Error compensation method and system for wire cutting device machining trajectory

Abstract

The application relates to the technical field of wire cutting, in particular to a wire cutting equipment machining track error compensation method and system, which comprises the following steps: synchronously collecting a multi-source deflection signal and a path curvature prior, and performing stamping storage to form a multi-source observation sequence with an arc length value as an index; performing dimension transformation to construct a two-dimensional path channel observation matrix, determining a judgment threshold value based on a curvature statistical distribution, and performing encryption sampling; introducing a curvature prior driven response supplement factor to improve filtering transfer logic, and reconstructing and separating to obtain a real deflection estimation sequence; and performing reverse projection decomposition on the real deflection estimation sequence to form a feedforward compensation increment, and superimposing the feedforward compensation increment on a servo command to complete compensation. The application effectively solves the problem of algorithm response lag at curvature mutation, and obviously improves the machining precision of special-shaped contours.

Description

Technical Field

[0001] This application relates to the field of wire EDM processing technology, specifically to an error compensation method and system for the processing trajectory of wire EDM equipment. Background Technology

[0002] Wire EDM is an irreplaceable core process in precision mold manufacturing. Its machining accuracy directly determines the contour reproduction of the mold cavity. During the EDM process, the electrode wire is subjected to the combined effects of discharge reaction force, wire tension fluctuation and coolant impact, which will produce a continuously changing mechanical deflection in the direction perpendicular to the wire feed. The deviation between this deflection and the actual cutting trajectory constitutes the most important source of systematic error in wire EDM.

[0003] To address the aforementioned flexure error, existing technologies generally employ a static filtering compensation scheme based on sampling at equal time intervals. This involves acquiring the electrode wire offset sensor signal at a fixed time step, smoothing it with a low-pass filter, extracting the flexure estimate, and then superimposing the compensation increment onto the servo command to correct the trajectory deviation.

[0004] However, the smoothing parameters of the filter remain static throughout the machining process and cannot detect changes in the geometric features of the machining path. When machining straight sections, static smoothing can effectively suppress discharge disturbance noise. However, when the toolpath enters a region of sudden curvature change such as a corner or a small arc, the electrode wire deflection will produce a real step jump as the path direction changes drastically. The static filter treats this real mechanical deflection step and discharge disturbance noise equally and smooths and suppresses it, resulting in response lag and amplitude attenuation of the compensation amount at the curvature change point. This leads to contour distortion phenomena such as corner collapse and small arc deformation at high feed rates, making it difficult to meet the high dynamic response requirements of precision mold machining for trajectory error compensation in irregular contour sections. Summary of the Invention

[0005] To address the existing technical problem of delayed response and resulting contour distortion at points of abrupt change in machining curvature in traditional time-domain filtering, this application provides an error compensation method and system for the machining trajectory of wire EDM equipment.

[0006] In a first aspect, this application provides an error compensation method for the processing trajectory of a wire EDM machine, comprising: synchronously acquiring multi-source flexural signals and path curvature a priori, and storing them by stamping them with arc length as an index to form a multi-source observation sequence; performing dimensional transformation on the multi-source observation sequence to construct a two-dimensional path channel observation matrix, and determining a curvature abrupt change judgment threshold based on the statistical distribution of path curvature to perform encrypted sampling; based on the path channel observation matrix, improving the filtering state transition logic by introducing a response supplementary factor driven by path curvature a priori, and reconstructing and separating to obtain a true flexural estimation sequence indexed by arc length; decomposing the true flexural estimation sequence by back projection along the path normal vector to form a feedforward compensation increment and superimposing it on the servo command to complete the error compensation.

[0007] By reorganizing multi-source observation data and reconstructing the dynamic response structure in the path dimension, the inherent defect of slow response at curvature change points in traditional time-domain smoothing is effectively solved, and the trajectory following accuracy of corners and small arc segments in the wire cutting process is significantly improved.

