A Method and System for Constructing Thermal Error Difference Models for Machine Tool Closed-Loop Feed Systems Based on STSVM

Abstract

This invention belongs to the field of machine tool thermal error modeling technology, and discloses a method and system for constructing a thermal error difference model for a machine tool's fully closed-loop feed system based on STSVM. The method includes acquiring the temperature and positioning error of the machine tool's fully closed-loop feed system under different operating conditions; based on the temperature and position information, using a Long Short-Term Memory (LSTM) network to extract temporal features of thermal characteristics, and using a multi-scale convolutional network to extract spatial features of thermal characteristics; fusing the temporal and spatial features through a fully connected layer to obtain spatiotemporal features; using the spatiotemporal features and the position information as input, and the corresponding positioning error as output, training a support vector machine to obtain a thermal error difference model. This method not only eliminates the dependence on temperature-sensitive point selection and variability issues of traditional models, but also improves generalization ability through spatiotemporal features, effectively overcomes the defect of deep learning easily getting trapped in local optima by utilizing the global optimization characteristics of SVM, and achieves high-precision thermal error prediction.

Description

Technical Field

[0001] This invention belongs to the field of machine tool thermal error modeling technology, and relates to a method and system for constructing thermal error difference models of a machine tool full closed-loop feed system based on STSVM. Background Technology

[0002] Machine tools are widely used in the manufacture of various parts requiring high precision and high quality machining of holes and cavities. During precision machining, the non-uniform distribution of the temperature field causes uneven deformation of related machine tool components, which causes the tool to deviate from the ideal machining position, thus generating thermal errors.

[0003] Modeling of machine tool thermal errors includes mechanistic analytical modeling and data-driven modeling. Mechanistic analytical modeling relies on numerical simulation or heat transfer-based methods to construct thermal error prediction models. However, it often requires making numerous assumptions during model building, leading to deviations between model boundary conditions and actual operating conditions. Data-driven modeling directly establishes the mapping relationship between temperature and error through empirical models. However, traditional data-driven modeling methods require the selection of temperature-sensitive points, and these selected points may exhibit collinearity, spurious correlation, and variability during machine tool operation, limiting the accuracy of model predictions. Existing data-driven models lack mechanistic support and are prone to getting trapped in local optima during training. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for constructing thermal error difference models of a machine tool's fully closed-loop feed system based on STSVM.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for constructing a thermal error difference model for a machine tool's fully closed-loop feed system based on STSVM, comprising the following steps: acquiring the temperature and positioning error of the machine tool's fully closed-loop feed system under different operating conditions; wherein, the positioning error includes the position information of the corresponding measurement point; based on the temperature and position information, using a long short-term memory network to extract temporal features of thermal characteristics, using a multi-scale convolutional network to extract spatial features of thermal characteristics, fusing the temporal features and spatial features through a fully connected layer to obtain spatiotemporal features; using the spatiotemporal features and the position information as input, and the corresponding positioning error as output, training a support vector machine to obtain a thermal error difference model.

[0006] Furthermore, the thermal error differential model is obtained through heterogeneous training of a convolutional neural network, a long short-term memory network, and a support vector machine, with the output of the support vector machine being... :

[0007] in, For positive Lagrange multipliers, For negative Lagrange multipliers, For kernel function, This is a bias term.

[0008] Furthermore, the temporal characteristic of the thermal properties is the thermal expansion of the grating ruler; the thermal expansion of the grating ruler is in... At that time :

[0009] in, The coefficient of thermal expansion of the grating ruler is... The initial temperature of the grating ruler. For thermal diffusivity, For the current moment, For time intervals, Here are the axial position coordinates of the grating ruler. For spatial integration variables, This is the total length of the grating ruler.

[0010] Furthermore, the spatial characteristic of the thermal properties is the thermal drift error; the thermal drift error is... :

[0011] in, Thermal drift of the fixed point of the grating ruler This refers to the thermal drift of the reading head relative to the center point of the tool.

[0012] Furthermore, the acquisition of the temperature of the machine tool's fully closed-loop feed system under different operating conditions includes: setting temperature sensors at the heat source and thermal error sensitive points of the machine tool's fully closed-loop feed system, converting the voltage signal collected by the temperature sensor into an analog signal through the gearbox, and inputting the analog signal to the industrial control computer.

