Probe compensation system and method for machining center Z-axis size instability

Abstract

The invention discloses a probe compensation system and method for machining center Z-axis size instability, and relates to the technical field of industrial machining control. A temperature sensor, a vibration sensor and a load sensor are installed at the key position of a Z-axis, and original measurement values of the temperature sensor and the vibration sensor are synchronously collected in the machining process; constructing a time sequence data set; extracting a chaos feature vector from the time sequence data, and predicting a Z-axis error value in real time based on a chaos time sequence model; chaotic feature extraction comprises the following processes of time delay determination, embedded dimension selection, phase space reconstruction and chaos parameter calculation. The chaos feature extraction and prediction model can effectively capture nonlinear dynamic errors which are difficult to deal with by a traditional method, and the error prediction accuracy is improved. A working condition correction factor and a timestamp are introduced in the compensation process for synchronous adjustment, it is ensured that a compensation value is matched with an actual machining working condition, and compensation real-time performance is enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial processing control, and in particular to a probe compensation system and method for unstable size of Z-axis of a machining center. BACKGROUND

[0002] As a high-precision machining equipment, the size precision of the Z-axis of the machining center directly determines the machining quality of the workpiece. In the actual machining process, the size stability of the Z-axis is reduced due to the thermal deformation of the screw and bearing, the wear of the mechanical parts caused by long-term operation, the change of the machining load and other factors, thereby causing the measurement error of the probe. If the error cannot be corrected in time, it will cause the size of the workpiece to be out of tolerance, increase the scrap rate, and affect the production efficiency and machining cost.

[0003] In the existing compensation methods for the unstable size of the Z-axis, the temperature compensation depends on the preset temperature-error corresponding relationship, which is difficult to cope with dynamic temperature changes under complex working conditions; the software calibration is usually an offline operation, which cannot respond to error fluctuations in the machining process in real time; and the linear correction is only suitable for simple linear error scenes, and has poor fitting effect on nonlinear errors caused by thermal deformation and mechanical wear. These methods generally lack real-time and adaptive ability, and it is difficult to meet the requirements of high-precision machining on the size stability of the Z-axis, so there is an urgent need for a technical solution that can accurately capture nonlinear error dynamics and real-time compensation. SUMMARY

[0004] To solve the above technical problems, the present application provides a probe compensation system and method for unstable size of Z-axis of a machining center. The following technical solutions are adopted:

[0005] A probe compensation method for unstable size of Z-axis of a machining center, comprising the following steps:

[0006] Step 1, install temperature sensors, vibration sensors and load sensors at key positions of the Z-axis, calibrate the accuracy of the temperature sensors, vibration sensors and load sensors, and build an initial error model of the Z-axis;

[0007] Step 2, synchronously collect the original measurement values of the temperature sensors and vibration sensors during the machining process, and build a time series data set after filtering and normalization processing;

[0008] Step 3, extract a chaotic feature vector from the time series data, and real-time predict the error value of the Z-axis based on the chaotic time series model; the chaotic feature extraction includes the following processes: time delay determination, embedding dimension selection, phase space reconstruction and chaotic parameter calculation;

[0009] Step 4, convert the predicted error into a compensation value, real-time adjust the measurement result of the probe and output to the CNC control system.

[0010] Optionally, in step 1, the temperature sensor is attached to both ends of the Z-axis lead screw, the outer ring of the ball bearing, and the motor housing; the vibration sensor is fixedly installed at the bottom of the Z-axis slide; and the load sensor is connected in series in the Z-axis motor power circuit.

[0011] Optionally, in the chaotic feature extraction process, the method for determining the time delay is: to use the autocorrelation function method to calculate the autocorrelation coefficient of the Z-axis position time series, and take the time interval when the coefficient first drops to 0.5 as the time delay τ.

[0012] The method for selecting the embedding dimension is as follows: using the false neighbor method, the embedding dimension is gradually increased from 1 to m. When the proportion of false neighbors is lower than a set proportion threshold, the optimal embedding dimension m is determined.

[0013] The method for reconstructing the phase space is: reconstructing the phase space of the Z-axis dynamic system based on τ and m;

[0014] The method for calculating chaotic parameters is as follows: the Wolf method is used to calculate the maximum Lyapunov exponent. If the maximum Lyapunov exponent is greater than a set threshold, chaotic characteristics are determined to exist. The correlation dimension is calculated by the GP algorithm to quantify the fractal structure of the system. The maximum Lyapunov exponent, correlation dimension, time delay τ, and embedding dimension m are recorded as chaotic feature vectors.

[0015] Optionally, in the phase space reconstruction method, the phase space of the Z-axis dynamic system is reconstructed using the following formula:

[0016] ;

[0017] in This is the preprocessed Z-axis position data. It refers to a specific point in time.

