Ultra-precision machining error compensation method based on multi-modal information fusion, medium and equipment

Abstract

The invention relates to the technical field of ultra-precision intelligent manufacturing, in particular to an ultra-precision machining error compensation method based on multi-modal information fusion, a medium and equipment. The method comprises the following steps: acquiring three-dimensional shape data of a workpiece through an in-situ measurement device, synchronously acquiring temperature field and vibration data, constructing a multi-modal space-time tensor, inputting a physical information neural network to decouple a static geometric error, a time-varying thermal drift error and a dynamic vibration error, and generating a four-dimensional correction tool path to realize reverse compensation. According to the method, physical constraints are embedded through a PINN model, the problem of false thermal drift misjudgment in a traditional method is solved, experiments show that the RMS error of free-form surface machining is reduced to 5 nm or below, and the precision is improved by 70% or above; and meanwhile, error decoupling is endowed with physical interpretability, the small sample generalization ability is enhanced, and a technical basis is provided for process optimization.

Description

Technical Field

[0001] This invention relates to the field of ultra-precision intelligent manufacturing technology, and in particular to ultra-precision machining error compensation methods, media and equipment based on multimodal information fusion. Background Technology

[0002] With the rapid development of aerospace, next-generation lithography, and biomedical imaging, the precision requirements for manufacturing freeform optical components (such as sinusoidal grids and microlens arrays) have increased to the nanometer level. Single-point diamond turning (SPDT) and slow-tool servo (STS) technologies are the mainstream processes for manufacturing such surfaces. However, inherent geometric errors of machine tools, thermal errors caused by ambient temperature fluctuations, and dynamic errors caused by high-frequency cutting forces have become bottlenecks limiting further improvements in machining precision.

[0003] Existing error compensation techniques are mainly divided into offline measurement compensation and in-situ measurement compensation. Offline measurement suffers from severe "repositioning error" (loss of reference), making it difficult to meet nanometer-level alignment requirements. While existing in-situ measurement techniques solve the reference problem, they have the following key technical drawbacks:

[0004] 1. Ignoring the time-varying nature of the measurement process: High-resolution in-situ scanning typically takes a long time (tens of minutes). During this period, the heat generated by the machine tool spindle rotation causes micron-level thermal drift in the structure. Existing techniques often assume that the measurement data is "static" and directly invert the measurement results for compensation. This leads to the misjudgment of "spurious thermal drift" generated during measurement as geometric errors on the workpiece surface, thus introducing new reverse errors during correction machining.

[0005] 2. Lack of physical mechanism support: Traditional error compensation models are mostly linear superpositions, lacking physical constraints on the error generation mechanism (thermal deformation, vibration), resulting in poor model generalization ability under small sample data and difficulty in coping with complex dynamic processing environments.

[0006] 3. Insufficient dynamic feasibility verification: The directly generated correction path may contain high-frequency components that exceed the machine tool servo bandwidth, which not only fails to compensate for errors but may also induce machine tool vibration.

[0007] Therefore, there is an urgent need for a new method that can distinguish between "true geometric error" and "time-varying thermal / dynamic error" and can compensate based on a mechanistic model. Summary of the Invention

[0008] To address one of the aforementioned technical problems, the present invention adopts the following technical solution:

[0009] According to one aspect of the present invention, a method for compensating for errors in ultra-precision machining based on multimodal information fusion is provided, the method comprising the following steps:

[0010] During the initial machining process, in-situ measuring devices integrated into the working space of the ultra-precision machine tool are used to acquire three-dimensional morphological data of the workpiece surface.

[0011] Temperature field data and vibration data during the initial machining process are collected synchronously by distributed temperature sensors and dynamic vibration sensors placed on the heat source parts and moving parts of the machine tool.

[0012] The three-dimensional topography data, temperature field data, and vibration data are aligned according to timestamps to construct a multimodal spatiotemporal tensor.

[0013] A multimodal spatiotemporal tensor is input into a physical information neural network to decouple and separate the total processing error into static geometric error, time-varying thermal drift error, and dynamic vibration error. The composite loss function of the physical information neural network is described. The following conditions must be met: .

[0014] in: This represents the total error of the network output and the mean square error of the measured data. The degree to which the thermal error components output by the network violate the heat conduction equation is calculated using automatic differentiation. The degree to which the dynamic error components output by the network violate the forced vibration equation is calculated using automatic differentiation. These are the weighting coefficients for the physical constraint terms.