[0008] Preferably, the synchronous acquisition of multi-source flexural signals and path curvature a priori, and the stamping and storage of these signals with the arc length value as an index to form a multi-source observation sequence, includes: synchronously acquiring electrode wire offset, discharge voltage, feed current, and wire tension reflecting the instantaneous state of the electrode wire, and extracting the path curvature a priori of the current tool position from the CNC interpolation kernel; the extraction operation and sensor sampling share the same clock reference, wherein the clock reference is provided by the interpolation cycle interrupt signal of the real-time kernel of the CNC system, and the sensor sampling interrupt and the interpolator output are both connected to the same rising edge of the interrupt signal, ensuring that both complete data latching at the same time at the hardware level; and uniformly stamping and storing all raw signals with the current arc length value as an index to form a multi-source observation sequence indexed by arc length.

[0009] Preferably, the step of performing dimensional transformation on the multi-source observation sequence to construct a two-dimensional path channel observation matrix includes: slicing and rearranging the original one-dimensional time series multi-source observation sequence according to arc length values ​​to construct the two-dimensional path channel observation matrix with the path arc length position as the row index and each sensor signal channel as the column index.

[0010] Preferably, the step of determining the curvature mutation judgment threshold based on the statistical distribution of path curvature and performing encrypted sampling includes: performing discrete statistics on the curvature sequence of all tool positions within the current CNC program segment, and calculating the mean curvature and the standard deviation of curvature; using the mean curvature plus three times the standard deviation of curvature as the curvature mutation judgment threshold; when the curvature of the current tool position is detected to exceed the curvature mutation judgment threshold, sending an encrypted trigger event to the sampling front end and increasing the sampling rate in the current arc length neighborhood.

[0011] By adaptively determining abrupt events and triggering encrypted sampling based on the statistical distribution of numerical control paths, the problem of losing true details of mechanical deflection due to insufficient spatial density in the curvature step region during equal-time interval sampling is effectively solved.

[0012] Preferably, after performing encrypted sampling to determine the curvature change judgment threshold based on the statistical distribution of path curvature, the method further includes: when the calculated curvature change judgment threshold is lower than the minimum resolvable curvature of the CNC system, determining that the current program segment is a straight line condition, and directly setting the subsequent response supplementary factor to zero.

[0013] Preferably, the step of improving the filter state transition logic by introducing a response supplementary factor driven by path curvature a priori, and reconstructing and separating to obtain the true deflection estimation sequence indexed by arc length, includes: migrating the filter recursive formula to the arc length domain, and replacing the single static gain at the gain position with an additive bivariate structure of the straight segment base gain and the response supplementary factor; when the curvature of the current tool position does not exceed the curvature mutation judgment threshold, the response supplementary factor is set to zero; when the curvature of the current tool position exceeds the curvature mutation judgment threshold, the response supplementary factor is calculated by multiplying the excess ratio of the curvature relative to the threshold with the straight segment base gain, using the excess ratio as the driving quantity.

[0014] By introducing a dynamic response structure that adaptively adjusts the path curvature a priori into the filter gain, the problem of accurately separating and restoring mechanical hysteresis deflection and high-frequency discharge disturbance within a single closed-loop framework is effectively solved.

[0015] Preferably, the basic gain of the straight segment is derived by acquiring the baseline offset observation sequence under the no-load wire calibration condition and calculating the ratio of the corresponding noise variance to the deflection process variance; when the original offset observation at a certain arc length position exceeds the physical range defined by the maximum allowable deflection of the machine tool electrode wire, it is determined as an invalid reading and the observation update term in the recursive formula of that arc length position is bypassed, and the estimated value of the previous arc length position is used to propagate forward.

[0016] Preferably, the step of back-projecting and decomposing the real deflection estimation sequence along the path normal vector to form a feedforward compensation increment and superimposing it on the servo command to complete the error compensation includes: back-projecting the real deflection estimation sequence along the path normal vector direction of the current tool position to decompose it to the machine tool feed coordinate system to form a feedforward compensation increment; and applying the feedforward compensation increment to the servo command output by the CNC interpolator in a superimposed manner.

