Gantry machine tool overall real-time thermal error compensation method based on CNN-BIGRU-A

Abstract

The application relates to the technical field of machine tools and specifically discloses a gantry machine tool overall real-time thermal error compensation method based on CNN-BIGRU-A, which comprises the following steps: acquiring historical temperature data and historical thermal deformation data of a main shaft and a transmission shaft; constructing a thermal error prediction model, training the thermal error prediction model according to the historical temperature data and the historical thermal deformation data, so as to obtain a trained thermal error prediction model; transmitting current temperature data of the main shaft and the transmission shaft to the trained thermal error prediction model, so as to output a current thermal error compensation value of the main shaft and a current thermal deformation calculation value of the transmission shaft; a numerical control system of the machine tool calculates a current thermal error compensation value of a target position point on the transmission shaft according to the coordinates of the target position point; and the numerical control system of the machine tool compensates according to the current thermal error compensation value of the main shaft and the current thermal error compensation value of the target position point on the transmission shaft. The application can realize a compensation strategy with high frequency and high synchronization.

Description

Technical Field

[0001] The present invention relates to the technical field of machine tools, and more particularly to a gantry machine tool overall real-time thermal error compensation method based on CNN-BIGRU-A. Background Art

[0002] Machine tools are critical equipment in the manufacturing industry, and their machining accuracy directly impacts product quality and production efficiency. However, due to the accumulation of heat generated by friction, cutting, and electrical equipment during machining, various machine tool components experience temperature fluctuations and thermal deformation. This thermal deformation can alter the machine tool's geometric accuracy, thereby affecting the dimensional and shape accuracy of machined parts. Research indicates that thermal errors in machine tools account for 60%-70% of their total errors, presenting a critical obstacle to improving machining accuracy. Therefore, research on thermal error compensation for machine tools is of great practical significance.

[0003] Currently, the most widely used method for reducing machine tool thermal errors is the error model compensation method. This involves analyzing, summarizing, and screening the temperatures of key heat sources in machine tools, studying the mapping relationship between machine tool thermal errors and key heat source temperatures, and obtaining an error model. Traditional error compensation methods such as kernel principal component analysis, multi-scale transformation, support vector machine (SVM), and backpropagation neural networks (BPNN) are only applicable to specific data sets and rely on relevant human experience when extracting features. This results in low generalization performance of the error model, making it difficult to meet the model robustness requirements under external environmental changes and multiple processing conditions. This, to a certain extent, restricts the development of thermal error compensation technology. On the other hand, traditional error compensation technology mainly targets the spindle, and there is little research on the thermal errors of the machine tool drive shaft and the overall thermal errors of the machine tool. Compensating for the overall error of the machine tool is also a major challenge. Summary of the Invention

[0004] The present invention aims to overcome the above-mentioned defects and provides a real-time thermal error compensation method for gantry machine tools based on CNN-BIGRU-A, which can realize a compensation strategy with high frequency and high synchronization.

[0005] As a first aspect of the present invention, a method for real-time thermal error compensation of a gantry machine tool based on CNN-BIGRU-A is provided, comprising the following steps:

[0006] Step S1: Acquire historical temperature data and historical thermal deformation data of the machine tool spindle, and simultaneously acquire historical temperature data and historical thermal deformation data of each measuring point of the machine tool transmission shaft;

[0007] Step S2: A thermal error prediction model is constructed by fusing a one-dimensional convolutional layer, a bidirectional gated neural network layer, an attention mechanism layer, and a fully connected layer; historical temperature data of the machine tool spindle and historical temperature data of each measuring point of the machine tool transmission shaft are used as input feature samples, and historical thermal deformation data of the machine tool spindle and historical thermal deformation data of each measuring point of the machine tool transmission shaft are used as output feature samples; the thermal error prediction model is trained based on the input feature samples and the output feature samples to obtain a trained thermal error prediction model;

[0008] Step S3: reading current temperature data of the machine tool spindle, transmitting the current temperature data of the machine tool spindle to the trained thermal error prediction model for prediction, so as to output a current thermal error compensation value of the machine tool spindle, and then transmitting the current thermal error compensation value of the machine tool spindle to the machine tool numerical control system, and the machine tool numerical control system performs compensation according to the current thermal error compensation value of the machine tool spindle;

[0009] Step S4: reading the current temperature data of each measuring point of the machine tool transmission shaft, transmitting the current temperature data of each measuring point of the machine tool transmission shaft to the trained thermal error prediction model for prediction, so as to output the current thermal deformation calculation value of each measuring point of the machine tool transmission shaft; then fitting the coordinates of each measuring point of the machine tool transmission shaft and the current thermal deformation calculation value of each measuring point into a one-variable multi-order equation to obtain the fitting coefficient of the machine tool transmission shaft; finally, transmitting the fitting coefficient of the machine tool transmission shaft to the machine tool numerical control system, the machine tool numerical control system substitutes the coordinates of the target position point on the machine tool transmission shaft into the one-variable multi-order equation to calculate the current thermal error compensation value of the target position point on the machine tool transmission shaft, and the machine tool numerical control system performs compensation according to the current thermal error compensation value of the target position point on the machine tool transmission shaft.

[0010] Furthermore, the acquisition of historical temperature data and historical thermal deformation data of the machine tool spindle further includes:

[0011] After the machine is heated for a specified time, a variable speed experiment of the machine tool spindle is carried out. After each set of experiments, the machine tool spindle is rotated at a constant speed for a specified time, and then the machine is stopped to detect the historical temperature data of the machine tool spindle through a temperature sensor and the historical thermal deformation data of the machine tool spindle through a laser interferometer;

[0012] The historical thermal deformation data of the machine tool spindle includes historical thermal deformation data of the machine tool spindle in the X, Y and Z directions.