[0018] Alternatively, the Wolf method can be used to calculate the maximum Lyapunov exponent:

[0019] In the phase space matrix X, for each state point Find the nearest neighbor point with the smallest Euclidean distance. ;

[0020] Tracking the distance to neighboring points as it evolves over time ; It is the i-th state point in phase space. It is a state point The nearest point, It is the Euclidean distance between the nearest points at time t;

[0021] Maximum Lyapunov index : , It is the instantaneous Lyapunov exponent at time t. , It calculates the total duration. It is the time step, if A value greater than 0.005 indicates the presence of chaotic characteristics.

[0022] Optionally, a method for real-time prediction of Z-axis error values ​​based on a chaotic time series model is:

[0023] Determine the current state point at the prediction time. In the phase space matrix X, select the... For the K nearest neighboring points, establish a linear regression model for the neighboring points, solve the coefficients using the least squares method, output the local predicted value, construct a BP neural network, input the local predicted value, the chaotic feature vector and the normalized value of the current temperature sensor, and output the final error predicted value.

[0024] Optionally, in step 4, the method for converting the prediction error into a compensation value is:

[0025] ;in It is a compensation value. It is the basic compensation value. It is the operating condition correction factor. This is the final error prediction value. ;

[0026] These are the normalized temperature values, vibration values, and load data, respectively.

[0027] Optional, align probe raw measurements and The timestamp will be synchronized with the compensation value C. =M(t)+C calculates the compensated measured value.

[0028] Optionally, the CNC controller calculates the Z-axis feed correction value after parsing the data. ; The S machining command presets the machining dimensions, which will cause the CNC to... The data is sent to the Z-axis servo driver in real time to adjust the servo motor speed and feed rate.

[0029] A probe compensation system for unstable Z-axis dimensions in machining centers is disclosed, comprising a probe compensation method for such unstable Z-axis dimensions in machining centers. The probe compensation system includes a trigger-type probe, a multi-sensor module, a high-speed data acquisition card, an industrial computer, a communication module, a CNC controller, and a Z-axis servo driver. The trigger-type probe is used to acquire raw measurement values ​​of the Z-axis machining dimensions. The multi-sensor module includes a temperature sensor, a vibration sensor, and a load sensor. The data input terminal of the high-speed data acquisition card is communicatively connected to the trigger probe and the multi-sensor module, respectively. The industrial computer is communicatively connected to the data output terminal of the high-speed data acquisition card, storing time-series datasets, chaotic feature vectors, and prediction model parameters. The CNC controller is communicatively connected to the industrial computer via a communication module. The industrial computer will use the compensated measurement values... and Z-axis feed correction value Real-time transmission is sent to the CNC controller. The Z-axis servo drive is communicatively connected to the CNC controller and receives data. And adjust the servo motor speed and feed rate.

[0030] In summary, the present invention has at least one of the following beneficial technical effects:

[0031] This invention provides a probe compensation system and method for unstable Z-axis dimensions in machining centers. It acquires key Z-axis operating condition data through multi-sensor fusion and extracts nonlinear error features using chaotic time series analysis, achieving high-precision real-time prediction of Z-axis errors. The chaotic feature extraction and prediction model effectively captures nonlinear dynamic errors that traditional methods struggle to handle, improving error prediction accuracy. During the compensation process, a condition correction factor is introduced and adjusted synchronously with the timestamp to ensure the compensation value matches the actual machining conditions, enhancing real-time compensation. The system utilizes existing probes and standard sensors, eliminating the need for expensive additional hardware and balancing cost-effectiveness. The overall solution enables adaptive Z-axis dimension compensation, dynamically optimizing model parameters according to changing operating conditions, significantly improving the stability of Z-axis dimensions in machining centers, reducing workpiece scrap rates, and increasing machining efficiency and quality. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating a probe compensation method for unstable Z-axis dimensions in a machining center, as described in this invention.

[0033] Figure 2 This is a schematic diagram of the component connection principle of a probe compensation system for unstable Z-axis dimensions in a machining center according to the present invention.

[0034] Explanation of reference numerals in the attached diagram: 1. Trigger probe; 21. Temperature sensor; 22. Vibration sensor; 23. Load sensor; 3. High-speed data acquisition card; 4. Industrial computer; 5. Communication module; 6. CNC controller; 7. Z-axis servo drive. Detailed Implementation

[0035] The present invention will be further described in detail below with reference to the accompanying drawings.

[0036] This invention discloses a probe compensation system and method for unstable Z-axis dimensions in machining centers.

[0037] Reference Figure 1 and Figure 2 Example 1: A probe compensation method for unstable Z-axis dimensions in a machining center, comprising the following steps:

[0038] Step 1: Install temperature sensor 21, vibration sensor 22 and load sensor 23 at key positions on the Z-axis, perform accuracy calibration on temperature sensor 21, vibration sensor 22 and load sensor 23, and construct the initial error model of the Z-axis.

[0039] Step 2: During the processing, the raw measurement values ​​of temperature sensor 21 and vibration sensor 22 are collected simultaneously, and a time series dataset is constructed after filtering and normalization.

