TWI448353B - Method and apparatus of machine tools for intelligently compensating thermal error - Google Patents

Description

Tool machine thermal error wisdom adapting device and method thereof

The invention relates to the thermal error adjustment of the machine tool, in particular to a tool machine thermal error intelligent adapting device and method thereof for warming machine judgment and adaptive learning.

The machine tool 10 is in the process of processing, no matter the motor, hydraulic system and mechanical friction are energy conversion, no matter the conversion route, most of them become heat, which causes temperature changes inside and around the body, and finally leads to the workpiece size during processing. Or the shape of the displacement error, referred to as thermal error (shown in Figure 1 to Figure 2).

The problem of thermal error is a problem that must always be faced in the development of precision machinery. According to the research literature [1], about 40-70% of the total error of tool machining is contributed by thermal error, which shows that it plays an absolutely crucial role in the impact of tool machining accuracy.

Traditionally, in order to reduce thermal error interference, it is necessary to start the machining operation after the temperature rise of the machine reaches a steady state (or warm state). This process relies on the judge to judge whether the warm-up is based on experience, as shown in Figure 3 and Figure 4.

Recently, the response strategy for tool machine thermal error can be divided into two ways. One is to use passive compensation method. By constructing the tool machine thermal error prediction model, the error amount is compensated in software mode; the other strategy is The active suppression method is used to reduce the amount of error generated during the design phase, and the purpose is to control or avoid the generation of thermal errors. The research and technical overview of active suppression and passive compensation are summarized in Tables 1 and 2 respectively.

Compared with the design method of active thermal suppression, it is more convenient and economical to adopt the method of thermal error software compensation. It does not directly remove or reduce the thermal error generated by the machine tool, but uses the experimental measurement results for operation analysis. This method is also widely used by foreign tool machine manufacturers, such as Mazak and Okuma in Japan and Mikron in Switzerland. Therefore, how to improve the existing compensation technology and develop a more accurate and reliable thermal error compensation method is a long-term goal of the tool machine industry.

However, from the past research results, it is found that the static compensation model using mathematical statistics can obtain good results for the thermal error of the steady state of the machine tool, but it is quite tricky for the thermal error of the transient state. It is not easy to handle, and domestic companies still cannot solve such problems.

Based on the above problems, the inventors have proposed a tool machine thermal error smart adaptation device and method thereof to overcome the deficiencies of the prior art.

Literature reference:

[1] Bryan, J. B., 1990, "International status of thermal error research," Annals of the CIRP 39/2, pp. 645-656.

[2]MAKINO website www . makino . co . jp

[3] Muto, A., 2005, "Machine tool with a feature for preventing a thermal deformation," U.S. Patent, No. 6,923,603.

[4]YASDA website www . yasda . co . jp

[5]OKUMA website www . okuma . co . jp

[6] Yintai Technology Co., Ltd. website www pmi -. Amt com.

[7]MORISEIKI website www . moriseiki . com/dixi/english/products/control . html

[8] Rahman, M., Mansur, MA, and Karim, MB, 2001, "Non-conventional materials for machine tool structures," JSME Int. J., Series C, Vol. 44, No. 1, pp.1 -11.

[9] Slocum, A.H., 1992, Precision Machine Design, Society of Manufacturing Engineers, 1st Edition.

[10] MAZAK website www.mazak.com

[11] Kobari, T. and Takada, R., 1999, "Shuttle table device," Japan patent, No. 11-267938.

[12] Kato, K. and ITO, T., 2006, "Machine tool and posture maintenance device," Japan patent, No. 2006-341328.

[13] Ramesh, R., Mannan, MA and Poo AN, 2000, “Error Compensation in Machine Tools-A Review Part II: Thermal Errors,” Int. J. Mach. Tools Manufact., Vol. 40, pp. 1257 -1284.

[14] Yang, JG, Ren, YQ, Liu, GL, Zhao, HT, Dou, XL, Chen, WZ, and He, SW, 2005, "Testing, variable selection and modeling of thermal errors on an INDEX-G200 turning Center," Int. J. Adv. Manuf. Technol., Vol. 26, pp. 814-818.

