1. Introduction
The technical level of CNC machine tools directly affects the development of the manufacturing industry. The accuracy and stability of CNC machine tools are crucial for product quality [
1]. The spindle system, whose accuracy is one of the core indexes to measure the performance of a CNC machine tool, is the end effector during the machining process of the workpiece [
2]. As a complex assembly composed of spindle, bearing supports, lubrication, cooling, and other component structures, its machining error is the result of the comprehensive effect from multiple factors related to the components and their interrelationships [
3]. Heat-related factors account for 40% to 70% of all factors which cause accuracy loss of the spindle system [
4,
5]. Starting from the factors affecting the thermal characteristics to study their comprehensive effect on the temperature and deformation of the spindle system is an effective method which can improve the performance of CNC machine tools.
In recent years, there have been numerous studies on the thermal characteristics and their influencing factors on the spindle system. Lee et al. [
6] researched the changes in temperature and deformation of the spindle system due to the influence of spindle speed by the finite element method and thermal analysis method. The relationship between thermal deformation and vibration was analyzed and modeled, which was verified experimentally. Raja et al. [
7] developed a coupled fluid–thermal model for spindles. The model can simulate the fluid–structural conjugate heat transfer of the spindle system. They measured the minimum deviation of the model through experiments as 7.6%. Marez et al. [
8] developed an approach based on TFs which is used for effective thermal error modeling for machine tools. This approach provides insight into the share of each source in the total machine thermal error through a combination of linear parametric models. Kaftan and Wegener et al. [
9] studied the adverse effects of internal and external heat sources on the complex non-symmetric structure of machine tools. They developed a novel method based on artistic intelligence that compensates for thermal errors associated with hidden boundary condition changes to address the drawbacks of traditional methods. Tan et al. [
10] studied the thermal characteristics of spindle bearings under preloading and achieved nonlinear prediction of thermal characteristics. The nonlinear changes in contact angle, contact force, preload, and stiffness were analyzed by considering the effects of contact thermal resistance, bearing parameters, lubricant viscosity, and time-varying temperature on the heat source. The proposed method significantly reduced the computational workload. Liu and Ma et al. [
11] proposed a closed-loop iterative modeling method. The heat generation of bearings and built-in motors, convection coefficient of bearing joints, and thermal contact resistance were calculated by modifying the heat source and thermal boundary conditions in each calculation step. By considering the comprehensive effects of lubricant viscosity changes and bearing thermal preload, the bearing heat generation was corrected. Xiang et al. [
12] proposed a data-driven prediction approach which can establish a dynamic linear model of spindle thermal error. The serious contradictions in traditional prediction methods have been resolved by predicting current thermal errors through historical temperature data without physical mechanism information. Wu et al. [
13] established a thermo-mechanical coupling analysis model for the spindle bearing system of machine tools. Through this model, they obtained the relationship between bearing preload and frictional heat generation, as well as between coolant and system thermal balance. Brecher, Wenkler, and Ihlenfeldt et al. [
14,
15] also explored the thermal related influencing factors of CNC machine tools in the research of thermal error modeling.
After the CNC machine tool is started, its state and surrounding environment exhibit nonlinear time-varying characteristics. The comprehensive effect of multiple changing factors is an important cause for the poor thermal characteristics and accuracy loss of CNC machine tools. At present, according to the research of many scholars, the research on factors such as heat sources, bearing loads, and coolant is relatively in-depth. The nonlinear and time-varying effects of these factors on the spindle have also been studied separately. However, there are few reports on solving the thermal characteristics of the spindle system under the comprehensive effects of multiple nonlinear time-varying factors such as heat sources, cooling conditions, surrounding environment, and heat transfer capability. Therefore, this paper establishes a numerical solving model for the nonlinear time-varying thermal characteristics of the spindle system. Based on the time-varying variables of the model, mathematical models of nonlinear time-varying factors such as friction torque generated by lubricant, convective heat transfer coefficient, and coolant and ambient temperature are constructed through theoretical derivation and data fitting. The time-varying temperature of the spindle system is solved taking into account multiple factors. Based on the time-varying value of the temperature, the deformation of the spindle at each moment is calculated. This paper provides a theoretical basis for the thermal characteristics study of spindle systems under complex working conditions.