[0040] Step 3: Extract chaotic feature vectors from time series data and predict Z-axis error values ​​in real time based on chaotic time series models; chaotic feature extraction includes the following processes: time delay determination, embedding dimension selection, phase space reconstruction, and chaotic parameter calculation;

[0041] Step 4: Convert the prediction error into a compensation value, adjust the probe measurement results in real time, and output them to the CNC control system.

[0042] In Example 2, in step 1, the temperature sensor 21 is attached to both ends of the Z-axis lead screw, the outer ring of the ball bearing, and the motor housing; the vibration sensor 22 is fixedly installed at the bottom of the Z-axis slide; and the load sensor 23 is connected in series in the Z-axis motor power circuit.

[0043] By adopting the above technical solution, the temperature sensor 21, vibration sensor 22, and load sensor 23 are respectively positioned to address the main causes of Z-axis error: the temperature sensor 21 is installed on the outer ring of the ball bearings at both ends of the lead screw and on the motor housing, as these parts are key sources of Z-axis thermal deformation, and can collect the temperature changes of the core heat-generating components in real time; the vibration sensor 22 is fixed to the bottom of the Z-axis slide, and can directly acquire the vibration state during slide movement, reflecting the vibration error caused by mechanical wear or assembly clearance; the load sensor 23 is connected in series in the Z-axis motor power circuit, and can indirectly monitor the fluctuation of machining load through changes in motor current. Accuracy calibration of the sensors is performed to eliminate the sensors' own errors and ensure the reliability of the collected data; an initial Z-axis error model is constructed, and an initial error benchmark is established through reference data, providing a comparative reference for subsequent real-time error prediction.

[0044] Simultaneous acquisition of raw measurements from multiple sensors during processing ensures the temporal correlation between data such as temperature, vibration, and load and the Z-axis motion state, avoiding errors in analysis caused by data asynchrony. Filtering removes noise signals introduced by electromagnetic interference and mechanical vibration, ensuring the data reflects the true working conditions. Normalization eliminates dimensional differences between different sensors, placing temperature, vibration, and load data within the same numerical range, facilitating unified calculation and analysis during subsequent chaotic feature extraction. Constructing a time-series dataset integrates discrete sensor data in chronological order, forming a continuous dynamic data sequence that meets the data continuity requirements of chaotic time-series analysis.

[0045] Step 3 utilizes chaos theory to capture the nonlinear dynamics of Z-axis errors. Errors in the Z-axis, caused by factors such as thermal deformation, mechanical wear, and load variations, exhibit nonlinear dynamic characteristics that are difficult to accurately describe using traditional linear analysis methods. In the chaotic feature extraction process, the time delay is determined using the autocorrelation function method to find the optimal interval reflecting the temporal correlation of the data, ensuring the independence of the state vectors during phase space reconstruction. The embedding dimension is selected using the spurious nearest neighbor method to determine the minimum dimension that can completely reconstruct the Z-axis dynamic system, avoiding information loss due to insufficient dimension or increased computational load due to excessive dimension. Phase space reconstruction maps the one-dimensional time series to a high-dimensional phase space, restoring the dynamic evolution trajectory of the Z-axis error. Chaotic parameter calculation uses the maximum Lyapunov exponent to determine whether the system is in a chaotic state, and quantifies the fractal structure of the system through the correlation dimension. These two parameters, along with the time delay and embedding dimension, constitute a chaotic feature vector, comprehensively characterizing the nonlinear dynamics of the Z-axis error. Based on this feature vector, a chaotic time series model is constructed. Combining the accurate capture of short-term dynamics by local linear prediction with the correction capability of the BP neural network for nonlinear deviations, high-precision real-time prediction of Z-axis errors is achieved.

[0046] When converting the predicted error into a compensation value, a working condition correction factor is incorporated to dynamically adjust the compensation value in real time according to changes in temperature, vibration, and load, ensuring that the compensation value matches the actual machining conditions. Aligning the timestamps of the probe's original measurement value and the error prediction value avoids compensation lag caused by data time differences, ensuring real-time compensation. The adjusted compensated measurement value is output to the CNC control system. The CNC calculates the Z-axis feed correction value based on the preset machining dimensions and sends it to the servo driver to adjust the motor speed and feed, realizing closed-loop control from error prediction to machining parameter correction, ultimately ensuring stable Z-axis dimensional accuracy.

[0047] In Example 3, the method for determining the time delay in the chaotic feature extraction process is as follows: the autocorrelation function method is used to calculate the autocorrelation coefficient of the Z-axis position time series, and the time interval when the coefficient first drops to 0.5 is taken as the time delay τ.

[0048] The method for selecting the embedding dimension is as follows: using the false neighbor method, the embedding dimension is gradually increased from 1 to m. When the proportion of false neighbors is lower than a set proportion threshold, the optimal embedding dimension m is determined.