[0013] Furthermore, the acquisition of the positioning error of the machine tool's fully closed-loop feed system under different working conditions includes: using a laser interferometer to measure the actual displacement of the machine tool's fully closed-loop feed system as it passes through the temperature measuring point, and the difference between the actual displacement and the ideal displacement is the positioning error; the difference between the positioning error and the initial cold-state error is the positioning error.

[0014] Furthermore, the acquisition of the initial cold state error includes: the X-axis feed system reciprocates N times in the initial experimental state, and the average value of the machine tool positioning error in the N runs is the initial cold state error.

[0015] Furthermore, the long short-term memory network includes a forget gate, an input gate, an output gate, and a storage unit; The output vector of the forget gate is :

[0016] in, For the sigmoid function, Here is the weight matrix for the forget gate. The short-term memory output from the model at the previous time step and passed over. In order to be in The length of time in short-term memory is the vector of independent variables input from the outside. For the bias term of the forget gate, This is the transpose operation of a matrix; The output vector of the input gate is :

[0017] in, Here is the weight matrix of the input gate. This is the bias term for the input gate; The output vector of the output gate is :

[0018] in, Let be the weight matrix of the output gate. For the output gate bias; The output vector of the storage unit is :

[0019]

[0020] in, The output vector of the output gate. This represents the current state of the cell. This represents the cell state at the previous moment. This is the output vector of the forget gate. The output vector of the input gate. To generate the weight matrix for candidate memory cells, Bias terms for generating candidate memory cells.

[0021] Furthermore, the objective function of the support vector machine is: :

[0022] in, For the weight vector, As a penalty factor, For the upper boundary slack variable, This is a lower boundary slack variable.

[0023] This invention also provides a thermal error difference modeling system for a machine tool fully closed-loop feed system based on STSVM, comprising: an acquisition module for acquiring the temperature and positioning error of the machine tool fully closed-loop feed system under different operating conditions; wherein the positioning error includes the position information of the corresponding measurement point; an extraction module for extracting temporal features of thermal characteristics using a long short-term memory network and extracting spatial features of thermal characteristics using a multi-scale convolutional network based on the temperature and position information, and fusing the temporal features and spatial features through a fully connected layer to obtain spatiotemporal features; and a training module for training a support vector machine with the spatiotemporal features and the position information as input and the corresponding positioning error as output to obtain a thermal error difference model.

[0024] Compared with the prior art, the present invention has the following beneficial technical effects: This invention presents a method for constructing a thermal error differential model for a machine tool's fully closed-loop feed system based on STSVM. It acquires the temperature and positioning error (including measurement point location information) under different operating conditions of the machine tool's fully closed-loop feed system, avoiding the prediction bias caused by incomplete data in traditional modeling. By accurately extracting temporal and spatial features of thermal characteristics through a Long Short-Term Memory (LSTM) network and a multi-scale convolutional network, respectively, and fusing these features through a fully connected layer, it overcomes the shortcomings of traditional data-driven modeling, which can only capture single-dimensional features and has insufficient feature utilization. The fused spatiotemporal features and location information are used as input, and the positioning error is used as output to train a support vector machine, constructing a thermal error differential model. This not only eliminates the dependence on temperature-sensitive point selection and variability issues of traditional models, but also improves generalization ability through spatiotemporal features. Utilizing the global optimization characteristics of SVM, it effectively overcomes the tendency of deep learning to get trapped in local optima, achieving high-precision thermal error prediction. Attached Figure Description

