Online evolutionary method for time-varying thermal error model of machine tool spindle based on long short-term memory network
Through the online evolution method of the machine tool spindle time-varying thermal error model based on long and short time memory network, the problem of poor adaptability of the time-varying thermal error model of CNC machine tools in complex environments is solved, and the long-term accuracy and production efficiency of the model are improved.
Patent Information
- Application Number
- CN202310065653.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-01-18
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2043-01-18
AI Technical Summary
The existing CNC machine tool time-varying thermal error model is difficult to adapt to environmental changes when facing complex processing conditions and long-term slow-varying working conditions, and relies on processing interval measurement data, affecting production efficiency and increasing the risk of failure.
The online evolution method of machine tool spindle time-varying thermal error model based on long and short time memory network is adopted to establish a model through the spindle temperature change speed and acceleration, and use the long and short time memory network to perform data-driven prediction to realize online update of model parameters and ensure that the model is in the long term. In-operation accuracy.
It improves the robustness and accuracy of the time-varying thermal error model, reduces the dependence on processing interval measurement, improves production efficiency, and reduces the failure rate of temperature measurement.
Smart Images

Figure CN115951633B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of error compensation for numerically controlled machine tools, and relates to an online evolution method for a time-varying thermal error model of a machine tool spindle based on a long short-term memory network. Background Art
[0002] Improving precision stability is of great significance to promoting the overall level of high-end CNC machine tools. Time-varying thermal error compensation is an effective means to improve the precision stability of CNC machine tools. The key to implementing time-varying thermal error compensation is to establish an accurate time-varying thermal error model. Since Switzerland discovered in 1933 that thermal deformation of machine tools is one of the main factors affecting positioning accuracy, a large number of time-varying thermal error models for CNC machine tools have been established, which has promoted the improvement of the precision stability of CNC machine tools. However, in the process of time-varying thermal error modeling, it is inevitable to use assumptions and introduce empirical formulas. Faced with complex machine tool processing conditions and long-term slow time-varying working conditions, how to ensure that the time-varying thermal error model adapts to the changes in the machine tool working environment under seasonal changes, and how to ensure that the time-varying thermal error model adapts to the thermal characteristics of the machine tool in the long term are key issues that need to be focused on and studied in depth.
[0003] In response to the above-mentioned time-varying thermal error compensation problem of machine tools, J. Mou et al. from the University of Illinois proposed an adaptive error compensation method that uses the error measured on the machine during machining intervals to update the error model. J. Ni et al. from the University of Michigan used the time-varying thermal error obtained by machining interval measurement to adjust the error model online. P. Blaser and J. Mayr et al. from the Swiss Federal Institute of Technology in Zurich used periodic machining interval measurement to obtain actual error data and regularly updated the parameters of the time-varying thermal error model. H. Liu et al. used the optimal temperature input data and the error data measured on the machine to adaptively update the model parameters. However, the current problems include: 1) relying on machining interval measurement error data, it is necessary to configure an error on-machine measurement device, such as a probe, a gauge block, etc.; 2) when measuring error data on the machine, it must be in a non-machining state, which inevitably affects production efficiency; 3) using temperature data as the judgment object for triggering error on-machine measurement or model update, it is necessary to set up multiple temperature measurement points, which not only increases the failure rate of temperature measurement, but also causes the problem of excessive wiring hindering the normal processing of the machine tool. Summary of the invention
