Lithium battery laser cutting trajectory energy following compensation method based on deep learning

Abstract

This invention relates to the field of lithium battery manufacturing technology and discloses a deep learning-based method for energy tracking compensation of lithium battery laser cutting trajectory. This method acquires target trajectory data and operating condition feature data, extracts multi-dimensional features to form a comprehensive feature vector, predicts trajectory deviation, generates a composite compensation amount, corrects control commands in real time, and collects feedback data for adaptive updates. This invention can effectively reduce electrode cutting accuracy errors, improve battery production consistency, and reduce scrap rates.

Description

Technical Field

[0001] This invention relates to the field of lithium battery manufacturing technology, and in particular to a deep learning-based method for energy tracking compensation of lithium battery laser cutting trajectory. Background Technology

[0002] As a core component in today's new energy field, the laser cutting process in the production of lithium batteries has a significant impact on battery performance, safety, and lifespan. Laser cutting technology is a key process in the processing of lithium battery electrodes, mainly used for the precise cutting and shaping of battery electrodes to meet battery design and performance requirements. Traditional laser cutting processes typically rely on fixed trajectory parameter settings, with a CNC system executing the cutting task according to a preset path.

[0003] With the diversification of lithium battery applications and the increasing performance requirements, laser cutting technology faces increasingly higher precision requirements. In actual production, factors such as the inhomogeneity of electrode materials, fluctuations in laser energy, and the dynamic response characteristics of mechanical systems often lead to deviations between the actual cutting trajectory and the preset target trajectory. These deviations affect the cutting quality of the electrode, and consequently, the overall performance of the battery.

[0004] The main problems with current lithium battery laser cutting technology include: Existing technologies struggle to effectively address trajectory errors caused by changes in material properties and dynamic equipment response, especially under high-speed cutting conditions where trajectory deviations due to system dynamics are more pronounced, failing to meet high-precision cutting requirements. Traditional compensation methods often employ static correction strategies, lacking the ability to perceive and analyze the real-time state of the process, making it difficult to cope with dynamic changes in production and resulting in poor compensation effects. Existing trajectory compensation methods generally lack adaptive learning capabilities, failing to self-adjust and update optimization strategies based on historical data and real-time feedback, hindering continuous optimization of process parameters and gradual improvement of accuracy.

[0005] With the development of intelligent manufacturing technology, applying artificial intelligence methods such as deep learning to the optimization of laser cutting trajectory compensation for lithium batteries is expected to solve the above problems, improve cutting accuracy and production efficiency, and is of great significance to promoting the progress of lithium battery manufacturing technology. Summary of the Invention

[0006] This invention provides a deep learning-based method for energy tracking compensation of lithium battery laser cutting trajectory, which can solve the problems in the prior art.

[0007] A first aspect of this invention provides a deep learning-based method for energy tracking compensation of lithium battery laser cutting trajectories, comprising: Acquire laser cutting target trajectory data and actual working condition characteristic data of lithium battery electrode sheets; Multi-dimensional feature extraction is performed on the aforementioned working condition characteristic data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system; The comprehensive feature vector is mapped to the laser cutting target trajectory data in the feature space to predictively quantify the trajectory deviation during the laser cutting process and obtain the trajectory deviation prediction result. Based on the spatial position deviation component and time delay deviation component in the trajectory deviation prediction result, a composite compensation amount including a position compensation vector and a timing adjustment parameter is generated using a multi-objective optimization method. The laser cutting control command is corrected in real time based on the composite compensation amount, and actual trajectory feedback data is collected during the cutting process; Error is calculated using the actual trajectory feedback data and the trajectory deviation prediction results. Based on the error calculation results, the parameters of the multi-dimensional feature extraction and the multi-objective optimization method are adaptively updated.

[0008] Multi-dimensional feature extraction is performed on the aforementioned operating condition characteristic data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system, including: The operating condition characteristic data are decomposed in the time domain to obtain material characteristic components characterizing the electrode material properties, energy characteristic components characterizing laser energy fluctuations, and response characteristic components characterizing the dynamic response of the mechanical system. Spatial scale transformation and texture structure extraction are performed on the material feature components to obtain material spatial features that reflect the surface morphology changes and material density distribution of the electrode. Frequency domain transformation and wave pattern identification are performed on the energy characteristic components to obtain energy frequency domain characteristics that reflect laser power stability and energy transfer efficiency; The dynamic characteristics of the response feature components are extracted to obtain the dynamic characteristics of the response reflecting mechanical inertia and response delay; The material spatial features, the energy frequency domain features, and the response dynamic features are fused to obtain the comprehensive feature vector.

[0009] The comprehensive feature vector is mapped to the laser cutting target trajectory data in a feature space to predictively quantify the trajectory deviation during the laser cutting process, resulting in a trajectory deviation prediction result, including: The comprehensive feature vector is spatially aligned with the laser cutting target trajectory data to obtain the feature-trajectory correspondence. Based on the aforementioned feature-trajectory correspondence, the temporal evolution path is determined. By extracting the evolutionary characteristics of each feature component in the comprehensive feature vector over time, a temporal evolution feature characterizing the development trend of trajectory deviation is obtained. Spatial evolution features are obtained by expanding the spatial dimension of the temporal evolution features, and trajectory spatial coordinates and trajectory direction information are extracted from the laser cutting target trajectory data to obtain the trajectory spatial coordinate system; The spatialized evolution features are matched with the trajectory space coordinate system in terms of dimension. Through multi-layer nonlinear transformation, the feature vectors in the feature space are mapped to their spatial positions in the trajectory space coordinate system to obtain the feature-position mapping result. Based on the feature-position mapping results, the deviation response intensity at each spatial location in the trajectory space is calculated to obtain the spatial distribution prediction of the trajectory deviation. Based on the spatial distribution prediction of the trajectory deviation, the deviation component is decomposed to obtain the trajectory deviation prediction result.

[0010] Based on the aforementioned feature-trajectory correspondence, the temporal evolution path is determined. By extracting the evolutionary characteristics of each feature component in the comprehensive feature vector over time, temporal evolution features characterizing the development trend of trajectory deviation are obtained, including: Based on the feature-trajectory correspondence, the correlation strength between each feature component and the trajectory deviation is extracted. Based on the correlation strength, the time series of each feature component in the comprehensive feature vector is weighted to obtain a weighted time series feature sequence. The weighted temporal feature sequence is divided into time windows and subjected to sliding processing to obtain evolutionary gradient information representing the temporal variation trend of the feature components; The evolution gradient information is time-aligned with the trajectory temporal information in the laser cutting target trajectory data to obtain the temporal evolution path; Based on the change amplitude sequence and change direction sequence of each feature component in the evolution gradient information, the evolution rate of each feature component is calculated. Based on the evolution rate, the time dimension of each feature component is extrapolated to calculate the predicted value of each feature component at future time. The predicted values ​​for future moments are compared with the feature component values ​​for the current moment to obtain the evolutionary offset of each feature component, and the temporal evolution trajectory of the feature component is determined based on the evolutionary offset. The temporal evolution characteristics are obtained by associating the time development trajectory with the time information in the laser cutting target trajectory data.