[0015] Based on the static geometric error, a four-way forward toolpath containing spatial coordinates (x, y, z) and machining time t is generated to spatially compensate for the static geometric error.

[0016] According to a second aspect of the present invention, a non-transitory computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the above-described method for compensating for ultra-precision machining errors based on multimodal information fusion.

[0017] According to a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for compensating for ultra-precision machining errors based on multimodal information fusion.

[0018] This invention has at least one of the following beneficial effects:

[0019] First, this invention effectively solves the problem of "false thermal drift misjudgment" caused by ignoring the time-varying nature of the measurement process in traditional in-situ measurement compensation methods by constructing a multimodal spatiotemporal tensor and introducing a Physical Information Neural Network (PINN) to decouple the total machining error. In existing technologies, in-situ scanning is time-consuming, during which machine tool thermal deformation continues to occur. However, the measurement data is directly used for compensation as static geometric errors, thus incorrectly superimposing dynamic thermal drift into the correction path, causing reverse overcutting. In this invention, temperature field and vibration signals are simultaneously acquired and aligned with three-dimensional topography data by timestamp to form a spatiotemporal correlation tensor. This tensor is then input into a PINN model with constraints embedded in the partial differential equation (PDE) of heat conduction and the ordinary differential equation (ODE) of structural dynamics. This allows the network to fit the measured total error while forcing its output thermal error components to satisfy the physical laws of thermal diffusion and the vibration components to conform to the dynamic response characteristics. Thus, the true static geometric error can be accurately separated from the mixed measurement signals, avoiding the misinterpretation of time-varying thermal drift as a geometric defect for compensation. This fundamentally eliminates the systematic bias introduced by thermal deformation within the measurement window, significantly improving the accuracy and reliability of compensation. Experiments show that for free-form surface machining, the root mean square error (RMS) can be reduced to below 5 nm, achieving an accuracy improvement of over 70% compared to traditional in-situ measurement methods.

[0020] Secondly, the physical information neural network architecture employed in this invention endows the error decoupling process with clear physical interpretability and strong generalization ability, overcoming the shortcomings of traditional pure data-driven models that lack mechanistic support and are prone to overfitting under small sample sizes. Existing compensation methods mostly rely on empirical linear superposition or black-box deep learning models, whose internal mapping relationships lack physical basis and are difficult to distinguish the essential differences of error sources, especially when sensor data is sparse or noise interference is strong, resulting in instability. In this invention, the loss function of PINN not only includes data fitting terms... It also explicitly introduces physical residual terms calculated by automatic differentiation. and These correspond to the degree of violation of the heat conduction equation and the forced vibration equation, respectively. This mechanism of embedding prior physical knowledge into the network training process allows the model to reasonably allocate error components based on the fundamental laws of thermal inertia and structural dynamics, even with limited measurement samples, thus ensuring that the decoupling results conform to the real physical process. This feature not only enhances the robustness of the model in complex dynamic processing environments but also provides an interpretable technical basis for error tracing and process optimization. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the system framework for ultra-precision machining error compensation based on multimodal information fusion.

[0023] Figure 2 This is a schematic diagram of the installation structure of a multi-axis ultra-precision machine tool and an in-situ measurement probe.

[0024] Figure 3 This is a schematic diagram of a sinusoidal grating surface.

[0025] Figure 4 This is a schematic diagram of a spiral scanning path used for in-situ measurements.

[0026] Figure 5 Point cloud map measured by ISSM.

[0027] Figure 6 A schematic diagram illustrating the principle of spatiotemporal alignment and enhanced tensor construction for multimodal data.

[0028] Figure 7 This is a schematic diagram of the network topology of a Physical Information Neural Network (PINN).

[0029] Figure 8 This is a schematic diagram illustrating the effect of error decoupling.

[0030] Figure 9 A schematic diagram comparing the effects of kinematic constraint path optimization. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] To achieve the above objectives, such as Figure 1 As shown, this invention provides a system for ultra-precision machining error compensation based on multimodal information fusion, the system comprising:

[0033] A multi-axis ultra-precision machine tool, such as Figure 2As shown, the machine tool is equipped with a position servo controlled spindle (C-axis) and at least two linear motion axes (X-axis and Z-axis) to perform slow tool servo (STS) machining.