[15] Kang, Y., Chang, CW, Huang, Y., Hsu, CL, Nieh, IF, 2007, "Modification of a neural network utilizing hybrid filters for the compensation of thermal deformation in machine tools," Int. J Mach. Tools Manufact., Vol. 47, pp. 376-387.

[16] Ramesh, R., Mannan, MA Poo, AN, and Keerthi, SS, 2003, "Thermal error measurement and modelling in machine tools. Part I. Influence of varying operating conditions," Int. J. Mach. Tools Manufact ., Vol. 43, pp.391-404.

[17] Ramesh, R., Mannan, MA Poo, AN, and Keerthi, SS, 2003, "Thermal error measurement and modelling in machine tools. Part II. Hybrid Bayesian Network-support vector machine model," Int. J. Mach . Tools Manufact., Vol. 43, pp. 405-419.

[18] Yang, H. and Ni, J., 2005, "Adaptive model estimation of machine-tool thermal errors based on recursive dynamic modeling strategy," Int. J. Mach. Tools Manufact., Vol. 45, pp.1 -11.

[19] Yang, H. and Ni, J., 2005, "Dynamic neural network modeling for nonlinear, nonstationary machine tool thermally induced error," Int. J. Mach. Tools Manufact., Vol. 45, pp. 455-465 .

[20] MIKRON website www.mikron.com

[21] FANUC website www.fanuc.co.jp

[22] Vapnik, V., Golowich, S. and Smola, AJ, 1997, "Support vector method for function approximation, regression estimation, and signal processing," in Advances in Neural Information Processing Systems, Vol. 9, pp. 281 -287

[23] D. Tax and R. Duin, “Support vector data description,” Machine Learning, vol. 54, pp. 45-66, 2004

[24] Yi-Hung Liu, Yu-Kai Huang, and Ming-Jui Lee, "Automatic inline-defect detection for TFT-LCD array process using locally linear embedding and support vector data description," Measurement Science and Technology, vol. 19 , August 2008

[25] Yi-Hung Liu, Szu-Hsein Lin, Yi-Ling Hsueh, and Ming-Jiu Lee, "Automatic target defect identification for TFT-LCD array process inspection using kernel fuzzy c-means based fuzzy SVDD ensemble", Expert Systems With Applications, vol. 40, 2008

The present invention aims to provide a thermal error model using Support Vector Regression (SVR) [22], and construct a steady state mode range using Support Vector Data Description (SVDD) [23] for immediate execution. Tool machine thermal error intelligent adjustment device and method for adjusting thermal error on line.

Another object of the present invention is to provide a machine tool thermal error smart adaptation device and method thereof that can perform incremental learning for temperature changes of different external environments.

In order to achieve the above object, the present invention provides a tooling machine thermal error intelligent adapting device, comprising: a warming machine feature description building unit, establishing a steady state mode range of a body temperature distribution vector of the machine tool; and constructing a thermal error model Unit, training a nonlinear thermal error model; and a warm-up judgment and thermal error calculation unit, by reading a plurality of temperature sensing signals on the machine tool and operating condition information of the machine tool, determining the machine tool In the warm state, the thermal error compensation amount of at least one node is calculated.

The tool error thermal intelligence adapting device further comprises an incremental learning unit, which can be adaptively adjusted to ensure the accuracy of the thermal error model.

In order to achieve the above object, the present invention further provides a method for intelligently adjusting the thermal error of a machine tool, the steps comprising: extracting thermal behavior data; selecting a warm state of the warming machine by an expert; constructing a warming machine characteristic description by using steady state temperature information; Describe the thermal behavior information that matches the steady state; construct the thermal error model with the thermal behavior information of the steady state; judge the line and perform the thermal error calculation.

Among them, the tool error thermal error adjustment method includes incremental learning to adapt to environmental changes and machine parameter drift. For similar concepts, please refer to the cited literature [24] [25].

While the invention has been described in terms of several preferred embodiments, the preferred embodiments of the present invention It is not intended that the invention be limited to the following drawings and embodiments.

Referring to FIG. 5, a block diagram of the thermal error intelligent adjustment device of the machine tool of the present invention is shown.