[0011] Based on the spatial position deviation component and time delay deviation component in the trajectory deviation prediction result, a composite compensation quantity including a position compensation vector and a timing adjustment parameter is generated using a multi-objective optimization method, including: Based on the spatial position deviation component and the time delay deviation component in the trajectory deviation prediction result, a set of deviation components is determined; The spatial position deviation component is spatially decomposed into a tangential deviation component along the cutting path and a normal deviation component perpendicular to the cutting path. The spatial position compensation target is then determined based on the tangential deviation component and the normal deviation component. The time delay deviation component is decomposed into laser triggering time deviation and cutting speed deviation, and a timing adjustment target is determined based on the laser triggering time deviation and the cutting speed deviation. After coupling the spatial location compensation target with the temporal adjustment target, an association constraint between spatial location deviation and time delay deviation is added to determine the coupling optimization relationship; Based on the aforementioned coupling optimization relationship, the spatial location compensation target and the temporal adjustment target are iteratively adjusted so that they simultaneously reach the convergence condition, thereby obtaining the composite compensation amount.

[0012] The time delay deviation component is decomposed into laser triggering time deviation and cutting speed deviation, and a timing adjustment target is determined based on the laser triggering time deviation and the cutting speed deviation, including: The time difference between the actual laser triggering time and the target triggering time is extracted from the time delay deviation component to obtain the laser triggering time deviation, and the adjustment amount of the laser triggering time is calculated based on the laser triggering time deviation; The speed difference between the actual cutting speed and the target cutting speed is extracted from the time delay deviation component to obtain the cutting speed deviation, and the adjustment amount of the cutting speed is calculated based on the cutting speed deviation; A correlation analysis was performed between the adjustment amount of the laser triggering time and the adjustment amount of the cutting speed. By establishing a constraint relationship between the laser triggering time adjustment and the cutting speed adjustment, the timing adjustment constraint conditions were obtained. The timing adjustment target is determined based on the timing adjustment constraints, wherein the timing adjustment target requires that the adjustment amount of the laser triggering time and the adjustment amount of the cutting speed simultaneously satisfy the timing adjustment constraints.

[0013] The laser cutting control commands are corrected in real time based on the composite compensation amount, and actual trajectory feedback data is collected during the cutting process, including: The location compensation vector and timing adjustment parameters are determined based on the composite compensation amount; The position compensation vector is superimposed on the trajectory position command component in the laser cutting control command. By correcting the coordinate values ​​in the trajectory position command component, the corrected trajectory position command is obtained. The timing adjustment parameters are integrated into the timing control command component of the laser cutting control command. By correcting the time parameters in the timing control command component, the corrected timing control command is obtained. The corrected trajectory position command is combined with the corrected timing control command to generate a real-time corrected laser cutting control command; The real-time corrected laser cutting control command is sent to the laser cutting execution device to drive the laser cutting execution device to perform the cutting operation; During the cutting process, sensors monitor the motion trajectory position and motion timing of the laser cutting device in real time to obtain the actual trajectory feedback data.

[0014] A second aspect of the present invention provides a deep learning-based energy tracking compensation system for lithium battery laser cutting trajectory, comprising: The first unit is used to acquire laser cutting target trajectory data of lithium battery electrode sheets and working condition characteristic data during the actual cutting process; The second unit is used to extract multi-dimensional features from the operating condition feature data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system. The third unit is used to perform feature space mapping between the comprehensive feature vector and the laser cutting target trajectory data, so as to predictively quantify the trajectory deviation during the laser cutting process and obtain the trajectory deviation prediction result. The fourth unit is used to generate a composite compensation amount containing a position compensation vector and a timing adjustment parameter based on the spatial position deviation component and the time delay deviation component in the trajectory deviation prediction result using a multi-objective optimization method. The fifth unit is used to perform real-time correction of the laser cutting control command based on the composite compensation amount, and to collect actual trajectory feedback data during the cutting process; The sixth unit is used to calculate the error using the actual trajectory feedback data and the trajectory deviation prediction result, and to adaptively update the parameters of the multi-dimensional feature extraction and the multi-objective optimization method based on the result of the error calculation.

[0015] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0017] This invention extracts multi-dimensional features from working condition characteristic data to form a comprehensive feature vector characterizing electrode material properties, laser energy fluctuations, and the dynamic response of the mechanical system. This achieves comprehensive perception and quantification of complex working condition factors, improving the targeting and accuracy of trajectory compensation. By employing feature space mapping technology to associate the comprehensive feature vector with target trajectory data, this invention achieves predictive quantification of trajectory deviations during laser cutting, fundamentally solving the problem of traditional methods only providing delayed compensation and improving the foresight of trajectory control. This invention innovatively decomposes trajectory deviation into spatial position deviation components and time delay deviation components, and generates a composite compensation quantity through multi-objective optimization, simultaneously considering spatial accuracy and temporal coordination, making the compensation strategy more comprehensive and effective. This invention establishes a closed-loop adaptive update mechanism based on actual trajectory feedback data, which can dynamically adjust feature extraction and optimization parameters according to actual execution results, achieving continuous optimization of the compensation algorithm, effectively responding to changes in working conditions during production, and improving robustness and adaptability. This invention integrates deep learning, multi-dimensional feature analysis, and multi-objective optimization techniques to form a complete prediction-compensation-feedback-optimization closed-loop system. Compared with traditional methods, it significantly improves the precision and consistency of lithium battery laser cutting, reduces the scrap rate, and increases production efficiency. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the energy tracking compensation method for lithium battery laser cutting trajectory based on deep learning, according to an embodiment of the present invention.

[0019] Figure 2 This is a flowchart illustrating the process of determining temporal evolution characteristics according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0021] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0022] Figure 1 This is a flowchart illustrating the energy tracking compensation method for lithium battery laser cutting trajectory based on deep learning, according to an embodiment of the present invention. Figure 1 As shown, the method includes: Acquire laser cutting target trajectory data and actual working condition characteristic data of lithium battery electrode sheets; Multi-dimensional feature extraction is performed on the aforementioned working condition characteristic data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system; The comprehensive feature vector is mapped to the laser cutting target trajectory data in the feature space to predictively quantify the trajectory deviation during the laser cutting process and obtain the trajectory deviation prediction result. Based on the spatial position deviation component and time delay deviation component in the trajectory deviation prediction result, a composite compensation amount including a position compensation vector and a timing adjustment parameter is generated using a multi-objective optimization method. The laser cutting control command is corrected in real time based on the composite compensation amount, and actual trajectory feedback data is collected during the cutting process; Error is calculated using the actual trajectory feedback data and the trajectory deviation prediction results. Based on the error calculation results, the parameters of the multi-dimensional feature extraction and the multi-objective optimization method are adaptively updated.

[0023] In one optional implementation, multi-dimensional feature extraction is performed on the operating condition characteristic data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system, including: The operating condition characteristic data are decomposed in the time domain to obtain material characteristic components characterizing the electrode material properties, energy characteristic components characterizing laser energy fluctuations, and response characteristic components characterizing the dynamic response of the mechanical system. Spatial scale transformation and texture structure extraction are performed on the material feature components to obtain material spatial features that reflect the surface morphology changes and material density distribution of the electrode. Frequency domain transformation and wave pattern identification are performed on the energy characteristic components to obtain energy frequency domain characteristics that reflect laser power stability and energy transfer efficiency; The dynamic characteristics of the response feature components are extracted to obtain the dynamic characteristics of the response reflecting mechanical inertia and response delay; The material spatial features, the energy frequency domain features, and the response dynamic features are fused to obtain the comprehensive feature vector.

[0024] First, during the lithium battery manufacturing process, characteristic data of the operating conditions are collected. This data includes real-time sensor data during laser cutting, electrode surface images, and mechanical system parameters. This data forms the basis for subsequent feature extraction and analysis.