[0034] An in-situ measurement probe 1 is integrated into the machine tool's workspace to measure the surface morphology of the workpiece without unloading it from the machine tool fixture, thereby maintaining a unified coordinate system between machining and measurement operations.

[0035] The multi-physics sensing network includes a distributed temperature sensor array arranged on the machine tool spindle bearing housing, linear motor coils and key hot spots of the bed structure, as well as dynamic vibration sensors arranged on the tool clamping end and spindle rotor, for synchronously acquiring instantaneous temperature field data and dynamic vibration data of the machine tool throughout the entire machining and measurement process.

[0036] The spatiotemporal data synchronization module is used to align three-dimensional topographic point cloud data, temperature field data and dynamic vibration data on the time axis based on the machine tool's position feedback signal, and construct a multimodal spatiotemporal tensor containing spatial coordinates and physical field states.

[0037] The intelligent edge computing unit communicates with the machine tool and various sensors, and incorporates a Physical Information Neural Network (PINN). The computing unit is configured to align spatial geometric point cloud data with thermal and dynamic vibration data in the time domain to construct a multimodal spatiotemporal tensor. This multimodal spatiotemporal tensor is then input into the PINN. Using the heat conduction equation and structural dynamics equation embedded in the network loss function as constraints, the measured total surface error is decoupled into static geometric error components, time-varying thermal drift error components, and dynamic vibration error components. Based on the decoupled static geometric error components, a dynamically corrected toolpath is generated, which is the ultra-precision machining error compensation method based on multimodal information fusion provided in the following embodiments.

[0038] Specifically, the ultra-precision machining error compensation method based on multimodal information fusion includes the following steps:

[0039] S1: During the initial machining process, the three-dimensional morphology data of the workpiece surface is obtained by using an in-situ measuring device integrated into the working space of the ultra-precision machine tool.

[0040] S2: Temperature field data and vibration data during the initial machining process are collected synchronously by distributed temperature sensors and dynamic vibration sensors arranged on the heat source parts and moving parts of the machine tool.

[0041] S3: Align the three-dimensional topography data, temperature field data, and vibration data according to timestamps to construct a multimodal spatiotemporal tensor.

[0042] S4: Input the multimodal spatiotemporal tensor into the physical information neural network to decouple and separate the total processing error into static geometric error, time-varying thermal drift error, and dynamic vibration error. The composite loss function of the physical information neural network. The following conditions must be met: .

[0043] in: This represents the total error of the network output and the mean square error of the measured data. The degree to which the thermal error components output by the network violate the heat conduction equation is calculated using automatic differentiation. The degree to which the dynamic error components output by the network violate the forced vibration equation is calculated using automatic differentiation. These are the weighting coefficients for the physical constraint terms.

[0044] The physical information neural network employs a two-stream feature extraction architecture:

[0045] Spatial Flow Branch: Graph Convolutional Networks (GCN) or PointNet are used to process geometric coordinate data and extract local curvature spatial features.

[0046] Temporal branch: Long Short-Term Memory (LSTM) or 1D-CNN is used to process temperature and vibration sequences and extract thermal inertia and vibration mode features;

[0047] Fusion layer: The spatial and temporal features are fused using an attention mechanism, and the decoupled error components are output.

[0048] S5: Based on the static geometric error, generate a four-way forward toolpath containing spatial coordinates (x, y, z) and machining time t to compensate for the static geometric error in reverse in space.

[0049] Specifically, the four-way maintenance toolpath The following conditions must be met:

[0050] .

[0051] in, To design curved surfaces, The static geometric error is obtained by decoupling.

[0052] S6: Power spectral density analysis is used to verify whether the corrected toolpath is within the dynamic capability range of the corresponding machine tool axis.

[0053] This embodiment employs the power spectral density (PSD) analysis method to obtain the spectral distribution of the Z-axis motion commands of the original toolpath and the modified toolpath by performing a Fourier transform. This spectral distribution is then compared with the known bandwidth limit of the machine tool's Z-axis servo system. Based on whether the spectral distribution of the tool trajectory is within the dynamic following spectrum range of the machine tool, it is determined whether the path is within the dynamic capability range of the machine tool, thus identifying any situations where the modified toolpath is not dynamically feasible for the machine tool axis.