[0013] Furthermore, the acquisition of historical temperature data and historical thermal deformation data of each measuring point of the machine tool transmission shaft further includes:

[0014] After the machine is heated for a specified time, a fixed-point test of the machine tool transmission shaft is carried out. Each measuring point of the machine tool transmission shaft is set respectively. After each measuring point of the machine tool transmission shaft is moved into position, the machine is stopped to detect the historical temperature data of each measuring point of the machine tool transmission shaft by a temperature sensor and the historical thermal deformation data of each measuring point of the machine tool transmission shaft by a laser interferometer.

[0015] The historical thermal deformation data of each measuring point of the machine tool transmission shaft includes the historical thermal deformation data of each measuring point of the machine tool transmission shaft in the X, Y and Z directions.

[0016] Furthermore, it also includes:

[0017] After the machine has been heated for 5 minutes, the fixed-point test of the machine tool transmission shaft is carried out;

[0018] The machine tool transmission shaft is measured for temperature and thermal deformation at intervals of L / N, where L is the length of the machine tool transmission shaft, N is the number of measuring points on the machine tool transmission shaft, and the feed speed of the machine tool transmission shaft is 1000 mm / min. The machine tool transmission shaft stops moving after each measuring point is in place.

[0019] After the current measuring point of the machine tool transmission shaft moves into position, it stops moving to detect the historical temperature data of the current measuring point of the machine tool transmission shaft through a temperature sensor and the historical thermal deformation data of the current measuring point of the machine tool transmission shaft through a laser interferometer; then the machine tool transmission shaft continues to move until the historical temperature data of subsequent measuring points of the machine tool transmission shaft and the historical thermal deformation data of subsequent measuring points are all detected.

[0020] Furthermore, the method of using the historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool transmission shaft as input feature samples and using the historical thermal deformation data of the machine tool spindle and the historical thermal deformation data of each measuring point of the machine tool transmission shaft as output feature samples further includes:

[0021] Standardizing the historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool transmission shaft to obtain a plurality of standardized historical temperature data;

[0022] A grey correlation analysis algorithm is used to calculate the correlation between each standardized historical temperature data and its corresponding historical thermal deformation data. Historical temperature data with a correlation exceeding a preset threshold value is screened out from the plurality of standardized historical temperature data. The historical temperature data with a correlation exceeding the preset threshold value is used as an input feature sample, and the historical thermal deformation data corresponding to the historical temperature data with a correlation exceeding the preset threshold value is used as an output feature sample. The calculation formula for the correlation is as follows:

[0023]

[0024] Where: x0(k) is the historical thermal deformation data, [x1(k), x2(k), ..., x t (k)] is the standardized historical temperature data, and ρ is the gray resolution coefficient.

[0025] Furthermore, the thermal error prediction model is constructed by integrating a one-dimensional convolutional layer, a bidirectional gated neural network layer, an attention mechanism layer, and a fully connected layer, and further includes:

[0026] The one-dimensional convolutional layer and the bidirectional gated neural network layer are used to extract and reconstruct features; the attention mechanism layer is used to redistribute the weight parameters of each feature in the form of dot product in the feature dimension; the fully connected layer is used to map the dimension to the output dimension and generate the final prediction result;

[0027] The one-dimensional convolution layer includes two convolution layers, a pooling layer and a normalization layer;

[0028] The bidirectional gating neural network layer uses two hidden layers and a bidirectional network design; the operating formula of the bidirectional gating neural network layer is as follows:

[0029] z t =f(W z ·[h t-1 ,x t ])

[0030] r t =f(W r ·[h t-1 ,x t ])

[0031]

[0032] Where: W z 、W r 、W h is the weight matrix; f() is the activation function Sigmoid, g() is the activation function Tanh; r t is the reset gate; z t is the update gate; x t is the input data; h t-1 is the hidden layer state of the previous time step; is the candidate state of the hidden layer at the current time step; h t is the hidden layer output of the current time step;

[0033] The attention mechanism layer uses a feature attention mechanism based on dot products to reweight the network output. The operating formula of the attention mechanism layer is as follows:

[0034]

[0035] wherein Q, K, V represent query matrix, key matrix, value matrix respectively, and dk is the feature dimension of the key matrix and the query matrix.

[0036] Further, the training of the thermal error prediction model according to the input feature sample and the output feature sample to obtain the trained thermal error prediction model further comprises:

[0037] dividing the data set composed of the input feature sample and the output feature sample into a training set, a validation set and a test set;

[0038] training the thermal error prediction model with the input feature sample and the output feature sample in the training set, verifying the performance of the thermal error prediction model with the input feature sample and the output feature sample in the validation set, and evaluating the performance index of the trained thermal error prediction model with the input feature sample and the output feature sample in the test set; wherein the performance index includes root mean square error RMSE and mean absolute error MAE, and the calculation formula is:

[0039]

[0040] wherein y i is the true label of the test set data, and f(x i ) is the predicted label of the test set data obtained by the thermal error prediction model.

[0041] Further, the current thermal error compensation value of the machine tool spindle includes the current thermal error compensation value of the machine tool spindle X, Y and Z directions.

[0042] Further, the step S4 further comprises:

[0043] Taking the X direction of the machine tool transmission shaft as an example, the current temperature data of N measuring points in the X direction of the machine tool transmission shaft is read, and the current temperature data of N measuring points is reduced in dimension to obtain the current temperature data of m measuring points, the dimension is [n,m], wherein n is the number of temperature sensors, i.e. the current temperature data of each measuring point is detected by n temperature sensors; the current temperature data with the dimension of [n,m] is averaged in the first dimension to obtain the average current temperature data of m measuring points, and the dimension is converted to [1,m];

[0044] The average current temperature data of m measuring points in the X direction of the machine tool transmission shaft is transmitted to the trained thermal error prediction model for prediction to output the current thermal deformation calculation value [e1, e2, …, em] of m measuring points in the X direction of the machine tool transmission shaft. m

[0045] ​Assume that the coordinates of m measuring points in the X direction of the machine tool transmission axis are [x1, x2, ..., x m ], the coordinates of m measuring points [x1, x2, ..., x m ] and the current thermal deformation calculation values ​​of m measuring points [e1, e2, ..., e m ] is fitted into a cubic equation to obtain the fitting coefficient matrix [a, b, c, d] of the machine tool transmission shaft;

[0046] Finally, the fitting coefficient matrix [a, b, c, d] of the machine tool transmission shaft is transmitted to the machine tool numerical control system. The machine tool numerical control system substitutes the coordinates of the target position point on the machine tool transmission shaft into the one-variable multi-order equation to calculate the current thermal error compensation value of the target position point on the machine tool transmission shaft. The machine tool numerical control system performs compensation based on the current thermal error compensation value of the target position point on the machine tool transmission shaft.