[0017] By injecting the actual deflection estimate into the servo command link outside the position loop in a feedforward manner, the problem of high-frequency oscillation induced by the servo bandwidth dependence of traditional post-feedback compensation is effectively solved, and the command following lag phenomenon under high feed speed is significantly improved.

[0018] Preferably, before applying the feedforward compensation increment to the servo command output by the CNC interpolator in a superimposed manner, the method further includes: applying a path domain slope limit to the feedforward compensation increment so that the change in the feedforward compensation increment between adjacent arc length steps does not exceed the displacement increment corresponding to the maximum allowable acceleration of the current servo axis.

[0019] In a second aspect, this application provides an error compensation system for the machining trajectory of a wire EDM machine, comprising: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the error compensation method for the machining trajectory of a wire EDM machine as described in any one of the first aspects is implemented.

[0020] This application constructs a spatiotemporally unified underlying data base by deeply integrating multi-source sensor data with CNC interpolation geometric information within the path arc length domain. Relying on the curvature event-driven adaptive encrypted sampling and dynamic response filtering framework, it achieves efficient separation of mechanical hysteresis deflection and discharge disturbance, ensuring the integrity and authenticity of feature information of each segment in the processing of irregular contours from the source.

[0021] The feedforward error compensation implemented in this application eliminates the dependence of traditional feedback control on servo bandwidth, enabling the compensation increment to be precisely and synchronously applied to the corresponding spatial position of the command trajectory. While ensuring smooth and jitter-free machining of straight segments, it also endows the system with the ability to respond quickly in curvature change sections, significantly improving distortion phenomena such as corner collapse and blunting, and providing a solid foundation for high-quality and efficient machining of precision molds. Attached Figure Description

[0022] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart illustrating an error compensation method for the machining trajectory of a wire EDM machine according to the present invention.

[0023] Figure 2 This diagram schematically illustrates a performance comparison between the error compensation method and system for the processing trajectory of a wire EDM machine according to the present invention and existing technologies. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] This invention discloses an error compensation method and system for the machining trajectory of a wire EDM machine, referring to... Figure 1 This includes steps S1-S4: S1. Multi-source flexural signals and path curvature are synchronously acquired.

[0027] In an optional embodiment, electrode wire offset displacement sensors are installed at the upper and lower guide wheel supports of the wire EDM machine. The installation reference for both sensors is the guide wheel axis, ensuring that the collected offset directly reflects the actual deflection displacement of the electrode wire relative to the theoretical wire feeding reference. Simultaneously, the system synchronously acquires the discharge gap voltage signal, the servo axis feed current signal, and the wire feeding tension signal. These three signals, together with the displacement sensor output, constitute a multi-source observation combination reflecting the instantaneous state of the electrode wire. All four raw signals are output as time-series data according to the machine tool's high-frequency sampling cycle. After front-end anti-aliasing filtering, they are sent to the data receiving buffer of the upper-level compensation controller to prevent spectral aliasing caused by high-frequency components of the discharge spark above the Nyquist frequency during subsequent processing.

[0028] Furthermore, while receiving data from multiple sensor sources, the compensation controller extracts the path tangent vector and path curvature of the current tool position from the CNC interpolation kernel in real time. The path curvature is parameterized using the arc length parameter along the machining path as the independent variable. This extraction operation shares the same clock reference with the sensor sampling interrupt. This clock reference is uniformly provided by the interpolation cycle interrupt signal of the CNC system's real-time kernel. Both the sensor sampling interrupt and the interpolator output are attached to the same rising edge of this interrupt signal, ensuring that both complete data latching at the same time at the hardware level. This ensures that the sensor readings corresponding to each sampling moment are strictly aligned in time with the tool position geometry information output by the interpolator. Since the servo axis undergoes acceleration and deceleration adjustments during wire EDM, if the sensor sampling and the interpolator output belong to different clock domains, a misalignment between the time axis and the path axis will occur during the acceleration and deceleration phases. This will cause the sensor readings at the same arc length position to not correspond correctly with the curvature a priori. By sharing the interpolation cycle interrupt signal as a unified trigger source, this misalignment is eliminated from the data source.