[0004] The present invention focuses on the urgent need for long-term and accurate compensation of time-varying thermal errors of CNC machine tools, and provides a new online evolution method of the time-varying thermal error model of a machine tool spindle based on a long short-term memory network. The online evolution of the spindle time-varying thermal elongation error model can be achieved through an online update mechanism of model parameters in which the long short-term memory network and the spindle time-varying thermal elongation error model correct each other, so as to ensure that the spindle time-varying thermal elongation error model is always accurate during the long-term operation of the machine tool.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is:
[0006] An online evolution method for the time-varying thermal error model of a machine tool spindle based on a long short-term memory network is proposed. Firstly, a spindle time-varying thermal elongation error model is established based on the spindle temperature change speed and acceleration. The spindle temperature and the ambient temperature are used as the model input to calculate the spindle time-varying thermal elongation error. Then, a data-driven model for predicting the spindle time-varying thermal elongation error is established based on the long short-term memory network. The spindle temperature and the ambient temperature are used as the model input to predict the spindle time-varying thermal elongation error. After that, the spindle time-varying thermal elongation error model is used to calculate the data set E. z and long short-term memory network prediction dataset E z 'consistency; finally, the model online update mechanism is implemented according to the consistency judgment result. z and E z ' is consistent, then use the temperature data set T S 、T E and error data set E z Update the long short-term memory network for the prediction of spindle time-varying thermal elongation error. If E z and E z 'If not consistent, use the temperature data set T S 、T E and error data set E z 'Updating the time-varying thermal elongation error model of the spindle; the method comprises the following steps:
[0007] The first step is to establish a time-varying thermal elongation error model of the spindle based on the spindle temperature change rate and acceleration;
[0008] The spindle temperature change speed and acceleration are used to calculate the steady-state value of the spindle time-varying thermal elongation error:
[0009]
[0010] In the formula, E zs (t) is the steady-state value of the time-varying thermal elongation error of the spindle at time t, λ v is the coefficient related to the speed of spindle temperature change, λ a is the coefficient related to the acceleration of spindle temperature change, T S (t) and T S (0) are the spindle temperature values at time t and the initial time, T E (t) and T E (0) are the ambient temperature values at time t and the initial time, respectively; T(t) is the net increase of the spindle temperature compared to the ambient temperature at time t;
[0011] The calculation formula of the time-varying thermal elongation error of the spindle based on the spindle temperature change speed and acceleration is written as:
[0012] Ez (t) = E z (t-Δt)+[E zs (t)-E z (t-Δt)]·(1-e -Δt / τ ) (2)
[0013] In the formula, E z (t) is the time-varying thermal elongation error of the spindle at time t, and τ is the thermal inertia coefficient of the spindle;
[0014] The coefficient λ related to the spindle temperature change speed is calculated by using a nonlinear quadratic programming algorithm. v , coefficient λ related to the acceleration of spindle temperature change a And the thermal inertia coefficient τ of the spindle are identified, and the objective function is:
[0015]
[0016] In the formula, E zt (t) is the measured value of the time-varying thermal elongation error of the spindle at time t, E z (t) is the calculated value of the time-varying thermal elongation error of the spindle at time t, τ min , vmin and λ amin are the parameters τ and λ respectively. v and λ a The lower limit of the constraint condition, τ max , vmax and λ amax The coefficients τ and λ are v and λ a The upper limit of the constraint condition.
[0017] The second step is to establish a long short-term memory network for the prediction of spindle time-varying thermal elongation error;
[0018] A data-driven model for predicting the time-varying thermal elongation error of the spindle is established using a long short-term memory network. The model input is the spindle temperature T S (t) and ambient temperature T E (t), the output is the predicted value of the time-varying thermal elongation error of the main axis E z ′(t), the model form is as follows:
[0019] E z ′(t)=f(K,T S (t),T E (t)) (4)
[0020] Where K is the weight matrix and bias vector parameter of the long short-term memory network, and the spindle temperature T is used. S (t) and ambient temperature T E(t) The measured value is the training data, and the measured value of the time-varying thermal elongation error of the main axis is E zt (t) Train the LSTM network for the labeled training set.