[0025] Figure 1 This is a flowchart of a method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to the present invention; Figure 2 This is a thermal analysis diagram of the grating ruler in an embodiment of the present invention; Figure 3 This is a simplified one-dimensional thermal deformation diagram of the grating ruler in an embodiment of the present invention; Figure 4 This is a thermal drift error diagram of the fixing point of the grating ruler in an embodiment of the present invention; Figure 5 This is a thermal drift error diagram of the grating scale heads in an embodiment of the present invention; Figure 6 This is a schematic diagram of the temperature sensor installation structure in an embodiment of the present invention; Figure 7 This is a schematic diagram of the laser interferometer installation structure in an embodiment of the present invention; Figure 8 This is a schematic diagram illustrating the principle of linear feed axis positioning error measurement in an embodiment of the present invention. Figure 9 This is a schematic diagram of the structure of a long short-term memory network in an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of a multi-scale convolutional neural network in an embodiment of the present invention; Figure 11 This is a schematic diagram of the parallel structure of a long short-term memory network and a multi-scale convolutional neural network in an embodiment of the present invention; Figure 12 This is a schematic diagram of the thermal error difference modeling method for a fully closed-loop machine tool feed system based on STSVM in an embodiment of the present invention. Figure 13 This is a curve showing the change in training loss value of the thermal error differential model in an embodiment of the present invention; Figure 14 This is a comparison chart of the thermal error predicted by different models and the actual measured thermal error in the embodiments of the present invention. In the chart, a is the thermal error predicted by the thermal error differential model, b is the thermal error predicted by CNN-LSTM, c is the thermal error predicted by CNN, d is the thermal error predicted by LSTM, and e is the thermal error predicted by SVM. Figure 15 The values ​​represent the prediction residuals of different models in the embodiments of the present invention. a is the prediction residual of the thermal error differential model, b is the prediction residual of CNN-LSTM, c is the prediction residual of CNN, d is the prediction residual of LSTM, and e is the prediction residual of SVM. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0027] Example 1 This invention discloses a method for constructing thermal error difference models of a machine tool's fully closed-loop feed system based on STSVM, such as... Figure 1As shown, the method includes acquiring the temperature and positioning error of the machine tool's fully closed-loop feed system under different operating conditions; wherein, the positioning error includes the position information of the corresponding measurement point; based on the temperature and position information, a long short-term memory network is used to extract the temporal features of the thermal characteristics, a multi-scale convolutional network is used to extract the spatial features of the thermal characteristics, and the temporal features and spatial features are fused through a fully connected layer to obtain spatiotemporal features; the spatiotemporal features and the position information are used as inputs, and the corresponding positioning error is used as output to train a support vector machine to obtain a thermal error differential model.

[0028] The thermal error of the grating ruler in the fully closed-loop feed system directly generates the positioning error. The thermal error of the fully closed-loop feed system includes the thermal expansion of the grating ruler and the thermal drift error.

[0029] The main factors affecting the temperature at the left end of the grating ruler are heat transfer from the motor and support bearings, and convection from the environment. The right end is primarily affected by the support bearings and ambient temperature. The grating ruler is fixed to the column at its center. Because both sides of the grating ruler are flexibly mounted without constraints, its expansion occurs from the center outwards. The grating ruler is affected by heat as follows... Figure 2 As shown. The length of the grating ruler is much larger than its diameter, so radial temperature changes are ignored, and axial temperature changes can be considered as a one-dimensional heat conductor. The simplified heat transfer model for one side of the grating ruler is as follows. Figure 3 As shown.

[0030] Starting from the origin of the coordinate system anchored at the midpoint of the left end surface of the shaft, the heat conduction equation satisfies:

[0031] in, The sign of the partial derivative. For temperature, Here are the axial position coordinates of the grating ruler. The time taken for heat conduction. For thermal diffusivity, The surface convective heat transfer coefficient is... The surface area of ​​the grid ruler participating in convective heat dissipation. The thermal conductivity of the grating ruler is... Let be the volume of the grating ruler.

[0032] Ignoring the transfer between the environment and the grating scale, the heat conduction equation can be simplified to:

[0033] in, The initial temperature of the grating ruler.

[0034] Solve using Fourier transform:

[0035] in, The initial temperature of the grating ruler. For spatial integration variables.

[0036] According to the theory of thermal expansion, the formula for the thermal deformation of a grating ruler with respect to temperature must satisfy:

[0037] When the time is At that time, the thermal expansion of the grating ruler is:

[0038] When the time is At that time, the thermal expansion of the grating ruler is:

[0039] in, The coefficient of thermal expansion of the grating ruler is... The thermal conductivity of the grating ruler is... The initial temperature of the grating ruler. For thermal diffusivity, For the current moment, For time intervals, Here are the axial position coordinates of the grating ruler. For spatial integration variables, This is the total length of the grating ruler.