[0025] The operating condition characteristic data was decomposed in the time domain into three key components. Wavelet decomposition was employed, using appropriate basis functions to break down the original signal into sub-signals of different frequency bands. Specifically, the db4 wavelet basis function was used to perform a five-level decomposition of the original data, yielding multiple wavelet coefficients. Analysis of the energy distribution characteristics of these coefficients categorized them into three types: material characteristic components, primarily found in high-frequency detail coefficients, reflecting changes in the microstructure of the electrode surface; energy characteristic components, mainly distributed in the mid-frequency coefficient range, reflecting periodic fluctuations in laser energy; and response characteristic components, primarily found in low-frequency approximation coefficients, representing the basic dynamic characteristics of the mechanical system.

[0026] This paper describes a method for spatial scale transformation and texture extraction of material feature components. First, the material feature components are reconstructed into two-dimensional spatial domain information, and the gray-level co-occurrence matrix method is used to analyze the surface texture features of the electrode. Four statistical indicators—contrast, correlation, energy, and homogeneity—are calculated to quantify the microstructure of the electrode surface. Then, a multi-scale analysis method is used to extract the density distribution information of the electrode at different resolutions. Using a Gaussian pyramid model, feature maps at three different scales are constructed, and the density gradient of the local region is calculated at each scale to obtain the material density distribution feature vector. Finally, the texture features and density distribution features are combined to form a complete spatial feature representation of the material.

[0027] Frequency domain transformation and fluctuation pattern identification are performed on the energy characteristic components. Short-time Fourier transform is used to convert the energy characteristic components to the frequency domain, and a Hanning window function with a window length of 256 points and an overlap rate of 50% is set. In the frequency domain representation, characteristic parameters such as power spectral density, dominant frequency components and their amplitudes, and bandgap energy ratios are extracted. By analyzing the stability of the spectrum, the spectral variance is calculated as a quantitative indicator of laser power stability. Furthermore, wavelet packet transform is used to refine the analysis of energy fluctuation patterns, identifying typical fluctuation patterns such as periodic fluctuations, random fluctuations, or mixed fluctuations. These features together constitute the energy frequency domain feature vector, reflecting the laser energy transfer efficiency and stability.

[0028] Dynamic characteristics are extracted from the response feature components to construct a simplified dynamic model of the mechanical system, including three basic parameters: mass, elasticity, and damping. The natural frequency and damping ratio of the system are estimated by analyzing the transient response characteristics in the response feature components. An autoregressive moving average model is used to identify dynamic characteristic parameters, including response time, rise time, and settling time. A response delay index is calculated to quantitatively describe the mechanical system's response speed to control commands. Furthermore, the differences in dynamic response under different operating conditions are analyzed, and an adaptive feature vector is constructed to reflect changes in dynamic performance under varying operating conditions. Finally, these features are merged into a response dynamic feature vector.

[0029] This paper fuses material spatial features, energy frequency domain features, and response dynamic features using a hierarchical fusion strategy. First, feature-level normalization is performed to ensure consistent feature dimensions. Then, principal component analysis is used to reduce feature dimensionality and remove redundant information. The selected features are then weighted and fused, with weights determined through offline training to reflect the contribution of each feature to cutting quality. The resulting comprehensive feature vector has 32 dimensions, encompassing core feature information from material, energy, and response aspects, comprehensively characterizing the multidimensional working conditions during laser cutting.

[0030] In practical applications, taking the laser cutting of a certain type of lithium battery electrode as an example, 500 sets of working condition data were collected. Through the aforementioned feature extraction process, the resulting comprehensive feature vector can effectively distinguish between normal and abnormal cutting states. Experiments show that the fault diagnosis model built based on this feature vector achieves a 95.8% accuracy rate in identifying five common cutting defects (burrs, slag, excessive heat-affected zone, incomplete cutting, and kerf offset), a 12.3% improvement compared to single feature analysis. The improvement in accuracy is particularly significant for defects caused by material inhomogeneity. This demonstrates the effectiveness of multi-dimensional feature extraction methods in improving the quality monitoring accuracy of lithium battery electrode laser cutting.

[0031] In one optional implementation, the integrated feature vector is mapped to the laser cutting target trajectory data in a feature space to predictively quantify the trajectory deviation during the laser cutting process, thereby obtaining a trajectory deviation prediction result, including: The comprehensive feature vector is spatially aligned with the laser cutting target trajectory data to obtain the feature-trajectory correspondence. Based on the aforementioned feature-trajectory correspondence, the temporal evolution path is determined. By extracting the evolutionary characteristics of each feature component in the comprehensive feature vector over time, a temporal evolution feature characterizing the development trend of trajectory deviation is obtained. Spatial evolution features are obtained by expanding the spatial dimension of the temporal evolution features, and trajectory spatial coordinates and trajectory direction information are extracted from the laser cutting target trajectory data to obtain the trajectory spatial coordinate system; The spatialized evolution features are matched with the trajectory space coordinate system in terms of dimension. Through multi-layer nonlinear transformation, the feature vectors in the feature space are mapped to their spatial positions in the trajectory space coordinate system to obtain the feature-position mapping result. Based on the feature-position mapping results, the deviation response intensity at each spatial location in the trajectory space is calculated to obtain the spatial distribution prediction of the trajectory deviation. Based on the spatial distribution prediction of the trajectory deviation, the deviation component is decomposed to obtain the trajectory deviation prediction result.

[0032] When mapping the comprehensive feature vector to the laser cutting target trajectory data in the feature space, spatial dimension alignment is first required. This process synchronizes the feature vector with the time dimension of the trajectory data, establishing a one-to-one correspondence between the two. Specifically, the timestamps of the collected process parameters and environmental data are matched with the timestamps of the trajectory sampling points to ensure that each feature vector corresponds to a specific position on the trajectory.

[0033] After spatial alignment, the temporal evolution path is determined based on the feature-trajectory correspondence. This step extracts temporal evolution features characterizing the trajectory deviation development trend by analyzing the trends of each component in the comprehensive feature vector over time. Specifically, a sliding window method can be used to calculate the rate of change, fluctuation amplitude, and periodic change pattern of each feature component within a fixed-length time window. For the laser power feature component, its average rate of change, standard deviation, and maximum fluctuation amplitude within a continuous time window can be calculated. These indicators effectively reflect the stability of the laser energy input and directly affect the accuracy of the cutting trajectory.

[0034] Next, the temporal evolution features are spatially expanded to obtain spatialized evolution features. This process transforms the time-series features into spatially distributed features to match the trajectory spatial coordinate system. This is achieved by converting the temporal features to the frequency domain using a Fourier transform, and then constructing a frequency-domain-space mapping model based on the spatial distribution characteristics of the trajectory. Simultaneously, trajectory spatial coordinates and orientation information are extracted from the laser cutting target trajectory data to establish a trajectory spatial coordinate system. This system comprises global coordinates (x, y, z) and local coordinates (tangential, normal, and subnormal), comprehensively describing the spatial geometric characteristics of the cutting trajectory.

[0035] When matching the spatialized evolutionary features with the trajectory space coordinate system, a multi-layer nonlinear transformation method is employed. This method constructs a three-layer neural network: the input layer receives the spatialized evolutionary features, the hidden layer performs nonlinear feature transformation, and the output layer maps to the trajectory space coordinates. During neural network training, historical trajectory deviation data is used as supervision information, and the network parameters are optimized through backpropagation, ultimately achieving an accurate mapping from the feature space to the trajectory space, resulting in a feature-position mapping.

[0036] Based on the feature-position mapping results, the deviation response intensity at each location in the trajectory space is calculated. By introducing a response function based on the feature-position mapping, the deviation sensitivity of each spatial location is calculated. The response function considers the magnitude of the eigenvalues, the rate of change, and the geometric constraints of the location to comprehensively evaluate the degree of deviation at each spatial point. For example, in regions where material hardness changes abruptly, even if the laser parameters are stable, the trajectory deviation can increase due to material heterogeneity, and the deviation response intensity in this region will be correspondingly increased.