[0021] The third step is to determine the spindle time-varying thermal elongation error model calculation data set E z And prediction dataset E z 'Consistency;
[0022] The collection period of spindle temperature and ambient temperature is t0. The spindle time-varying thermal elongation error model established in the first step calculates the spindle time-varying thermal elongation error in real time. When the time reaches the consistency judgment period T0, the data set calculated by the spindle time-varying thermal elongation error model is E z ; The spindle temperature and ambient temperature time series data set T obtained in the T0 period S and T E Input the long short-term memory network established in the second step to obtain the spindle time-varying thermal elongation error prediction data set E z ';
[0023] The consistency evaluation method (Bland-Altman method) is used to calculate the data set E according to the model of the spindle time-varying thermal elongation error stored in the consistency judgment period T0. z and long short-term memory network prediction dataset E z The mean A of the difference between the two sets of data and the width B of the consistency boundaries are used to judge the consistency of the two sets of data; the threshold A0 of the mean difference between the model calculation data set and the long short-term memory network prediction data set and the threshold B0 of the width between the consistency boundaries are obtained through experiments. The calculation methods of A, B, A0 and B0 are:
[0024]
[0025] Where m is the data volume of the model-calculated value and the long short-term memory network-predicted value of the spindle time-varying thermal elongation error within the consistency judgment period T0; n is the data volume of the model-calculated value and the long short-term memory network-predicted value of the spindle time-varying thermal elongation error obtained from the experiment when solving the thresholds A0 and B0; E z (t i ) and E z (t j ) are t i and t j E is the model calculation value of the time-varying thermal elongation error of the spindle; z ′(t i ) and E z ′(t j ) respectively correspond to t i and t j Time length short-term memory network prediction value;
[0026] The spindle time-varying thermal elongation error model based on spindle temperature change speed and acceleration is used in the working process. When the model calculates the data set E z and long short-term memory network prediction dataset E z When the mean value A of the difference between the two values and the width B of the consistency boundary satisfy: A>A0 or B>B0, the model is considered to calculate the data set E z and long short-term memory network prediction dataset E z ' is less consistent, that is, the accuracy of the spindle time-varying thermal elongation error model is considered to be worse; otherwise, it is considered that the model calculation data set E z and long short-term memory network prediction dataset E z 'Good consistency.
[0027] The fourth step is to update the time-varying thermal elongation error model of the spindle online according to the consistency judgment result;
[0028] According to the judgment result of the third step, if E z and E z 'If the consistency is poor, the spindle temperature time series set T obtained within the period T0 is used to judge the consistency S , ambient temperature time series set T E and the time series set E predicted by the long short-term memory network z 'Online identification of the parameters λ of the time-varying thermal elongation error model of the spindle v , a and τ;
[0029] If E z and E z 'If the consistency is good, the spindle time-varying thermal elongation error model is considered to be accurate. To ensure the consistency, the spindle temperature, ambient temperature and spindle time-varying thermal elongation error information within the judgment period T0 are memorized. The spindle temperature time series set T obtained within the T0 period is used. S , ambient temperature time series set T E and the spindle time-varying thermal elongation error model calculation data set E z The training set composed of the two layers is used to update the long short-term memory network obtained in the second step online.
[0030] Beneficial effects of the present invention:
[0031] (1) The spindle time-varying thermal elongation error model based on the spindle temperature change rate and acceleration has the ability to evolve online, ensuring the accuracy of the model in long-term operation and improving the robustness of the time-varying thermal error model;
[0032] (2) The implementation method of using the spindle time-varying thermal error model to accurately compensate for the long-term time-varying thermal error of the CNC machine tool spindle is clarified, providing an effective solution for the long-term robust compensation of the time-varying thermal error of the machine tool spindle;
[0033] (3) The method provided by the present invention has a certain degree of universality, and the model online evolution method provided by the present invention can be extended to the data-driven model online evolution in other fields. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 Flowchart for the implementation of the online evolutionary approach for the time-varying thermal error model of a machine tool spindle.
[0035] Figure 2 Schematic diagram of online evolution of the time-varying thermal error model of machine tool spindle. DETAILED DESCRIPTION
[0036] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings.
[0037] Taking the spindle of a certain type of horizontal milling and boring machine as an example, the implementation method of the present invention is described in detail. The CNC system of the horizontal milling and boring machine is a Huazhong 8 CNC system, the spindle speed range is 20r / min to 5000r / min, the cooling system temperature threshold is set to 25°C, and the spindle motor and the spindle are connected by a pulley. The implementation process of the online evolution method of the time-varying thermal error model of the machine tool spindle is as follows Figure 1 shown.