[0040] grating ruler in thermal expansion and The thermal expansion of the grating ruler is closely related to the thermal expansion over time, meaning that the current thermal expansion is influenced by the previous thermal expansion state. As time progresses, the axial thermal expansion of the grating ruler gradually accumulates. Therefore, the thermal expansion of the grating ruler exhibits temporal characteristics.

[0041] like Figure 4 As shown, the main heat source of the X-axis feed system is concentrated on the left side of the column, and the grating ruler is installed close to the motor drive end. Therefore, the uneven temperature gradient generated on the column surface after the machine tool runs will drive the upper grating ruler fixing point away from the motor drive end, which manifests as an overall thermal drift towards the rear bearing housing. The column has a structure with a constrained lower end and a free upper end, so the thermal drift error of the lower grating can be approximately ignored. Figure 5 As shown, the X-axis feed system's reading head is fixedly installed close to the leadscrew nut, which easily leads to localized high temperatures and thermal expansion deformation. Furthermore, the reading point is not in the same vertical position as the tool center point, and heating can cause the actual reading point of the grating ruler to deviate to the ideal position to the left. ,and The offset will cause the reading head to read a position smaller than the actual position, which, after closed-loop feedback, will cause the tool center point to deviate to the right along the X-axis. .

[0042] The thermal drift error of the X-axis feed system is :

[0043] in, Thermal drift of the fixed point of the grating ruler This refers to the thermal drift of the reading head relative to the center point of the tool.

[0044] In addition to the heat generated by the motor and the front bearing, the heat generated by the rear bearing and the guide rail slider can also affect the temperature field of the column through heat conduction, thus affecting the thermal drift of the grating ruler fixing point. Similarly, The thermal error of the linear encoder is not only related to the heat generated by the lead screw nut, but also to nearby heat-generating components such as the guide rail slider and the Y-axis motor, which exacerbate the thermal deformation of the connection between the reading head and the slide. Therefore, the thermal error of the linear encoder is affected by heat sources in different spatial locations of the machine tool, giving the thermal drift error spatial characteristics.

[0045] The temperature and positioning error of the machine tool's fully closed-loop feed system under different operating conditions are obtained. The temperature is obtained by installing temperature sensors at heat sources and thermal error-sensitive points within the system, converting the voltage signals collected by the temperature sensors into analog signals via a gearbox, and then inputting these analog signals to the industrial control computer. The positioning error is obtained by measuring the actual displacement of the system as it passes through temperature measurement points using a laser interferometer; the difference between the actual displacement and the ideal displacement is the positioning error; the difference between the positioning error and the initial cold-state error is also considered the positioning error.

[0046] Specifically, PT100 platinum resistance magnetic temperature sensors are placed at the main heat sources and thermal error-sensitive parts (such as grating rulers, guide rails, lead screws, and nuts). The specific arrangement of the temperature measuring points is as follows: Figure 6As shown, T1 measures the ambient temperature, T2, T8, and T15 measure the motor temperature, T3, T7, and T17 measure the temperature at the front support bearing end, T13 and T20 measure the temperature at the rear support bearing end, T10, T11, and T12 measure the temperature of the grating ruler housing, T6, T9, T18, and T21 measure the temperatures on both sides of the slide, T4, T14, T16, and T22 measure the temperatures on both sides of the guide rail, and T5 and T19 measure the nut temperature. The voltage signal from the PT100 sensor is converted into a standard analog signal by a transmitter. The NI data acquisition card inputs the standard analog signal to the industrial control computer. A program is written in the LabVIEW software on the industrial control computer to achieve real-time acquisition and storage of temperature data. A Renishaw XL80 laser interferometer is used to measure the positioning error of the feed system.

[0047] The environmental compensator, interferometer, and reflector are arranged as follows: Figure 7 The laser interferometer, mounted at the indicated location on the machine tool accessory, emits a helium-neon laser and measures the machine tool displacement in real time using the principle of interference. The laser interferometer is connected to a host computer via a USB interface, and the Capture software installed on the host computer records error information in real time. In this embodiment, the working stroke of the X-axis dual-drive feed system is 500mm. Eleven temperature measuring points are set at 50mm intervals. Machine tool G-code is written, and the laser interferometer measures the actual displacement of the machine tool as the feed system passes the measuring points. The difference between the actual displacement and the ideal displacement is the positioning error of the current measuring point.

