Dynamic precision compensation control method based on intelligent PCB drilling machine

Abstract

The application relates to a dynamic precision compensation control method based on an intelligent PCB drilling machine, which comprises the following steps: collecting a dynamic parameter set in real time through a multi-source sensor group installed on a drilling main shaft, the dynamic parameter set comprising displacement deviation caused by main shaft axial deformation, thermal deformation quantity analyzed based on interlayer temperature gradient, and frequency spectrum features reflecting tool-material coupling vibration, dynamically synthesizing a spatial trajectory correction instruction and a motion parameter optimization instruction according to the multi-dimensional compensation vector and current drilling process parameters, wherein the trajectory correction instruction is reconstructed through a B-spline curve to realize real-time interpolation of a path, a motion controller receives the spatial trajectory correction instruction and the motion parameter optimization instruction, at least three compensation iterations are completed in a single drilling cycle, a linear motor and a main shaft motor are driven to execute online precision compensation, and the compensation instruction execution delay is less than 2 ms.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a dynamic precision compensation control method based on an intelligent PCB drilling machine. Background Technology

[0002] Drilling quality directly affects part performance and assembly reliability. Currently, drilling processing faces significant technical challenges. Displacement deviation caused by spindle axial deformation, thermal deformation caused by interlayer temperature gradient, and spectral characteristics generated by tool-material coupled vibration interact to cause unstable hole diameter accuracy and decreased hole wall quality. The interaction error between thermomechanical coupling effect and vibration is particularly prominent in high-speed machining. Traditional static compensation is difficult to adapt to dynamic changes, making it difficult to meet high-standard requirements for machining accuracy.

[0003] Existing technology relies on tool wear prediction and dynamic balance correction, combined with visual inspection and genetic algorithm to optimize parameters. However, it focuses on blade processing and lacks a real-time decoupling and dynamic compensation mechanism for spindle deformation, thermal deformation and vibration in drilling, and cannot cope with the complex dynamic environment of composite material drilling.

[0004] The core issue addressed by this technology is how to decouple the interactive errors of spindle deformation, thermomechanical coupling effect, and tool-material vibration in real time during high-precision drilling, in order to achieve stable control of hole diameter accuracy and hole wall quality. Especially in the machining of multi-layer composite materials, the difference in thermal expansion coefficient and the dynamic changes of high-speed machining exacerbate the accumulation of errors. There is an urgent need for a method that can generate low-latency compensation commands in real time, while ensuring that multiple iterative corrections are completed within a single drilling cycle, in order to adapt to the trajectory deviation and vibration suppression requirements under complex machining conditions, thereby improving machining efficiency and part quality. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the purpose of this application is to provide a dynamic accuracy compensation control method based on an intelligent PCB drilling machine.

[0006] The dynamic precision compensation control method based on an intelligent PCB drilling machine described in this application includes the following steps:

[0007] S101. Real-time acquisition of dynamic parameter sets through a multi-source sensor group installed on the drill spindle;

[0008] S102, the dynamic parameter set includes displacement deviation caused by spindle axial deformation, thermal deformation based on interlayer temperature gradient analysis, and spectral characteristics reflecting tool-material coupled vibration;

[0009] S103. Input the dynamic parameter set into the online incremental training machine learning model. The model uses a neural network architecture with embedded physical constraints to simultaneously decouple the interaction error between the thermo-mechanical coupling effect and the vibration transmission chain, and generates a multi-dimensional compensation vector in real time. The multi-dimensional compensation vector includes: a spatial trajectory offset to offset the deformation accumulation error, a dynamic correction coefficient for the spindle speed to suppress resonance, and an axial feed acceleration compensation value to optimize the hole wall quality.

[0010] S104. Based on the multidimensional compensation vector and the current drilling process parameters (including plate stacking structure, target hole diameter and hole depth), dynamically synthesize spatial trajectory correction instructions and motion parameter optimization instructions, wherein the trajectory correction instructions reconstruct the real-time interpolation path through B-spline curves.

[0011] S105. The motion controller receives the spatial trajectory correction instruction and motion parameter optimization instruction, completes at least 3 compensation iterations within a single drilling cycle, and drives the linear motor and spindle motor to perform online accuracy compensation, wherein the execution delay of the compensation instruction is less than 2ms.