[0037] Finally, based on the spatial distribution prediction of the trajectory deviation, deviation components are decomposed, and the predicted trajectory deviation is decomposed into three components in the local coordinate system: tangential deviation, normal deviation, and subnormal deviation. Tangential deviation represents the progress deviation along the trajectory's forward direction, normal deviation represents the depth deviation perpendicular to the trajectory plane, and subnormal deviation represents the lateral deviation perpendicular to the forward direction within the trajectory plane. Through component decomposition, laser parameters can be adjusted specifically, such as adjusting laser power to correct depth deviation, adjusting scanning speed to correct progress deviation, and optimizing the focus position to correct lateral deviation.

[0038] After obtaining the trajectory deviation prediction results, one approach is to establish an energy-trajectory joint compensation mechanism to achieve dual closed-loop control of cutting quality. Specifically, the decomposed normal deviation component is correlated with the density distribution information in the material spatial characteristics. When the predicted normal deviation value exceeds 0.025 mm and the material density gradient shows a high-density region at the bottom layer, it is determined to be a "potential incomplete cutting" risk state. At this time, based on the original trajectory compensation, the laser power in this area is simultaneously increased by 14%-19%, and the cutting speed is reduced by 11%-16% to ensure that the electrode is completely cut through. Simultaneously, the power spectral density change in the energy frequency domain characteristics is monitored. When the frequency band energy ratio is detected to be continuously lower than 0.70 and the normal deviation shows an increasing trend, the energy compensation amplitude is further increased to a maximum of 22%. Conversely, when the predicted sub-normal deviation value exceeds 0.030 mm and the material spatial characteristics show an abnormal increase in edge texture energy (exceeding 0.85), combined with the vibration signal analysis in the response dynamic characteristics, it is determined to be a "overcutting accompanied by burr and heat-affected zone expansion" state. Immediately afterward, the laser energy is reduced by 18%-26% and the cutting speed is increased by 15%-22%, while the focusing position is optimized to compensate for the sub-normal deviation. This joint compensation mechanism has a decision cycle of 12ms and achieves coordinated adjustment of laser power, cutting speed, and focusing position through a fuzzy PID controller. In actual testing, this method improved trajectory accuracy by 31.2%, reduced the heat-affected zone width by 28.5%, reduced the incomplete cutting defect rate by 92.3%, and controlled the burr height at the cutting edge to within 0.04mm, significantly improving the overall quality indicators of electrode cutting.

[0039] In practical applications, taking lithium battery laser cutting as an example, when the ambient humidity rises from 45% to 60%, the system can identify that this change will lead to an increase in normal deviation. Through feature-position mapping, a maximum depth deviation of approximately 0.03mm is predicted in the wafer edge region. Based on this, the laser power in that region is automatically increased to control the deviation within an acceptable range, effectively ensuring the stability of the cutting quality.

[0040] This method can not only predict the spatial distribution of trajectory deviation, but also provide accurate quantitative information on deviation through component decomposition, providing a scientific basis for real-time parameter adjustment in the laser cutting process, and significantly improving production efficiency and product quality.

[0041] In one optional implementation, the temporal evolution path is determined based on the feature-trajectory correspondence. By extracting the evolutionary characteristics of each feature component in the comprehensive feature vector over time, temporal evolution features characterizing the development trend of trajectory deviation are obtained, including: Based on the feature-trajectory correspondence, the correlation strength between each feature component and the trajectory deviation is extracted. Based on the correlation strength, the time series of each feature component in the comprehensive feature vector is weighted to obtain a weighted time series feature sequence. The weighted temporal feature sequence is divided into time windows and subjected to sliding processing to obtain evolutionary gradient information representing the temporal variation trend of the feature components; The evolution gradient information is time-aligned with the trajectory temporal information in the laser cutting target trajectory data to obtain the temporal evolution path; Based on the change amplitude sequence and change direction sequence of each feature component in the evolution gradient information, the evolution rate of each feature component is calculated. Based on the evolution rate, the time dimension of each feature component is extrapolated to calculate the predicted value of each feature component at future time. The predicted values ​​for future moments are compared with the feature component values ​​for the current moment to obtain the evolutionary offset of each feature component, and the temporal evolution trajectory of the feature component is determined based on the evolutionary offset. The temporal evolution characteristics are obtained by associating the time development trajectory with the time information in the laser cutting target trajectory data.

[0042] Figure 2 This is a schematic diagram illustrating the process of determining temporal evolution characteristics according to an embodiment of the present invention. Figure 2 As shown, firstly, the correlation strength between each feature component and the trajectory deviation is extracted based on the feature-trajectory correspondence. The Pearson correlation coefficient is used to calculate the correlation between the feature component and the trajectory deviation, denoted as the correlation coefficient r, with a value range of [-1, 1]. By calculating the correlation coefficient between the time series data of the feature component and the time series data of the trajectory deviation, a numerical value reflecting the correlation strength is obtained. For feature components in the laser cutting process, such as tool vibration frequency and cutting speed, the correlation coefficient between them and the trajectory deviation is calculated, and feature components with an absolute correlation coefficient greater than a threshold (e.g., 0.6) are selected as key features. Based on the obtained correlation strength, the time series of each feature component in the comprehensive feature vector is weighted to obtain a weighted time series feature sequence. The weighting method is to use the correlation strength as a weight coefficient and multiply it by the corresponding feature component, thereby highlighting the feature components with strong correlation to the trajectory deviation and suppressing the feature components with weak correlation.

[0043] Next, the weighted time-series feature sequence is divided into time windows and subjected to sliding processing to obtain evolutionary gradient information representing the temporal variation trend of the feature components. The window size is set according to the characteristics of the laser cutting process, such as setting the window width to 50 time points and the sliding step size to 10 time points. Within each window, the rate of change of the feature components is calculated, i.e., the difference between adjacent time points divided by the time interval. For the rate of change sequence within the window, its statistical characteristics, such as the average rate of change and the variance of the rate of change, are calculated to form an evolutionary gradient feature vector. The sliding window is calculated sequentially to obtain the evolutionary gradient sequence that changes over time, reflecting the variation trend of the feature components in different time periods.

[0044] The evolutionary gradient information is time-aligned with the trajectory temporal information in the laser cutting target trajectory data to obtain a temporal evolution path. Time alignment uses timestamp matching to map the evolutionary gradient sequence of feature components to the trajectory deviation data one-to-one according to timestamps. When timestamps do not perfectly match, linear interpolation is used to fill in the gaps, ensuring a strict temporal correspondence between the feature evolution information and the trajectory deviation data. The aligned temporal data constitutes the temporal evolution path, describing the dynamic relationship between changes in feature components and trajectory deviation.

[0045] The evolution rate of each feature component is calculated based on the change amplitude and direction sequences of each feature component in the evolution gradient information. The change amplitude sequence refers to the absolute change of the feature component between adjacent time windows, while the change direction sequence indicates whether the change is positive or negative. The evolution rate is calculated by dividing the change amplitude by the time window interval, taking the change direction into account. For feature components that are steadily increasing or decreasing, their evolution rates remain relatively stable; while for feature components that fluctuate, their evolution rates also fluctuate accordingly. Based on the evolution rate, each feature component is extrapolated over time to calculate its predicted value at future times. The extrapolation method uses linear prediction or polynomial fitting, predicting the value of the feature component within a future period (e.g., 100 time points) based on the historical trend of the evolution rate.