The machine tool thermal error intelligent adapting device 1 of the present embodiment functions on a power tool 10 mainly comprising a warm-up feature description construction unit 2, a thermal error model construction unit 3, a warm-up judgment and thermal error operation unit 4 and a Incremental learning unit 5.

Referring to FIG. 6, the warm-up feature description construction unit 2 establishes a steady-state mode range of the body temperature distribution vector of the machine tool with the support vector data description (SVDD), that is, the support vector data description (SVDD) is confirmed in the machine tool. The temperature state of each node a~d reaches a steady state until the high dimensional spatial feature vector distribution of the support vector data description (SVDD) is the same as the preset steady state mode range.

The support vector data description (SVDD) is described in detail below.

■ Purpose of SVDD: Estimate the judgment function D ( x ) from the training data, where

■ SVDD model use: Give the input variable x to determine whether x falls within the range that can be explained by the previous data; D ( x ) 0 means applicable, D ( x )>0 then no.

■ Modeling process:

1). Set all the combinations in the parameters σ, C to verify the range, select one of the groups (σ, C );

2). Perform k-fold and randomly divide the steady-state experimental data into k groups, and take k -1 group as model training data (denoted as Another group plus transient data for testing (denoted as

3). For the training and verification combination of the h group, solve the quadratic programming problem

Equivalent to solving the dual problem,

Decide out { } (support vector, SVs, defined as "the corresponding α _i ≠ 0"), N _SV (the number of support vectors) and α _i

4). Select any support vector with ξ _i =0 , calculate R

5). Bring the test data into the model test and get

Calculate the F measure

6). Repeat steps 3~5 for the different training and verification combinations of k groups, average all errors, define The classification performance of the combination of parameters;

7) Repeat steps 1~6 for all parameter combinations to select the combination of parameters that best perform the classification ( );

8). Complete SVDD modeling.

Referring to FIG. 7, after the machine reaches a steady state, the prediction is performed, and the thermal error model construction unit 3 trains a nonlinear thermal error model by using support vector regression (SVR). The support vector regression (SVR) has fewer operational parameters and is predicted. In terms of accuracy, a lower MAPE (mean absolute percentage error) can be obtained, and its core spirit is the kernel method.

The following is a detailed description of Support Vector Regression (SVR). .

■ SVR purpose: Estimate the function y = f ( x ) from the training data, where:

In the above formula, each training instance has its corresponding ξ _i and Used to determine whether the training instance can fall outside the scope of ε. The role of C is to adjust whether the training model is over or under adapted (overfitting or underfitting). When the core definition is clear, the following three parameters can be adjusted:

Gamma: Adjust the std of the Gaussian kernel function, which is σ in the above equation.

C: The weighting measure used to adjust the error term during training can determine overfitting or underfitting.

Epsilon: The size of the error tolerance band, which is ε in the above equation.

■ SVR model use: For the input variable x , the value f ( x ) of the corresponding output variable is predicted.

■ Modeling process:

1). Set all the combinations of parameters σ, C , ε in the range to be verified, and select one of them (σ, C , ε);

2). Perform k-fold, randomly divide the experimental data into k groups, and take k -1 group as model training data (denoted as Another group for testing purposes (denoted as

3). For the training and verification combination of the h group, solve the quadratic programming problem

Equivalent to solving the dual problem,

Decide out { } (support vector, defined as "the corresponding -α _i ≠0』), N _SV (number of support vectors), With α _i ;

4). Choose any ξ _i =0 or =0 support vector , calculate b ,

5). Calculation error

6). Repeat steps 3~5 for the different training and verification combinations of k groups, average all errors, define The prediction error of the combination of the parameters of this group;

7). Repeat steps 1~6 for all parameter combinations to select the parameter combination when the prediction error is the smallest ( );

8). Complete SVR modeling.

Referring to FIG. 8 , the warm-up determination and thermal error calculation unit 4 calculates at least one of the temperature sensing signals of each node and the operating condition information of the machine tool on the tool machine to determine the warm-up state of the machine tool. The thermal error compensation amount of one node is processed.