2. Establishment of the Nonlinear Time-Varying Thermal Characteristics Model
During the operation of CNC machine tools, the heat transfer process of mechanical components in the spindle system follows the control equation shown in Equation (1) [
16].
where
ρ is the material density of the component,
CP is specific heat,
T is temperature,
t is time,
k is thermal conductivity, and
S is source terms.
Based on the finite volume method, this study will use hybrid grids to partition the control volume of the physical model for the spindle system. For each control volume, the central difference scheme and time implicit format are introduced. The pressure-based solver of Fluent 2022 R2 software will be used for solving. The least square cell-based and second-order upwind are adopted for spatial discretization. Therefore, within the allowable range of discretization error and numerically controllable, Equation (1) can be expanded as
where
and
are the average temperature of the local control volume
P at the
i-th and
i−1th time step,
VP is the volume of
P,
M is the number of adjacent control volumes to
P,
is the average temperature of control volume
N adjacent to
P at the
i-th time step,
kPN is the harmonic mean of thermal conductivity between control volume
P and
N,
is the distance between the center of
P and
N,
APN is the heat transfer surface area between
P and
N,
s is the intensity of the heat source,
is the
i-th time step size, and
.
The heat flux on both sides of the interface between the spindle system and the external fluid zone in contact with the system is equal under the state of fluid–solid coupling heat transfer [
17]. According to Fourier’s law and Newton’s flow equation, Equation (3) can be obtained.
where
L is the distance along the vector normal to the interface,
h is the convective heat transfer coefficient, and
is the temperature difference between the fluid and wall surface.
Assuming the temperature of fluid in contact with the spindle system is
TF, Equation (4) can be established within the allowable range of discretization error and numerically controllable.
where
kp is the thermal conductivity of
P,
is the temperature of the wall surface at the
i-th time step,
is the distance between the center of
P and the wall surface, and
is the convective heat transfer coefficient of the fluid.
In components, there is obviously Equation (5), as follows.
Take
. According to Equations (4) and (5) above, Equation (6) can be obtained.
It is assumed that the thermal conductivity of the spindle system material is constant. The nonlinear time-varying thermal characteristics of spindle systems should be considered from heat generation and heat dissipation. The heat generation is reflect in the intensity of the heat sources whose values at the
i-th time step can be expressed as
in the control equation. And the heat dissipation is usually related to the external ambient conditions and the capabilities to transfer heat outward, which is mainly reflected in the convective heat transfer coefficient and the temperature of the fluid in contact with the system. Their values at the
i-th time step can be expressed as
and
in the control equation. By combining Equations (2) and (6), the nonlinear time-varying thermal characteristics model shown in Equation (7) can be derived through substituting
,
, and
.
where
J is the number of convective heat transfer surfaces of
P, and
APF is the convective heat transfer surface area of control volume
P.
The heat transfer equation shown in Equation (7) is the model for the nonlinear time-varying thermal characteristics of the spindle system. According to the model, , , and are the variables which can affect the value of the temperature of control volume P and its adjacent control volume N. They are usually nonlinear and time-varying. As the definite solution conditions for the temperature of the spindle system, the models of nonlinear and time-varying factors related to heat source intensity, convective heat transfer coefficient, and the temperature of fluid in contact with the system should be constructed and substituted into the solution.
4. Experimental Platform Construction and Data Detection
This paper conducts experimental research on the G1160 CNC machine tool. This CNC machine tool adopts No. 32 cooling lubricating oil, whose cooling channel is spiral shaped. The motor drives the spindle through a synchronous belt. The spindle is equipped with 7014 C angular contact ball bearings with DT installation method. The internal structure of the G1160 CNC machine tool spindle system is shown in
Figure 1.