[0049] The method for reconstructing the phase space is: reconstructing the phase space of the Z-axis dynamic system based on τ and m;

[0050] The method for calculating chaotic parameters is as follows: the Wolf method is used to calculate the maximum Lyapunov exponent. If the maximum Lyapunov exponent is greater than a set threshold, chaotic characteristics are determined to exist. The correlation dimension is calculated by the GP algorithm to quantify the fractal structure of the system. The maximum Lyapunov exponent, correlation dimension, time delay τ, and embedding dimension m are recorded as chaotic feature vectors.

[0051] Example 4: In the phase space reconstruction method, the phase space of the Z-axis dynamic system is reconstructed using the following formula:

[0052] ;

[0053] in This is the preprocessed Z-axis position data. It refers to a specific point in time.

[0054] Example 5: The Wolf method for calculating the maximum Lyapunov exponent is as follows:

[0055] In the phase space matrix X, for each state point Find the nearest neighbor point with the smallest Euclidean distance. ;

[0056] Tracking the distance to neighboring points as it evolves over time ; It is the i-th state point in phase space. It is a state point The nearest point, It is the Euclidean distance between the nearest points at time t;

[0057] Maximum Lyapunov index : , It is the instantaneous Lyapunov exponent at time t. , It calculates the total duration. It is the time step, if A value greater than 0.005 indicates the presence of chaotic characteristics.

[0058] By employing the above technical solutions, the core of chaotic feature extraction is to restore the dynamic essence of the Z-axis error system through mathematical transformation. The time delay is determined using the autocorrelation function method. Since the autocorrelation coefficient of the Z-axis position time series reflects the correlation of data at different times, the time interval τ at which the coefficient first drops to 0.5 avoids redundancy between adjacent data while preserving effective temporal correlation, ensuring the independence and information integrity of the state vector in subsequent phase space reconstruction. The embedding dimension is selected using the spurious neighbor method, gradually increasing from low dimension to m. When the proportion of spurious neighbors is below a threshold, it indicates that dimension m can fully accommodate the system's dynamic information, avoiding the loss of system features due to excessively low dimension and preventing excessively high dimension from increasing the computational load. Phase space reconstruction, based on τ and m, maps the one-dimensional time series to a high-dimensional space, allowing the evolution trajectory of the Z-axis error to unfold in the high-dimensional space and restoring its nonlinear dynamic law. In the calculation of chaotic parameters, the maximum Lyapunov exponent is used to quantify the system's sensitivity to initial conditions using the Wolf method. An exponent greater than a threshold indicates the presence of chaotic characteristics in the system, requiring a chaotic model rather than a nonlinear model for description. The correlation dimension is used to quantify the fractal structural complexity of the system using the GP algorithm, reflecting the nonlinearity of the error dynamics. Recording these four parameters as chaotic feature vectors can comprehensively characterize the chaotic characteristics of the Z-axis error, providing core input for subsequent predictions.

[0059] The essence of the phase space reconstruction formula Xt = Xt, Xt+τ, ..., Xt+m−1τ is to transform a one-dimensional Z-axis position time series into a high-dimensional state vector. Xt represents the preprocessed Z-axis position data at time t. By introducing a time delay τ, the position data at times t, t+τ, ..., t+(m-1)τ are combined into the state vector Xt, making the system dynamics originally hidden in the one-dimensional sequence explicit in the high-dimensional phase space. Each state vector Xt corresponds to a point in the phase space, and the set of these points constitutes the evolution trajectory of the Z-axis error system. Through this trajectory, the dynamic changes of the error can be observed intuitively, laying the foundation for subsequent finding of neighboring points and analyzing the trajectory evolution trend, which is a prerequisite for chaotic time series prediction.

[0060] The maximum Lyapunov exponent is a core indicator for determining whether a system is chaotic and quantifying the degree of chaos. The Wolf method calculates it by tracking the evolution of neighboring points in the phase space. In the phase space matrix X, for each state point Xi, the nearest neighbor Xj with the smallest Euclidean distance is found. Since neighboring points in a chaotic system will exponentially separate over time, tracking the evolution of their distance dit reflects the system's expansion characteristics. The instantaneous Lyapunov exponent λit is obtained by calculating the exponential growth rate of the distance within adjacent time steps ∆t, i.e., λit = 1 / ∆t × ln(dit + ∆t / dit), reflecting the local expansion rate of the system at that moment. The maximum Lyapunov exponent λ is obtained by averaging the instantaneous exponents over the total time T. If λ is greater than 0.005, it indicates that the system is sensitive to initial conditions, and the error evolution exhibits chaotic characteristics, requiring a chaotic time series model for prediction; otherwise, it is a non-chaotic system, and a traditional model can be used. This method directly relates to the core characteristics of chaotic systems, providing a scientific basis for the selection of subsequent error prediction models.

[0061] Example 6: The method for real-time prediction of Z-axis error values ​​based on a chaotic time series model is as follows:

[0062] Determine the current state point at the prediction time. In the phase space matrix X, select the... A linear regression model is established for the K nearest neighboring points. The coefficients are solved by the least squares method, and the local predicted value is output. A BP neural network is constructed, and the local predicted value, chaotic feature vector and the normalized value of the current temperature sensor 21 are input. The BP neural network outputs the final error predicted value.