[0047] Furthermore, before step S1, the method further includes:

[0048] A three-dimensional model of the machine tool is constructed, and the heat-sensitive areas of the machine tool spindle and the machine tool drive shaft are divided according to the heat conduction characteristics of the machine tool. Temperature collection points are selected in the heat-sensitive areas of the machine tool spindle and the machine tool drive shaft, and temperature sensors are installed at the temperature collection points. The historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool drive shaft are detected by the temperature sensor, and the detected historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool drive shaft are transmitted to the host computer.

[0049] The present invention provides a real-time thermal error compensation method for gantry machine tools based on CNN-BIGRU-A, which has the following beneficial effects: first, sensors are installed in the heat-sensitive area of ​​the machine tool, and data transmission software and hardware strategies such as adapters and serial port acquisition cards are prepared; second, spindle speed variation experiments and transmission shaft fixed measurement point experiments are conducted to collect temperature data and thermal deformation; then, temperature features are screened and reduced using error correlation analysis. A thermal error prediction model is then constructed by integrating algorithms such as 1DCNN, BIGRU, and an attention mechanism, and deep feature extraction is performed on the temperature data. Finally, according to the developed thermal error compensation software, the spindle thermal error compensation value is directly transmitted to the machine tool's numerical control system. The coordinates of each transmission shaft measurement point and the calculated thermal deformation value of each measurement point are fitted into a single-variable multi-order equation. The fitting coefficients are transmitted to the machine tool's numerical control system, and the thermal error compensation value of the target position on the machine tool's transmission shaft is calculated based on the real-time coordinates of the target position for real-time compensation. This can achieve a high-frequency and highly synchronized compensation strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention, but do not constitute a limitation of the present invention.

[0051] Figure 1 This is a flow chart of the overall real-time thermal error compensation method for gantry machine tools based on CNN-BIGRU-A provided by the present invention.

[0052] Figure 2 This is a graph of historical thermal deformation data in the three directions of X, Y, and Z of the machine tool spindle collected in the machine tool spindle speed change experiment provided by the present invention.

[0053] Figure 3 This is a structural diagram of the thermal error prediction model provided by the present invention.

[0054] Figure 4(a)-Figure 4(c) This is a diagram showing the thermal error prediction effect of the existing MLR model in the main axis X, Y, and Z directions during the training phase.

[0055] Figure 5(a)-Figure 5(c) This is the prediction effect diagram of the thermal error in the X, Y, and Z directions of the main axis of the existing MLR model during the testing phase.

[0056] Figure 6(a)-Figure 6(c) This is a diagram showing the thermal error prediction effect of the existing ANN model in the main axis X, Y, and Z directions during the training phase.

[0057] Figure 7(a)-Figure 7(c) This is a diagram showing the thermal error prediction effect of the existing ANN model in the main axis X, Y, and Z directions during the testing phase.

[0058] Figure 8(a)-Figure 8(c) This is a diagram showing the thermal error prediction effects of the CNN_BiGRU_A thermal error prediction model in the training phase in the main axis X, Y, and Z directions.

[0059] Figure 9(a)-Figure 9(c) This is a diagram showing the thermal error prediction effects of the CNN_BiGRU_A thermal error prediction model in the test phase in the main axis X, Y, and Z directions.

[0060] Figure 10(a)-Figure 10(f) This is a diagram of the thermal error compensation interface provided by the present invention. DETAILED DESCRIPTION

[0061] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of a CNN-BIGRU-A-based method for real-time thermal error compensation for gantry machine tools. It should be understood that the described embodiments are only a subset of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are considered within the scope of protection of the present invention.

[0062] It should be noted that the terms "first," "second," and the like in the specification and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate for the embodiments of the present invention described herein. In addition, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatuses.

[0063] In this embodiment, a real-time thermal error compensation method for a gantry machine tool based on CNN-BIGRU-A is provided. Figure 1 As shown, the overall real-time thermal error compensation method of the gantry machine tool based on CNN-BIGRU-A includes the following steps:

[0064] Step S1: Acquire historical temperature data and historical thermal deformation data of the machine tool spindle, and simultaneously acquire historical temperature data and historical thermal deformation data of each measuring point of the machine tool transmission shaft;

[0065] It should be noted that, before step S1, the following steps are also included:

[0066] A three-dimensional model of the machine tool is constructed, and the heat-sensitive areas of the machine tool spindle and the machine tool drive shaft are divided according to the heat conduction characteristics of the machine tool. Temperature collection points are selected in the heat-sensitive areas of the machine tool spindle and the machine tool drive shaft, and temperature sensors are installed at the temperature collection points. The historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool drive shaft are detected by the temperature sensor, and the detected historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool drive shaft are transmitted to the host computer through an adapter, a serial port acquisition card, etc.

[0067] In an embodiment of the present invention, 39 temperature sensors are installed in the heat-sensitive areas of the machine tool spindle and the machine tool transmission shaft, respectively. The installation locations are the front and rear bearings of the spindle (2 temperature sensors), the coolant inlet and outlet (2 temperature sensors), the left and right bed (6 temperature sensors), the left and right grating scales in the X direction (12 temperature sensors), the left and right grating scales in the Y direction (7 temperature sensors), the left and right grating scales in the Z direction (8 temperature sensors), the environment and the workbench (2 temperature sensors). The data are transmitted to the host computer through three serial port acquisition cards with 10 channels, and then the host computer returns the predicted current thermal error compensation value to the CNC machine tool for real-time compensation.