[0048] In the initial experimental state, the X-axis feed system reciprocated three times, and the average of the machine tool positioning errors from these three runs was taken as the initial cold-state error (geometric error). The positioning error was measured every 30 X-axis reciprocations. The thermally induced positioning error was obtained by subtracting the initial cold-state error from the final positioning error. Figure 8 As shown. The standard operating temperature of this machine tool is 20℃. Different ambient temperatures and different feed speeds (3m / min, 4m / min, 5m / min) were set within the range of 19℃-26℃, and corresponding temperature and positioning error data were collected. Twelve sets of experiments (K1-K12) were conducted, and the specific experimental conditions are as follows: K1-4m / min-20℃, K2-5m / min-20℃, K3-3m / min-22℃, K4-5m / min-22℃, K5-4m / min-24℃, K6-3m / min-26℃, K7-3m / min-18℃, K8-4m / min-16℃, K9-5m / min-24℃, K10-3m / min-24℃, K11-4m / min-22℃, K12-3m / min-20℃.

[0049] Temporal features of thermal characteristics are extracted using a long short-term memory network, spatial features of thermal characteristics are extracted using a multi-scale convolutional network, and the temporal and spatial features are fused through a fully connected layer to obtain spatiotemporal features.

[0050] Long Short-Term Memory (LSTM) networks, such as Figure 9 As shown, it includes a forget gate, an input gate, an output gate, and a storage unit.

[0051] The output vector of the forget gate is :

[0052] in, For the sigmoid function, Here is the weight matrix for the forget gate. The short-term memory output from the model at the previous time step and passed over. In order to be in The length of time in short-term memory is the vector of independent variables input from the outside. For the bias term of the forget gate, This is the transpose operation of a matrix; The output vector of the input gate is :

[0053] in, Here is the weight matrix of the input gate. This is the bias term for the input gate; The output vector of the output gate is :

[0054] in, Let be the weight matrix of the output gate. For the output gate bias; The output vector of the storage unit is :

[0055]

[0056] in, The output vector of the output gate. This represents the current state of the cell. This represents the cell state at the previous moment. This is the output vector of the forget gate. The output vector of the input gate. To generate the weight matrix for candidate memory cells, Bias terms for generating candidate memory cells.

[0057] CNN networks can extract spatial features through repeated convolution operations. However, the CNN hot error prediction model, which mainly uses a single convolutional kernel structure, limits the model's spatial perception range of the feature matrix. This invention uses convolutional kernels of different sizes to construct a multi-scale CNN network, also known as a multi-scale convolutional network (MCNN), to perform multi-level perception of spatial features, such as... Figure 10 As shown, MCNN uses 1×1, 3×3 and 5×5 convolutional kernels to extract spatial features of thermal properties in multiple layers.

[0058] When CNN and LSTM are combined sequentially, information loss can occur due to the feature extraction method, such as... Figure 11 As shown, this invention fuses the temporal features of thermal characteristics extracted by the Long Short-Term Memory Network and the spatial features of thermal characteristics extracted by the Multi-Scale Convolutional Network through a fully connected layer to obtain spatiotemporal features.

[0059] Using the spatiotemporal features and location information as inputs, and the corresponding positioning error as output, a support vector machine is trained, such as... Figure 12 As shown, the thermal error differential structure model is obtained. Support Vector Machine (SVM) is a statistical model based on minimizing structural risk and cannot be directly used in the training of deep learning algorithms.

[0060] The objective function of the support vector machine is :

[0061] in, For the weight vector, As a penalty factor, For the upper boundary slack variable, This is a lower boundary slack variable.

[0062]

[0063]

[0064]

[0065]

[0066] in, Let be the predicted value of the thermal error differential model for the i-th input feature. This represents the actual true value corresponding to the i-th input feature. For insensitive loss threshold, These are slack variables at the upper boundary.

[0067] The input dataset for constructing the MCNN-LSTM-SVM (STSVM) thermal error differential prediction model based on multi-condition temperature-thermal error data is divided into training and validation sets in an 8:2 ratio.