[0046] By comparing the predicted values ​​for future timeframes with the current value of the feature components, the evolutionary offset of each feature component is obtained. This offset reflects the degree of change of the feature component within the prediction time period and is calculated as the difference between the predicted and current values. The magnitude and direction of the evolutionary offset are analyzed to determine the trend of the feature component's change. When the evolutionary offset exceeds a preset threshold (e.g., 20% of the current value), the feature component is considered to have a significant trend. Based on the evolutionary offset, the temporal trajectory of the feature component is determined. This trajectory consists of continuous time points from the current timeframe to the prediction time and their corresponding feature values, visually demonstrating the change pattern of the feature component over time.

[0047] Finally, the temporal evolution trajectory is correlated with the temporal information in the laser cutting target trajectory data to obtain temporal evolution characteristics. The correlation method involves mapping the temporal evolution trajectory of the feature components to the trajectory deviation data in the time dimension, analyzing the time lead of feature changes before trajectory deviation changes. By calculating the cross-correlation function, the time lag relationship between feature component changes and trajectory deviation changes is determined. Feature components with large time lead are considered early warning indicators of trajectory deviation. The temporal evolution characteristics include the evolution rate, evolution direction, evolution offset, and time lead relationship with trajectory deviation of the feature components, comprehensively describing the impact of feature component changes over time on trajectory deviation, providing a basis for trajectory deviation prediction and control.

[0048] In one optional implementation, based on the spatial position deviation component and the time delay deviation component in the trajectory deviation prediction result, a composite compensation quantity including a position compensation vector and a timing adjustment parameter is generated using a multi-objective optimization method, including: Based on the spatial position deviation component and the time delay deviation component in the trajectory deviation prediction result, a set of deviation components is determined; The spatial position deviation component is spatially decomposed into a tangential deviation component along the cutting path and a normal deviation component perpendicular to the cutting path. The spatial position compensation target is then determined based on the tangential deviation component and the normal deviation component. The time delay deviation component is decomposed into laser triggering time deviation and cutting speed deviation, and a timing adjustment target is determined based on the laser triggering time deviation and the cutting speed deviation. After coupling the spatial location compensation target with the temporal adjustment target, an association constraint between spatial location deviation and time delay deviation is added to determine the coupling optimization relationship; Based on the aforementioned coupling optimization relationship, the spatial location compensation target and the temporal adjustment target are iteratively adjusted so that they simultaneously reach the convergence condition, thereby obtaining the composite compensation amount.

[0049] During laser cutting, the effects of spatial position and time delay need to be comprehensively considered. Multi-target optimization is used to generate composite compensation quantities to ensure that the laser accurately acts on the target position.

[0050] First, a set of stereoscopic deviation components is determined based on the spatial position deviation component and the time delay deviation component in the trajectory deviation prediction results. In practical applications, a high-precision vision sensor is used to obtain the deviation between the actual and ideal positions of the cutting trajectory, while a time-series analysis module obtains the time delay data of laser triggering. These data together constitute the set of stereoscopic deviation components, containing all deviation information in both spatial and temporal dimensions. For example, during silicon wafer cutting, the vision system detects a deviation of 3 micrometers along the X-axis and 2 micrometers along the Y-axis, while the laser triggering is delayed by 0.5 milliseconds.

[0051] Next, the spatial position deviation components are spatially decomposed into tangential deviation components along the cutting path and normal deviation components perpendicular to the cutting path. A local coordinate system is established along the tangential direction of the cutting path, and the original spatial position deviation is projected onto the tangential and normal directions through coordinate transformation. The tangential deviation represents the advance or lag along the cutting path, while the normal deviation represents the degree of deviation perpendicular to the cutting path. This decomposition method allows for more precise design of corresponding compensation strategies for deviations in different directions. Based on the decomposed deviation components, the spatial position compensation objective is determined, typically expressed as an optimization objective function that makes both the tangential and normal deviations close to zero.

[0052] For the time delay deviation component, time-series decomposition divides it into two parts: laser trigger timing deviation and cutting speed deviation. Laser trigger timing deviation reflects the time difference between the laser emission command and the actual laser output, while cutting speed deviation represents the difference between the actual cutting speed and the set speed. A time-series deviation model can be constructed by analyzing historical trigger records and speed monitoring data. For example, under specific operating conditions, a fixed 2-millisecond trigger delay is found in the laser system, and the actual cutting speed is approximately 5% lower than the set value. Based on these decomposition results, a time-series adjustment target is determined, typically expressed as minimizing a weighted sum function of the laser trigger delay and speed deviation.

[0053] Coupled with the spatial location compensation objective and the temporal adjustment objective, a crucial step is the optimization of spatial location and time delay. Since there is an inherent relationship between spatial location and time delay—for example, time delay can cause spatial location shifts—the two cannot be optimized independently. Therefore, a constraint linking spatial location deviation and time delay deviation is added to the coupled optimization relationship. This constraint is typically expressed as a function of tangential location deviation and time delay; for example, tangential deviation equals cutting speed multiplied by time delay. By establishing a comprehensive optimization model, both types of objective functions and their constraints are integrated into a single framework.

[0054] Finally, based on the coupled optimization relationship, an iterative optimization algorithm is used to adjust the position compensation vector and the timing adjustment parameters. In each iteration, the spatial compensation vector and the timing adjustment parameters are updated simultaneously, the degree to which the current solution satisfies the two optimization objectives is evaluated, and the search direction is adjusted according to the evaluation results. The iteration stops when both the spatial position compensation objective and the timing adjustment objective simultaneously reach the preset convergence conditions, and the final composite compensation amount is output. For example, when both the tangential deviation and the normal deviation are less than 1 micrometer, and the laser triggering time deviation is less than 0.1 milliseconds, the convergence condition is considered to have been met.

[0055] In practical applications, such as the lithium battery cutting process, this composite compensation amount can simultaneously adjust the mechanical motion trajectory and laser triggering timing, effectively addressing complex deviations caused by factors such as mechanical inertia and control delays, and significantly improving laser cutting accuracy. The composite compensation amount generated by the above method includes a spatial position compensation vector and timing adjustment parameters, which can guide the subsequent trajectory execution system to make real-time adjustments, ensuring precise positioning of the laser's point of action.

[0056] In one optional implementation, the time delay deviation component is decomposed into laser triggering time deviation and cutting speed deviation, and a timing adjustment target is determined based on the laser triggering time deviation and the cutting speed deviation, including: The time difference between the actual laser triggering time and the target triggering time is extracted from the time delay deviation component to obtain the laser triggering time deviation, and the adjustment amount of the laser triggering time is calculated based on the laser triggering time deviation; The speed difference between the actual cutting speed and the target cutting speed is extracted from the time delay deviation component to obtain the cutting speed deviation, and the adjustment amount of the cutting speed is calculated based on the cutting speed deviation; A correlation analysis was performed between the adjustment amount of the laser triggering time and the adjustment amount of the cutting speed. By establishing a constraint relationship between the laser triggering time adjustment and the cutting speed adjustment, the timing adjustment constraint conditions were obtained. The timing adjustment target is determined based on the timing adjustment constraints, wherein the timing adjustment target requires that the adjustment amount of the laser triggering time and the adjustment amount of the cutting speed simultaneously satisfy the timing adjustment constraints.

[0057] First, time delay data during the laser cutting process is collected. A high-precision timing sensor records the actual moment of each laser trigger and the cutting machine's speed parameters, while also recording the preset target trigger moment and target cutting speed. The collected data undergoes preliminary processing to remove outliers and standardize it into a dataset based on a unified time reference.