[0012] Further, in step S101, the real-time acquisition of dynamic parameter sets through a multi-source sensor group installed on the drill spindle includes:

[0013] Multi-source sensors collect spindle speed, vibration, and temperature data in real time to form a dynamic stream. After denoising and standardization, the first data stream is generated, laying the foundation for analysis. If the parameters exceed the threshold, the anomaly detection algorithm is triggered to mark the abnormal state.

[0014] Based on the abnormal state label set, the first data stream is classified by support vector machine to accurately identify fault modes. Fault trend features are extracted by combining time series analysis and input into long short-term memory network to predict future parameter changes. By comparing the deviation between the predicted data stream and the real-time data stream, the system operating status is dynamically evaluated.

[0015] Further, in step S102, the dynamic parameter set includes displacement deviation caused by spindle axial deformation, thermal deformation based on interlayer temperature gradient analysis, and spectral characteristics reflecting tool-material coupled vibration, including:

[0016] Multi-source sensors collect displacement deviation, thermal deformation, and spectral characteristics in real time to generate a data stream. After filtering, denoising, and normalization, a standardized data stream is formed to build a high-quality analysis basis. When any parameter exceeds the threshold, the isolated forest algorithm generates an abnormal feature set.

[0017] This feature set drives time series analysis to extract dynamic features of faults, forming a fault feature set. Support vector machine accurately classifies fault types based on this feature set. Based on the fault type and feature set, time series analysis extracts trend features.

[0018] The future parameter prediction stream is generated by trend feature set, and the deviation analysis between real-time data and prediction stream drives state assessment. The assessment results feed back into threshold setting and feature extraction strategy.

[0019] Further, in step S103, the dynamic parameter set is input into an online incrementally trained machine learning model. This model, through a neural network architecture embedded with physical constraints, synchronously decouples the interaction error between the thermomechanical coupling effect and the vibration transmission chain, and generates a multi-dimensional compensation vector in real time. This multi-dimensional compensation vector includes: a spatial trajectory offset to offset the cumulative deformation error, a dynamic correction coefficient for the spindle speed to suppress resonance, and an axial feed acceleration compensation value to optimize the hole wall quality, including:

[0020] Multi-source sensors collect displacement, temperature gradient, and vibration spectrum in real time to generate raw data streams. After wavelet denoising, threshold monitoring is triggered. When the parameters exceed the limit, principal component analysis is initiated to extract the core feature set and construct a feature basis under physical constraints.

[0021] Based on the feature set, an online incremental neural network is embedded in the thermodynamic-vibration coupling model to decouple the interaction error between the heat engine effect and the vibration transmission chain, and generate a multi-dimensional compensation vector that includes spatial trajectory offset, rotational speed correction and acceleration compensation.

[0022] By extracting trend features through time series analysis of the compensated parameter set, a future parameter prediction stream is generated, and the deviation between real-time data and the prediction stream drives the state assessment.

[0023] Further, in step S104, the dynamic synthesis of spatial trajectory correction instructions and motion parameter optimization instructions based on the multidimensional compensation vector and current drilling process parameters (including plate laminate structure, target hole diameter, and hole depth), wherein the trajectory correction instructions reconstruct the real-time interpolation path through B-spline curves, including:

[0024] Based on multidimensional compensation vectors and process parameters such as plate laminate structure, pore size, and pore depth, key features are extracted through principal component analysis to drive B-spline curve reconstruction and generate real-time path interpolation data stream.

[0025] When the interpolation parameters deviate from the threshold, the Kalman filter is triggered to optimize the path in real time. Based on this, the spatial trajectory correction instruction set is calculated. Combined with the process parameters, the motion trend features are extracted by time series analysis to generate the motion parameter optimization instruction set and update the interpolation data stream in real time.

[0026] Further, in step S105, the motion controller receives the spatial trajectory correction command and the motion parameter optimization command, completes at least three compensation iterations within a single drilling cycle, and drives the linear motor and spindle motor to perform online accuracy compensation, wherein the execution delay of the compensation command is less than 2ms, including:

[0027] The motion controller parses spatial trajectory correction instructions and motion parameter optimization instructions, extracts trajectory coordinate sets and parameter adjustment sets, and divides the drilling cycle into multiple time periods based on time series segmentation to generate time period segmentation data.

