Numerical control machine tool error intelligent compensation method based on spatio-temporal feature fusion

Abstract

The invention discloses a numerical control machine tool error intelligent compensation method based on spatio-temporal feature fusion, and the method has the following advantages: through spatio-temporal feature fusion and a physical constraint model, decoupling analysis of multi-source errors such as thermal errors and force errors is realized, through a physical model, the thermal errors and the force errors are separated, and a data driving model corrects nonlinear residual errors; the modeling precision is improved, and the modeling error is reduced; based on a rolling optimization strategy of model prediction control, error compensation can be dynamically carried out on the numerical control machine tool in real time, compensation parameter updating can be completed in a polar period of time, and the method adapts to rapid changes of machining working conditions; a hybrid model of physical prior and data driving is combined, and all-working-condition positioning error control is carried out, so that the error is controlled within a precise error range, and the precision is greatly improved; the physical model provides mechanism explanation of error generation, a data driving model and correction of supplementary nonlinear errors, and the engineering credibility of the compensation strategy is greatly improved through the error compensation strategy combining the two.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of intelligent manufacturing, and in particular to a numerical control machine tool error intelligent compensation method based on space-time feature fusion. BACKGROUND

[0002] In modern manufacturing, numerical control machine tools as core production equipment, its running stability directly affects the production efficiency and product quality. The existing numerical control machine tool error compensation technology faces the following core challenges: (1) Difficulty in modeling multi-source error coupling: thermal deformation, mechanical wear, servo hysteresis and other error sources interact with each other, and traditional linear models are difficult to describe their nonlinear coupling relationship.

[0003] (2) Insufficient processing of time-varying characteristics: factors such as tool wear and workpiece material removal during machining cause error characteristics to change dynamically over time, and static compensation parameters cannot be adapted.

[0004] (3) Poor interpretability of data-driven models: pure deep learning models lack physical constraints, resulting in insufficient engineering verifiability of the compensation strategy.

[0005] (4) Real-time control precision and response speed conflict: complex model reasoning takes a long time, making it difficult to meet the microsecond-level compensation requirements in high-speed machining scenarios.

[0006] Therefore, how to design a numerical control machine tool error compensation method with high control precision, fast response speed and the ability to couple multi-source data is a technical problem to be solved. SUMMARY

[0007] Therefore, it is necessary to provide a numerical control machine tool error intelligent compensation method based on space-time feature fusion in view of the above technical problems. The application constructs a multi-source data space-time feature fusion model to analyze the space-time distribution law of the error, designs a hybrid error prediction model combining physical constraints and data-driven, and develops an adaptive compensation algorithm based on model predictive control to realize dynamic optimization of the compensation parameters.

[0008] The application embodiment provides a numerical control machine tool error intelligent compensation method based on space-time feature fusion, which comprises the following steps: S1: acquiring multi-source data of a numerical control machine tool, wherein the multi-source data comprises displacement data, temperature data and vibration signal data collected by laser sensors, temperature sensors and acceleration sensors arranged at key positions of the numerical control machine tool; S2: performing feature extraction on the displacement data, the temperature data and the vibration signal data based on a space-time feature extraction algorithm to obtain a time feature vector and a space feature vector; S3: obtaining a spatio-temporal fusion feature vector based on the time feature vector, the space feature vector, and a preset spatio-temporal feature fusion model; S4: constructing a linear error model of the numerical control machine tool, i.e., a thermal error model and a force error model, based on data correlation between the displacement data, the temperature data, and the vibration signal data; S5: constructing a nonlinear error model of the numerical control machine tool based on the spatio-temporal fusion feature vector and a long short-term memory network; S6: obtaining total error prediction data based on the thermal error model, the force error model, and the nonlinear error model; S7: constructing an objective function to optimize the total error prediction data to obtain optimal compensation error; S8: compensating for errors of the numerical control machine tool based on the optimal compensation error and updating the nonlinear error model.

[0009] Preferably, step S2 comprises: S21: performing wavelet packet decomposition on the displacement data and the vibration signal data to obtain a time feature vector; S22: constructing a spatial correlation matrix of temperature data and displacement data based on a structure topology of the numerical control machine tool, and extracting spatial features through a graph convolution network to obtain a space feature vector.