[0029] Specifically, after completing the aforementioned synchronous acquisition, the controller stamps and stores the electrode wire offset, discharge voltage, feed current, wire tension, and path curvature a priori data of each sampling point, indexed by the current arc length, forming a set of multi-source observation sequences indexed by arc length. This stamping operation is the foundation of all subsequent signal reconstruction steps. Its core lies in converting the original time-axis-based sensor data stream into a spatial data stream indexed by the path arc length, so that the deflection observation is no longer an isolated time series, but a spatial sequence bound to the path geometric information. Subsequent filtering, separation, and compensation operations are all performed in the arc length coordinate system, rather than in the time coordinate system.

[0030] For example, the sampling cycle can be set to once per millisecond, corresponding to an arc length resolution on the order of micrometers at the machine tool feed speed, which can meet the spatial resolution requirements for trajectory error compensation in precision mold processing. The cutoff frequency of the front-end anti-aliasing filter can be objectively set according to the frequency band difference between the discharge frequency and the mechanical deflection response frequency, preserving mechanical deflection information while suppressing high-frequency components of the discharge.

[0031] In this way, the controller obtains a set of physically traceable multi-source arc length index observation sequences, providing a complete and spatiotemporally aligned underlying data foundation for subsequent dimension reconstruction and curvature event-driven processing, effectively avoiding the problem of mismatch between curvature a priori and deflection observations caused by misalignment of time axis and path axis.

[0032] S2, Path-dimensional slicing and curvature event-driven encrypted sampling.

[0033] In an optional embodiment, the controller performs a time-to-path dimensional transformation on the obtained arc-length indexed multi-source observation sequence. Specifically, the original one-dimensional time series is sliced ​​and rearranged according to the arc-length value to construct a two-dimensional path channel observation matrix with the path arc-length position as the row index and each sensor signal channel as the column index. After this rearrangement, the electrode wire offset, discharge voltage, feed current, wire tension, and path curvature prior at the same arc-length position are naturally aligned in the same row, which facilitates subsequent steps to jointly analyze and process multiple signals on the same arc-length section without having to perform time alignment operations again during processing.

[0034] Furthermore, based on the dimensional transformation, the system introduces a curvature event-driven encrypted sampling method to address the inherently insufficient sampling density at curvature abrupt change points in equal-interval sampling. The system first performs discrete statistics on the curvature sequences of all tool positions within the current CNC program segment, calculates the mean and standard deviation of curvature, and then determines the curvature abrupt change threshold according to the following formula:

[0035] In the formula, This is the average curvature of all tool points in the current CNC program segment. This represents the standard deviation of the corresponding curvature sequence. Based on the 3-sigma criterion of path curvature distribution, tool positions falling outside the mean plus three times the standard deviation are statistically considered significant curvature abrupt changes. The threshold is objectively determined by the curvature distribution of the current machining program segment.

[0036] Specifically, when the CNC interpolator detects that the curvature of the current tool position exceeds the calculated abrupt change threshold during real-time interpolation, it sends an encryption trigger event to the sampling front-end. Upon receiving the trigger event, the sampling front-end locally increases the sampling rate in the neighborhood of the current arc length, thereby maintaining the basic sampling rate on straight segments and automatically increasing sampling density at corners and small arc segments. This adaptive sampling density allocation method enables the deflection signal in curvature abrupt change segments to have higher spatial resolution in the path matrix, avoiding the problem of losing true deflection step details due to undersampling at abrupt change points in equal-interval sampling.