Referring to FIG. 9, when the external environment temperature changes, such as the machine tool moves to other work site operations, the incremental learning unit 5 can describe the original data of the support vector data description (SVDD) of the construction unit 2 for the warm-up feature, and the heat. The support vector regression (SVR) of the error model construction unit 3 receives the new data of the environmental temperature change, and confirms whether the incremental learning is performed by comparing the warm error judgment and the thermal error calculation unit 4 with the known predetermined thermal error data. Parameter adjustments ensure the accuracy of the thermal error model.

The detailed operation flow will be detailed later.

Referring to FIG. 10, a method for intelligently adjusting the thermal error of the machine tool according to the present invention includes the steps of: step S1: extracting thermal behavior data; step S2: selecting a warm state by an expert; and step S3: constructing a warming machine with steady state temperature information. Feature description; Step S4: Filtering the thermal behavior information according to the steady state with the warm-up feature description; Step S5: constructing the thermal error model with the thermal behavior information of the steady state; Step S6: determining the online warm-up and performing the thermal error calculation; Step S7 : Incremental learning, to adapt to environmental changes and parameter drift.

The thermal behavior in step S1 includes temperature and positioning error, etc., the expert selection system in step S2 can be used for classification in the historical data, and the warm-up feature description in step S3 is described by support vector data (SVDD). In the manner of construction, the thermal error model of step S5 is constructed by support vector regression (SVR), and the online warm-up judgment of step S6 is performed by reading information such as temperature sensing signals on the machine tool and operating conditions of the machine tool.

Please refer to FIG. 11 for description of the single-axis feeding machine 11; temperature sensors T1~T24 are arranged in each part of the machine tool, and the position of the parts is shown in FIG. 12 to FIG. 18; the thermal behavior information extraction plan is as follows:

Itinerary planning: 1100mm full stroke.

Feed speed planning: There are four feed speeds of 9m/min, 18m/min, 27m/min and 36m/min.

Measurement time planning: In order to analyze the thermal error behavior mode of the transient and steady state of the system, it is measured every 15 minutes after starting up, and the transient information is taken. After a long time (taking 150 minutes as an example), Change it to measure every 30 minutes and retrieve the steady state information of the system.

Please refer to FIG. 19, which is a flow chart showing a method for extracting thermal behavior data. The thermal behavior data extraction method step of the present invention comprises: step SA1: selecting a feed speed; step SA2: setting a machine tool parameter and setting a temperature sensor at a plurality of nodes of the machine tool; step SA3: the machine tool is turned on and enters an operation mode; Step SA4: Whether the operation mode reaches 150 minutes (predetermined operation time), if not, enters the measurement mode every 15 minutes (first time interval) (step SA41); if yes, enters every 30 minutes (second time interval) Measuring mode (step SA42); step SA5: performing temperature measurement on each node of the machine tool; and step SA6: storing data.

Wherein, step SA6 can manually fill in the form or automatically record by the temperature capture program.

Referring to FIGS. 12-18, the machine tool 10 includes a saddle 11, a screw 12, a slide rail 13, a servo motor 14, a base 15, and a ground screw 16. The information of the temperature variable includes an external heat source (ambient temperature) and an internal heat source. Therefore, for the recording of temperature variables, there are twenty-five items, one is the external ambient temperature, and twenty-four structural body temperatures, wherein twenty-four structural body temperatures include: the slider end temperature of five saddles 11 , a nut seat (at the screw 12) temperature, eleven base 15 rails (slide 13) temperature, four base 15 foot end structure (foot screw 16) temperature, a motor seat bearing end temperature, The temperature of a tailstock bearing end, and the temperature of a motor interface seat, that is, the nodes of the twenty-four temperature sensors T1~T24; in addition, according to the thermodynamic dynamics, the temperature change rate (derivative) should be included in the reference. .

The thermal displacement related variables are planned according to the travel of the vehicle. In order to match the position of the nodes in the segmentation model constructed by the virtual sensing, the experiment sets the measurement of the positioning accuracy at intervals of 125 mm, and the measurement position (mm) They are: 131, 256, 381, 506, 631, 756, 881, and 1006, as shown in FIG.