[0010] Preferably, step S3 comprises: S31: obtaining an adaptive fusion weight based on the time feature vector and the space feature vector; S32: obtaining a spatio-temporal fusion feature vector based on the time feature vector, the space feature vector, the adaptive fusion weight, and a preset spatio-temporal feature fusion model.

[0011] Preferably, the preset spatio-temporal feature fusion model is represented by the following formula: (1) ; wherein, is the spatio-temporal fusion feature vector, is the time feature vector, is the space feature vector, is the adaptive fusion weight, is represented by the following formula: (2) ; wherein, and are learnable parameter matrices.

[0012] Preferably, step S4 comprises: S41: based on the thermal conductivity coefficient and the material expansion coefficient of the numerical control machine tool, calculate the displacement error caused by temperature change, and construct a thermal error model; S42: according to the cutting force and the structural flexibility matrix of the numerical control machine tool, calculate the force deformation error, and construct a force error model.

[0013] Preferably, the thermal error model is represented by the following formula: (3) ; Wherein, is the error caused by temperature change, is the thermal expansion coefficient of the material of the numerical control machine tool, is the temperature change, is the length vector of the key components of the machine tool, is the time; The force error model is represented by the following formula: (4) ; Wherein, is the force deformation error, is the structural flexibility matrix, is the cutting force vector, .

[0014] Preferably, step S5 comprises: based on the space-time fusion feature vector and the long short-term memory network, the above linear error model of the numerical control machine tool is constructed into a covering nonlinear error model; The nonlinear error is represented by the following formula: (5) ; Wherein, LSTM is a long short-term memory network.

[0015] Preferably, the total error prediction data is calculated by the following formula: (6) ; Wherein, is the total error prediction data.

[0016] Preferably, step S7 comprises: S71: based on the total error minimization and the control smoothness as the target, construct a target function; S72: solve the target function by QP programming algorithm to obtain the optimal compensation parameter; The target function is represented by the following formula: (7) ; The constraint condition of the target function is: (8) ; (9); wherein, is a compensation parameter vector, is a prediction horizon, is a control horizon, is a control quantity smoothness weight.

[0017] Preferably, step S8 comprises: S81: compensating multi-source data of the numerical control machine tool based on the optimal compensation parameter and updating to obtain an actual error value after compensation; S82: updating network parameters of the nonlinear error model through a gradient descent algorithm and a residual of total error prediction data and the actual error value.

[0018] Compared with the prior art, the present application has the following beneficial effects: (1) Multi-source error decoupling modeling: through spatiotemporal feature fusion and a physical constraint model, decoupling analysis of thermal errors, force errors and other multi-source errors is realized, thermal errors and force errors are separated through a physical model, a data-driven model corrects nonlinear residuals, modeling accuracy is improved, and modeling errors are reduced; (2) Enhanced dynamic adaptability: based on a rolling optimization strategy of model predictive control, the numerical control machine tool can be compensated in real time and dynamically, and the system can complete compensation parameter updating within a very short time, adapting to rapid changes in machining conditions; (3) Improved compensation accuracy: a hybrid model combining physical priori and data-driven is used to control positioning errors in all conditions within an accurate error range, and the accuracy is significantly improved compared with traditional methods; (4) Enhanced interpretability: a physical model, i.e., a linear model, provides an engineering explanation of error generation, a data-driven model, i.e., a nonlinear model, supplements correction of nonlinear errors, and the combination of the two error compensation strategies greatly improves the engineering credibility of the compensation strategy. BRIEF DESCRIPTION OF DRAWINGS

[0019] The exemplary embodiments of this application can be more fully understood with reference to the following drawings in which:

[0020] Figure 1 A flowchart of a numerical control machine tool error intelligent compensation method based on spatiotemporal feature fusion according to an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0021] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood, and so that the full scope of the present disclosure is conveyed to those skilled in the art.

[0022] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0023] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as there is no conflict between them.