[0028] Calculate the drive signal set for the linear motor and spindle motor for each time period;

[0029] When the signal set exceeds the threshold, the sliding window filter optimizes the signal value in real time. The controller generates a compensation command based on this, drives the actuator to complete online accuracy compensation, generates compensation execution data, and calculates the command execution delay by comparing timestamps.

[0030] The dynamic precision compensation control method based on an intelligent PCB drilling machine described in this application has the advantage of addressing the problems in complex multilayer board drilling scenarios, such as deformation accumulation error, resonance, and hole wall quality degradation caused by the interaction of thermomechanical coupling effect and vibration transmission chain. It integrates a multi-source sensor group on the drilling spindle to collect dynamic parameter sets in real time, including displacement deviation, thermal deformation, and spectral characteristics. These parameters are then input into a neural network model embedded with physical constraints to simultaneously decouple thermomechanical coupling and vibration error, generating a multi-dimensional compensation vector. Based on this vector, the invention dynamically synthesizes spatial trajectory correction instructions and motion parameter optimization instructions. The real-time interpolation path is reconstructed using B-spline curves, driving the motion controller to complete at least three compensation iterations within a single drilling cycle. The instruction execution delay is less than 2 milliseconds, achieving precise compensation for spatial trajectory offset, rotational speed correction coefficient, and feed acceleration. This invention significantly improves the accuracy and hole wall quality of drilling complex multilayer boards, reduces resonance effects, and ensures efficient and stable processing performance. Attached Figure Description

[0031] Figure 1 This is the flow chart of the dynamic precision compensation control method based on an intelligent PCB drilling machine described in this application. Figure 1 ;

[0032] Figure 2 This is the flow chart of the dynamic precision compensation control method based on an intelligent PCB drilling machine described in this application. Figure 2 . Detailed Implementation

[0033] like Figures 1-2 As shown, the dynamic precision compensation control method based on an intelligent PCB drilling machine described in this application includes:

[0034] like Figures 1-2 As shown, S101, dynamic parameter sets are collected in real time through a multi-source sensor group installed on the drill spindle.

[0035] Furthermore, in step S101, the rotational speed, vibration, and temperature data of the drill spindle are collected through a multi-source sensor group to generate a dynamic data stream;

[0036] The dynamic data stream is denoised and standardized using data preprocessing methods to obtain the first data stream;

[0037] If any parameter in the first data stream exceeds the preset threshold, the abnormal state is determined by the anomaly detection algorithm, and an anomaly tag set is obtained.

[0038] Based on the anomaly label set, the first data stream is classified using the support vector machine algorithm to determine the fault mode;

[0039] The dynamic changing trends of failure modes are extracted using time series analysis methods to obtain a trend feature set;

[0040] By using a trend feature set and a long short-term memory network, future parameter changes are predicted to obtain a predictive data stream.

[0041] Based on the deviation between the predicted data stream and the real-time data stream, the system operating status is determined, and the status assessment result is obtained.

[0042] Specifically, in step S101, a set of dynamic parameters is collected in real time by a multi-source sensor group installed on the drill spindle. The specific implementation method includes sensor data acquisition, data preprocessing, feature extraction and analysis.

[0043] First, an accelerometer, a temperature sensor, and a torque sensor are selected to collect spindle vibration frequency, temperature, and torque data, respectively. The sampling frequency is set to 1000Hz to ensure the capture of high-frequency dynamic signals.

[0044] For example, an accelerometer collects spindle vibration and outputs a signal range of ±50g; a temperature sensor measures the spindle operating temperature in the range of 0-150℃; and a torque sensor records torque changes from 0-500Nm. The data is transmitted to an industrial computer in real time at 16-bit resolution via a high-speed data acquisition card.

[0045] Next, the collected data is preprocessed. The time-domain signal is converted into a frequency-domain signal using the Fast Fourier Transform (FFT) algorithm, and the main frequency component of the vibration signal is extracted. For example, if the main frequency is in the range of 200-300Hz, the stability of the spindle operation is judged. Temperature data is filtered by moving average to remove noise. The filtering window is set to 10 sampling points to remove outliers (such as temperature changes exceeding 5°C / s). Torque data is processed by wavelet transform to remove high-frequency interference. The Daubechies wavelet basis is selected, and the decomposition level is 5. Low-frequency signals are retained for load analysis.