The spindle runs at a speed of 6000 r/min. The intelligent thermal characteristic detection and compensation instrument for the CNC machine tool spindles is adopted to detect data of the spindle system. The working temperature of this instrument is 4–50 °C, and the sampling frequency is 5 s per time. The system temperature and ambient temperature are detected through a PT100 temperature sensor manufactured by Heraeus in Germany. It has a 16 bit A/D converter and a measurement accuracy of 0.4%. The cooling oil temperature
Toil of the CNC machine tool is detected and read using the RCO-15PTS oil cooler digital temperature display instrument which is equipped with the CNC machine tool and produced by Ruike Refrigeration Plant Co., Ltd, Dongguan, China. To detect the spindle deformation, a BT40 inspection rod with a diameter of 40 mm and a length of 300 mm is clamped onto the spindle. The axial deformation of the inspection rod is detected by the intelligent thermal characteristic detection and compensation instrument for CNC machine tool spindles, using the CPL230 sensor produced by LION PRECISION in America (Minneapolis, MN, USA). It includes a high-precision capacitive displacement probe and a compact multi-channel driver. As a non-contact sensor, its detection range is 125–375 μm and the working temperature is 4–50 °C. The RMS resolution of the sensor is 18 nm. The construction of an experimental platform for the data detection of the G1160 CNC machine tool spindle system and the instruments are shown in
Figure 2.
Taking into account the structure of the spindle system, as well as the heat source and heat dissipation, the temperature sensors numbered T1–T8 were arranged according to
Figure 3. The temperature values detected by the T1–T8 sensor are
Ti,
i = 1, 2, … The temperature data of the spindle end, spindle bearings, flange, spindle body, both sides of the spindle box, and spindle box support plate are detected separately. A temperature sensor numbered T9 is placed in the air to detect the ambient temperature
Tair simultaneously. The sensor used to detect the axial displacement
Zs of the inspection rod is numbered Z. The settings for sensor positions are shown in
Figure 3 and the description of the sensor positions is provided in
Table 1.
The detected sample values of
Tair and
Toil used for fitting at each time point are shown in
Figure 4.
5. The Solution of Thermal Characteristics Based on Multiple Time-Varying Factors
It can be seen from Equations (10) and (11) that the friction torque generated by the lubricant is affected by the characteristics of kinematic viscosity. According to reference [
26], the nonlinear time-varying characteristics of the No. 32 oil can be described using Equation (18).
The torque
M1 generated by the load on the bearing can be calculated according to Equation (19) [
27].
where
f1 is a coefficient determined by the structure and load of the bearing, and
P1 is the calculated load for the bearing torque. For angular contact ball bearings,
f1 and
P1 can be calculated according to Equation (20) [
28].
where
F0 and
C0 are the equivalent static load and rated static load of the bearing, and
Fr and
Fa are the radial and axial load of the bearing.
F0 is calculated according to Equation (21).
where
X0 and
Y0 are equivalent static load coefficients.
If
P1 obtained from Equation (20) is less than
Fr, then
P1 =
Fr. The load on the bearing of the spindle installing the inspection rod mainly includes the bearing preload
Fp, the synchronous belt compression spindle force
FQ, and the gravity
G of the rotating component.
Fp provided by the enterprise is 700 N.
G can be obtained by summing up the masses of each component.
FQ is calculated according to Equation (22) [
29].
where
KA is the operating condition coefficient,
Pm is the nominal power,
d1 is the diameter of the small pulley, and
n1 is the small pulley speed.
According to
Figure 4, the ambient temperature rise range is only 2.9 °C during the detection period. The air property parameters in the convective heat transfer coefficient are all monotonic functions with air temperature as the independent variable. Equation (23) is constructed to determine the sensitivity of multiple air property parameters to the convective heat transfer coefficient. Low-sensitivity property parameters can be set as constants to simplify calculations.
where
Sg is the sensitivity,
is the extreme difference of the convective heat transfer coefficient affected by a certain property parameter within the detection temperature range,
is the time averaged value of the convective heat transfer coefficient during the experimental period,
g(
Tmax) and
g(
Tmin) are the peak values of a certain property parameter at the maximum and minimum detection values of experimental temperature, respectively, and
te is the end time of the experimental period.