[0024] With reference to Figure 1 The present embodiment discloses an intelligent error compensation method for a numerical control machine tool based on spatio-temporal feature fusion, comprising the following steps: S1: acquiring multi-source data of the numerical control machine tool, the multi-source data comprising displacement data, temperature data and vibration signal data collected by laser sensors, temperature sensors and acceleration sensors deployed at key positions of the numerical control machine tool; In a preferred embodiment, a plurality of types of sensors can be deployed at key positions of the numerical control machine tool, and the sensor types of the present application are as follows: (1) Laser sensor, specifically a laser interferometer, the sampling frequency is selected according to the needs, the laser interferometer measures the displacement data of each axis of the numerical control machine tool in real time ; (2) Temperature sensor, specifically an optical fiber temperature sensor, for collecting temperature data of the numerical control machine tool ; (3) Acceleration sensor, specifically an accelerometer, for acquiring vibration signal data of the numerical control machine tool .

[0025] S2: performing feature extraction on the displacement data, temperature data and vibration signal data based on a spatio-temporal feature extraction algorithm to obtain a time feature vector and a space feature vector; Specifically, step S2 comprises: S21: Perform wavelet packet decomposition on the displacement data and the vibration signal data to obtain the time feature vector; Specifically, wavelet packet decomposition is performed on time-series data such as displacement and vibration signals to extract the energy features of each segment, thereby obtaining a time feature vector. ; (10); (11); in, For the first There are 1 wavelet packet components, M is the number of decomposition levels, and N is the number of sampling points for each segment; Indicates the first The signal energy within a sub-band is considered higher if the energy is higher, indicating a more significant signal component within that band (e.g., a sudden increase in energy in a certain frequency band of a vibration signal may correspond to wear of mechanical parts). Energy characteristics of major subbands (e.g., before selection) (each low-frequency sub-band) is combined to form a time feature vector. This is used for subsequent time-series modeling (such as LSTM input). Wavelet packet decomposition transforms the displacement and vibration signals from the time domain to the time-frequency domain, extracting the energy feature vector. It can effectively characterize the dynamic changes in the machine tool's operating status, providing key inputs for subsequent error prediction and compensation. In practical applications, the decomposition parameters need to be optimized in conjunction with the specific characteristics of the equipment to balance feature accuracy and computational efficiency.

[0026] S22: Based on the structural topology diagram of CNC machine tools, construct a spatial correlation matrix between temperature data and displacement data, and extract spatial features through graph convolutional networks to obtain spatial feature vectors.

[0027] Specifically, temperature changes in CNC machine tools cause displacement changes in related components. Therefore, this spatial correlation matrix is ​​constructed based on prior experience and a spatial relationship formed using an expert database built from large datasets. Spatial features are then extracted using methods such as Generative Convolutional Networks (GCNs) to obtain spatial feature vectors. Specifically, a machine tool structure topology diagram is constructed, where nodes represent temperature measurement points and axis coordinates, and edges represent heat conduction paths. The spatial correlation between the temperature field and error is extracted using GCN. (12); in, It is an adjacency matrix with dimension . , Represents a node and nodes The thermal conductivity between nodes (Temperature measuring points) and nodes There is a heat conduction path between the (axis coordinates), then (thermal conductivity, unit ); if there is no direct connection, ; the diagonal element (not considering self-connection); wherein, is a degree matrix, , used to normalize the adjacency matrix to prevent gradient explosion / vanishing; wherein, is a temperature data matrix with dimensions , is a characteristic number, such as temperature value, temperature rise rate, etc. is a learnable weight matrix with dimensions , optimized by backpropagation, output feature dimension; is an activation function, commonly used form , to introduce non-linear characteristics. Equation (12) fuses the local features (temperature values) of nodes (temperature measurement points, axis coordinates) with global structural information (heat conduction paths) through GCN, extracting spatial correlation features of temperature field and displacement error. For example: how does the temperature change of the main shaft box affect the X-axis positioning accuracy through the bed structure.

[0028] By constructing the spatial correlation matrix and GCN, this step converts the complex spatial relationship between the machine tool temperature field and the displacement error into a calculable feature vector, providing a structured input for subsequent error prediction. In practical applications, the adjacency matrix parameters need to be optimized in combination with the specific machine tool structure, and the spatial correlation model needs to be continuously updated through online learning.