[0046] Subsequently, the feature extraction stage extracts the root mean square (RMS) value from the vibration signal, calculated using the following formula:

[0047] Where n represents the total number of sampling points, x i This represents the vibration acceleration value at the i-th sampling point;

[0048] For example, an RMS value exceeding 2g indicates abnormal spindle vibration; temperature feature extraction calculates the average temperature over 30 seconds; if it exceeds 100℃, an overheat warning is triggered; torque feature analysis calculates the peak torque; if it exceeds 400Nm, it indicates excessive load.

[0049] Finally, the analysis process employs the Support Vector Machine (SVM) algorithm, training a model based on extracted features. The kernel function is the Radial Basis Function (RBF), with parameter C set to 1.0 and γ set to 0.1. The spindle status is classified as normal, slightly abnormal, or severely abnormal, achieving an accuracy rate of over 95%. All data is transmitted in real-time to a cloud database via industrial Ethernet. Combined with historical data, trend analysis is performed to predict spindle lifespan. For example, if the vibration RMS value increases by 10% per week, it is predicted that the spindle needs maintenance within 30 days. The above method forms a closed loop through sensors, algorithms, and analysis to ensure real-time monitoring and optimization of the spindle's operating status.

[0050] like Figures 1-2 As shown in S102, the dynamic parameter set includes displacement deviation caused by spindle axial deformation, thermal deformation based on interlayer temperature gradient analysis, and spectral characteristics reflecting tool-material coupled vibration.

[0051] Furthermore, in step S102, the displacement deviation, thermal deformation and spectral characteristics of the spindle are collected in real time by a multi-source sensor group to generate a real-time data stream;

[0052] The real-time data stream is denoised using a filtering method, and a standardized data stream is generated using a normalization method.

[0053] If any parameter of displacement deviation, thermal deformation or spectral characteristics in the standardized data stream exceeds a preset threshold, the isolated forest algorithm is used to determine the abnormal state and obtain an abnormal feature set.

[0054] Based on the abnormal feature set, dynamic change features related to the fault are extracted through time series analysis to obtain the fault feature set;

[0055] The support vector machine algorithm is used to classify the fault feature set and determine the fault type;

[0056] By using fault types and fault feature sets, time series analysis methods are employed to extract changing trends and obtain trend feature sets.

[0057] Based on the trend feature set, a predictive data stream of future parameter changes is generated to determine the system's operating status and obtain the status assessment result.

[0058] Specifically, in step S102, by installing a high-precision displacement sensor, an infrared thermometer, and an acoustic sensor on the drill spindle, the displacement deviation caused by the axial deformation of the spindle, the thermal deformation caused by the interlayer temperature gradient, and the spectral characteristic data of the tool-material coupled vibration are collected in real time.

[0059] The displacement sensor acquires the axial displacement of the spindle at a sampling frequency of 2000Hz, with a range of ±0.5mm and a resolution of 0.001mm. The data is transmitted to the edge computing device through a 24-bit resolution data acquisition module. The infrared thermometer monitors the temperature gradient of the contact surface between the spindle and the bearing, with a range of -20 to 200°C, an accuracy of ±0.5℃, and a sampling frequency of 500Hz.

[0060] Acoustic sensors capture high-frequency acoustic signals of tool-material coupled vibrations in the range of 10-50kHz. Data is transmitted at 16-bit resolution. In the data preprocessing stage, displacement data is smoothed for noise using a Kalman filter algorithm with a filter gain of 0.8. Outliers with displacement abrupt changes exceeding 0.01mm / s are removed. Temperature gradient data is processed by a first-order low-pass filter with a cutoff frequency of 10Hz. The temperature difference between adjacent measurement points is calculated. For example, a difference exceeding 2℃ / mm is considered a risk of thermal deformation. The acoustic signal is converted to the time-frequency domain using a short-time Fourier transform (STFT) with a window length of 1024 points and an overlap rate of 50%. Energy spectrum features in the 10-30kHz frequency band are extracted.