[0058] When extracting the laser trigger time deviation from the time delay deviation component, the difference between the actual laser trigger time and the preset target trigger time is calculated for each cutting process. Specifically, if the target trigger time of the laser in a certain cutting process is T_target, and the actual trigger time is T_actual, then the laser trigger time deviation ΔT is calculated as: ΔT = T_actual - T_target. Based on statistical analysis methods, trend analysis is performed on the trigger time deviation data in multiple consecutive cutting processes to determine whether the deviation exhibits systematic shift or random fluctuation characteristics. For systematic shifts, the adjustment amount for the laser trigger time is calculated as -ΔT, that is, the trigger timing is adjusted in the opposite direction to make the actual trigger time approach the target time. For random fluctuations, the fluctuation range is calculated and an allowable range for the trigger time is set. Fluctuations within this range are considered normal, while those exceeding the range are adjusted accordingly.

[0059] When extracting the cutting speed deviation from the time delay deviation component, the actual cutting speed is compared with the target cutting speed. If the target cutting speed is V_target and the actual measured cutting speed is V_actual, then the cutting speed deviation ΔV is calculated as: ΔV = V_actual - V_target. Based on the magnitude and direction of the speed deviation, and combined with the dynamic characteristics of the cutting machine, the adjustment amount of the cutting speed is calculated. Typically, the adjustment of the cutting speed needs to consider acceleration limitations and mechanical response characteristics; therefore, the calculation of the adjustment amount needs to include a gradual adjustment strategy to avoid sudden speed changes that could lead to instability in the mechanical system.

[0060] When analyzing the correlation between the adjustment of the laser triggering time and the adjustment of the cutting speed, a constraint relationship model is established between the two. This model reflects the complementary effect between the triggering time adjustment and the speed adjustment. For example, when the laser triggering is advanced by 10 microseconds and the cutting speed is increased by 5 millimeters per second, the effects of the two adjustments on the cutting position cancel each other out. Specifically, a functional relationship is established: f(ΔT_adj, ΔV_adj) = 0, where ΔT_adj represents the adjustment of the triggering time and ΔV_adj represents the adjustment of the cutting speed.

[0061] Based on the established timing adjustment constraints, and considering the cutting accuracy requirements and system stability, the optimal timing adjustment target is determined. The adjustment target must simultaneously satisfy: ① the error after adjusting the laser trigger time is within the allowable range; ② the fluctuation after adjusting the cutting speed does not exceed the threshold for stable equipment operation; ③ the combined effect of the two adjustments can effectively eliminate time delay deviation.

[0062] In practical applications, a weighted optimization strategy can be adopted. For example, for fine cutting processes with extremely high timing accuracy requirements, a higher weight can be given to the adjustment of the trigger time; while for continuous high-speed cutting scenarios, it is more inclined to maintain the stability of the cutting speed, so a smaller adjustment range is given to the speed adjustment.

[0063] Iterative optimization algorithms, such as gradient descent or genetic algorithms, are used to search for the optimal combination of adjustment parameters within the constraint space. Specifically, the objective function can be set as: min(w1•|ΔT_adj| + w2•|ΔV_adj|), where w1 and w2 are weight coefficients reflecting the cost or importance of adjustments at the trigger time and at the velocity, respectively. By solving this minimization problem, the optimal combination of adjustment values ​​that satisfies the constraints is obtained.

[0064] After determining the timing adjustment target, the adjustment parameters are applied to the control system, and the adjustment effect is verified in the next cutting process. Based on the verification results, the parameters need to be further fine-tuned until the time delay deviation is effectively controlled within the target range.

[0065] In a specific application scenario, suppose that during the laser cutting of a lithium battery, a systematic offset of 0.05 mm in the cutting line position is found. Analysis reveals that this offset consists of two parts: a 5-microsecond laser trigger delay and a 2% increase in cutting speed. According to this method, the trigger timing deviation is first extracted as 5 microseconds, and the adjustment is calculated as -5 microseconds; the speed deviation is extracted as 2% of the target speed, and the adjustment is calculated as a reduction of 1.96% of the current speed. Through correlation analysis, it is found that under these process conditions, advancing the trigger timing by 1 microsecond is equivalent to a 0.4% reduction in speed in terms of impact on the cutting position. Therefore, the deviation can be eliminated entirely by advancing the trigger timing, entirely by reducing the cutting speed, or a combination of both. Ultimately, considering the balance between production efficiency and accuracy, advancing the trigger timing by 3 microseconds while simultaneously reducing the cutting speed by 0.8% is determined as the optimal timing adjustment target.

[0066] In one optional implementation, the laser cutting control command is corrected in real time based on the composite compensation amount, and actual trajectory feedback data is collected during the cutting process, including: The location compensation vector and timing adjustment parameters are determined based on the composite compensation amount; The position compensation vector is superimposed on the trajectory position command component in the laser cutting control command. By correcting the coordinate values ​​in the trajectory position command component, the corrected trajectory position command is obtained. The timing adjustment parameters are integrated into the timing control command component of the laser cutting control command. By correcting the time parameters in the timing control command component, the corrected timing control command is obtained. The corrected trajectory position command is combined with the corrected timing control command to generate a real-time corrected laser cutting control command; The real-time corrected laser cutting control command is sent to the laser cutting execution device to drive the laser cutting execution device to perform the cutting operation; During the cutting process, sensors monitor the motion trajectory position and motion timing of the laser cutting device in real time to obtain the actual trajectory feedback data.

[0067] In the laser cutting process, the acquired composite compensation amount first needs to be decomposed into two dimensions of compensation information: position compensation vector and timing adjustment parameters.

[0068] After obtaining the composite compensation amount, it is transformed into a form suitable for the control system through mathematical transformation. Specifically, the position compensation vector can be represented as (Δx, Δy, Δz) coordinate correction value, which is used to correct the spatial position during laser cutting; the timing adjustment parameters include acceleration adjustment coefficient k_acc, velocity correction factor k_v, and delay compensation value t_delay, which are used to optimize the time characteristics of the motion.

[0069] When superimposing the position compensation vector onto the trajectory position command, a direct vector superposition method is used. Assuming the coordinates of a point in the original trajectory position command are P(x, y, z), the corrected coordinates P' can be expressed as: P'(x', y', z') = P(x+Δx, y+Δy, z+Δz); For complex curved trajectories, the difference in compensation amounts at different locations must be considered. Therefore, a segmented compensation strategy is adopted. The overall trajectory is divided into several small segments, and a corresponding position compensation vector is applied to each segment to ensure a smooth transition of the entire trajectory.

[0070] Regarding timing adjustments, the timing adjustment parameters are integrated into the timing control commands, primarily achieved by modifying acceleration / deceleration parameters, motion duration, and switching point times. For example, if the original command specifies an execution time of t, a velocity of v, and an acceleration of a certain trajectory, the adjusted parameters become: t' = t + t_delay, v' = v × k_v, a' = a × k_acc. This adjustment effectively improves the dynamic characteristics of the laser cutting process and reduces errors caused by inertia, vibration, and other factors.

[0071] When combining the corrected trajectory position command with the corrected timing control command, it is necessary to ensure that their spatiotemporal correspondence is correct. This can be achieved by establishing a unified spatiotemporal mapping relationship, that is, at each time t, the laser head should be located at the corresponding position point P'(t) on the corrected trajectory. In actual operation, a complete control command flow is generated by constructing a command data packet containing the corrected position sequence and the corresponding timestamp.

[0072] After generating the laser cutting control commands with real-time correction, the commands are sent to the execution device through the control system interface. These commands must conform to the device's communication protocol, including correct command headers, data segments, and checksums. Real-time communication is used to ensure timely delivery and execution of commands, minimizing the impact of communication delays.