The invention adopts support vector regression (SVR) as a thermal error prediction model, and uses support vector data description (SVDD) to construct a steady state mode range of the body temperature distribution vector of the machine tool to define the thermal error prediction model. Applicability. SVDD is a novel machine learning algorithm whose purpose is to construct a minimum-volume hypersphere to surround the training set. Since the hypersphere is constructed in the feature space, the surface of the sphere changes in the input space. Very flexible. For a new machine tool body temperature distribution vector variable, we only need to calculate its distance to the center of the ball, we can calculate the confidence index of the warm-up state corresponding to this input: the greater the distance, the input and training The greater the difference in the set, the smaller the confidence index of the output. There are two advantages to using SVDD: 1) its solution does not have a local minimum problem, because its dual problem is also a QP problem, so there is no need to manually set the threshold, and 2) regardless of the distribution of the input data training set, With SVDD you can find a boundary that can tightly surround it. The detailed technical details have been revealed as before.

According to the above structure and method, the thermal displacement model and the support vector data description (SVDD) hypersphere are established by using the temperature sensor set in the various components of the machine tool to obtain the thermal displacement behavior data. It can judge the warm-up state under the variation of the ambient temperature, and predict the compensation information directly on the line to compensate the thermal error.

Although the present invention has been explained in connection with the preferred embodiments, it is not intended to limit the invention. It should be noted that many other similar embodiments can be constructed in accordance with the teachings of the present invention, and are within the scope of the present invention.

1. . . Tool machine thermal error wisdom adapting device

10. . . Machine tool

11. . . Saddle

12. . . Screw

13. . . Slide rail

14. . . Servo motor

15. . . Base

16. . . Foot screw

2. . . Warm-up feature description building unit

3. . . Thermal error model construction unit

4. . . Warm-up judgment and thermal error calculation unit

5. . . Incremental learning unit

a~d. . . node

T1~T24. . . Temperature sensor

Steps S1 to S7 According to the present invention, the thermal error intelligent adjustment method of the machine tool

Steps SA1 to SA6 According to the extraction heat behavior step of the present invention

Figure 1 shows a schematic diagram of thermal error of a conventional machine tool.

Figure 2 shows another thermal error diagram of a conventional machine tool.

Figure 3 shows a graph of the thermal error of a conventional machine tool.

Figure 4 shows a graph of a conventional machine tool that determines the thermal error by the operator.

Figure 5 is a block diagram showing the thermal error intelligent adjustment device of the machine tool of the present invention.

Figure 6 is a schematic diagram showing the steady state description of the thermal error intelligent adaptation device of the machine tool of the present invention in support of vector data description.

Fig. 7 is a graph showing the thermal error intelligent adjustment device of the machine tool of the present invention for supporting the vector regression to describe the steady state prediction error amount.

FIG. 8 is a schematic diagram showing the judgment operation of the warm-up judgment and thermal error calculation unit in the thermal error intelligent adjustment device of the machine tool of the present invention.

FIG. 9 is a schematic diagram showing the judgment of the incremental learning unit in the thermal error intelligent adjustment device of the machine tool of the present invention.

Figure 10 is a flow chart showing the method of intelligently adjusting the thermal error of the machine tool of the present invention.

Fig. 11 is a schematic view showing the stroke planning of the uniaxial feed tooling machine of the present invention.

Fig. 12 is a structural view showing the uniaxial feeding machine of the present invention as an example.

Figure 13 shows one of the distribution diagrams of the present invention for deploying a temperature sensor in a single axis feed tool.

Figure 14 shows the second distribution of the temperature sensor of the present invention in a single-axis feed tool machine.

Figure 15 shows the third distribution of the temperature sensor of the present invention in a single-axis feed tool machine.

Figure 16 shows the fourth distribution of the temperature sensor of the present invention in a single-axis feed tool machine.

Figure 17 is a view showing the distribution of the temperature sensor of the present invention in a single-axis feed tool machine.

Figure 18 is a diagram showing the distribution of the temperature sensor of the present invention in a single-axis feed tool machine.

Figure 19 is a flow chart showing the steps of extracting thermal behavior data in the thermal error intelligent adaptation method of the machine tool of the present invention.