[0054] S7: If the modified toolpath is not feasible due to machine axis dynamics, the optimizer based on jerk constraints will modify the modified toolpath.

[0055] The optimizer with acceleration constraints satisfies the following conditions:

[0056] .

[0057] Constraints: .

[0058] in, This is the optimized and corrected toolpath. This is a revised toolpath before optimization. This represents the servo bandwidth limit for the corresponding machine tool axis in the machine tool system. The third derivative of the optimized corrected toolpath in the machine tool axis displacement direction is the jerk.

[0059] The dynamic feasibility verification defined by S6 and S7 involves establishing an optimization model before performing corrective machining. This model aims to minimize path following error and uses the maximum speed, maximum acceleration, and maximum jerk of each axis of the machine tool as constraints. The generated four-axis corrective toolpath is then smoothed, and high-frequency compensation components that exceed the machine tool servo bandwidth are removed.

[0060] This invention introduces a feasibility verification mechanism based on the dynamic limits of the machine tool servo after generating the corrected toolpath. This effectively avoids the risk of machine tool chatter induced by high-frequency compensation commands and solves the problem of insufficient dynamic feasibility verification in existing technologies. Traditional methods often directly invert the decoupled error to generate the compensation path without considering whether the path is within the bandwidth of the machine tool motion system. When the original error contains high-frequency components, such paths may contain acceleration or jerk changes that exceed the tracking capability of the servo system, leading to tracking lag or even structural resonance during actual execution. In this invention, although the corrected path is generated based on static geometric errors, it still needs to be verified by a dynamic feasibility analysis module. This module uses the maximum speed, acceleration, and jerk of each axis of the machine tool as hard constraints to smooth the path or filter out high-frequency components. This ensures that the final four-dimensional path is both faithful to the actual geometric defects and fully compatible with the dynamic performance boundaries of the machine tool, thereby ensuring the stability and safety of the machining process while achieving high-precision compensation.

[0061] Specifically, the relevant methods and systems in the above embodiments will be described and explained in detail below:

[0062] 1. System Hardware Configuration and Multimodal Sensing Network

[0063] The implementation system architecture of the present invention is as follows: Figure 1 As shown, the core platform is an ultra-precision single-point diamond lathe with high rigidity and high thermal stability (e.g., Precitech Nanoform 250). To achieve physical field sensing throughout the entire machining and measurement process, this system constructs a heterogeneous sensor network based on the traditional machine tool:

[0064] In-situ measurement unit: A high-resolution white light interferometric probe (e.g., Keyence SI-F80 series) integrated on the side of the Z-axis slide. Its configuration is used to acquire point cloud data P(x, y, z) of the surface topography without unloading the workpiece, thereby ensuring the physical unity of the machining and measurement coordinate systems.

[0065] Thermal monitoring unit: A fiber optic grating (FBG) temperature sensing network with 8 measuring points is constructed. Sensors are respectively positioned at the front / rear spindle bearing housings, inside the X / Z axis linear motor coils, key structural points on the machine tool bed, and the cutting fluid inlet. The FBG sensors have electromagnetic interference resistance characteristics, and the sampling rate is set to 1Hz to capture temperature field changes T(t) that cause thermal drift in the machine tool.

[0066] Dynamic monitoring unit: A triaxial high-frequency piezoelectric accelerometer (e.g., Kistler 8763B) mounted on the root of the tool holder and the spindle rotor housing, with a sampling rate set to 20kHz. It is used to capture micro-vibration signals A(t) during the cutting process in real time to identify interference from spindle rotation error and environmental vibration on the measurement.

[0067] Spatiotemporal synchronization module: All sensor data are synchronized with the position feedback signal of the machine tool grating ruler at the nanosecond level via the IEEE 1588 PTP (Precision Time Protocol), ensuring that each spatial measurement point can accurately correspond to the physical field state at the time of measurement.

[0068] 2. Mathematical framework before compensating toolpath generation

[0069] 2.1 The following example demonstrates the fabrication of a sinusoidal grid surface. Please refer to [the documentation / reference]. Figure 3 , Figure 3 This is a schematic diagram of a sinusoidal grating surface, which is composed of the superposition of two orthogonal sine waves. Its ideal geometry can be described by the following mathematical function:

[0070]

[0071] in and These are the amplitudes in the X and Y directions, respectively. and For the corresponding wavelength, and and This is the phase shift.