[0073] During the cutting process, multiple sensors are used to monitor the equipment's operating status in real time. Position sensors (such as grating rulers and encoders) are used to capture the actual spatial coordinates of the laser head; timing sensors (such as high-precision clocks and synchronization signal detectors) record the times when key events occur. After preprocessing (including filtering, calibration, and spatiotemporal synchronization), these sensor data form standardized actual trajectory feedback data.

[0074] Feedback data acquisition follows a sufficiently high sampling frequency, typically 5-10 times the control cycle, to ensure the capture of rapid changes in the trajectory. The acquired data includes timestamps, three-dimensional spatial coordinates, motion states (such as velocity and acceleration), forming a complete trajectory description.

[0075] The collected feedback data also needs to undergo quality assessment, removing outliers and filling in missing data. This can be achieved through statistical analysis and predictive algorithms, ensuring the continuity and reliability of the feedback data. The processed feedback data is not only used for online evaluation of the current cutting quality but also serves as an important basis for subsequent compensation optimization.

[0076] Based on the above, a complete real-time correction mechanism for laser cutting control commands was established. This mechanism makes full use of composite compensation information to precisely adjust the control commands from both position and timing dimensions, and further verifies the correction effect through real-time feedback data, forming a closed-loop control that effectively improves the accuracy and stability of laser cutting.

[0077] In a preferred embodiment, this invention combines underlying PSO (Peak Power and Overlap Rate) control technology to achieve constant control of single-pulse peak power and overlap rate during variable-speed laser cutting. During laser cutting, the cutting effect is primarily determined by two core parameters: single-pulse peak power... And the overlap ratio N. According to the laser cutting principle, the relationship between these two parameters and other process parameters is as follows: ; ; ; ; in, Peak power, For single-pulse energy, Where is the pulse width, N is the overlap ratio, D is the spot diameter, and d is the pulse spacing. Where f is the laser power, f is the frequency, and V is the cutting speed.

[0078] In actual cutting processes, when the cutting speed V changes (such as during complex conditions like acceleration, deceleration, and rotation during tab cutting), traditional methods require simultaneous adjustment of laser power and frequency, which is computationally complex and makes it difficult to guarantee consistent cutting quality. This invention utilizes PSO control technology to implement an automatic compensation mechanism at the underlying level. The target peak power is set. Then, based on real-time speed Automatically calculate the required frequency: ; The pulse spacing d is determined by the set overlap rate N and the spot diameter D.

[0079] Based on the calculated frequency Automatically adjust laser power: ; With this control method, the peak power and overlap rate of the single pulse remain constant regardless of the cutting speed, ensuring consistent cutting quality at high speed, low speed, and variable speed.

[0080] In the user interface of the control system, the operator only needs to directly set the peak energy of the single pulse. (e.g., 20.0 μJ), overlap ratio N (e.g., 0.90), target cutting speed (e.g., 1.20 m / s) and the velocity curves of each segment along the cutting path. The laser power is automatically calculated and adjusted in real time. The underlying parameters such as frequency f are eliminated, avoiding cumbersome power, frequency, and speed conversions and greatly simplifying the process parameter setting process.

[0081] For lithium battery tab cutting, this invention supports one-click import of cutting parameters from a pre-set Excel spreadsheet (.xlsx format) by process engineers. This includes tab geometry parameters (length, width, corner radius, etc.), single-pulse energy and overlap rate of each trajectory segment, cutting speed of each segment, and coordinates of the start and end points. After import, a complete cutting trajectory is automatically generated, and the laser power and frequency parameters of each segment are calculated based on the PSO control principle. This eliminates the need for manual input by operators, reducing human intervention time and input errors, and enabling even operators with no prior experience to accurately set process parameters.

[0082] This invention reorganizes commonly used functional interfaces, simultaneously displaying a visual preview of the electrode geometry, parameter settings for each trajectory segment (single-pulse energy, overlap rate, velocity, etc.), a superimposed display of the real-time cutting trajectory and the actual feedback trajectory, and real-time monitoring and compensation display of trajectory deviation within a single interface. Operators can complete all operations, including trajectory design, parameter setting, real-time monitoring, and deviation compensation, on a single interface, reducing the unpleasant experience caused by interface switching and improving operational efficiency.

[0083] In the application of laser cutting of electrode tabs in a lithium battery production line, the PSO control technology and deep learning trajectory energy following compensation method of this invention were adopted. During the acceleration of the cutting speed from 0.5 m / s to 2.0 m / s, the deviation in cutting quality (including kerf width, heat-affected zone, and burr height) was controlled within ±3%, while the deviation of traditional methods reached over ±15%. Through a one-click import function, the electrode process parameter setting time was shortened from 15-20 minutes using traditional methods to less than 2 minutes, and the parameter input error rate decreased from 8% to 0. In the cutting process of complex electrode trajectories (including multiple straight lines, arcs, and corners), the combination of trajectory deviation prediction by deep learning and automatic adjustment of power frequency by PSO control improved the cutting position accuracy by 42% and reduced the heat-affected zone width by 35%. By combining trajectory deviation prediction by deep learning with automatic adjustment of underlying parameters of PSO control technology, this invention achieves intelligent, precise, and simplified laser cutting processes, significantly improving the quality and production efficiency of lithium battery electrode cutting.

[0084] The deep learning-based lithium battery laser cutting trajectory energy tracking compensation system of this invention includes: The first unit is used to acquire laser cutting target trajectory data of lithium battery electrode sheets and working condition characteristic data during the actual cutting process; The second unit is used to extract multi-dimensional features from the operating condition feature data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system. The third unit is used to perform feature space mapping between the comprehensive feature vector and the laser cutting target trajectory data, so as to predictively quantify the trajectory deviation during the laser cutting process and obtain the trajectory deviation prediction result. The fourth unit is used to generate a composite compensation amount containing a position compensation vector and a timing adjustment parameter based on the spatial position deviation component and the time delay deviation component in the trajectory deviation prediction result using a multi-objective optimization method. The fifth unit is used to perform real-time correction of the laser cutting control command based on the composite compensation amount, and to collect actual trajectory feedback data during the cutting process; The sixth unit is used to calculate the error using the actual trajectory feedback data and the trajectory deviation prediction result, and to adaptively update the parameters of the multi-dimensional feature extraction and the multi-objective optimization method based on the result of the error calculation.

[0085] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0086] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0087] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based energy tracking compensation method for lithium battery laser cutting trajectory, characterized in that, include: Acquire laser cutting target trajectory data and actual working condition characteristic data of lithium battery electrode sheets; Multi-dimensional feature extraction is performed on the aforementioned working condition characteristic data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system; The comprehensive feature vector is mapped to the laser cutting target trajectory data in the feature space to predictively quantify the trajectory deviation during the laser cutting process and obtain the trajectory deviation prediction result. Based on the spatial position deviation component and time delay deviation component in the trajectory deviation prediction result, a composite compensation amount including a position compensation vector and a timing adjustment parameter is generated using a multi-objective optimization method. The laser cutting control command is corrected in real time based on the composite compensation amount, and actual trajectory feedback data is collected during the cutting process; Error is calculated using the actual trajectory feedback data and the trajectory deviation prediction results. Based on the error calculation results, the parameters of the multi-dimensional feature extraction and the multi-objective optimization method are adaptively updated.