[0061] In the feature extraction stage, the peak deviation of the displacement data is calculated using the following formula:

[0062] Where D max Indicates the peak displacement deviation, x i Let represent the instantaneous displacement value of the i-th sampling point, and n represent the total number of sampling points;

[0063] If the peak value exceeds 0.1 mm, it indicates significant axial deformation;

[0064] The thermal deformation is calculated by integrating the temperature gradient, using the following formula:

[0065] Where H represents thermal deformation, L represents the bearing contact surface length, L0 represents the reference length, ΔT represents the position temperature change, and dx represents the length micro-element;

[0066] The integration interval is the bearing contact surface length. A thermal deformation exceeding 0.05mm triggers an early warning. The peak energy of the acoustic signal spectrum is extracted. If the energy exceeds 0.02J near 20kHz, it indicates abnormal tool vibration.

[0067] The analysis process uses the random forest algorithm, based on the extracted displacement, thermal deformation and spectral features, with a tree depth of 10, a number of trees of 100, and the classification axis status as stable, slightly biased or severely biased.

[0068] Data is transmitted to the cloud via a 5G network. Combined with historical data analysis, the weekly growth rate of displacement deviation is calculated. For example, if the growth rate is 0.005 mm / week, it is estimated that the equipment needs to be calibrated within 60 days. All processing is automatically executed through edge computing devices, forming a real-time monitoring closed loop.

[0069] like Figures 1-2 As shown in S103, the dynamic parameter set is input into the online incremental training machine learning model. This model uses a neural network architecture with embedded physical constraints to simultaneously decouple the interaction error between the thermo-mechanical coupling effect and the vibration transmission chain, and generates a multi-dimensional compensation vector in real time. The multi-dimensional compensation vector includes: a spatial trajectory offset for offsetting deformation accumulation error, a dynamic correction coefficient for the spindle speed to suppress resonance, and an axial feed acceleration compensation value to optimize the hole wall quality.

[0070] Furthermore, in step S103, a set of dynamic parameters, including spindle displacement, temperature gradient and vibration spectrum, is collected in real time by a multi-source sensor group to generate a raw data stream;

[0071] The original data stream is denoised using wavelet transform to obtain a denoised data stream. If any parameter in the denoised data stream exceeds a preset threshold, principal component analysis is used to extract the main features to obtain a feature set.

[0072] Based on the feature set, an online incrementally trained neural network model is adopted, physical constraints are embedded, the interaction error between the thermo-mechanical coupling effect and the vibration transmission chain is decoupled, and a multi-dimensional compensation vector is generated.

[0073] By using multidimensional compensation vectors, the spatial trajectory offset, spindle speed correction coefficient, and axial feed acceleration compensation value are calculated to obtain the compensation parameter set.

[0074] Based on the compensation parameter set, time series analysis is used to extract the parameter change trend to obtain the trend feature set.

[0075] By using trend feature sets, a predictive data stream of future parameter changes is generated to determine the system's operating status and obtain status assessment results.

[0076] Specifically, in step S103, the dynamic parameter set is input into the online incrementally trained machine learning model, which adopts a neural network architecture with embedded physical constraints and is processed by a high-performance edge computing device to simultaneously decouple the interaction error between the thermomechanical coupling effect and the vibration transmission chain, generating a multi-dimensional compensation vector. The dynamic parameter set includes spindle displacement deviation, temperature distribution data and vibration spectrum characteristics. The data is acquired at a frequency of 1000Hz through a high-bandwidth interface, the displacement deviation range is ±0.3mm, the resolution is 0.002mm, the temperature distribution range is 0 to 150℃, the accuracy is ±0.3℃, and the vibration spectrum covers 5-40kHz.

[0077] The neural network model uses a Long Short-Term Memory (LSTM) network with 128 hidden layers and a learning rate of 0.001. Physical constraints are embedded in the loss function using the coefficient of thermal expansion (0.000012 / ℃) and the material stiffness matrix to ensure decoupling accuracy.

[0078] The decoupling process first involves wavelet transform denoising of the displacement deviation, using the Daubechies4 wavelet basis and a decomposition level of 5, eliminating noise energy below 0.001 mm. 2 The components are used to generate a smooth displacement sequence. The temperature distribution data is extracted using a two-dimensional convolutional neural network to extract spatial gradient features. The convolutional kernel size is 3x3 and the stride is 1. The maximum temperature gradient is calculated. If it exceeds 1.5℃ / mm, it is marked as having a significant thermal effect.