[0072] 2.2 To machine such non-rotationally symmetric surfaces, this invention employs Slow Tool Servo (STS) technology. This technology achieves complex spatial trajectory generation by synchronously controlling the linear axes (X-axis, Z-axis) of the machine tool and the position servo of the spindle (C-axis). To ensure complete coverage of the workpiece surface and efficient machining, the toolpath is planned as an Archimedean spiral. The discrete control points of this ideal toolpath ( , (zi) is generated according to the following steps:

[0073] In the machine tool's native cylindrical coordinate system, the radial coordinate of the i-th control point polar coordinates Defined by the following formula:

[0074]

[0075]

[0076] in, Let the workpiece radius be 1. For feed rate, Main spindle speed (RPM). This refers to the number of discrete control points set per spindle revolution.

[0077] Convert the above cylindrical coordinates to Cartesian coordinates. , ):

[0078]

[0079] By substituting Cartesian coordinates into the surface definition equation, the corresponding ideal Z-axis coordinates can be calculated. :

[0080]

[0081] 2.3 This invention employs a Z-axis-only tool radius compensation strategy. For the machine tool structure used in this embodiment, the X-axis slide carries the entire C-axis spindle system, and its moment of inertia is much greater than that of the Z-axis slide, which only carries the tool post. This results in the X-axis having a significantly lower dynamic response bandwidth than the Z-axis. This invention ensures that the X-axis performs only smooth, uniform motion by constraining all high-frequency, complex compensation movements to the Z-axis, which has superior dynamic performance. This proactively avoids servo errors introduced by insufficient X-axis dynamic tracking capability. The compensated tool center position is ( , ), where the radial position remains unchanged, i.e. = New Z-axis position By superimposing a Z-axis compensation amount at the ideal position To obtain, that is = + The amount of compensation The calculations aim to ensure that the arc profile of the tool is precisely tangent to the theoretical profile on the radial section. The value of is obtained by solving the following implicit system of equations at each control point i:

[0082]

[0083]

[0084] In this system of equations The local slope angle at the new tangent point. This is the first derivative of the surface profile with respect to the radial section. Because... If both sides of the equation appear simultaneously, the system of equations needs to be solved using numerical algorithms such as Newton's iteration method.

[0085] 3. Core Algorithm Flow and Implementation Steps

[0086] 3.1 Multimodal Data Acquisition and Spatiotemporal Tensor Construction: After the initial machining is completed using the compensation path described in step 2.3, the ISSM probe measures the actual machined surface morphology without unloading the workpiece. The measurement path can be similar to the machining path, such as... Figure 4 The spiral scanning strategy shown generates, as follows: Figure 5 The point cloud map shown is used to collect point cloud data M from a series of measurement points. The key innovation lies in the spatiotemporal alignment of the data. Because the probe scans point by point, the first point in the point cloud... and the Nth point At different times and Measurement. The system not only records spatial coordinates, but also constructs a... Figure 6 The "enhanced measurement tensor" is shown. For each measurement point i, the system constructs a feature vector. :

[0087]

[0088] in, , is the instantaneous temperature vector at that moment. It is the instantaneous vibration feature vector at that moment. The final dataset contains "all processing information" at the time the error occurred.

[0089] Processing

[0090] 3.2 Physical Information-Driven Error Decoupling: Input the above-mentioned enhanced tensor into, for example... Figure 7 The Physical Information Neural Network (PINN) shown is designed to solve the following inverse problem:

[0091] in, It is the real geometric error (static) that needs to be compensated. It is a spurious thermal error (time-varying) introduced during the measurement process. It is vibration noise (time-varying).

[0092] The network architecture adopts a dual-stream design:

[0093] Spatial Flow:

[0094] A graph convolutional network (GCN) is used to process geometric coordinates (x,y,z) to extract local curvature features of the workpiece surface.

[0095] Time Flow

[0096] Long Short-Term Memory (LSTM) networks are used to process time series data (T,A) to capture thermal inertia and vibration modes.