[0037] Next, to address the degradation condition where the mean curvature and standard deviation of curvature approach zero in straight or near-straight line program segments, thus causing the mutation threshold to approach zero, the system implements a physical fallback mechanism. When the calculated mutation threshold is lower than the minimum resolvable curvature of the CNC system, the system automatically determines the current program segment as a straight line condition, sets the curvature response supplementary factor in subsequent S3 to zero, and degenerates the filter into a steady-state Kalman filter, ensuring stable operation of the system under any program segment type.

[0038] For example, in a typical CNC program containing straight line segments and small circular arc transition segments, the curvature of the tool position point in the straight line segment is close to zero, while the curvature of the small circular arc segment is significantly higher than that in the straight line segment. The statistical distribution of the curvature sequence shows obvious bimodal characteristics. At this time, the threshold calculated according to the 3 sigma criterion can effectively identify the tool position point in the small circular arc segment as a mutation point and trigger encrypted sampling, while the tool position point in the straight line segment maintains the basic sampling rate. The overall sampling resources are concentrated in the curvature mutation segment.

[0039] Thus, through dimensional transformation and curvature event-driven encrypted sampling, the path channel observation matrix has sufficient spatial resolution in curvature abrupt change sections and maintains a reasonable sampling density in straight line sections. This provides input data with a reasonable information density distribution for subsequent curvature a priori-driven deflection component separation and reconstruction, eliminating the information loss problem caused by under-sampling at abrupt change points in equal-interval sampling from a data perspective.

[0040] S3, the deflection and disturbance components are separated and reconstructed by the prior knowledge-driven method.

[0041] In an optional embodiment, for the constructed path channel observation matrix, the controller uses the existing state update formula of the standard one-dimensional Kalman filter as an improved benchmark, replacing the static gain with a dynamic response structure driven by the path curve a priori, enabling the filter to simultaneously perform the functions of steady-state smoothing of straight segments and rapid following of abrupt transitions within the same recursive framework. As the improved benchmark, the existing state update formula of the standard one-dimensional Kalman filter is as follows:

[0042] In the formula, For the current deflection estimate, For current observations, This is the Kalman gain; the gain in this formula It is determined solely by the static ratio of the observed noise variance to the process noise variance, and is completely independent of the path curvature. Therefore, it cannot quickly follow the actual deflection jump at corners. The filter will smoothly suppress both the actual mechanical deflection step and the discharge disturbance noise, resulting in a response hysteresis and amplitude attenuation of the compensation amount at the curvature abrupt change point.

[0043] Furthermore, this step improves the existing formula by shifting the recursive expression from the time domain to the arc length domain, and introduces a response supplement factor driven by the path curvature a priori at the gain position. The improved recursive expression for deflection estimation is as follows:

[0044] In the formula, arc length Estimate the actual deflection of the electrode wire at that location. For the original offset observation at the corresponding arc length, The path step size, The base gain for the straight line segment. This is a response supplement factor driven by the path curvature a priori. Compared to the single static gain in the original formula. In comparison, the improvement lies in using add The additive bi-term structure replaces the original single-term gain, causing the gain to degenerate to a steady-state value in the straight segment and to adaptively increase with the a priori curve in the abrupt change segment.

[0045] Specifically, response supplementary factors The value of is given by the following relationship:

[0046] The logic for constructing this relation consists of two branches. When the curvature of the current blade point does not exceed the mutation determination threshold, the value within the parentheses is negative, and after a factorial operation... If the result is zero, the recurrence relation degenerates into This is the steady-state Kalman form, where the smoothing and noise reduction capabilities of straight line segments are fully preserved. When the curvature of the current blade location exceeds the mutation detection threshold, the proportion of curvature exceeding the threshold is used as the objective driving quantity. As the overshoot ratio increases linearly, the filter's tracking speed of the actual flexural step increases accordingly; the more significant the overshoot, the faster the tracking. As The multiplicative reference allows the response scale of the abrupt change segment to be naturally anchored above the steady-state scale of the straight segment, without the need to introduce independent new parameters.