[0068] Preferably, the acquiring of historical temperature data and historical thermal deformation data of the machine tool spindle further includes:

[0069] After the machine is heated for a specified time, a variable speed experiment of the machine tool spindle is carried out. After each set of experiments, the machine tool spindle is rotated at a constant speed for a specified time, and then the machine is stopped to detect the historical temperature data of the machine tool spindle through a temperature sensor and the historical thermal deformation data of the machine tool spindle through a laser interferometer;

[0070] Wherein, the historical thermal deformation data of the machine tool spindle includes the historical thermal deformation data of the machine tool spindle in the X, Y and Z directions. Figure 2 As shown, it can be seen that there is a strong correlation between the three.

[0071] In the embodiment of the present invention, the machine is warmed up for two hours, and the machine tool spindle is subjected to a variable speed experiment with a speed gradient of 1000 r / min (1000 r / min, 2000 r / min, ..., 8000 r / min, ..., 2000 r / min, 1000 r / min), each speed lasting for 5 minutes. After that, the machine is stopped and temperature data is collected using a temperature sensor and thermal deformation data is collected using a laser interferometer.

[0072] Preferably, the step of obtaining the historical temperature data and the historical thermal deformation data of each measuring point of the machine tool transmission shaft further includes:

[0073] After the machine is heated for a specified time, a fixed-point test of the machine tool transmission shaft is carried out. The measurement points in the X, Y, and Z directions of the machine tool transmission shaft are set respectively. After the measurement points of the machine tool transmission shaft are moved into position, the machine is stopped to detect the historical temperature data of each measurement point of the machine tool transmission shaft through a temperature sensor and the historical thermal deformation data of each measurement point of the machine tool transmission shaft through a laser interferometer.

[0074] The historical thermal deformation data of each measuring point of the machine tool transmission shaft includes the historical thermal deformation data of each measuring point of the machine tool transmission shaft in the X, Y and Z directions.

[0075] Specifically, it also includes:

[0076] After the machine has been heated for 5 minutes, the fixed-point test of the machine tool transmission shaft is carried out;

[0077] Taking the X direction of the machine tool drive shaft as an example, temperature and thermal deformation measurements are performed every L / N interval in the X direction of the machine tool drive shaft, where L is the length of the machine tool drive shaft, N is the number of measuring points in the X direction of the machine tool drive shaft, and the feed speed of the machine tool drive shaft is 1000mm / min. Each measuring point in the X direction of the machine tool drive shaft moves to its position and then stops moving.

[0078] After the current measuring point in the X direction of the machine tool transmission axis moves to a position, the machine tool transmission axis stops moving, and detects the historical temperature data of the current measuring point in the X direction of the machine tool transmission axis by a temperature sensor and the historical thermal deformation data of the current measuring point in the X direction of the machine tool transmission axis by a laser interferometer; the machine tool transmission axis continues to move until the historical temperature data and the historical thermal deformation data of subsequent measuring points in the X direction of the machine tool transmission axis are all detected;

[0079] The Y-direction thermal error experiment and Z-direction thermal error experiment of the machine tool drive shaft are carried out with the same experimental idea as above.

[0080] Step S2: A thermal error prediction model is constructed by fusing a one-dimensional convolutional layer (1DCNN), a bidirectional gated neural network layer (BIGRU), an attention mechanism layer (Attention), and a fully connected layer; the historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool transmission shaft are used as input feature samples, and the historical thermal deformation data of the machine tool spindle and the historical thermal deformation data of each measuring point of the machine tool transmission shaft are used as output feature samples. The thermal error prediction model is trained according to the input feature samples and the output feature samples to obtain a trained thermal error prediction model;

[0081] Preferably, Figure 3 As shown, the thermal error prediction model is constructed by integrating a one-dimensional convolutional layer, a bidirectional gated neural network layer, an attention mechanism layer, and a fully connected layer, and further includes:

[0082] The one-dimensional convolutional layer and the bidirectional gated neural network layer are used to extract and reconstruct features; the attention mechanism layer is used to redistribute the weight parameters of each feature in the form of dot product in the feature dimension, so that the network can better focus on information related to error compensation in the input data to make more accurate predictions; the fully connected layer is used to map the dimension to the output dimension and generate the final prediction result;

[0083] The one-dimensional convolution layer includes two convolution layers, a pooling layer and a normalization layer;

[0084] 1DCNN is a variant of convolutional neural networks (CNNs) specifically designed for processing one-dimensional sequential data, such as time series and text data. Unlike traditional two-dimensional CNNs used in image processing, 1DCNNs capture local patterns and features within the input sequence by applying a one-dimensional convolution operation. In 1DCNNs, the convolution kernel is one-dimensional and slides along the length of the input sequence. Similar to two-dimensional CNNs, 1DCNNs also extract features from the data by learning the weights of the convolution kernel. These features may include local patterns, trends, and periodicity within the sequence. By stacking multiple convolutional and pooling layers, the network can gradually extract higher-level, more abstract features. By sliding the convolution kernel and performing element-by-element multiplication and accumulation operations on the one-dimensional input data, the network is able to learn the convolution kernel parameters for different features, such as edge detection and texture detection. 1DCNNs offer several advantages when processing sequential data. First, they can automatically learn and extract features from sequences, eliminating the need for hand-crafted feature extractors. Second, 1DCNNs can capture the relationship between local patterns and global structure, leading to a better understanding of sequential data. In addition, 1DCNN can also flexibly adapt to sequence data of different lengths and complexities by adjusting the size and number of convolution kernels. Assuming the input matrix size is w, the convolution kernel size is k, the stride is s, and the number of zero-padding layers is p, the size of the feature map generated after one-dimensional convolution is:

[0085]

[0086] Pooling layers, also known as downsampling, play a role in convolutional neural networks by reducing dimensionality and computational parameters, extracting important features, and maintaining translation invariance. Similar to convolution, pooling uses a sliding pooling kernel to extract features. Each time it slides over a region, the maximum value within that region is taken as the pooled output. This method is called max pooling. Alternatively, the mean of the outputs can be calculated, called mean pooling. After several layers of convolution and pooling operations, the resulting feature map is sequentially expanded row by row and concatenated into a vector.