[0038] (1) Establish a time-varying thermal elongation error model of the spindle based on the spindle temperature change rate and acceleration;
[0039] A magnetic temperature sensor is used to obtain the spindle temperature and ambient temperature, with a temperature measurement accuracy of ±0.1℃ (5℃~45℃). An eddy current displacement sensor is used to measure the spindle axial time-varying thermal elongation error. The eddy current sensor is DT3005-U1-M-C1, with a linear range of 1mm, a resolution of 0.5μm, a sensitivity value of approximately 111um / V, and a data acquisition card of NI-USB6281. The spindle operates at a speed of 1300r / min for 4h, then stops for 5h. The temperature and error acquisition cycle is set to 10s, and the spindle temperature, ambient temperature and spindle time-varying thermal elongation error are obtained; due to the large amount of data, the temperature and error data are displayed at intervals of 5min as shown in Table 1:
[0040] Table 1 Spindle temperature, ambient temperature and spindle time-varying thermal elongation error
[0041]
[0042] The nonlinear quadratic programming algorithm is used to calculate the coefficient λ related to the spindle temperature change speed by using equations (1) to (3) and the spindle temperature, ambient temperature and spindle time-varying thermal elongation error test results in Table 1. v , coefficient λ related to the acceleration of spindle temperature changea The identification result is [τ,λ v ,λ a ]=[-9.437×10 5 ,1.996×10 4 ,1.68].
[0043] (2) Establish a long short-term memory network for the prediction of spindle time-varying thermal elongation error;
[0044] The number of layers of the LSTM network was set to 5, the learning rate was 0.01, and the gradient threshold was set to 1. The spindle temperature and ambient temperature measurements for the first 6 hours during the temperature and error test were used as training sets, and the spindle time-varying thermal elongation error measurements for the first 6 hours were used as labels to train the LSTM network and obtain a LSTM network model for spindle time-varying thermal elongation error prediction.
[0045] The spindle temperature, ambient temperature and spindle time-varying thermal elongation error measurement values in the last 3 hours of the temperature and error test were used as the test set to test the modeling effect of the long short-term memory network model. The test set data and test results are shown in Table 2:
[0046] Table 2 Test set data and test results
[0047]
[0048] From the test results, it can be seen that the difference between the model prediction results and the actual measurement results of the spindle time-varying thermal elongation error ranges from -1.4μm to 1.1μm. The results show that the long short-term memory network has a good effect on predicting the spindle time-varying thermal elongation error.
[0049] (3) Determine the spindle time-varying thermal elongation error model calculation data set E z And prediction dataset E z 'Consistency;
[0050] The collection period of spindle temperature and ambient temperature is set to t0 = 10s, and the consistency judgment period is set to T0 = 1h. The spindle time-varying thermal elongation error model calculates the spindle time-varying thermal elongation error in real time. When the time reaches the consistency judgment period T0 = 1h, the data set calculated by the spindle time-varying thermal elongation error model is E z ; The spindle temperature and ambient temperature time series data set T within 1 hour S and T E Input the long short-term memory network to obtain the spindle time-varying thermal elongation error prediction data set E z '; Calculation data set E of the spindle time-varying thermal elongation error model for 5 consecutive consistency judgment cycles z and long short-term memory network prediction dataset E z'As shown in Table 3:
[0051] Table 3 Calculation and prediction data sets of time-varying thermal elongation error of the spindle
[0052]
[0053] Using the data in Table 3, the thresholds A0 and B0 are calculated according to formula (5):
[0054]
[0055] Take A0 = 0.2, B0 = 1.7; use Bland-Altman method to determine E z and E z 'Consistency: Make the spindle time-varying thermal elongation error model based on spindle temperature change speed and acceleration work normally, and store the model calculation data set E within the consistency judgment cycle z and long short-term memory network prediction dataset E z 'As shown in Table 4:
[0056] Table 4 Calculation and prediction data sets of spindle time-varying thermal elongation error within the consistency judgment period
[0057]
[0058] Using the data in Table 4, the mean A and the width B between the consistency boundaries are calculated according to formula (5):
[0059]
[0060] At this time, the width B between the mean A and the consistency boundary satisfies: A>A0, B>B0, and the model calculation data set E z and long short-term memory network prediction dataset E z ' has poor consistency, that is, the accuracy of the spindle time-varying thermal elongation error model is poor.