[0063] In Example 7, step 4, the method for converting the prediction error into a compensation value is as follows:

[0064] ;in It is a compensation value. It is the basic compensation value. It is the operating condition correction factor. This is the final error prediction value. ;

[0065] These are the normalized temperature values, vibration values, and load data, respectively.

[0066] Example 8: Aligning the original satellite measurements of the probe and The timestamp will be synchronized with the compensation value C. =M(t)+C calculates the compensated measured value.

[0067] By employing the above technical solution, the core of the prediction method is to leverage the short-term predictability and local linearity and global nonlinearity of chaotic systems, combined with multi-dimensional information, to achieve accurate error prediction. In chaotic systems, neighboring state points in phase space will evolve along similar trajectories in the short term; therefore, determining the current state point at the prediction time is crucial. Then, select with The evolution of the K nearest neighbors can be approximated by the evolution of these neighbors. The future trend of change. A linear regression model is established for the neighborhood points, and the coefficients are solved by the least squares method. This can quickly fit the linear evolution relationship of the local area and output the local predicted value, balancing computational efficiency and short-term prediction accuracy.

[0068] However, the Z-axis error exhibits overall nonlinear dynamics, and single local linear predictions are prone to bias. Therefore, a BP neural network is constructed for correction. The local predicted value, chaotic feature vector, and the normalized value of the current temperature sensor 21 are used as inputs. The chaotic feature vector comprehensively characterizes the nonlinear nature of the system, and the normalized temperature value reflects real-time key operating conditions. The combination of these three allows the neural network to capture nonlinear deviations not covered by local linear predictions. The network corrects errors through its nonlinear fitting ability, ultimately outputting a high-precision final error prediction value, achieving a complementary advantage between linear prediction and nonlinear correction.

[0069] The calculation logic for the compensation value is based on basic error matching plus dynamic adaptation to operating conditions, ensuring that the compensation value accurately corresponds to the actual error and machining conditions. Basic compensation value The final error prediction value y is obtained by multiplying it by the compensation gain K, which is directly related to the magnitude of the prediction error. This allows the compensation value to initially offset the predicted Z-axis error. The existence of K allows for flexible adjustment of the basic compensation strength to adapt to the error sensitivity of different machine tools.

[0070] The working condition correction factor α is designed to cope with the dynamic changes in working conditions during the machining process. middle, , , These correspond to the normalized temperature, vibration, and load data, respectively. These three types of data are the core causes of Z-axis error. When the temperature increases, vibration intensifies, or the load fluctuates, the α value increases accordingly, affecting the final compensation value. Synchronous adjustments are made to avoid insufficient or excessive compensation due to changes in operating conditions, ensuring that the compensation value always matches the real-time operating conditions and improving the adaptability and accuracy of compensation.

[0071] Establish an accurate correspondence between prediction error and measured values. (Original probe measurements) Compared with the final error prediction value Timestamp alignment is necessary because the Z-axis state changes dynamically during processing, and time differences can cause the compensation value to misalign with the actual measurement scenario, leading to compensation lag or deviation. By aligning the timestamps, we ensure that each original measurement value matches the prediction error at the corresponding moment, laying the foundation for accurate compensation.

[0072] Compensated measurement value pass The calculated value is added to C, and the compensation value is directly superimposed on the original measurement value to offset the influence of Z-axis error on the measurement result, thus obtaining accurate data reflecting the true machining dimensions. This calculation method is simple and efficient, and can quickly complete the measurement value correction, providing a reliable dimensional basis for subsequent adjustment commands issued by the CNC control system, ensuring that the compensation action is linked with the measurement and machining processes in real time.

[0073] Example 9: After parsing the data, the CNC controller 6 calculates the Z-axis feed correction value. ; The S machining command presets the machining dimensions, which will cause the CNC to... The data is sent to the Z-axis servo driver 7 in real time to adjust the servo motor speed and feed rate.

[0074] By adopting the above technical solution, the CNC controller 6, as the control core of the machining center, analyzes the received compensated measurement values. Its core is through formula Calculate the Z-axis feed correction value. Where S is the target machining dimension preset in the machining command. The difference between the two is the measured value that reflects the true processing state after error compensation. It directly quantifies the deviation between the current machining size and the target size, and clarifies the magnitude and direction of the feed rate adjustment.

[0075] To ensure real-time dimensional correction, the CNC controller needs to send the calculated ∆Z to the Z-axis servo drive in real time. The Z-axis servo drive, acting as the actuator, receives ∆Z and adjusts the servo motor's speed and feed rate to change the Z-axis motion parameters. When the deviation is positive, the feed rate is increased; when the deviation is negative, the feed rate is decreased, thus bringing the actual machining dimension of the Z-axis closer to the preset S.

[0076] The entire process achieves a complete closed loop from error prediction and measurement compensation to machining parameter adjustment, ensuring that dimensional deviations in each machining step can be corrected in a timely manner, ultimately guaranteeing the stability and accuracy of Z-axis machining dimensions.