[0087] BatchNormalization is a commonly used technique in deep learning, aiming to accelerate the training process of the model and improve the generalization ability of the model. It normalizes the activation of the intermediate layers of the network on each mini-batch of data, so that the input distribution of each layer is stable, and the internal covariate shift is reduced. Internal covariate shift refers to the change of the input distribution of each layer of the neural network, which makes the update of network parameters difficult, and thus affects the training effect of the model. The BatchNormalization layer is usually added after the convolutional layer or fully connected layer of the network as part of the network structure. During the training process, it reduces the internal covariate shift by normalizing each mini-batch of data, thereby accelerating the convergence speed of the model. Dropout layer is a regularization technique used to avoid overfitting, which randomly discards a part of the neurons in the network, so that the network is an incomplete sub-network during each forward propagation, thereby reducing the dependence between neurons, improving the generalization ability of the deep learning model, and reducing the risk of overfitting.

[0088] wherein the bidirectional gated neural network layer uses two hidden layers and a bidirectional network design; the GRU network is a variant of recurrent neural network (RNN), which not only overcomes the problem of gradient disappearance or gradient explosion in traditional RNN network when processing long-term dependencies, but also effectively avoids the problems of complex internal structure, long training time and overfitting risk of LSTM network. The GRU network combines the input gate and the forget gate of LSTM into an update gate to reduce the number of model parameters and improve the training efficiency of the model. The running formula of the bidirectional gated neural network layer is as follows:

[0089] z t =f(W z ·[h t-1 ,x t ])

[0090] r t =f(W r ·[h t-1 ,x t ])

[0091]

[0092] wherein: W z , W r , W h are weight matrices; f() is the activation function Sigmoid, g() is the activation function Tanh; r t is the reset gate; z t is the update gate; x t is the input data; ht-1 is the hidden layer state of the previous time step; is the candidate state of the hidden layer at the current time step; h t is the hidden layer output of the current time step;

[0093] Among them, the attention mechanism layer uses a feature attention mechanism based on dot product to reweight the network output. The model uses a convolutional neural network for feature extraction and is optimized through BatchNormalization and dropout layers to enhance the performance of the model.

[0094] The purpose of introducing the attention mechanism is to allow the neural network to focus on the relevant parts when processing the input data, so that it can automatically learn and selectively focus on the important information in the input, thereby improving the performance and generalization ability of the model. Common attention mechanisms include self-attention mechanism, channel attention mechanism and spatial attention mechanism. In this study, we adopted the self-attention mechanism. Depending on the application object, the self-attention mechanism can be divided into two types: based on the time step dimension and based on the output / hidden layer feature dimension. The self-attention mechanism based on the time step dimension aims to capture the relationship between different time steps in the sequence data, while the self-attention mechanism based on the output / hidden layer feature dimension focuses on capturing the correlation between features at different positions in the input sequence. In this study, we used a self-attention mechanism based on dot product. The operating formula of the attention mechanism layer is as follows:

[0095]

[0096] Among them, Q, K, and V represent the query matrix, key matrix, and value matrix respectively, and their dimensions are (n,d k )、(m,d k )、(m,d v) , n is the number dimension of the query, m is the number of key values, d k is the characteristic dimension of the key matrix and query matrix, and dv is the characteristic dimension of the value;

[0097] The calculation process is:

[0098] ① Calculate the dot product of the query vector and the key vector to obtain the matching score matrix S;

[0099]

[0100] Among them, q i T is the transpose of the query vector, k j is the key vector.

[0101] ②Perform Softmax normalization on the score matrix to obtain the attention weight matrix A;

[0102]

[0103] Among them, S ij is the matching score matrix, d k It is the feature dimension of key and query.

[0104] ③ Add the attention weight matrix A and the value matrix V to obtain the context tensor matrix C.

[0105]

[0106] Among them, A ij is the attention weight matrix, v j is a value matrix.

[0107] Preferably, the method of using the historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool transmission shaft as input feature samples and using the historical thermal deformation data of the machine tool spindle and the historical thermal deformation data of each measuring point of the machine tool transmission shaft as output feature samples further includes:

[0108] Standardizing the historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool transmission shaft to obtain a plurality of standardized historical temperature data; wherein the historical temperature data are all mapped to the interval [0,1];

[0109]

[0110] Where: x is the historical temperature data collected, x min is the minimum value of the historical temperature data collected, x max It is the maximum value of the historical temperature data collected.

[0111] A grey correlation analysis algorithm is used to calculate the correlation between each standardized historical temperature data and its corresponding historical thermal deformation data, and the historical temperature data with a correlation exceeding a preset threshold is screened out from the plurality of standardized historical temperature data to achieve a dimensionality reduction operation on the historical temperature data, which can better capture the global and local features of the historical temperature data, thereby achieving the purpose of reconstructing features, reducing model parameters and computational complexity, and improving model accuracy;

[0112] The historical temperature data (historical temperature data after dimensionality reduction) whose correlation exceeds the preset threshold is used as the input feature sample, and the historical thermal deformation data corresponding to the historical temperature data (historical temperature data after dimensionality reduction) whose correlation exceeds the preset threshold is used as the output feature sample; wherein, the calculation formula of the correlation is as follows:

[0113]

[0114] Where: x0(k) is the historical thermal deformation data, that is, the target sequence, [x1(k), x2(k), ..., x t (k)] is the standardized historical temperature data, that is, the comparison sequence, and ρ is the gray resolution coefficient, which ranges from [0,1] and is generally 0.5.