[0068] The parameters of the thermal error differential model are optimized using a grid search method. Specifically, the grid search method is used to optimize the temporal length of the input to the thermal error differential model, the number of neurons in the LSTM hidden layer, the number of neurons in the fully connected layer, the learning rate, the width of the SVM kernel function, and the SVM regularization coefficient. Actual thermal error in each sample With predicted thermal error The root mean square (MSE) error is used as the fitness function of the grid search method.

[0069]

[0070] The hyperparameters that minimize the fitness function on the validation set are selected as the optimal hyperparameters for the thermal error differential model. Through grid search, the time series lengths of the MCNN and LSTM modules are found to be 30 and 16, respectively, with a learning rate of 0.12, 32 neurons in the LSTM hidden layer, a kernel width of 0.85 for the SVM, an SVM regularization coefficient of 9, and 16 neurons in the fully connected layer. Figure 13 The figure shows the convergence curve of the loss value of the thermal error differential model during training after the parameters are determined. It converges at approximately the 300th iteration, with a convergent MSE loss value of around 0.05.

[0071] A thermal error differential model was trained using training and validation sets, and its predictive performance for positioning errors was validated on a test set. Thermal characteristic experiments were conducted under variable speed conditions to obtain the dataset K13-(3-5-4-3.5-4.5)-24℃, and the predictive performance of the thermal error differential model was evaluated using the K13 prediction set. The predicted thermal error range of the thermal error differential model is -0.06 to 8.62 μm, which is very close to the actual measured thermal error range of -0.08 to 8.76 μm. Figure 14 As shown in Figure a, the predicted curve is basically consistent with the actual thermal error change curve. Figure 14 The prediction thermal error range of the CNN-LSTM model shown in b is -0.32 to 9.16 μm. Figure 14As shown in Figures c, d, and e, the thermal error prediction ranges of the CNN, LSTM, and SVM models significantly deviate from the actual range, recorded as -0.10 to 9.90 μm, 0.15 to 10.10 μm, and 0.15 to 9.65 μm, respectively. The thermal error predicted by the SVM model differs significantly from the actual error curve, indicating that SVM struggles to train on high-dimensional data, leading to underfitting. The prediction curve of the CNN model exhibits considerable fluctuations due to the lack of temporal features during training. The prediction curves of the CNN-LSTM and LSTM models are similar, but because the LSTM model cannot extract the spatial features of thermal information, the trend of thermal error changes in the later stages of the experiment differs greatly from the actual thermal error. Furthermore, the serial structure of the CNN-LSTM model causes mutual interference during feature extraction, resulting in inaccurate thermal error prediction. The STSVM model overcomes the limitations of the above algorithms, extracting the temporal and spatial features of thermal information more comprehensively, and using SVM to solve the global optimum for this regression task, demonstrating superior prediction accuracy.

[0072] like Figure 15 As shown in Figure a, the prediction residuals of the thermal error differential model are within ±1 μm, improving the positioning accuracy of the machine tool. Figure 15 The maximum prediction residuals of the CNN-LSTM, CNN, LSTM and SVM models shown in b, c, d and e are 1.5μm, 1.57μm, 1.77μm and 1.54μm, respectively, and their prediction performance is lower than that of the thermal error differential model.

[0073] Example 2 This invention discloses a thermal error difference modeling system for a machine tool fully closed-loop feed system based on STSVM, comprising an acquisition module, an extraction module, and a training module.

[0074] The acquisition module is used to acquire the temperature and positioning error of the machine tool's fully closed-loop feed system under different operating conditions; wherein, the positioning error includes the position information of the corresponding measurement point; the extraction module is used to extract the temporal features of thermal characteristics using a long short-term memory network and the spatial features of thermal characteristics using a multi-scale convolutional network based on the temperature and position information, and to fuse the temporal and spatial features through a fully connected layer to obtain spatiotemporal features; the training module is used to train a support vector machine with the spatiotemporal features and the position information as input and the corresponding positioning error as output to obtain a thermal error differential model.