[0077] Example 10: A probe compensation system for unstable Z-axis dimensions in a machining center. This system implements a probe compensation method for unstable Z-axis dimensions in a machining center. The probe compensation system includes a trigger-type probe 1, a multi-sensor module, a high-speed data acquisition card 3, an industrial computer 4, a communication module 5, a CNC controller 6, and a Z-axis servo driver 7. The trigger-type probe 1 is used to acquire the raw measurement values ​​of the Z-axis machining dimensions. The multi-sensor module includes a temperature sensor 21, a vibration sensor 22, and a load sensor 23. The data input terminal of the high-speed data acquisition card 3 is communicatively connected to the trigger probe 1 and the multi-sensor module, respectively. The industrial computer 4 is communicatively connected to the data output terminal of the high-speed data acquisition card 3, storing time-series datasets, chaotic feature vectors, and prediction model parameters. The CNC controller 6 is communicatively connected to the industrial computer 4 via a communication module 5. The industrial computer 4 will use the compensated measurement values... and Z-axis feed correction value Real-time transmission is sent to the CNC controller 6. The Z-axis servo driver 7 is communicatively connected to the CNC controller 6 and receives data. And adjust the servo motor speed and feed rate.

[0078] By adopting the above technical solution, the trigger probe 1 and the multi-sensor module constitute the sensing layer of the system. The trigger probe 1 directly collects the original measurement value of the Z-axis machining dimension, providing basic dimensional data for error analysis. The temperature sensor 21, vibration sensor 22, and load sensor 23 / 2 respectively collect relevant data on the load fluctuation of the Z-axis thermal deformation vibration state, comprehensively capturing the core working condition information that causes dimensional errors, and providing multi-source data support for subsequent error prediction.

[0079] The high-speed data acquisition card 3 plays a key bridging role in data transmission. Its data input end is connected to the trigger-type probe multi-sensor module, which can synchronously acquire various types of raw data, avoiding error analysis deviations caused by asynchronous data acquisition. At the same time, it transmits the integrated raw data to the industrial computer 4 in real time, ensuring the efficiency and integrity of data transmission.

[0080] Industrial computer 4, as the data processing core of the system, receives data transmitted by high-speed data acquisition card 3, stores time series dataset chaotic feature vectors and prediction model parameters, and runs core algorithms such as data preprocessing, chaotic feature extraction, error prediction, and compensation value calculation to transform the original data into compensated measurement values ​​and Z-axis feed correction values, thus completing the transformation from data to effective control parameters.

[0081] The communication module 5 establishes a real-time data transmission channel between the industrial computer 4 and the CNC controller 6, ensuring that the compensated measurement values ​​and Z-axis feed correction values ​​are transmitted quickly and accurately, avoiding data delays that affect the timeliness of compensation, and ensuring the real-time issuance of control commands.

[0082] The CNC controller 6 and the Z-axis servo drive 7 constitute the system's execution layer. After receiving the data transmitted by the industrial computer 4, the CNC controller 6 completes instruction parsing and logical judgment in combination with the machining requirements. The Z-axis servo drive 7, as the direct actuator, receives the Z-axis feed correction value issued by the CNC controller 6 and realizes dynamic correction of the Z-axis machining parameters by adjusting the servo motor speed and feed.

[0083] The entire system achieves a closed-loop control architecture through a collaborative process involving a perception layer for data acquisition, a processing layer for analysis and transformation, a transmission layer for real-time linkage, and an execution layer for precise adjustments. This ensures that every step of the compensation method can be efficiently implemented through the corresponding module, ultimately achieving stable control of the Z-axis dimension of the machining center.

[0084] The implementation principle of the present invention is illustrated below through specific embodiments:

[0085] Taking the machining of aluminum alloy box parts by a vertical machining center as an example, the required depth of the hole in the Z-axis direction of this part is 20mm, with an allowable error of ±0.002mm. The Z-axis travel of the machining center is 0-500mm. The following compensation method and system are used to achieve stable dimensional control.

[0086] Trigger probe 1 is a high-precision contact probe with a measurement resolution of 0.0001mm, installed at the end of the machining center spindle. In the multi-sensor module, temperature sensor 21 is a PT100 high-precision sensor, which is attached to both ends of the Z-axis lead screw, the outer ring of the ball bearing, and the motor housing with high-temperature resistant adhesive; vibration sensor 22 is a piezoelectric accelerometer, which is fixed to the bottom of the Z-axis slide with bolts; load sensor 23 is a Hall current sensor, connected in series in the Z-axis servo motor power supply circuit.

[0087] The high-speed data acquisition card 3 is a 16-bit resolution acquisition card, and its data input terminals are connected to the trigger probe 1 and the multi-sensor module via shielded cables. The industrial computer 4 uses an industrial-grade Core i7 processor, equipped with 64GB of memory and a 1TB solid-state drive, and pre-installed with data processing and algorithm execution software. The communication module 5 uses an EtherCAT bus communication unit, through which the industrial computer 4 establishes communication with the FANUC 0i-MF model CNC controller 6. The Z-axis servo drive 7 is a servo drive unit matched with the CNC controller and is directly connected to the Z-axis servo motor.