[0115] Preferably, the step of training the thermal error prediction model according to the input feature samples and the output feature samples to obtain a trained thermal error prediction model further includes:

[0116] The data set consisting of the input feature samples and the output feature samples is divided into a training set, a validation set, and a test set in a ratio of 6:2:2;

[0117] The thermal error prediction model is trained with the input feature samples and output feature samples in the training set, the performance of the thermal error prediction model is verified with the input feature samples and output feature samples in the validation set, and the performance indicators of the trained thermal error prediction model are evaluated with the input feature samples and output feature samples in the test set. LSTM, 1DCNN, GRU, and 1DCNN-BIGRU are simultaneously used to predict machine tool thermal errors and the performance is compared with 1DCNN_BIGRU_A to verify the superiority of the model; wherein the performance indicators include root mean square error (RMSE) and mean absolute error (MAE), and their calculation formulas are as follows:

[0118]

[0119] Among them, y i is the true label of the test set data, f(x i ) is the predicted label of the test set data obtained by the thermal error prediction model.

[0120] The CNN_BiGRU_A thermal error prediction model used in the present invention is compared with various network models to evaluate the performance of the model. Figure 4(a)-Figure 4(c) 、 Figure 5(a)-Figure 5(c) 、 Figure 6(a)-Figure 6(c) 、 Figure 7(a)-Figure 7(c) 、 Figure 8(a)-Figure 8(c) and Figure 9(a)-Figure 9(c) As shown, it can be seen that the thermal error prediction values ​​of the MLR model, ANN model and CNN_BiGRU_A thermal error prediction model on the training set are basically consistent with the true values, but the thermal error prediction values ​​of the MLR model and ANN model on the test set are significantly different from the true values, which proves that their generalization ability is insufficient. However, the CNN_BiGRU_A thermal error prediction model used in the present invention shows an ideal effect in thermal error prediction on the test set.

[0121] Step S3: reading current temperature data of the machine tool spindle, transmitting the current temperature data of the machine tool spindle to the trained thermal error prediction model for prediction, so as to output a current thermal error compensation value of the machine tool spindle, and then transmitting the current thermal error compensation value of the machine tool spindle to the machine tool numerical control system, and the machine tool numerical control system performs compensation according to the current thermal error compensation value of the machine tool spindle;

[0122] Preferably, the current thermal error compensation value of the machine tool spindle includes the current thermal error compensation value of the machine tool spindle in the X, Y, and Z directions, with a dimension of [1,3]. The software transmits the current thermal error compensation value of the machine tool spindle in the X, Y, and Z directions directly to the machine tool CNC system through the NC-Link protocol.

[0123] Step S4: reading the current temperature data of each measuring point of the machine tool transmission shaft, transmitting the current temperature data of each measuring point of the machine tool transmission shaft to the trained thermal error prediction model for prediction, so as to output the current thermal deformation calculation value of each measuring point of the machine tool transmission shaft; then fitting the coordinates of each measuring point of the machine tool transmission shaft and the current thermal deformation calculation value of each measuring point into a one-variable multi-order equation to obtain the fitting coefficient of the machine tool transmission shaft; finally, transmitting the fitting coefficient of the machine tool transmission shaft to the machine tool numerical control system, the machine tool numerical control system substitutes the coordinates of the target position point on the machine tool transmission shaft into the one-variable multi-order equation to calculate the current thermal error compensation value of the target position point on the machine tool transmission shaft, and the machine tool numerical control system performs compensation according to the current thermal error compensation value of the target position point on the machine tool transmission shaft.

[0124] Preferably, the step S4 further includes:

[0125] Taking the X direction of the machine tool drive shaft as an example, read the current temperature data of N measuring points in the X direction of the machine tool drive shaft, and perform dimensionality reduction on the current temperature data of the N measuring points to obtain the current temperature data of m measuring points. The dimension is [n, m], where n is the number of temperature sensors. That is, the current temperature data of each measuring point is detected by n temperature sensors. The current temperature data of the dimension [n, m] is averaged in the first dimension to obtain the average current temperature data of the m measuring points. The dimension is converted to [1, m].

[0126] The average current temperature data of m measuring points in the X direction of the machine tool transmission axis is transmitted to the trained thermal error prediction model for prediction, so as to output the current thermal deformation calculation value [e1, e2, ..., e m ];

[0127] Assume that the coordinates of m measuring points in the X direction of the machine tool transmission axis are [x1, x2, ..., x m ], the coordinates of m measuring points [x1, x2, ..., x m] and the current thermal deformation calculation values ​​of m measuring points [e1, e2, ..., e m ] is fitted into a cubic equation to obtain the fitting coefficient matrix [a, b, c, d] of the machine tool transmission shaft;

[0128] Finally, the fitting coefficient matrix [a, b, c, d] of the machine tool transmission shaft is transmitted to the machine tool numerical control system. The machine tool numerical control system substitutes the coordinates of the target position point on the machine tool transmission shaft into the one-variable multi-order equation to calculate the current thermal error compensation value of the target position point on the machine tool transmission shaft. The machine tool numerical control system performs compensation based on the current thermal error compensation value of the target position point on the machine tool transmission shaft.

[0129] In addition, the remaining functions of the software of the present invention also include temperature acquisition point setting, real-time display of temperature data, error compensation value display, data acquisition, etc.

[0130] As shown in Figure 10(a), the temperature sensor data window displays real-time data on the device's operating temperature. As shown in Figure 10(b), the operating status window counts and displays the number of normal and abnormal temperature sensor data acquisitions. As shown in Figure 10(c), the spindle thermal error prediction window displays the model-predicted spindle thermal error compensation values. This window displays the thermal error compensation values ​​for the spindle's X, Y, and Z directions in the same window. As shown in Figure 10(d), the drive shaft X prediction error window displays the model-predicted thermal error compensation value for the drive shaft in the X direction. As shown in Figure 10(e), the drive shaft Y prediction error window displays the model-predicted thermal error compensation value for the drive shaft in the Y direction. As shown in Figure 10(f), the drive shaft Z prediction error window displays the model-predicted thermal error compensation value for the drive shaft in the Z direction.