[0075] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

Claims

1. A method for constructing a thermal error difference model of a machine tool's fully closed-loop feed system based on STSVM, characterized in that, Includes the following steps: The temperature and positioning error of the machine tool's closed-loop feed system under different operating conditions are obtained; wherein, the positioning error includes the position information of the corresponding measurement point; Based on the temperature and location information, a long short-term memory network is used to extract the temporal features of the thermal characteristics, a multi-scale convolutional network is used to extract the spatial features of the thermal characteristics, and the temporal and spatial features are fused through a fully connected layer to obtain spatiotemporal features. Using the spatiotemporal features and location information as inputs, and the corresponding positioning error as output, the support vector machine is trained to obtain the thermal error differential model.

2. The method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to claim 1, characterized in that: The thermal error heterogeneous model is obtained by heterogeneous training of a convolutional neural network, a long short-term memory network, and a support vector machine. The output of the support vector machine is... : in, For positive Lagrange multipliers, For negative Lagrange multipliers, For kernel function, This is a bias term.

3. The method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to claim 2, characterized in that: The temporal characteristics of the thermal properties are the thermal expansion of the grating ruler; The thermal expansion of the grating ruler is within At that time : in, The coefficient of thermal expansion of the grating ruler is... The initial temperature of the grating ruler. For thermal diffusivity, For the current moment, For time intervals, Here are the axial position coordinates of the grating ruler. For spatial integration variables, This is the total length of the grating ruler.

4. The method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to claim 3, characterized in that: The spatial characteristic of the thermal properties is thermal drift error; The thermal drift error is : in, Thermal drift of the fixed point of the grating ruler This refers to the thermal drift of the reading head relative to the center point of the tool.

5. The method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to claim 1, characterized in that: The acquisition of the temperature of the machine tool's fully closed-loop feed system under different operating conditions includes: Temperature sensors are installed at the heat source and thermal error sensitive points of the machine tool's closed-loop feed system. The voltage signal collected by the temperature sensor is converted into an analog signal by the gearbox and then input to the industrial control computer.

6. The method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to claim 5, characterized in that: The method for obtaining the positioning error of the machine tool's fully closed-loop feed system under different operating conditions includes: The actual displacement of the machine tool's closed-loop feed system passing through the temperature measuring point is measured using a laser interferometer. The difference between the actual displacement and the ideal displacement is the positioning error. The difference between the positioning error and the initial cold-state error is the positioning error.

7. The method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to claim 6, characterized in that: The acquisition of the initial cold state error includes: In the initial state of the experiment, the X-axis feed system reciprocates N times, and the average of the machine tool positioning error in N runs is the initial cold state error.

8. The method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to claim 1, characterized in that: The long short-term memory network includes a forget gate, an input gate, an output gate, and a storage unit; The output vector of the forget gate is : in, For the sigmoid function, Here is the weight matrix for the forget gate. The short-term memory output from the model at the previous time step and passed over. In order to be in The length of time in short-term memory is the vector of independent variables input from the outside. For the bias term of the forget gate, This is the transpose operation of a matrix; The output vector of the input gate is : in, Here is the weight matrix of the input gate. This is the bias term for the input gate; The output vector of the output gate is : in, Let be the weight matrix of the output gate. For the output gate bias; The output vector of the storage unit is : wherein, is an output vector of the output gate, is a cell state at a current time, is a cell state at a previous time, is an output vector of the forget gate, is an output vector of the input gate, is a weight matrix for generating a candidate memory cell, is a bias term for generating a candidate memory cell.

9. The method for constructing thermal error difference models of a machine tool fully closed-loop feed system based on STSVM according to claim 1, characterized in that: The objective function of the support vector machine is: : in, For the weight vector, As a penalty factor, For the upper boundary slack variable, This is a lower boundary slack variable.

10. A thermal error difference modeling system for a machine tool fully closed-loop feed system based on STSVM, characterized in that, include: Acquisition module: used to acquire the temperature and positioning error of the machine tool's closed-loop feed system under different operating conditions; wherein, the positioning error includes the position information of the corresponding measurement point; Extraction module: Based on the temperature and location information, it uses a long short-term memory network to extract the temporal features of thermal characteristics, uses a multi-scale convolutional network to extract the spatial features of thermal characteristics, and fuses the temporal and spatial features through a fully connected layer to obtain spatiotemporal features; Training module: Used to train the support vector machine by taking the spatiotemporal features and the location information as inputs and the corresponding positioning error as output, to obtain the thermal error differential model.

Images