[0079] The vibration spectrum was analyzed using Fast Fourier Transform (FFT) with a window length of 2048 points. The power spectral density of the 15-25kHz frequency band was extracted. Vibration anomalies were considered when the peak energy exceeded 0.015J. The decoupled features were input into the online incremental training module. The model updated the weights every minute. The incremental data batch size was 64. The parameters were adjusted based on the gradient descent optimization algorithm.

[0080] The multidimensional compensation vector includes the formula for calculating the spatial trajectory offset.

[0081] Where S represents the spatial trajectory offset, Δx i This represents the displacement deviation in the i-th direction, where n represents the spatial dimension. Adjustments are triggered when the offset exceeds 0.08 mm, along with a dynamic correction coefficient for the spindle speed (based on the peak vibration frequency, with an adjustment range of ±5%; if the frequency is close to 18 kHz, the coefficient is set to 0.95) and an axial feed acceleration compensation value (based on the hole wall roughness prediction model; when the target roughness Ra < 0.8 μm, the acceleration is adjusted to 1.2 mm / s²). 2 All calculations are performed on the edge device with millisecond-level latency, and the compensation vector is transmitted to the CNC system in real time via industrial Ethernet to form closed-loop control logic.

[0082] like Figures 1-2 As shown, in step S104, based on the multidimensional compensation vector and the current drilling process parameters (including the plate stack structure, target hole diameter and hole depth), a spatial trajectory correction command and a motion parameter optimization command are dynamically synthesized. The trajectory correction command reconstructs the real-time interpolation path through a B-spline curve.

[0083] Furthermore, in step S104, key features are extracted using principal component analysis through multidimensional compensation vectors and drilling process parameters, including the plate laminate structure, target hole diameter, and hole depth, to obtain a feature set;

[0084] Based on the feature set, a B-spline curve reconstruction method is used to generate a real-time path interpolation data stream;

[0085] If any parameter in the real-time path interpolation data stream deviates from the preset threshold, the interpolation path is adjusted by Kalman filtering to obtain an optimized path data stream.

[0086] Based on the optimized path data flow, spatial trajectory correction instructions are calculated to obtain the trajectory correction instruction set;

[0087] By using the trajectory correction instruction set and combining drilling process parameters, time series analysis is used to extract the trend of motion parameter changes and obtain a trend feature set.

[0088] Based on the trend feature set, motion parameter optimization instructions are generated to obtain the optimization instruction set;

[0089] By optimizing the instruction set and updating the real-time path interpolation data stream, the final corrected data stream is obtained.

[0090] Specifically, in step S104, based on the multidimensional compensation vector and the drilling process parameters, the system first receives the plate laminate structure data.

[0091] For example, a four-layer carbon fiber composite material and a two-layer aluminum alloy (thicknesses of 2.5 mm and 1.0 mm respectively) are used to construct a target hole with a diameter of 6.0 mm and a depth of 8.0 mm. The process is handled by a high-performance edge computing device. The trajectory correction command uses a B-spline curve to reconstruct the path, and a multi-dimensional compensation vector is input (including a spatial offset of 0.05 mm, a spindle speed correction coefficient of 0.98, and a feed acceleration compensation value of 1.1 mm / s). 2 Combined with the plate laminate stiffness matrix (carbon fiber stiffness modulus 150GPa, aluminum alloy 70GPa), the B-spline curve is calculated with 5th order basis function and 10 control points. The node vector is evenly distributed in [0,1]. The offset is fitted by the least squares method to generate a smooth interpolation path. The path deviation is controlled within ±0.01mm.

[0092] The motion parameter optimization command is based on the target hole diameter and hole depth, calculating the spindle speed (initially set at 8000 rpm) and feed rate (initially 2.0 mm / s). If the rate of change of the stiffness of the laminated interface exceeds 20 GPa / mm, the system adjusts the feed rate using the Newton-Raphson iteration method with an iteration step size of 0.1 mm / s. The convergence condition is that the hole wall roughness Ra < 0.7 μm. The speed optimization uses power spectrum analysis to extract the 10-20 kHz vibration frequency band during drilling. If the peak power density exceeds 0.01 J, the speed is reduced by 2% until the power density drops below 0.008 J. All commands are transmitted to the CNC system via industrial Ethernet at a frequency of 500 Hz to update the path and parameters in real time, ensuring machining accuracy.