[0088] Step 1: After the sensors are installed, calibrate the temperature sensor 21 by placing it in a standard constant temperature chamber ranging from -10℃ to 80℃, recording the output values ​​at different temperatures, and completing the calibration. Vibration sensor 22 undergoes sensitivity calibration by inputting vibration signals from 5Hz to 1kHz through a standard vibration table. Load sensor 23 undergoes calibration by inputting current from 0.5A to 5A through a standard current source. Place a 20mm standard gauge block at the center of the worktable, and repeatedly measure 10 times along the Z-axis using the trigger probe 1. Take the difference between the average measured value and the size of the standard gauge block as the initial error. Combine this with 10 sets of position data collected during the Z-axis idle run to construct the Z-axis initial error model.

[0089] During step 2, the high-speed data acquisition card 4 synchronously acquires raw data from the temperature sensor 21, vibration sensor 22, load sensor 23, and trigger probe 1. Kalman filtering is used to remove noise from the acquired data, and then the temperature, vibration, and load data are normalized to the 0-1 range using the min-max method. All preprocessed data are integrated in chronological order to construct a time-series dataset, which is then stored in the industrial computer 4.

[0090] Step 3 extracts chaotic feature vectors from the time series data. The autocorrelation coefficient of the Z-axis position time series is calculated using the autocorrelation function method. When the coefficient first drops to 0.5, the time delay τ is determined to be 10ms. The embedding dimension is gradually increased using the false neighbor method. When the proportion of false neighbors is less than 5%, the optimal embedding dimension m is determined to be 4. Based on τ and m, the one-dimensional Z-axis position data is transformed into a high-dimensional state vector according to the phase space reconstruction formula, and the phase space is reconstructed. The maximum Lyapunov exponent is calculated using the Wolf method, yielding a value of 0.02 (1 / ms). A value greater than 0.005 indicates the presence of chaotic characteristics in the system. The correlation dimension is calculated to be 2.8 using the GP algorithm. The correlation dimension τ and m of the maximum Lyapunov exponent are recorded as chaotic feature vectors.

[0091] Determine the current state point Xk at the prediction time, select 20 nearest neighbor points to Xk in the phase space matrix, establish a linear regression model for the neighbor points, and obtain the local predicted value by solving the coefficients using the least squares method. Construct a BP neural network with 6 input neurons, 2 hidden layers (32 and 16 neurons respectively), and 1 output neuron. Input the chaotic feature vector of the local predicted value and the normalized value of the current temperature sensor 21 into the network, and output the final error predicted value.

[0092] Step 4: Calculate the compensation value according to the compensation value calculation formula. The compensation gain K is initially set to 1.0. The basic compensation value is calculated based on the final error prediction value. Then, the working condition correction factor α is calculated by combining the normalized temperature, vibration, and load data to finally obtain the compensation value C. The timestamps of the original measured value Mt and the final error prediction value of the trigger probe are extracted. Data with a time deviation exceeding 5ms are aligned using linear interpolation, and the compensated measured value Mcompt is calculated according to the formula.

[0093] Industrial computer 4 transmits Mcompt to CNC controller 6 via EtherCAT bus. CNC controller 6 calculates the Z-axis feed correction value ∆Z according to the formula. When Mcompt is 19.998mm, ∆Z is 0.002mm. CNC controller sends ∆Z to Z-axis servo driver 7 in real time. After receiving it, Z-axis servo driver 7 increases the feed of Z-axis servo motor, so that the actual machining depth of Z-axis approaches 20mm.

[0094] During the machining process, every 10 parts are machined, a standard calibration piece is measured using a trigger probe to verify the compensation effect and ensure that the Z-axis hole depth is stable within the range of 20mm±0.002mm, meeting the machining accuracy requirements of the parts.

[0095] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A probe compensation method for unstable Z-axis dimensions in machining centers, characterized in that, Includes the following steps: Step 1: Install temperature sensor (21), vibration sensor (22) and load sensor (23) at key positions on the Z-axis, perform accuracy calibration on temperature sensor (21), vibration sensor (22) and load sensor (23), and construct initial error model of Z-axis; Step 2: During the processing, the raw measurement values ​​of the temperature sensor (21) and vibration sensor (22) are collected simultaneously, and a time series dataset is constructed after filtering and normalization. Step 3: Extract chaotic feature vectors from the time series data, and predict the Z-axis error value in real time based on the chaotic time series model; the chaotic feature extraction includes the following process: Time delay determination, embedding dimension selection, phase space reconstruction, and chaotic parameter calculation; Step 4: Convert the prediction error into a compensation value, adjust the probe measurement results in real time, and output them to the CNC control system.