[0029] S3: Based on the time feature vector, the space feature vector, and the preset spatio-temporal feature fusion model, obtain the spatio-temporal fusion feature vector; Specifically, step S3 includes: S31: Based on the time feature vector and the space feature vector, obtain an adaptive fusion weight; S32: Based on the time feature vector, the space feature vector, the adaptive fusion weight, and the preset spatio-temporal feature fusion model, obtain the spatio-temporal fusion feature vector.

[0030] In this embodiment, the preset spatio-temporal feature fusion model is represented by the following formula: (1); wherein, is the spatio-temporal fusion feature vector, is the time feature vector, is the space feature vector, is the adaptive fusion weight; formula (1) fuses the time feature vector and the space feature vector through the adaptive weight The optimal fusion of multimodal information is achieved by dynamically balancing the contributions of temporal characteristics (reflecting dynamic changes in processing) and spatial characteristics (characterizing the relationship between temperature field and structural deformation). The mechanism of dynamic adjustment it demonstrates is as follows: (1) During high-speed cutting (vibration is dominant). The value increases, At this point, the spatiotemporal feature model focuses on time features.

[0031] (2) When thermal deformation is dominant (temperature field changes slowly). The value increases, At this point, the spatiotemporal feature fusion model focuses on spatial features.

[0032] In this embodiment, the fusion weights of temporal and spatial features are calculated using an attention mechanism: (2); Weight Dynamic adjustments, for example, emphasizing time characteristics during high-speed cutting and emphasizing spatial characteristics when thermal deformation is dominant; and Given a learnable parameter matrix, nonlinear transformations are applied to the temporal and spatial features respectively. The transformed features are mapped to the [0,1] interval using the softmax function to ensure that the sum of the weights of the temporal and spatial feature vectors is 1.

[0033] Through an adaptive attention mechanism, this invention achieves dynamic fusion of temporal and spatial features, enabling the error prediction model to automatically focus on the currently dominant error source (such as vibration or thermal deformation). In practical applications, it is recommended to adjust the model according to the machine tool type and machining process. and The initial value is determined, and the fusion strategy is continuously optimized through online learning.

[0034] S4: Based on the data correlation between displacement data, temperature data and vibration signal data, construct a linear error model for the CNC machine tool: a thermal error model and a force error model; Specifically, step S4 includes: S41: Based on the thermal conductivity and material expansion coefficient of CNC machine tools, calculate the displacement error caused by temperature changes and construct a thermal error model; S42: Based on the cutting force and structural compliance matrix of the CNC machine tool, calculate the force deformation error and construct a force error model.

[0035] The thermal error model is represented by the following formula: (3); in, Errors caused by temperature changes The material thermal expansion coefficient of the numerical control machine tool, is a temperature change, measured in real time by a temperature sensor; is a machine tool key component length vector, is time. Equation (3) represents the geometric size change caused by the thermal expansion of the material, which directly affects the accuracy of the shaft positioning.

[0036] In the preferred embodiment, when the component installation angle is not 0, the projection effect needs to be considered, and at this time is represented by the following formula: ; In the preferred embodiment, when there are multiple heat sources, is represented by the following formula: ; wherein the force error model is represented by the following formula: (4); wherein, is the force deformation error, is the structural flexibility matrix, reflecting the ability of the machine tool to resist deformation, and the flexibility matrix is the inverse matrix of the stiffness matrix, and the engineering acquisition method has the following two kinds: (1) finite element analysis (FEA): simulate the force deformation of the machine tool through software such as ANSYS, and extract the flexibility coefficient; (2) experimental identification: apply a known force to the machine tool, measure the corresponding displacement, and estimate ; is the cutting force vector, measured by a cutting force model or a force gauge, ; Equation (4) represents the elastic deformation of the machine tool structure under the action of the cutting force, which causes the relative position of the tool and the workpiece to deviate.