[0097] Physical drug beam training

[0098] To accurately separate the components from underdetermined measurement data, this invention introduces a physical information loss function. :

[0099] Among them, data fitting term Ensure that the sum of the three separated terms equals the measured value;

[0100] Thermophysical constraint terms Using automatic differentiation technology, the thermal error component of the network output is forced. Satisfies the partial differential equation of heat conduction:

[0101]

[0102] in :express Rate of change over time (rate of heating or cooling); This indicates heat diffusion due to thermal conduction. α is the thermal diffusivity of the material. : Indicates an internal heat source. This constraint requires that the thermal error must change continuously and slowly over time to prevent the network from misinterpreting high-frequency noise as thermal drift.

[0103] Dynamic constraints Forced dynamic error components It conforms to the forced vibration equation, ensuring that its frequency characteristics are consistent with the accelerometer readings.

[0104] like Figure 8 As shown, the trained model can successfully decompose the measured shape distortion error into true geometric error (retained) and spurious thermal drift and dynamic error (rejected). The toolpath used for machining correction consists of a large number of control points, the positions of which typically do not completely coincide with discrete error measurement points.

[0105] Therefore, a method is needed to determine the error value at any point on the surface in order to generate a continuous and smooth correction toolpath. This invention employs a bilinear interpolation algorithm to generate a continuous error function from a discrete measurement error grid. For any toolpath point P where the error to be calculated... If it falls on a measurement point with four known errors ( , ), ( , ), ( , ) and Within the constructed rectangular grid, the error value E(P) can be calculated through the following two interpolation steps:

[0106] First linear interpolation (along) direction):

[0107]

[0108]

[0109] Second linear interpolation (along) direction):

[0110]

[0111] 3.3 Four-dimensional path synthesis: Generated corrected toolpath instructions Compensation for static geometric errors:

[0112]

[0113] in For the final execution instructions (the revised toolpath); The surface is designed theoretically, which is the shape the workpiece should have in an ideal state (e.g., an ideal sinusoidal mesh function). The minus sign is used here because the compensation logic is reversed: in the region where the error is positive (i.e., material out of tolerance), the tool needs to cut deeper; in the region where the error is negative (i.e., material under-tolerance), the tool needs to cut shallower. The true static geometric error is the error "purified" from the measurement data and fixed on the workpiece surface by the Physical Information Neural Network (PINN). It eliminates the thermal drift artifact during measurement and retains only the true surface defects.

[0114] Because the machine tool spindle expands due to heat during machining, the length of this expansion will cause corresponding thermal deformation errors on the final machined surface. Therefore, to further improve the accuracy of the final compensation path, it is preferable to incorporate the error caused by the thermal expansion deformation of the machine tool spindle into the corresponding machining command. The length change caused by this expansion deformation varies over time and is directly related to relevant machining process parameters. The machining process parameters to be corrected (including spindle speed, feed rate, and machining time) are input into a trained neural network model, which then infers the thermal dynamic evolution trend of the machine tool spindle during the correction machining process. This neural network model can be trained based on historically collected process parameters and thermal parallel data of the spindle over time. Under these circumstances, the generated toolpath command... It not only compensates for static geometric errors, but also pre-sets dynamic biases to counteract predicted thermal drift:

[0115]

[0116] in, The predicted future thermal drift error is the amount of machine tool thermal deformation that will occur during the correction process, as predicted by the neural network model. Of course, other existing methods can also be used to predict the thermal deformation of the machine tool spindle, such as by establishing a mapping table between machining parameters and the thermal deformation trend of the machine tool spindle.

[0117] 3.4 Dynamic Feasibility Verification and Kinematic Optimization: Power Spectral Density (PSD) analysis was used to verify the modified machining path. Whether it is within the dynamic capability range of the machine tool. If there is a dynamic infeasibility problem in correcting the path, this invention introduces an optimizer based on jerk constraints to establish an optimization model:

[0118]

[0119] Constraints:

[0120] in To optimize the path; This refers to the servo bandwidth limit of the machine tool system. For example... Figure 9 As shown, the optimizer can automatically smooth sharp abrupt changes in the path by "shaving off peaks and filling valleys" to ensure that the machine tool does not generate secondary vibrations when performing compensation.

[0121] 4. Specific comparative experiments

[0122] The specific steps of this comparative experiment are explained in detail below.

[0123] Step S1001: Initial Machining. Using STS technology and the aforementioned Z-axis-only tool radius compensation strategy, a sinusoidal grid is machined on the workpiece. Specific machining and tool parameters are shown in Tables 1 and 2.