[0047] Next, the basic gain in the improved recursive formula is... The method of determination will be explained. Instead of relying on manual empirical values, a baseline offset observation sequence without discharge interference is acquired under no-load wire EDM calibration conditions. The noise variance and deflection process variance of this sequence are calculated, and the ratio of these two variances is objectively derived. Essentially, this is equivalent to the steady-state Kalman gain under calibration conditions. This calibration process is automatically executed during the machine tool startup initialization phase. The parameters are stored in the compensation controller parameter area and can be directly called during subsequent processing without the need for readjustment before each processing cycle.

[0048] For example, a typical toolpath with a transition from a straight line to a small circular arc is substituted into the calculation: assuming the base gain under the calibration condition. The value is 0.2. After entering the small arc segment, if the ratio of the curvature of the current blade point to the mutation judgment threshold reaches 2.5, then the response supplementary factor is activated. At this time, the instantaneous response gain Compared to the significant amplification of 0.2 on a straight segment, the filter can quickly follow the actual jump in mechanical deflection at abrupt changes. Meanwhile, because the discharge spark disturbance does not rise synchronously with the path curvature a priori, the curvature a priori does not exceed the abrupt change judgment threshold at the time of the disturbance. Keep it at zero, the disturbance component is still... The corresponding steady-state smoothing suppression thus achieves the separation of the causes of mechanical hysteresis deflection components and discharge disturbance components within the same recursive framework.

[0049] Furthermore, to address potential temporary sensor signal failures or communication interruptions during the processing, the system is equipped with fault-tolerant handling mechanisms. This includes measures to mitigate the impact of changes in the original offset at a given arc length position. When the reading exceeds the physical range defined by the maximum allowable deflection of the machine tool electrode wire, it is determined to be an invalid reading. The observation update term in the recursive formula of the arc length position is bypassed, and the deflection estimate directly uses the estimated value of the previous arc length position to propagate forward. The update is restarted after the valid observation is restored to avoid invalid observations contaminating the deflection estimation sequence.

[0050] Thus, after the deflection component and disturbance component are separated and reconstructed by the curvature prior driving, the controller obtains a set of true deflection estimation sequences of the electrode wire indexed by arc length. This sequence is stable in the straight segment and responds quickly in the curvature change segment. It effectively distinguishes mechanical hysteresis deflection from discharge disturbance noise at the causal level, and significantly improves the problem of response lag and amplitude attenuation at corners in traditional time-domain filtering.

[0051] S4, Deflection estimation feedforward injection servo setpoint.

[0052] In an optional embodiment, the controller back-projects the already separated and reconstructed true deflection estimation sequence indexed by arc length along the path normal direction of the current tool position, decomposing the deflection estimate into the machine tool's X-axis and Y-axis feed coordinate systems to form a feedforward compensation increment strictly synchronized with the current tool position. The path normal vector is provided synchronously by the CNC interpolation kernel while outputting the tool position tangent vector, sharing the same arc length clock reference with the deflection estimation sequence. Therefore, the compensation increment is not spatially phase misaligned with the command trajectory, enabling the application of the correct direction correction at the correct path position.

[0053] Furthermore, the feedforward compensation increment is applied to the servo command output by the CNC interpolator in a superimposed manner, and the superposition is completed before being sent to the driver. The entire compensation link is located before the servo position loop. Feedforward injection does not occupy servo bandwidth, and the application of compensation amount does not depend on the response speed of the position loop. Therefore, there will be no compensation lag due to servo bandwidth limitations in high feed rates or large curvature sections.

[0054] Specifically, because the flexure estimation sequence is stable in straight segments and responds rapidly in segments with abrupt curvature changes, the feedforward compensation increment exhibits a natural characteristic of being stable in straight segments and rapidly following curves. In straight segments, With a zero flexural gain, the deflection estimate is smoothly updated using the steady-state Kalman gain, resulting in a gradual change in the feedforward increment and preventing the introduction of high-frequency jitter components into the servo commands. At corners and small arc segments, The servo axis is raised first by the curve, and the flexure estimation quickly follows the actual mechanical flexure step. The feedforward increment responds in a timely manner, so that the servo axis can obtain sufficient correction at the point of curvature change, effectively suppressing corner collapse and small arc deformation.