2. The method according to claim 1, characterized in that, Multi-dimensional feature extraction is performed on the aforementioned operating condition characteristic data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system, including: The operating condition characteristic data are decomposed in the time domain to obtain material characteristic components characterizing the electrode material properties, energy characteristic components characterizing laser energy fluctuations, and response characteristic components characterizing the dynamic response of the mechanical system. Spatial scale transformation and texture structure extraction are performed on the material feature components to obtain material spatial features that reflect the surface morphology changes and material density distribution of the electrode. Frequency domain transformation and wave pattern identification are performed on the energy characteristic components to obtain energy frequency domain characteristics that reflect laser power stability and energy transfer efficiency; The dynamic characteristics of the response feature components are extracted to obtain the dynamic characteristics of the response reflecting mechanical inertia and response delay; The material spatial features, the energy frequency domain features, and the response dynamic features are fused to obtain the comprehensive feature vector.

3. The method according to claim 1, characterized in that, The comprehensive feature vector is mapped to the laser cutting target trajectory data in a feature space to predictively quantify the trajectory deviation during the laser cutting process, resulting in a trajectory deviation prediction result, including: The comprehensive feature vector is spatially aligned with the laser cutting target trajectory data to obtain the feature-trajectory correspondence. Based on the aforementioned feature-trajectory correspondence, the temporal evolution path is determined. By extracting the evolutionary characteristics of each feature component in the comprehensive feature vector over time, a temporal evolution feature characterizing the development trend of trajectory deviation is obtained. Spatial evolution features are obtained by expanding the spatial dimension of the temporal evolution features, and trajectory spatial coordinates and trajectory direction information are extracted from the laser cutting target trajectory data to obtain the trajectory spatial coordinate system; The spatialized evolution features are matched with the trajectory space coordinate system in terms of dimension. Through multi-layer nonlinear transformation, the feature vectors in the feature space are mapped to their spatial positions in the trajectory space coordinate system to obtain the feature-position mapping result. Based on the feature-location mapping results, the deviation response intensity at each spatial location in the trajectory space is calculated to obtain the spatial distribution prediction of the trajectory deviation. Based on the spatial distribution prediction of the trajectory deviation, the deviation component is decomposed to obtain the trajectory deviation prediction result.

4. The method according to claim 3, characterized in that, Based on the aforementioned feature-trajectory correspondence, the temporal evolution path is determined. By extracting the evolutionary characteristics of each feature component in the comprehensive feature vector over time, temporal evolution features characterizing the development trend of trajectory deviation are obtained, including: Based on the feature-trajectory correspondence, the correlation strength between each feature component and the trajectory deviation is extracted. Based on the correlation strength, the time series of each feature component in the comprehensive feature vector is weighted to obtain a weighted time series feature sequence. The weighted temporal feature sequence is divided into time windows and subjected to sliding processing to obtain evolutionary gradient information representing the temporal variation trend of the feature components; The evolution gradient information is time-aligned with the trajectory temporal information in the laser cutting target trajectory data to obtain the temporal evolution path; Based on the change magnitude sequence and change direction sequence of each feature component in the evolution gradient information, the evolution rate of each feature component is calculated. Based on the evolution rate, the time dimension extrapolation of each feature component is performed to calculate the predicted value of each feature component at future time. The predicted values ​​for future moments are compared with the feature component values ​​for the current moment to obtain the evolutionary offset of each feature component, and the temporal evolution trajectory of the feature component is determined based on the evolutionary offset. The temporal evolution characteristics are obtained by associating the time development trajectory with the time information in the laser cutting target trajectory data.

5. The method according to claim 1, characterized in that, Based on the spatial position deviation component and time delay deviation component in the trajectory deviation prediction result, a composite compensation quantity including a position compensation vector and a timing adjustment parameter is generated using a multi-objective optimization method, including: Based on the spatial position deviation component and time delay deviation component in the trajectory deviation prediction result, a set of deviation components is determined; The spatial position deviation component is spatially decomposed into a tangential deviation component along the cutting path and a normal deviation component perpendicular to the cutting path. The spatial position compensation target is then determined based on the tangential deviation component and the normal deviation component. The time delay deviation component is decomposed into laser triggering time deviation and cutting speed deviation, and the timing adjustment target is determined based on the laser triggering time deviation and the cutting speed deviation. After coupling the spatial location compensation target with the temporal adjustment target, an association constraint between spatial location deviation and time delay deviation is added to determine the coupling optimization relationship; Based on the aforementioned coupling optimization relationship, the spatial location compensation target and the temporal adjustment target are iteratively adjusted so that they simultaneously reach the convergence condition, thereby obtaining the composite compensation amount.

6. The method according to claim 5, characterized in that, The time delay deviation component is decomposed into laser triggering time deviation and cutting speed deviation, and a timing adjustment target is determined based on the laser triggering time deviation and the cutting speed deviation, including: The time difference between the actual laser triggering time and the target triggering time is extracted from the time delay deviation component to obtain the laser triggering time deviation, and the adjustment amount of the laser triggering time is calculated based on the laser triggering time deviation; The speed difference between the actual cutting speed and the target cutting speed is extracted from the time delay deviation component to obtain the cutting speed deviation, and the adjustment amount of the cutting speed is calculated based on the cutting speed deviation; A correlation analysis was performed between the adjustment amount of the laser triggering time and the adjustment amount of the cutting speed. By establishing a constraint relationship between the laser triggering time adjustment and the cutting speed adjustment, the timing adjustment constraint conditions were obtained. The timing adjustment target is determined based on the timing adjustment constraints, wherein the timing adjustment target requires that the adjustment amount of the laser triggering time and the adjustment amount of the cutting speed simultaneously satisfy the timing adjustment constraints.

7. The method according to claim 1, characterized in that, The laser cutting control commands are corrected in real time based on the composite compensation amount, and actual trajectory feedback data is collected during the cutting process, including: The location compensation vector and timing adjustment parameters are determined based on the composite compensation amount; The position compensation vector is superimposed on the trajectory position command component in the laser cutting control command. By correcting the coordinate values ​​in the trajectory position command component, the corrected trajectory position command is obtained. The timing adjustment parameters are integrated into the timing control command component of the laser cutting control command. By correcting the time parameters in the timing control command component, the corrected timing control command is obtained. The corrected trajectory position command is combined with the corrected timing control command to generate a real-time corrected laser cutting control command; The real-time corrected laser cutting control command is sent to the laser cutting execution device to drive the laser cutting execution device to perform the cutting operation; During the cutting process, sensors monitor the motion trajectory position and motion timing of the laser cutting device in real time to obtain the actual trajectory feedback data.

8. A deep learning-based lithium battery laser cutting trajectory energy following compensation system, used to implement the method as described in any one of claims 1-7, characterized in that, include: The first unit is used to acquire laser cutting target trajectory data of lithium battery electrode sheets and working condition characteristic data during the actual cutting process; The second unit is used to extract multi-dimensional features from the operating condition feature data to obtain a comprehensive feature vector characterizing the electrode material properties, laser energy fluctuations, and dynamic response of the mechanical system. The third unit is used to perform feature space mapping between the comprehensive feature vector and the laser cutting target trajectory data, so as to predictively quantify the trajectory deviation during the laser cutting process and obtain the trajectory deviation prediction result. The fourth unit is used to generate a composite compensation amount containing a position compensation vector and a timing adjustment parameter based on the spatial position deviation component and the time delay deviation component in the trajectory deviation prediction result using a multi-objective optimization method. The fifth unit is used to perform real-time correction of the laser cutting control command based on the composite compensation amount, and to collect actual trajectory feedback data during the cutting process; The sixth unit is used to calculate the error using the actual trajectory feedback data and the trajectory deviation prediction result, and to adaptively update the parameters of the multi-dimensional feature extraction and the multi-objective optimization method based on the result of the error calculation.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images