[0037] In this embodiment, by combining the physical law of heat conduction and the principle of structural mechanics, the error is modeled and analyzed physically, which has stronger explainability and extrapolation ability compared with pure data-driven methods. In practical applications, it is recommended to recalibrate the model parameters every quarter to adapt to the performance changes caused by the aging of the machine tool.

[0038] S5: Constructing a nonlinear error model of the numerical control machine tool based on the spatiotemporal fusion feature vector and the long short-term memory network; Specifically, the above-mentioned linear error model of the numerical control machine tool based on the spatiotemporal fusion feature vector and the long short-term memory network is a covering nonlinear error model; wherein the nonlinear error is represented by the following formula: (5); wherein, LSTM is a long short-term memory network. The long short-term memory network (LSTM) is a kind of time recurrent neural network, which is specially designed to solve the long-term dependence problem existing in general RNN (recurrent neural network), and all RNNs have a chain form of repeated neural network modules. In the standard RNN, the repeated structure module only has a very simple structure, such as a tanh layer.

[0039] Through the LSTM network, the nonlinear component in the numerical control machine tool error is effectively captured, and the modeling accuracy of complex dynamic processes such as thermal accumulation and servo lag is significantly improved. In practical application, it is recommended to combine an online learning strategy (such as fine-tuning the model once an hour) to adapt to the error characteristic drift caused by factors such as tool wear and environmental temperature changes.

[0040] The nonlinear error modeling of the embodiment has the following advantages: (1) capturing complex dynamics: such as the hysteresis effect of the servo system, friction nonlinearity, and thermal-structural coupling deformation; (2) handling non-stationarity: adapting to changes in machining parameters (such as sudden changes in feed rate) or environmental disturbances (such as temperature fluctuations); (3) fusing multi-source features: directly using a spatiotemporal fusion feature vector , integrating temperature, vibration, cutting force and other multi-dimensional information.

[0041] S6: obtaining total error prediction data based on the thermal error model, the force error model and the nonlinear error model; Specifically, the total error prediction value is obtained by combining the physical model (linear model) and the data-driven model (nonlinear model): (6) ; wherein, is the total error prediction data.

[0042] The principle of the total error prediction calculation model of the embodiment is as follows: the physical model (thermal error model, force error model) is based on prior knowledge and is good at capturing interpretable linear errors; the data-driven model (nonlinear model) learns the nonlinear residuals (such as friction nonlinearity and thermal-structural coupling effect) that are not covered by the physical model through LSTM. The thermal error component reflects the geometric deformation caused by material thermal expansion (such as spindle elongation and guide rail bending); the force error component represents the elastic deformation caused by cutting force (such as tool holder deflection and worktable sinking); the nonlinear component captures time-varying characteristics (such as servo system dynamic response), uncertainties (such as measurement noise) and unmodeled dynamics (such as cooling liquid temperature fluctuations). The fusion of the above two in the embodiment realizes the advantages of heat conduction law, structural mechanics and data-driven residual correction, making the error prediction more accurate.

[0043] By fusing the physical model and the data-driven model, the application realizes high-precision prediction of the error of the numerical control machine tool, and the advantage is remarkable especially under complex working conditions (such as variable cutting parameters and long-time machining). In actual application, it is suggested to adjust the fusion strategy according to the type of the machine tool and the machining precision requirement, for example, the depth of the LSTM network can be increased for a precision grinding machine, and the physical model can be simplified for a common milling machine to improve real-time performance.

[0044] S7: constructing a target function to optimize the total error prediction data, and obtaining optimal compensation error; Specifically, step S7 includes: S71: constructing a target function based on the minimization of the total error and the smoothness of the control; S72: solving the target function by a QP planning algorithm to obtain optimal compensation parameters; Wherein, the target function is represented by the following formula: (7); Wherein, the first term of formula (7) is an error tracking term, which minimizes the deviation of the prediction error and the compensation amount ; the second term is a control smoothing term, which punishes the drastic change of the compensation amount The greater, the more smooth the control is. , usually , the prediction of the future 10 time points , the control of the future 5 time points.

[0045] Specifically, the constraint condition of the target function is: (8); (9); Wherein, is a compensation parameter vector, is a prediction time domain, is a control time domain, is a control amount smoothness weight. Formula (8) is a physical limit of the compensation amount, and formula (9) is a compensation rate limit.