[0093] In the logic chain, the sheet metal stack data determines the stiffness constraint, which affects the distribution of control points on the B-spline curve. The aperture and depth guide the initial values ​​of motion parameters, and vibration analysis further optimizes the rotational speed and feed rate, forming a closed-loop control.

[0094] like Figures 1-2 As shown, in step S105, the motion controller receives the spatial trajectory correction command and motion parameter optimization command, completes at least 3 compensation iterations within a single drilling cycle, and drives the linear motor and spindle motor to perform online accuracy compensation, wherein the execution delay of the compensation command is less than 2ms.

[0095] Further, in step S105, the motion controller obtains the spatial trajectory correction instruction and motion parameter optimization instruction from the above, parses the instruction data stream, extracts the trajectory coordinate set and parameter adjustment set, and obtains the instruction parsing result;

[0096] Based on the command parsing results, the drilling cycle is divided into at least three time periods using a time series segmentation method, generating time period segmentation data;

[0097] By dividing the data into time periods, the motion controller calculates the real-time drive signals of the linear motor and the spindle motor for each time period to obtain a set of drive signals;

[0098] If any signal in the driving signal set exceeds the preset threshold, the signal value is adjusted by a sliding window filtering method to obtain an optimized signal set.

[0099] Based on the optimized signal set, the motion controller generates real-time compensation commands to drive the linear motor and spindle motor to perform online accuracy compensation and obtain compensation execution data.

[0100] By compensating for the execution data and using a timestamp comparison method, the instruction execution delay is calculated, and it is determined whether the delay is lower than a preset threshold to obtain the delay judgment result.

[0101] If the delay exceeds the preset threshold based on the delay judgment result, the transmission frequency of the drive signal set is adjusted, an updated signal set is generated, and the compensation instruction generation step is repeated.

[0102] Specifically, in step S105, after the motion controller receives the spatial trajectory correction instruction and the motion parameter optimization instruction, it first parses the instruction data through the high-performance DSP chip. The instruction includes a spatial offset of 0.03mm, a feed speed correction value of 1.5mm / s, and a spindle speed adjustment coefficient of 0.95. The data is encapsulated in JSON format, and the parsing time is controlled within 0.5ms. After parsing, the controller calls the pre-stored plate stiffness model.

[0103] For example, a three-layer glass fiber composite material (stiffness modulus 120 GPa, thickness 2.0 mm) combined with a single-layer titanium alloy (stiffness modulus 110 GPa, thickness 1.2 mm) is used. The stiffness distribution at the drilling point is calculated by finite element analysis to generate a stiffness gradient matrix with a resolution of 0.01 mm. Based on this, the controller uses a cubic spline interpolation algorithm to reconstruct the motion trajectory, sets a fourth-order basis function, controls 8 control points, and the node vectors are uniformly distributed in the interval [0,1], with the interpolation error controlled within ±0.005 mm.

[0104] The trajectory correction command is converted into a linear motor drive signal through the inverse kinematics model. The motor response time is less than 0.8ms and the drive current accuracy reaches 0.1A. The motion parameter optimization command initializes the spindle speed to 10000rpm and the feed rate to 1.8mm / s based on the target hole diameter of 5.5mm and hole depth of 7.0mm.

[0105] The controller analyzes the spindle vibration signal through Fourier transform and extracts the 5-15kHz frequency band. If the vibration amplitude exceeds 0.02mm, the feed speed is adjusted using the gradient descent method with a step size of 0.05mm / s. The convergence condition is that the vibration amplitude is below 0.015mm.

[0106] In each compensation iteration, the controller uses Kalman filtering to fuse sensor data (accelerometer resolution 0.01 m / s²). 2 The system predicts the motor's motion state, generates new drive signals, and controls the total time for three iterations within 1.5ms. The instructions are transmitted to the linear motor and spindle motor via industrial Ethernet at a frequency of 1000Hz, ensuring that the execution delay is less than 2ms.