[0061] (4) updating the time-varying thermal elongation error model of the spindle online according to the consistency judgment result;
[0062] The spindle temperature time series set T obtained within the period T0 is used to judge the consistency S , ambient temperature time series set T E and the time series set E predicted by the long short-term memory network z 'Online identification of the parameters λ of the time-varying thermal elongation error model of the spindle v , a and τ, the updated model parameters are [τ,λ v ,λ a ]=[-8.757×10 5 ,1.243×104 , [1.889]; Obtain the calculation data set and prediction data set of the spindle time-varying thermal elongation error during the consistency judgment period by using the updated model, as shown in Table 5:
[0063] Table 5 Calculation data set and prediction data set of spindle time-varying thermal elongation error during the consistency judgment period
[0064]
[0065] Using the data in Table 5, calculate the width B between the mean value A and the consistency boundary according to Equation (5):
[0066]
[0067] At this time, the width B between the mean value A and the consistency boundary satisfies: A < A0 and B < B0, and it is considered that the consistency between the model calculation data set E z and the long short-term memory network prediction data set E z ' is good, that is, the accuracy of the spindle time-varying thermal elongation error model is good.
[0068] The online evolution principle of the machine tool spindle time-varying thermal error model based on the long short-term memory network is as Figure 2 shown. The spindle time-varying thermal elongation error model is used to calculate the spindle time-varying thermal error calculation data set E z , and the long short-term memory network is used to calculate the spindle time-varying thermal error prediction data set E z '. When the time reaches the consistency judgment period T0, judge the consistency between the calculation data set E z and the prediction data set E z '; if the consistency between the two is good, then use the spindle and ambient temperature data sets T S and T E and the spindle time-varying thermal error calculation data set E z to update the long short-term memory network; if the consistency between the two is poor, then use the spindle and ambient temperature data sets T S and T E and the spindle time-varying thermal error prediction data set E z ' to update the spindle time-varying thermal error model.
[0069] According to the above example, the calculation accuracy of the spindle time-varying thermal error model after online parameter update is good, indicating that the present invention has a good effect on ensuring the accuracy of the time-varying thermal error model.
[0070] The above embodiments only represent the implementation manners of the present invention, but should not be construed as limiting the scope of the present invention patent. It should be noted that for those skilled in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention.
Claims
1. An online evolution method for a time-varying thermal error model of a machine tool spindle based on a long short-term memory network, characterized in that: The online evolution method first establishes a spindle time-varying thermal elongation error model based on the spindle temperature change speed and acceleration, takes the spindle temperature and the ambient temperature as the model input, and calculates the spindle time-varying thermal elongation error; then, a data-driven model for predicting the spindle time-varying thermal elongation error is established based on the long short-term memory network, takes the spindle temperature and the ambient temperature as the model input, and predicts the spindle time-varying thermal elongation error; then, the spindle time-varying thermal elongation error model is determined to calculate the data set E z and long short-term memory network prediction dataset E z 'consistency; finally, the model online update mechanism is implemented according to the consistency judgment result. z and E z ' is consistent, then use the temperature data set T S , T E and error data set E z Update the long short-term memory network for the prediction of spindle time-varying thermal elongation error. If E z and E z 'If not consistent, use the temperature data set T S , T E and error data set E z 'Update the time-varying thermal elongation error model of the spindle; The method comprises the following steps: The first step is to establish a time-varying thermal elongation error model of the spindle based on the spindle temperature change rate and acceleration; The spindle temperature change speed and acceleration are used to calculate the steady-state value of the spindle time-varying thermal elongation error: In the formula, E zs (t) is the steady-state value of the time-varying thermal elongation error of the spindle at time t, λ v is the coefficient related to the speed of spindle temperature change, λ a is the coefficient related to the acceleration of spindle temperature change, T S (t) and T S (0) are the spindle temperature values at time t and the initial time, T E (t) and T E (0) are the ambient temperature values at time t and the initial time, respectively; T(t) is the net increase of the spindle temperature compared to the ambient temperature at time t; The calculation formula of the time-varying thermal elongation error of the spindle based on the spindle temperature change speed and acceleration is written as: E z (t)=E z (t-Δt)+[E zs (t)-E z (t-Δt)]·(1-e -Δt / τ ) (2) In the formula, E z (t) is the time-varying thermal elongation error of the spindle at time t, and τ is the thermal inertia coefficient of the spindle; The coefficient λ related to the spindle temperature change speed is calculated by using a nonlinear quadratic programming algorithm. v , coefficient λ related to the acceleration of spindle temperature change a And the thermal inertia coefficient τ of the spindle are identified, and