1. . . Tool machine thermal error wisdom adapting device

2. . . Warm-up feature description building unit

3. . . Thermal error model construction unit

4. . . Warm-up judgment and thermal error calculation unit

5. . . Incremental learning unit

Claims (7)

A tool machine thermal error intelligent adjustment device is provided in a machine tool comprising: a warm-up feature description construction unit, which uses a support vector data description (SVDD) to confirm that the complex node temperature state of the machine tool reaches a steady state, Until the support vector data description (SVDD) has the same high-dimensional spatial feature vector distribution and the preset steady-state mode range, the support vector data description (SVDD) estimates the judgment function D ( x ) from a training data, wherein , x is the input variable, N _SV is the number of support vectors in SVDD, , The support vector in SVDD, α _i , α _j is the weight value of the corresponding support vector, R , The best SVDD model parameter after training; a thermal error model construction unit, which uses a support vector regression (SVR) to estimate the function y = f ( x ) from the training data, wherein , training a nonlinear thermal error model, x- series input variables, number of support vectors in the N _SV- based SVR, Is the support vector in the SVR, , α _i is the weight value corresponding to the support vector, b , The best SVR model parameters after training; and a warm-up judgment and thermal error calculation unit, by reading a plurality of temperature sensing signals on the machine tool and operating condition information of the machine tool, determining the warm-up of the machine tool State, calculate the thermal error compensation amount of at least one node of the machine tool. For example, the tool machine thermal error intelligent adjustment device described in claim 1 further includes an incremental learning unit, which can be adaptively adjusted to ensure the accuracy of the thermal error model. The machine tool thermal error intelligent adjustment device according to claim 1, wherein each component on the machine tool is provided with a plurality of temperature sensors. The utility model relates to a tool machine thermal error intelligent adapting method thereof, which is applied to a tool machine thermal error intelligent adapting device according to the first application of the patent application scope, the steps comprising: extracting thermal behavior data, wherein the thermal behavior comprises a machine tool temperature and positioning Error; the warm-up state is selected by the expert, wherein the expert selection system can select the classification in a historical data; the warm-state feature description is constructed with the steady-state temperature information, wherein the warm-up feature description is described by a support vector data ( The SVDD) method is constructed; the thermal behavior information corresponding to the steady state is filtered by the warm-up feature description, and the support vector data description (SVDD) is obtained by estimating the judgment function D ( x ) from a training data, wherein Enter the variable x to determine if x falls within the range that can be explained by the previous data; D ( x ) 0 means applicable, D ( x )>0, no, x is an input variable, and N _SV is the number of support vectors in SVDD. , The support vector in SVDD, α _i , α _j is the weight value of the corresponding support vector, R , The optimal SVDD model parameters after training; the thermal error model is constructed with the thermal behavior information of the steady state, and the thermal error model is constructed via a support vector regression (SVR). The support vector regression (SVR) is estimated from the training data. The function y = f ( x ), where x is the input variable, the number of support vectors in the N _SV system SVR, Is the support vector in the SVR, , α _i is the weight value corresponding to the support vector, b , The best SVR model parameters after training; online warm-up judgment and thermal error calculation, online warm-up judgment is performed by reading the temperature sensing signal on the machine tool and the operating condition information of the machine tool. For example, the method for intelligently adjusting the thermal error of the machine tool described in the fourth application of the patent scope includes incremental learning, which is adapted to environmental changes by parameter drift. The method for intelligently adjusting the thermal error of the machine tool according to claim 5, wherein the step of extracting the thermal behavior data further comprises: selecting a feed speed; setting a machine tool parameter and setting temperature sensing at a plurality of nodes of the machine tool; The machine tool is turned on and enters the operation mode; whether the operation mode reaches a predetermined operation time, and if not, enters a measurement mode of the first time interval; if yes, enters a measurement mode of the second time interval; wherein The first time interval is less than the second time interval; temperature measurement is performed on each node of the machine tool; and the data is stored. The method for intelligently adjusting the thermal error of the machine tool according to claim 6 of the patent application, wherein the step of storing the data comprises filling in a manual record or Automatically recorded by the temperature capture program.