spectral detection is initiated.

[0203] Step S6 also includes:

[0204] Incremental update priority: When the ablation depth error exceeds the material single-pulse removal threshold, the spectrum coupling coefficient is forced to be updated first.

[0205] Historical snapshot anchoring: Save the average value of pulse cluster features before parameter switching as a reference for weight update direction constraint.

[0206] Step S7 also includes:

[0207] Tracing tree construction algorithm: Using the abnormal node as the root, generate diagnostic paths in descending order of relation edge confidence;

[0208] Confidence-based dynamic reward and punishment: Successful diagnosis adds a score equal to the reciprocal of the time taken to resolve the fault, while misdiagnosis decreases exponentially with the number of consecutive occurrences.

[0209] The sensor acquisition in step S1 uses conventional photoelectric conversion technology, and the emergency template in step S8 calls the standard parameter preset library. All threshold settings comply with international laser processing safety standards.

[0210] By deeply collaborating with dynamic knowledge graphs and deep learning models, accurate analysis and real-time optimization of multi-domain parameter coupling relationships in pulsed lasers are achieved. Through cross-domain coupling quantization technology using dynamic knowledge graphs, the interactive influences of time-domain, spatial-domain, and frequency-domain parameters are transformed into an online-updable coupling coefficient matrix, solving the problem of traditional methods struggling to dynamically analyze strong parameter coupling. A condition-triggered computational architecture based on a hybrid model, dynamically activating submodules using a heterogeneous LSTM-CNN-GNN network, significantly reduces computational load while maintaining prediction accuracy. Based on a dual-loop verification and incremental learning system, physical conflict parameters are intercepted through knowledge graph rule verification and adversarial robustness testing, combined with an error-driven hierarchical update strategy to ensure continuous system optimization.

[0211] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0212] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A pulsed laser control system integrating dynamic knowledge graphs and deep learning, characterized in that, Including sequential connections: The module includes a dynamic knowledge graph construction module, a deep learning hybrid model module, and a dual-closed-loop collaborative optimization module. The input of the dynamic knowledge graph construction module is connected to the time-domain sensor, spatial-domain sensor and frequency-domain sensor of the pulsed laser system, and is used to collect pulse width, repetition frequency, spot size, working distance and laser wavelength parameters in real time. The input of the deep learning hybrid model module is connected to the output of the dynamic knowledge graph construction module, which is used to process multi-domain parameters and output predicted values ​​of processing accuracy, ablation depth and surface quality. The input end of the dual-closed-loop collaborative optimization module is connected to the output end of the deep learning hybrid model module, its control end is connected to the actuator of the pulsed laser system, and its feedback end is connected to the output end of the sensor. The dynamic knowledge graph construction module includes: The entity relationship definition unit associates the time domain parameters with the pulse width and repetition frequency entity, the spatial domain parameters with the spot size and working distance entity, and the frequency domain parameters with the laser wavelength entity. A cross-domain coupling strength quantization unit generates dynamic coupling coefficient matrices between the time-space domain, the time-frequency domain, and the spatial-frequency domain. The real-time update unit dynamically adjusts the relationships between entity nodes and the coupling coefficient matrix based on sensor data. The cross-domain coupling strength quantization unit operates in the following manner: Initialize the coupling coefficient matrix based on the pulsed laser transmission equation; The coupling coefficient matrix is ​​trained and updated using historical data collected by photodetectors and spot analyzers. When an abnormality in machining accuracy is detected, real-time correction of the coupling coefficient matrix is ​​triggered. The dual-closed-loop collaborative optimization module includes: The parameter optimizer uses an evolutionary algorithm to search for the optimal combination of parameters that satisfies the target performance. The controller outputs the verified parameter combination to the actuator of the pulsed laser system. The incremental learner updates the knowledge graph coupling coefficient matrix and model weights based on real-time performance data fed back from the sensors. The incremental learner operates as follows: When the deviation between the actual processing accuracy and the model prediction exceeds the limit, the coupling coefficient matrix of the knowledge graph is updated first. If the bias persists, an incremental learning algorithm is used to adjust the weights of the fully connected layers in the deep learning hybrid model.

2. The pulsed laser control system integrating dynamic knowledge graph and deep learning according to claim 1, characterized in that: The deep learning hybrid model module contains parallel connections: The temporal processing unit employs a long short-term memory network with temporal convolutional kernels to process sequence features such as pulse width and repetition frequency. The spatial processing unit uses a convolutional neural network with a spatial transformation layer to process the spatial distribution characteristics of spot size and working distance. The frequency domain processing unit employs a graph neural network, embedding the feature vectors of laser wavelength nodes in the knowledge graph.

3. The pulsed laser control system integrating dynamic knowledge graph and deep learning according to claim 2, characterized in that: The output of the deep learning hybrid model module is connected to a dual verification unit, which includes: A knowledge graph rule validator verifies whether parameter combinations violate entity relationship constraints. To combat robustness testers, perturbation samples are generated in the neighborhood of optimized parameters and performance fluctuations are detected.

4. The pulsed laser control system integrating dynamic knowledge graph and deep learning according to claim 3, characterized in that: The dual verification unit performs: If a parameter combination is deemed invalid by the knowledge graph rule validator, the combination is discarded. If the adversarial robustness tester detects performance fluctuations exceeding a threshold, it triggers model weight fine-tuning.

5. The pulsed laser control system integrating dynamic knowledge graph and deep learning according to claim 4, characterized in that, Also includes: The fault diagnosis unit, whose input is connected to the dynamic knowledge graph construction module, locates the cross-domain coupled fault source by tracing back the entity relationship path when the processing accuracy is abnormal.

6. A pulsed laser modulation method based on any one of the systems of claims 1-5, characterized in that, Includes the following steps: Step a: Collect time-domain, spatial-domain, and frequency-domain parameters in real time using sensors; Step b: Dynamically update the entity relationships and coupling coefficient matrix of the knowledge graph; Step c: Predict the system performance of parameter combinations using a deep learning hybrid model; Step d: After double verification, output the optimal parameters to the pulsed laser system; Step e: Update the knowledge graph and model weights based on real-time performance feedback.

Images