[0046] Specifically, the quadratic programming (QP) algorithm is a mathematical method for solving optimization problems with quadratic target functions and linear constraint conditions. It has wide application in portfolio optimization, machine learning, engineering control and other fields. Through QP optimization, the application realizes optimal control of error compensation, which suppresses the drastic change of the control signal while ensuring tracking accuracy. In actual application, it is suggested to adjust 、 and and periodically (e.g., weekly) update the prediction model to adapt to time-varying factors such as tool wear.

[0047] S8: Compensate the error of the numerical control machine tool based on the optimal compensation error and update the nonlinear error model.

[0048] Specifically, step S8 includes: S81: Compensate the multi-source data of the numerical control machine tool based on the optimal compensation parameter and update to obtain the actual error value after compensation; Specifically, the optimal compensation parameter is solved online by the QP algorithm After that, the control amount of the first time step is executed to realize the rolling horizon control: (13) ; wherein, is the error weight matrix, is the control weight matrix.

[0049] S82: Update the network parameters of the nonlinear error model by the gradient descent algorithm and the residual of the total error prediction data and the actual error value.

[0050] In this embodiment, the actual error data after compensation is collected in real time The residual of the actual error data and the total error prediction data is calculated: (14) ; Next, the parameters of the hybrid model are updated by the gradient descent method , that is, the network parameters of the LSTM are updated: (15) ; wherein, is the learning rate, in this embodiment, it can be set to trigger the online training of the model once every 100 ms to continuously optimize the prediction accuracy. The scheme of the present application updates the parameters of the error model after the error compensation of the numerical control machine tool system, which facilitates the execution of the next error compensation, and such setting makes the compensation process continuously depend on the foregoing parameter process, thereby realizing the real-time and dynamic compensation of the system. The system can complete the compensation parameter update within a very short time and adapt to the rapid changes of the machining conditions.

[0051] The present application realizes real-time and dynamic optimization of error compensation, and is particularly suitable for precise machining scenes with frequent changes in working conditions. In actual deployment, it is recommended to use hardware acceleration (such as GPU / FPGA) to process LSTM inference and gradient calculation, and periodically perform full-amount model training to maintain long-term stability.

[0052] Compared with the prior art, the present application has the following beneficial effects: (1) Multi-source error decoupling modeling: Through the spatio-temporal feature fusion and the physical constraint model, the decoupling analysis of thermal error, force error and other multi-source errors is realized, and the thermal error and the force error are separated through the physical model, the nonlinear residual is corrected through the data-driven model, the modeling accuracy is improved, and the modeling error is reduced; (2) Dynamic adaptability enhancement: Based on the rolling optimization strategy of model predictive control, the error compensation of the numerical control machine tool can be dynamically compensated in real time, and the system can complete the compensation parameter update within a very short time, and adapt to the rapid change of the machining condition; (3) Compensation accuracy improvement: Combined with the hybrid model of physical priori and data-driven, the positioning error control in all conditions is realized, and the error is controlled in the accurate error range, so that the accuracy is obviously improved compared with the traditional method; (4) Enhanced interpretability: The physical model, i.e. linear model, provides the mechanical explanation of error generation, the data-driven model, i.e. nonlinear model, supplements the correction of nonlinear error, and the error compensation strategy combining the two greatly improves the engineering credibility of the compensation strategy.

[0053] It should be noted that the flowcharts and block diagrams in the drawings show the possible implementation architecture, function and operation of the system, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment or a part of code, which contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks can also occur in different order from that marked in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be realized by a special hardware-based system for executing the specified function or action, or can be realized by a combination of special hardware and computer instructions.

[0054] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0055] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented by other manners. The apparatus embodiments described above are merely illustrative, for example, the division of the units is merely a logical function division, and actual implementation can have another division manner, and for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some communication interfaces, apparatuses or units, which can be electrical, mechanical or other forms.

[0056] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e., they can be located in one place or distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0057] In addition, the functional units in each embodiment of the present application can be integrated into one processing unit, or each unit can be physically present alone, or two or more units can be integrated into one unit.