[0107] In the logic chain, the stiffness gradient matrix guides the trajectory interpolation algorithm, the aperture and depth initialize the motion parameters, vibration analysis drives parameter optimization, and Kalman filtering ensures iterative convergence, forming a closed-loop control.

[0108] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. A dynamic precision compensation control method based on an intelligent PCB drilling machine, characterized in that, include: A set of dynamic parameters is collected in real time by a multi-source sensor group installed on the drill spindle. The set of dynamic parameters includes displacement deviation caused by spindle axial deformation, thermal deformation based on interlayer temperature gradient analysis, and spectral characteristics reflecting tool-material coupled vibration. The multi-source sensor group includes at least a displacement sensor for acquiring the displacement deviation, a temperature sensor array for acquiring the temperature gradient, and an acceleration sensor for acquiring the vibration spectrum. The dynamic parameter set is input into an online incrementally trained machine learning model. The machine learning model is a long short-term memory network with embedded physical constraints, including the coefficient of thermal expansion and the material stiffness matrix. The machine learning model simultaneously decouples the interaction error between the thermo-mechanical coupling effect and the vibration transmission chain, and generates a multi-dimensional compensation vector in real time. Based on the multidimensional compensation vector and the current drilling process parameters, including the plate laminate structure, target hole diameter and hole depth, and based on the stiffness modulus distribution of the plate laminate structure, the real-time interpolation path is reconstructed through B-spline curves, and spatial trajectory correction instructions are dynamically synthesized. Based on the power spectral density analysis of the vibration spectrum and the hole wall roughness prediction model, motion parameter optimization instructions are dynamically synthesized. The motion controller receives the spatial trajectory correction command and motion parameter optimization command, completes at least 3 compensation iterations within a single drilling cycle, and drives the linear motor and spindle motor to perform online accuracy compensation, wherein the entire delay from the generation of the compensation command to its execution is less than 2ms.

2. The dynamic precision compensation control method based on an intelligent PCB drilling machine according to claim 1, characterized in that, The multi-source sensor group includes an acceleration sensor, a temperature sensor, and a torque sensor; The real-time data acquisition process includes: The sensor data is converted into a frequency domain signal using Fast Fourier Transform, and then denoised using moving average filtering and wavelet transform. Extract the root mean square feature of the vibration signal, the average temperature feature of the temperature data, and the peak torque feature of the torque data.

3. The dynamic precision compensation control method based on an intelligent PCB drilling machine according to claim 1, characterized in that, The multi-source sensor group also includes a displacement sensor, an infrared thermometer, and an acoustic sensor; Displacement deviation is smoothed using a Kalman filter algorithm, thermal deformation is analyzed by calculating the temperature difference between adjacent measuring points, and spectral characteristics are extracted using a short-time Fourier transform to obtain the time-frequency domain energy spectrum.

4. The dynamic precision compensation control method based on an intelligent PCB drilling machine according to claim 1, characterized in that, The machine learning model is a long short-term memory network with 128 hidden layers. The physical constraints include the thermal expansion coefficient and material stiffness matrix embedded in the loss function. The weights are updated once per minute during online incremental training, and the incremental data batch size is 64.

5. The dynamic precision compensation control method based on an intelligent PCB drilling machine according to claim 1, characterized in that, The process of dynamically synthesizing spatial trajectory correction instructions includes: Based on the stiffness modulus distribution of the laminated plate structure, the path is reconstructed using B-spline curves and fifth-order basis functions. When the interpolation parameters deviate from the threshold, Kalman filtering is triggered to optimize the path in real time.

6. The dynamic precision compensation control method based on an intelligent PCB drilling machine according to claim 1, characterized in that, The process of generating the motion parameter optimization instructions includes: The spindle speed and feed rate are initialized based on the target hole diameter and hole depth. The spindle speed is dynamically adjusted based on the power spectral density analysis of the vibration spectrum. The axial feed acceleration is iteratively optimized through the hole wall roughness prediction model.

7. The dynamic precision compensation control method based on an intelligent PCB drilling machine according to claim 1, characterized in that, The compensation iteration process includes: dividing the drilling cycle into multiple time periods through time series segmentation, calculating the drive signal set of the linear motor and spindle motor for each time period, and using sliding window filtering to optimize the signal value in real time when the signal set exceeds the threshold.

Images