According to reference [
30], the natural convection heat transfer coefficient is much smaller than the forced convection heat transfer coefficient. Considering the actual radiation heat dissipation, the equivalent natural convection heat transfer coefficient is set to 9.7 W/(m
2·°C).
Tair(
t) and
Toil(
t) are the important nonlinear time-varying factors affecting the thermal characteristics of the spindle system. The fitting degree of their mathematical models for experimental data can affect the accuracy of the calculation results. The fitting degree is evaluated using root mean square error (
RMSE) and
R-
square. For this study,
RMSE was calculated according to Equation (24) [
31].
where
n is the number of data samples, and
T(
ti) and
are the detected value of the experiment and the fitting value of the mathematical model at the
i-th time point.
The calculation of
R-
square is shown as Equation (25) [
32].
where
is the average value of the detected temperature.
The rational function fitting method is used to fit the detected data of
Tair and
Toil. Both the Numerator degree and Denominator degree are set to 1. The fitted mathematical models are shown in Equation (26) and Equation (27), respectively.
where
and
are the fitted mathematical models of ambient temperature and cooling oil temperature. The
RMSE and
R-
square of
and
are shown in
Table 2, which reflects a high degree of fit in
and
.
According to the numerical table of thermophysical properties of air in Appendix 2 of Reference [
33], the values of each parameter at different temperatures can be obtained by linear interpolation. According to Equation (23), the sensitivity solution results of air property parameters are shown in
Table 3. The small variation in air property parameters, especially Pr
air, due to the small range of ambient temperature rise detected in this study can be seen from the table.
The SST k-Omega turbulence model was used for solving in this study. The solution process adopts the SIMPLE strategy. In transient calculations, the values of time-varying factors and their parameters at each time step will be solved and updated by the solver in two ways: the time-varying factors are directly solved by substituting time steps into their time functions or the time function of property parameters; the factor values for the next time step are solved and updated based on the temperature values obtained from the current time step. It is worth noting that the different materials of different components and the different thermal characteristics of the same component at different positions can affect the waste heat dissipation after the last operation of the CNC machine tool. Therefore, there is a certain gradient in the initial temperature field of the spindle system, for which the initial detected temperature varies at different detection positions. Due to the focus of this study on the process of multiple factors affecting thermal characteristics, before solving, the spindle system is initialized with 7.4 °C, which is the lowest initial temperature value of each sensor detection position.
7. Comparison of Thermal Characteristics Solution with Non Time-Varying Factors
The traditional methods often consider changes in the heat source conditions of the spindle system but rarely comprehensively consider the external environment, internal cooling conditions, and the heat transfer ability caused by their temperature during the actual operation of the CNC machine tool. Therefore, the ambient temperature and cooling oil temperature of the spindle system are set as non time-varying values in this study. In order to compare the thermal characteristics value at the last moment, separately, the ambient temperature and cooling oil temperature are set to the detected 12.3 °C and 16.4 °C, respectively. The thermal characteristics of each point in the spindle system are solved. The temperature comparison of the detection values and solution values affected by time-varying factors and by non time-varying factors at the final moment, as well as their relative errors, are shown in
Figure 9.
It can be seen from
Figure 9 that except for T5, the relative errors of the solution values affected by time-varying factors are closer to 0 than the solution values affected by non time-varying factors. This indicates that the proposed solution method considering the comprehensive effect of time-varying factors has more advantages in solving the accuracy of temperature in the spindle system.
The axial displacement curves of the detection values and solution values affected by time-varying factors and by non time-varying factors, as well as their relative error curves are shown in
Figure 10. It can be seen that due to the constant ambient and cooling conditions, the axial displacement curve affected by non time-varying factors takes a relatively short time to approach a steady state, about 2500 s. However, in reality, the steady state is not achieved even at 7000 s. The change trend of axial displacement is not as close to the actual axial displacement rise, as solved by considering the effect of time-varying factors. From the perspective of calculation accuracy, at 7000 s, the relative error of the solution value calculated by non time-varying factors is −5.13% The value obtained by time-varying factors is −1.24%. The above analysis indicates that the nonlinear time-varying thermal characteristics model has advantages in reflecting the trend of numerical changes and the accuracy of result solving.