[0124] Table 1

[0125]

[0126] Table 2

[0127]

[0128] Step S1002: Multimodal Data Acquisition. After initial machining, the workpiece remains stationary. The white light interferometer probe (Keyence SI-F80) integrated into the machine tool is controlled to scan the workpiece surface along a spiral path. Simultaneously, a fiber optic grating (FBG) temperature sensor array and a piezoelectric accelerometer (Kistler 8763B) synchronously record the temperature field and vibration data throughout the entire measurement process (approximately 12 minutes). The scanning parameters are shown in Table 3.

[0129] Table 3

[0130]

[0131] Step S1003: Data Alignment and Error Decoupling: The intelligent edge computing unit reads the data collected in step S302. First, based on a unified timestamp (IEEE 1588 PTP), each spatial measurement point P(x,y,z) is aligned with the temperature T(t) and vibration A(t) at that moment, constructing an enhanced measurement tensor. Then, this tensor is input into a pre-trained PINN network. Network analysis reveals a strong correlation between the 100nm height difference in the measurement data and the 0.3°C temperature rise curve of the main axis during scanning (consistent with the physical constraints of thermal expansion), therefore it is determined to be... (False thermal errors) are identified and removed. The final extracted true geometric error is then determined. .

[0132] Step S1004: Four-dimensional prediction path generation: The system generates a four-dimensional correct path. This path is for The geometric errors are compensated for.

[0133] Step S1005: Motion optimization and dynamic feasibility verification. Before execution, the kinematic optimizer checks the generated path. For example... Figure 9As shown, it was found that the compensation command for a certain local ripple caused the Z-axis acceleration to momentarily exceed 0.6g (close to the servo limit). The optimizer uses the Jerk algorithm to minimize the jerk and slightly smooth the command curve at that point (sacrificing 1nm of theoretical accuracy) to ensure that the actual acceleration is controlled within 0.5g.

[0134] Step S1006: Correction machining. Download the corrected toolpath, which has been optimized and dynamically verified, to the machine tool controller and perform a final correction machining operation.

[0135] Step S1007: Final Verification. To verify the effect, three sets of comparative experiments were set up, and the results are shown in Table 4.

[0136] Table 4

[0137]

[0138] Results analysis:

[0139] Limitations of Method B: Although Method B eliminates the repositioning error, it misjudges the 45nm thermal expansion during the measurement process as a geometric error, resulting in a reverse wedge shape on the corrected surface and the RMS value remaining at 18.2nm.

[0140] Advantages of the present invention: The present invention (method C) eliminates thermal drift artifacts through PINN decoupling, and the final RMS value reaches 4.8nm, which is 73.6% higher than that of B, realizing true nanoscale deterministic manufacturing.

[0141] In summary, this invention significantly breaks through the accuracy limits of traditional in-situ measurements through integrated system design and deep decoupling compensation closed-loop of physical information.

[0142] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0143] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, mobile terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0144] In an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0145] Those skilled in the art will understand that various aspects of the present invention can be implemented as systems, methods, or program products. Therefore, various aspects of the present invention can be specifically implemented in the following forms: entirely in hardware, entirely in software (including firmware, microcode, etc.), or in a combination of hardware and software, collectively referred to herein as “circuit,” “module,” or “system.”

[0146] An electronic device according to this embodiment of the invention. The electronic device is merely an example and should not be construed as limiting the functionality or scope of the embodiments of the invention.

[0147] Electronic devices are manifested in the form of general-purpose computing devices. The components of an electronic device may include, but are not limited to: at least one processor, at least one memory, and a bus connecting different system components (including memory and processor).

[0148] The memory stores program code that can be executed by a processor, causing the processor to perform the steps described in the "Exemplary Methods" section above, according to various exemplary embodiments of the present invention.

[0149] The storage may include readable media in the form of volatile storage, such as random access memory (RAM) and / or cache memory, and may further include read-only memory (ROM).

[0150] The storage may also include programs / utilities having a set (at least one) of program modules, including but not limited to: an operating system, one or more applications, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0151] A bus can represent one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus that uses any of the various bus architectures.

[0152] The electronic device can also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with the electronic device, and / or any device that enables the electronic device to communicate with one or more other computing devices (e.g., routers, modems, etc.). This communication can be performed via input / output (I / O) interfaces. Furthermore, the electronic device can communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter. The network adapter communicates with other modules of the electronic device via a bus. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0153] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible embodiments, various aspects of the present invention may also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of the present invention described in the "Exemplary Methods" section above.