2. The probe compensation method for unstable Z-axis dimensions in a machining center according to claim 1, characterized in that, In step 1, the temperature sensor (21) is attached to both ends of the Z-axis lead screw, the outer ring of the ball bearing and the motor housing; the vibration sensor (22) is fixedly installed at the bottom of the Z-axis slide; and the load sensor (23) is connected in series to the Z-axis motor power circuit.

3. The probe compensation method for unstable Z-axis dimensions in a machining center according to claim 2, characterized in that, In the chaotic feature extraction process, the method for determining the time delay is as follows: the autocorrelation function method is used to calculate the autocorrelation coefficient of the time series of the Z-axis position, and the time interval when the coefficient first drops to 0.5 is taken as the time delay τ. The method for selecting the embedding dimension is as follows: using the false neighbor method, the embedding dimension is gradually increased from 1 to m. When the proportion of false neighbors is lower than a set proportion threshold, the optimal embedding dimension m is determined. The method for reconstructing the phase space is: reconstructing the phase space of the Z-axis dynamic system based on τ and m; The method for calculating chaotic parameters is as follows: the Wolf method is used to calculate the maximum Lyapunov exponent. If the maximum Lyapunov exponent is greater than a set threshold, chaotic characteristics are determined to exist. The correlation dimension is calculated by the GP algorithm to quantify the fractal structure of the system. The maximum Lyapunov exponent, correlation dimension, time delay τ, and embedding dimension m are recorded as chaotic feature vectors.

4. The probe compensation method for unstable Z-axis dimensions in a machining center according to claim 3, characterized in that: In the phase space reconstruction method, the phase space of the Z-axis dynamic system is reconstructed using the following formula: ； in This is the preprocessed Z-axis position data. It refers to a specific point in time.

5. A probe compensation method for unstable Z-axis dimensions in a machining center according to claim 4, characterized in that: The Wolf method for calculating the maximum Lyapunov exponent is as follows: In the phase space matrix X, for each state point Find the nearest neighbor point with the smallest Euclidean distance. ; Tracking the distance to neighboring points as it evolves over time ; It is the i-th state point in phase space. It is a state point The nearest point, It is the Euclidean distance between the nearest points at time t; Maximum Lyapunov index : , It is the instantaneous Lyapunov exponent at time t. , It calculates the total duration. It is the time step, if A value greater than 0.005 indicates the presence of chaotic characteristics.

6. A probe compensation method for unstable Z-axis dimensions in a machining center according to claim 5, characterized in that: The method for real-time prediction of Z-axis error values ​​based on chaotic time series models is: Determine the current state point at the prediction time. In the phase space matrix X, select the... The nearest K neighborhood points are used to establish a linear regression model for the neighborhood points. The coefficients are solved by the least squares method, and the local predicted value is output. A BP neural network is constructed, and the local predicted value, chaotic feature vector and normalized value of the current temperature sensor (21) are input. The BP neural network outputs the final error predicted value.

7. A probe compensation method for unstable Z-axis dimensions in a machining center according to claim 6, characterized in that: In step 4, the method for converting the prediction error into a compensation value is as follows: ;in It is a compensation value. It is the basic compensation value. It is the operating condition correction factor. This is the final error prediction value. ; These are the normalized temperature values, vibration values, and load data, respectively.

8. A probe compensation method for unstable Z-axis dimensions in a machining center according to claim 7, characterized in that: Align probe raw measurement values and The timestamp will be synchronized with the compensation value C. =M(t)+C calculates the compensated measured value.

9. A probe compensation method for unstable Z-axis dimensions in a machining center according to claim 8, characterized in that: The CNC controller (6) calculates the Z-axis feed correction value after parsing the data. ; The S machining command presets the machining dimensions, which will cause the CNC to... The data is sent to the Z-axis servo driver (7) in real time to adjust the servo motor speed and feed rate.

10. A probe compensation system for unstable Z-axis dimensions in machining centers, characterized in that: To implement the probe compensation method for unstable Z-axis dimensions in a machining center as described in claim 9, the probe compensation system includes a trigger probe (1), a multi-sensor module, a high-speed data acquisition card (3), an industrial computer (4), a communication module (5), a CNC controller (6), and a Z-axis servo driver (7). The trigger probe (1) is used to acquire the original measurement values ​​of the Z-axis machining dimensions. The multi-sensor module includes a temperature sensor (21), a vibration sensor (22), and a load sensor (23). The data input terminal of the high-speed data acquisition card (3) is connected to the trigger probe (1) and the multi-sensor module respectively. The industrial computer (4) is connected to the data output terminal of the high-speed data acquisition card (3) and stores time series datasets, chaotic feature vectors, and prediction model parameters. The CNC controller (6) is connected to the industrial computer (4) through the communication module (5). The industrial computer (4) will use the compensated measurement values... and Z-axis feed correction value The data is transmitted in real time to the CNC controller (6). The Z-axis servo drive (7) is communicatively connected to the CNC controller (6) and receives data. And adjust the servo motor speed and feed rate.

Images