[0131] The present invention proposes a real-time thermal error compensation method for gantry machine tools based on CNN-BIGRU-A. First, the heat-sensitive areas are divided according to the heat conduction characteristics of the machine tool, temperature sensors are installed in the heat-sensitive areas, and a software and hardware strategy for data transmission is established; then, thermal error experiments on the main shaft and transmission shaft are carried out respectively to collect the temperature signals and thermal error signals of each measuring point, and the thermal error-related temperature characteristics of the main shaft and transmission shaft are screened out by using grey correlation analysis; then, a thermal error prediction model is constructed by integrating algorithms such as 1DCNN, BIGRU, and attention mechanism, and the model is used to perform deep feature extraction and error prediction on the temperature data; finally, according to the development of thermal error compensation software, the main shaft error is directly transmitted to the machine tool CNC system, the transmission shaft measuring point error and the measuring point coordinates are fitted into a one-variable multi-order equation, the fitting coefficient is transmitted to the CNC system, and real-time compensation is performed according to the real-time position calculation of the machine tool.

[0132] In response to the problem that machine tool thermal errors are difficult to compensate, the present invention proposes a real-time thermal error compensation method for gantry machine tools based on CNN-BIGRU-A, and carries out relevant experimental verification according to this method. The experimental verification proves the effectiveness of the algorithm model in the present invention.

[0133] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment, it is not intended to limit the present invention. Any technician familiar with the present profession can make slight changes or modifications to equivalent embodiments using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A real-time thermal error compensation method for gantry machine tools based on CNN-BIGRU-A, characterized in that: The steps include: Step S1: Acquire historical temperature data and historical thermal deformation data of the machine tool spindle, and simultaneously acquire historical temperature data and historical thermal deformation data of each measuring point of the machine tool transmission shaft; Step S2: A thermal error prediction model is constructed by fusing a one-dimensional convolutional layer, a bidirectional gated neural network layer, an attention mechanism layer, and a fully connected layer; historical temperature data of the machine tool spindle and historical temperature data of each measuring point of the machine tool transmission shaft are used as input feature samples, and historical thermal deformation data of the machine tool spindle and historical thermal deformation data of each measuring point of the machine tool transmission shaft are used as output feature samples; the thermal error prediction model is trained based on the input feature samples and the output feature samples to obtain a trained thermal error prediction model; Step S3: reading current temperature data of the machine tool spindle, transmitting the current temperature data of the machine tool spindle to the trained thermal error prediction model for prediction, so as to output a current thermal error compensation value of the machine tool spindle, and then transmitting the current thermal error compensation value of the machine tool spindle to the machine tool numerical control system, and the machine tool numerical control system performs compensation according to the current thermal error compensation value of the machine tool spindle; Step S4: reading the current temperature data of each measuring point of the machine tool transmission shaft, transmitting the current temperature data of each measuring point of the machine tool transmission shaft to the trained thermal error prediction model for prediction, so as to output the current thermal deformation calculation value of each measuring point of the machine tool transmission shaft; Then, the coordinates of each measuring point of the machine tool transmission shaft and the current thermal deformation calculation value of each measuring point are fitted into a one-variable multi-time equation to obtain the fitting coefficient of the machine tool transmission shaft; finally, the fitting coefficient of the machine tool transmission shaft is transmitted to the machine tool numerical control system, and the machine tool numerical control system substitutes the coordinates of the target position point on the machine tool transmission shaft into the one-variable multi-time equation to calculate the current thermal error compensation value of the target position point on the machine tool transmission shaft, and the machine tool numerical control system performs compensation according to the current thermal error compensation value of the target position point on the machine tool transmission shaft.

2. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 1, characterized in that: The acquisition of historical temperature data and historical thermal deformation data of the machine tool spindle also includes: After the machine is heated for a specified time, a variable speed experiment of the machine tool spindle is carried out. After each set of experiments, the machine tool spindle is rotated at a constant speed for a specified time, and then the machine is stopped to detect the historical temperature data of the machine tool spindle through a temperature sensor and the historical thermal deformation data of the machine tool spindle through a laser interferometer; The historical thermal deformation data of the machine tool spindle includes historical thermal deformation data of the machine tool spindle in the X, Y and Z directions.

3. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 1, characterized in that: The acquisition of historical temperature data and historical thermal deformation data of each measuring point of the machine tool transmission shaft also includes: After the machine is heated for a specified time, a fixed-point test of the machine tool transmission shaft is carried out. Each measuring point of the machine tool transmission shaft is set respectively. After each measuring point of the machine tool transmission shaft is moved into position, the machine is stopped to detect the historical temperature data of each measuring point of the machine tool transmission shaft by a temperature sensor and the historical thermal deformation data of each measuring point of the machine tool transmission shaft by a laser interferometer. The historical thermal deformation data of each measuring point of the machine tool transmission shaft includes the historical thermal deformation data of each measuring point of the machine tool transmission shaft in the X, Y and Z directions.

4. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 3, characterized in that: Also includes: After the machine has been heated for 5 minutes, the fixed-point test of the machine tool transmission shaft is carried out; The machine tool transmission shaft is measured for temperature and thermal deformation at intervals of L / N, where L is the length of the machine tool transmission shaft, N is the number of measuring points on the machine tool transmission shaft, and the feed speed of the machine tool transmission shaft is 1000 mm / min. The machine tool transmission shaft stops moving after each measuring point is in place. After the current measuring point of the machine tool transmission shaft moves into position, it stops moving to detect the historical temperature data of the current measuring point of the machine tool transmission shaft through a temperature sensor and the historical thermal deformation data of the current measuring point of the machine tool transmission shaft through a laser interferometer; then the machine tool transmission shaft continues to move until the historical temperature data of subsequent measuring points of the machine tool transmission shaft and the historical thermal deformation data of subsequent measuring points are all detected.

5. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 1, characterized in that: The method of using the historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool transmission shaft as input feature samples and using the historical thermal deformation data of the machine tool spindle and the historical thermal deformation data of each measuring point of the machine tool transmission shaft as output feature samples further includes: Standardizing the historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool transmission shaft to obtain a plurality of standardized historical temperature data; A grey correlation analysis algorithm is used to calculate the correlation between each standardized historical temperature data and its corresponding historical thermal deformation data. Historical temperature data with a correlation exceeding a preset threshold value is screened out from the plurality of standardized historical temperature data. The historical temperature data with a correlation exceeding the preset threshold value is used as an input feature sample, and the historical thermal deformation data corresponding to the historical temperature data with a correlation exceeding the preset threshold value is used as an output feature sample. The calculation formula for the correlation is as follows: Where: x0(k) is the historical thermal deformation data, [x1(k), x2(k), ..., x t (k)] is the standardized historical temperature data, and ρ is the gray resolution coefficient.

6. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 1, characterized in that: The thermal error prediction model is constructed by integrating a one-dimensional convolutional layer, a bidirectional gated neural network layer, an attention mechanism layer, and a fully connected layer, and further includes: The one-dimensional convolutional layer and the bidirectional gated neural network layer are used to extract and reconstruct features; the attention mechanism layer is used to redistribute the weight parameters of each feature in the form of dot product in the feature dimension; the fully connected layer is used to map the dimension to the output dimension and generate the final prediction result; The one-dimensional convolution layer includes two convolution layers, a pooling layer and a normalization layer; The bidirectional gating neural network layer uses two hidden layers and a bidirectional network design; the operating formula of the bidirectional gating neural network layer is as follows: z t =f(W z ·[h t-1 ,x t ]) r t =f(W r ·[h t-1 ,x t ]) Where: W z 、W r 、W h is the weight matrix; f() is the activation function Sigmoid, g() is the activation function Tanh; r t is the reset gate; z t is the update gate; x t is the input data; h t-1 is the hidden layer state of the previous time step; is the candidate state of the hidden layer at the current time step; h t is the hidden layer output of the current time step; The attention mechanism layer uses a feature attention mechanism based on dot products to reweight the network output. The operating formula of the attention mechanism layer is as follows: Among them, Q, K, and V represent the query matrix, key matrix, and value matrix respectively. k is the characteristic dimension of the key matrix and query matrix.

7. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 1, characterized in that: The step of training the thermal error prediction model according to the input feature samples and the output feature samples to obtain a trained thermal error prediction model further includes: Dividing the data set consisting of the input feature samples and the output feature samples into a training set, a validation set, and a test set; The thermal error prediction model is trained using the input feature samples and output feature samples in the training set, the performance of the thermal error prediction model is verified using the input feature samples and output feature samples in the validation set, and the performance indicators of the trained thermal error prediction model are evaluated using the input feature samples and output feature samples in the test set; wherein the performance indicators include the root mean square error (RMSE) and the mean absolute error (MAE), and their calculation formulas are: Among them, y i is the true label of the test set data, f(x i ) is the predicted label of the test set data obtained by the thermal error prediction model.

8. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 1, characterized in that: The current thermal error compensation value of the machine tool spindle includes the current thermal error compensation values ​​of the machine tool spindle in the X, Y, and Z directions.

9. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 1, characterized in that: The step S4 further includes: Taking the X direction of the machine tool transmission shaft as an example, the current temperature data of N measuring points in the X direction of the machine tool transmission shaft is read and the current temperature data of the N measuring points is reduced in dimension to obtain the current temperature data of m measuring points. The dimension is [n, m], where n is the number of temperature sensors, that is, the current temperature data of each measuring point is detected by n temperature sensors. The current temperature data of the dimension [n, m] is averaged in the first dimension to obtain the average current temperature data of the m measuring points, and the dimension is converted to [1, m]. The average current temperature data of m measuring points in the X direction of the machine tool transmission axis is transmitted to the trained thermal error prediction model for prediction, so as to output the current thermal deformation calculation value [e1, e2, ..., e m ]; Assume that the coordinates of m measuring points in the X direction of the machine tool transmission axis are [x1, x2, ..., x m ], the coordinates of m measuring points [x1, x2, ..., x m ] and the current thermal deformation calculation values ​​of m measuring points [e1, e2, ..., e m ] is fitted into a cubic equation to obtain the fitting coefficient matrix [a, b, c, d] of the machine tool transmission shaft; Finally, the fitting coefficient matrix [a, b, c, d] of the machine tool transmission shaft is transmitted to the machine tool numerical control system. The machine tool numerical control system substitutes the coordinates of the target position point on the machine tool transmission shaft into the one-variable multi-order equation to calculate the current thermal error compensation value of the target position point on the machine tool transmission shaft. The machine tool numerical control system performs compensation based on the current thermal error compensation value of the target position point on the machine tool transmission shaft.

10. The method for compensating the overall real-time thermal error of a gantry machine tool based on CNN-BIGRU-A according to claim 1, characterized in that: Before step S1, the method further includes: A three-dimensional model of the machine tool is constructed, and the heat-sensitive areas of the machine tool spindle and the machine tool drive shaft are divided according to the heat conduction characteristics of the machine tool. Temperature collection points are selected in the heat-sensitive areas of the machine tool spindle and the machine tool drive shaft, and temperature sensors are installed at the temperature collection points. The historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool drive shaft are detected by the temperature sensor, and the detected historical temperature data of the machine tool spindle and the historical temperature data of each measuring point of the machine tool drive shaft are transmitted to the host computer.

Images