[0154] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0155] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0156] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0157] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0158] Furthermore, the accompanying drawings are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes shown in the above drawings do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0159] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0160] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for compensating for errors in ultra-precision machining based on multimodal information fusion, characterized in that, The method includes the following steps: During the initial processing, in-situ measuring devices integrated into the working space of ultra-precision machine tools are used to acquire three-dimensional morphological data of the workpiece surface; Temperature field data and vibration data during the initial machining process are collected synchronously by distributed temperature sensors and dynamic vibration sensors arranged on the heat source parts and moving parts of the machine tool. The three-dimensional topography data, temperature field data, and vibration data are aligned according to timestamps to construct a multimodal spatiotemporal tensor; The multimodal spatiotemporal tensor is input into a physical information neural network to decouple and separate the total processing error into static geometric error, time-varying thermal drift error, and dynamic vibration error; the composite loss function of the physical information neural network... The following conditions must be met: ; in: The network outputs the total error and the mean square error of the measured data. The degree to which the thermal error components output by the network violate the heat conduction equation is calculated using automatic differentiation; The degree of violation of the forced vibration equation by the dynamic error components output by the network is calculated using automatic differentiation; These are the weighting coefficients for the physical constraint terms; Based on the static geometric error, a four-way forward toolpath containing spatial coordinates (x, y, z) and machining time t is generated to spatially compensate for the static geometric error.

2. The method according to claim 1, characterized in that, After generating a four-way maintenance toolpath containing spatial coordinates (x, y, z) and machining time t, the method further includes: Power spectral density analysis was used to verify whether the corrected toolpath was within the dynamic capability range of the corresponding machine tool axis. If the modified toolpath is not feasible due to machine axis dynamics, then the optimizer based on jerk constraints will modify the modified toolpath. The optimizer for the acceleration constraint satisfies the following condition: ； Constraints: ; in, The optimized and corrected toolpath; The toolpath was corrected before optimization; This represents the servo bandwidth limit for the corresponding machine tool axis in the machine tool system. The third derivative of the optimized corrected toolpath in the machine tool axis displacement direction is the jerk.

3. The method according to claim 1, characterized in that, The physical information neural network adopts a two-stream feature extraction architecture: Spatial Flow Branch: Use graph convolutional networks or point cloud networks to process geometric coordinate data and extract local curvature spatial features; Temporal branch: Long short-term memory networks or one-dimensional convolutional networks are used to process temperature and vibration sequences to extract thermal inertia and vibration mode features; Fusion layer: The spatial and temporal features are fused using an attention mechanism, and the decoupled error components are output.

4. The method according to claim 1, characterized in that, The four maintenance positive tool paths The following conditions must be met: ； in, To design curved surfaces, The static geometric error is obtained by decoupling.

5. The method according to claim 1, characterized in that, The ultra-precision machine tool is equipped with a position servo controlled spindle and at least two linear motion axes to perform slow tool servo machining. The in-situ measurement device includes an in-situ measurement probe integrated into the working space of the machine tool. It is used to measure the surface morphology of the workpiece without unloading it from the machine tool fixture, thereby maintaining a unified coordinate system between machining and measurement operations.

6. The method according to claim 5, characterized in that, The temperature sensor is a fiber optic grating sensor, and the heat source of the machine tool includes the front and rear bearing seats of the spindle, the linear motor coil winding, and the structural points of the machine tool bed; the vibration sensor is a high-frequency piezoelectric accelerometer or a MEMS accelerometer; the moving parts include the tool clamping end and the spindle rotor.

7. The method according to claim 5, characterized in that, The initial machining employs a spindle-only tool radius compensation strategy, where all tool radius compensation movements are strictly limited to the spindle to avoid introducing dynamic tracking errors on linear motion axes.

8. The method according to claim 1, characterized in that, The in-situ measuring device uses a continuous spiral scanning path to acquire three-dimensional morphological data of the workpiece surface.

9. A non-transitory computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the ultra-precision machining error compensation method based on multimodal information fusion as described in any one of claims 1 to 8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the ultra-precision machining error compensation method based on multimodal information fusion as described in any one of claims 1 to 8.

Images