the objective function is: In the formula, E zt (t) is the measured value of the time-varying thermal elongation error of the spindle at time t, E z (t) is the calculated value of the time-varying thermal elongation error of the spindle at time t, τ min , vmin and λ amin are the parameters τ and λ respectively. v and λ a The lower limit of the constraint condition, τ max , vmax and λ amax The coefficients τ and λ are v and λ a The upper limit of the constraint condition; The second step is to establish a long short-term memory network for the prediction of spindle time-varying thermal elongation error; A data-driven model for predicting the time-varying thermal elongation error of the spindle is established using a long short-term memory network. The model input is the spindle temperature T S (t) and ambient temperature T E (t), the output is the predicted value of the time-varying thermal elongation error of the main axis E z ′(t), the model form is as follows: E z ′(t)=f(K,T S (t),T E (t)) (4) Where K is the weight matrix and bias vector parameter of the long short-term memory network, and the spindle temperature T is used. S (t) and ambient temperature T E (t) The measured value is the training data, and the measured value of the time-varying thermal elongation error of the main axis is E zt (t) Train a long short-term memory network for the labeled training set; The third step is to determine the spindle time-varying thermal elongation error model calculation data set E z And prediction dataset E z 'Consistency; The collection period of spindle temperature and ambient temperature is t0. The spindle time-varying thermal elongation error model established in the first step calculates the spindle time-varying thermal elongation error in real time. When the time reaches the consistency judgment period T0, the data set calculated by the spindle time-varying thermal elongation error model is E z ; The spindle temperature and ambient temperature time series data set T obtained in the T0 period S and T E Input the long short-term memory network established in the second step to obtain the spindle time-varying thermal elongation error prediction data set E z '; The consistency evaluation method is used to calculate the data set E according to the model of the spindle time-varying thermal elongation error stored in the consistency judgment period T0. z and long short-term memory network prediction dataset E z 'The mean value A of the difference between them and the width B of the consistency boundaries, combined with the threshold value A0 of the mean value of the difference between the model calculation data set and the long short-term memory network prediction data set and the threshold value B0 of the width between the consistency boundaries, judge the consistency of the two sets of data; The spindle time-varying thermal elongation error model based on spindle temperature change speed and acceleration is used in the working process. When the model calculates the data set E z and long short-term memory network prediction dataset E z When the mean value A of the difference between the two values and the width B of the consistency boundary satisfy: A>A0 or B>B0, the model is considered to calculate the data set E z and long short-term memory network prediction dataset E z ' is less consistent, that is, the accuracy of the spindle time-varying thermal elongation error model is considered to be worse; otherwise, it is considered that the model calculation data set E z and long short-term memory network prediction dataset E z 'Good consistency; The fourth step is to update the time-varying thermal elongation error model of the spindle online according to the consistency judgment result; According to the judgment result of the third step, if E z and E z 'If the consistency is poor, the spindle temperature time series set T obtained within the period T0 is used to judge the consistency S , ambient temperature time series set T E and the time series set E predicted by the long short-term memory network z 'Online identification of the parameters λ of the time-varying thermal elongation error model of the spindle v , a and τ; If E z and E z 'If the consistency is good, the spindle time-varying thermal elongation error model is considered to be accurate. To ensure the consistency, the spindle temperature, ambient temperature and spindle time-varying thermal elongation error information within the judgment period T0 are memorized. The spindle temperature time series set T obtained within the T0 period is used. S , ambient temperature time series set T E and the spindle time-varying thermal elongation error model calculation data set E z The training set composed of the two layers is used to update the long short-term memory network obtained in the second step online.
2. The online evolution method of a time-varying thermal error model of a machine tool spindle based on a long short-term memory network according to claim 1 is characterized in that: In the third step, the threshold A0 of the mean difference between the model calculation data set and the long short-term memory network prediction data set and the threshold B0 of the width between the consistency boundaries are obtained through experiments. The calculation method of A, B, A0 and B0 is: Wherein, m is the data volume of the model calculation value of the spindle time-varying thermal elongation error and the long short-term memory network prediction value within the consistency judgment period T0; n is the data volume of the model calculation value of the spindle time-varying thermal elongation error and the long short-term memory network prediction value obtained from the experiment when solving the thresholds A0 and B0; E z (t i ) and E z (t j ) are t i and t j E is the model calculation value of the time-varying thermal elongation error of the spindle; z ′(t i ) and E z ′(t j ) respectively correspond to t i and t j Time length short-term memory network prediction value.