[0055] Next, to address the issue of instantaneous jumps in servo commands potentially caused by rapid increases in the feedforward increment at curvature abrupt change points, the system imposes a path domain slope limit on the feedforward increment. This means that between adjacent arc length steps, the change in the feedforward increment does not exceed the displacement increment corresponding to the maximum allowable acceleration of the current servo axis. When the rate of change of the curvature a priori-driven deflection estimate exceeds this limit, the feedforward increment gradually approaches the target value according to the slope limit, preventing rapid jumps in the deflection estimate from directly translating into step shocks in the servo command. This ensures corner following performance while maintaining the stable operation of the servo system.

[0056] Furthermore, the entire compensation closure is completed in the arc length dimension. Once the machining program finishes execution, the deflection correction of the entire path of the current part is completed, eliminating the need for additional tracking secondary finishing processes after machining. For multi-cutting processes, each cut independently executes the complete compensation process from S1 to S4, and the deflection estimation parameters for each cut are independent of each other to accommodate the differences in discharge energy and wire tension under different cutting numbers.

[0057] For example, for machining programs of irregular contours with multiple sharp corners and small arcs commonly found in precision molds, the system completes the statistical calculation of the curvature sequence of the entire program segment before machining, determines the threshold for abrupt change, and executes the complete compensation link from S1 to S4 in real time during machining. The feedforward increment is adaptively adjusted according to the curvature change of the tool position point, responds quickly at sharp corners and small arcs, and outputs smoothly in straight segments, without the need for manual intervention to adjust the compensation parameters throughout the process.

[0058] In this way, by injecting the deflection estimate into the servo setpoint in a feedforward manner, the compensation link effectively transforms the real deflection estimate obtained from the separation and reconstruction into the actual correction action of the servo axis without occupying the servo bandwidth or introducing additional hysteresis. This objectively suppresses the distortion of irregular contours such as corner collapse, small arc deformation, and blunting of sharp corners from the cause level, and has direct engineering value for improving the trajectory accuracy in the wire cutting processing of precision molds and irregular parts.

[0059] Figure 2 This diagram schematically illustrates a performance comparison between the error compensation method and system for the machining trajectory of a wire EDM machine according to the present invention and existing technologies. It shows the convergence state of trajectory errors at corners and small arc segments compared to existing static time-domain filtering methods, verifying the significant technical effect of the present solution in improving machining accuracy and suppressing corner collapse distortion.

[0060] This invention also discloses an error compensation system for the machining trajectory of a wire EDM machine, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an error compensation method for the machining trajectory of a wire EDM machine according to the present invention.

[0061] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for error compensation in the machining trajectory of a wire EDM machine, characterized in that, include: Simultaneously acquire multi-source flexural signals and path curvature priors, and store them using arc length values ​​as indexes to form a multi-source observation sequence; A two-dimensional path channel observation matrix is ​​constructed by performing a dimensional transformation on the multi-source observation sequence, and a curvature abrupt change judgment threshold is determined based on the statistical distribution of path curvature to perform encrypted sampling; Based on the path channel observation matrix, the filtering state transition logic is improved by introducing a response supplement factor driven by the path curvature prior, and the true deflection estimation sequence indexed by arc length is reconstructed and separated. The true deflection estimation sequence is decomposed by back-projection along the path normal vector to form a feedforward compensation increment, which is then superimposed on the servo command to complete the error compensation.

2. The error compensation method for the machining trajectory of a wire EDM machine according to claim 1, characterized in that, The simultaneous acquisition of multi-source flexural signals and path curvature priors, and the storage of these signals using arc length as an index, forms a multi-source observation sequence, including: The electrode wire offset, discharge voltage, feed current and wire tension are collected synchronously to reflect the instantaneous state of the electrode wire, and the path curve prior of the current tool position is extracted from the CNC interpolation kernel. The extraction operation and sensor sampling share the same clock reference; All raw signals are uniformly stamped and stored using the current arc length as an index, forming a multi-source observation sequence indexed by arc length.