[0058] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various storage medium that can store program codes.

[0059] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the description of the present application.

Claims

1. A numerical control machine tool error intelligent compensation method based on spatio-temporal feature fusion, characterized in that, The method comprises the following steps: S1: acquiring multi-source data of a numerical control machine tool, wherein the multi-source data comprises displacement data, temperature data and vibration signal data collected by laser sensors, temperature sensors and acceleration sensors arranged at key positions of the numerical control machine tool; S2: performing feature extraction on the displacement data, the temperature data and the vibration signal data based on a space-time feature extraction algorithm to obtain a time feature vector and a space feature vector; S3: obtaining a space-time fusion feature vector based on the time feature vector, the space feature vector and a preset space-time feature fusion model; S4: constructing a linear error model of the numerical control machine tool, i.e., a thermal error model and a force error model, based on data correlation between the displacement data, the temperature data and the vibration signal data; S5: constructing a nonlinear error model of the numerical control machine tool based on the space-time fusion feature vector and a long short-term memory network; S6: obtaining total error prediction data based on the thermal error model, the force error model and the nonlinear error model; S7: constructing an objective function to optimize the total error prediction data to obtain optimal compensation error; S8: compensating for errors of the numerical control machine tool based on the optimal compensation error and updating the nonlinear error model.

2. The method of claim 1, wherein, Step S2 comprises: S21: performing wavelet packet decomposition on the displacement data and the vibration signal data to obtain a time feature vector; S22: constructing a space correlation matrix of temperature data and displacement data based on a structure topology graph of the numerical control machine tool and extracting space features through a graph convolution network to obtain a space feature vector.

3. The method of claim 2, wherein, Step S3 comprises: S31: obtaining adaptive fusion weights based on the time feature vector and the space feature vector; S32: obtaining a space-time fusion feature vector based on the time feature vector, the space feature vector, the adaptive fusion weights and a preset space-time feature fusion model.

4. The method of claim 3, wherein, The preset space-time feature fusion model is represented by the following formula: （1）； wherein, is a spatio-temporal fusion feature vector, is a temporal feature vector, is a spatial feature vector, is an adaptive fusion weight, is expressed by the following equation: （2）； wherein, and are learnable parameter matrices.

5. The method of claim 4, wherein, Step S4 comprises: S41: calculating displacement errors caused by temperature changes based on a thermal conductivity coefficient and a material expansion coefficient of the numerical control machine tool to construct a thermal error model; S42: calculating force deformation errors based on a cutting force and a structure flexibility matrix of the numerical control machine tool to construct a force error model.

6. The method of claim 5, wherein, The thermal error model is represented by the following formula: （3）； wherein, is the error due to temperature variation, is the material thermal expansion coefficient of the CNC machine, is the temperature variation, is the machine key components length vector, is the time; The force error model is represented by the following formula: （4）； wherein, is the force deformation error, is the structural flexibility matrix, is the cutting force vector, .

7. The method of claim 6, wherein, Step S5 comprises: constructing a nonlinear error model covering the linear error model of the numerical control machine tool based on the space-time fusion feature vector and a long short-term memory network; The nonlinear error is represented by the following formula: （5）； Wherein, LSTM is a long short-term memory network.

8. The method of claim 7, wherein, The total error prediction data is calculated by the following formula: （6）； wherein, is the total error prediction data.

9. The method of claim 8, wherein, Step S7 comprises: S71: constructing an objective function based on total error minimization and control smoothness as the target; S72: solving the objective function through a QP planning algorithm to obtain optimal compensation parameters; The objective function is represented by the following formula: （7）； The constraint condition of the objective function is: （8）； （9）； wherein is a compensation parameter vector, is a prediction time horizon, is a control time horizon, is a control quantity smoothness weight.

10. The method of claim 9, wherein, Step S8 comprises: S81: Compensate the multi-source data of the numerical control machine tool based on the optimal compensation parameter and update to obtain an actual error value after compensation; S82: Update the network parameters of the nonlinear error model by gradient descent algorithm and the residual error between the total error prediction data and the actual error value.

Images