3. The error compensation method for the machining trajectory of a wire EDM machine according to claim 1, characterized in that, The step of performing dimensional transformation on the multi-source observation sequence to construct a two-dimensional path channel observation matrix includes: slicing and rearranging the original one-dimensional time series multi-source observation sequence according to the arc length value to construct the two-dimensional path channel observation matrix with the path arc length position as the row index and each sensor signal channel as the column index.

4. The error compensation method for the machining trajectory of a wire EDM machine according to claim 1, characterized in that, The process of determining the curvature abrupt change threshold based on the statistical distribution of path curvature and performing encrypted sampling includes: Discretely statistically analyze the curvature sequence of all tool positions within the current CNC program segment, and calculate the mean curvature and standard deviation of curvature; The mean curvature plus three times the standard deviation of curvature is used as the threshold for determining curvature abrupt changes; When the curvature of the current tool point is detected to exceed the curvature change judgment threshold, an encryption trigger event is sent to the sampling front end to increase the sampling rate in the current arc length neighborhood.

5. The error compensation method for the machining trajectory of a wire EDM machine according to claim 4, characterized in that, After performing encrypted sampling to determine the curvature mutation judgment threshold based on the statistical distribution of path curvature, the method further includes: when the calculated curvature mutation judgment threshold is lower than the minimum resolvable curvature of the CNC system, the current program segment is determined to be a straight line condition, and the subsequent response supplementary factor is directly set to zero.

6. The error compensation method for the machining trajectory of a wire EDM machine according to claim 1, characterized in that, The improved filtering state transition logic, by introducing a response supplementary factor driven by path curvature a priori, reconstructs and separates the true deflection estimation sequence indexed by arc length, including: The filter recursion is shifted to the arc length domain, and at the gain position, the single static gain is replaced by an additive bivariate structure of the line segment base gain and the response supplement factor. When the curvature of the current blade position does not exceed the curvature change determination threshold, the response supplementary factor is zero; When the curvature of the current tool point exceeds the curvature abrupt change judgment threshold, the response supplement factor is calculated by multiplying the excess ratio of the curvature relative to the threshold with the basic gain of the straight line segment.

7. The error compensation method for the machining trajectory of a wire EDM machine according to claim 6, characterized in that, The basic gain of the straight segment is derived by acquiring the baseline offset observation sequence under the no-wire calibration condition and calculating the ratio of the corresponding noise variance to the flexural process variance. When the original offset observation at a certain arc length position exceeds the physical range defined by the maximum allowable deflection of the machine tool electrode wire, it is determined to be an invalid reading and the observation update term in the recursive formula of that arc length position is bypassed, and the estimated value of the previous arc length position is used to propagate forward.

8. The error compensation method for the machining trajectory of a wire EDM machine according to claim 1, characterized in that, The step of decomposing the true deflection estimation sequence by back projection along the path normal vector to form a feedforward compensation increment and superimposing it on the servo command to complete error compensation includes: The true deflection estimation sequence is back-projected along the path normal direction of the current tool position and decomposed into the machine tool feed coordinate system to form a feedforward compensation increment. The feedforward compensation increment is applied to the servo commands output by the CNC interpolator in a superimposed manner.

9. The error compensation method for the machining trajectory of a wire EDM machine according to claim 8, characterized in that, Before applying the feedforward compensation increment to the servo command output by the CNC interpolator in a superimposed manner, the method further includes: applying a path domain slope limit to the feedforward compensation increment so that the change in the feedforward compensation increment between adjacent arc length steps does not exceed the displacement increment corresponding to the maximum allowable acceleration of the current servo axis.

10. An error compensation system for the machining trajectory of a wire EDM machine, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement an error compensation method for the machining trajectory of a wire EDM device according to any one of claims 1-9.

Images