JP2006055918A - Method of correcting thermal deformation error of machine tool - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To highly precisely correct a machining error caused by the thermal deformation of a machine tool by using a neural network and using input data different from those used in a conventional method. <P>SOLUTION: In the method of correcting the machining error caused by the thermal deformation of the machine tool by using the neural network 1 estimating correction values δ<SB>x</SB>, δ<SB>y</SB>and δ<SB>z</SB>with temperature data t<SB>0</SB>, t<SB>1</SB>, ... of machine parts at the time of machining as input values, the time of delay for correcting the calculation error of the quantity of correction based on the room temperature and the hour-room temperature data, which are obtained by measuring the room temperature at each hour on a plurality of recent days are registered, and the room temperature T<SB>0</SB>at the time of machining T<SB>d</SB>is found from the hour-room temperature data to make input data on an input layer. Further, several patterns a, b, c, ... of variation in time from the actuation of the machine are registered, and the variation in time δ<SB>f</SB>computed from the selected pattern and the elapsed time T<SB>s</SB>at the time of machining is made input data on the input layer. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a method for correcting a machining error of a workpiece caused by thermal deformation of a machine tool over time, and relates to the above method for inferring a thermal deformation amount of a machine tool at a machining time using a neural network.

  When machining a workpiece with a machine tool, the temperature of each part of the machine rises due to the rotation of the spindle, the temperature rise of the hydraulic oil, and the machining heat. The temperature rise of each part accompanying the operation of the machine tool and the change of the room temperature with the passage of time change the machine temperature, and the thermal expansion accompanying this causes a machining error in the workpiece machined by the machine. Therefore, in order to make this machining error as small as possible, the room temperature and the temperature of the main part of the machine are measured, the amount of thermal deformation of the machine tool is estimated based on the measured value, and the NC device is based on the estimated value. Correction of the given tool post and table coordinates is performed.

  As a method of estimating the amount of thermal deformation of the machine tool, work the workpiece under various conditions in advance, create a correction formula from the room temperature at that time and the temperature of each part of the machine, and the error that occurred in the workpiece at that time, A correction method based on the correction formula is generally performed by an NC device or a computer exchanging data with the NC device, and is corrected by a neural network using room temperature and the temperature of each part of the machine as input data. A method for inferring the amount of deformation of a machine by using a computer has also been proposed.

For example, Patent Document 1 discloses the amount of thermal deformation of a machine tool based on a neural network theory from the outputs of a plurality of temperature measuring means arranged in each part of the machine tool and the outputs of position measuring means installed on a table. A data table to be predicted is created, the amount of thermal deviation of the machine tool is predicted based on the data table from the output of each temperature measuring means at the time of machining the workpiece, and the movement of the machine tool is corrected based on the prediction result A correction method has been proposed.
Japanese Patent Laid-Open No. 6-8107 Toshimichi Moriwaki, 1 other, “Prediction of Thermal Deformation of Machining Center by Neural Network”, Transactions of the Japan Society of Mechanical Engineers (C) June 1992, Vol. 58, No. 550, p.246-251

  The neural network infers the output value from a plurality of changing input data according to a predetermined inference process, and repeats the operation of comparing the inferred output value with the actual value. It is to raise.

  By using this neural network theory for inferring the amount of thermal deformation of machine tools, compared to the general method of estimating the amount of thermal deformation using a formula that was created based on a limited amount of test data. Thus, it is considered possible to correct the thermal deformation error with higher accuracy.

  However, if the type of input data given to the neural network is not appropriate, the inference results cannot be expected to exceed a certain limit, and in order to make the inference value converge more quickly with higher accuracy. It is essential to provide the neural network with the appropriate type of input data.

  It is an object of the present invention to make it possible to correct a workpiece machining error with higher accuracy by using input data different from the conventional one in correction of a machining error due to thermal deformation of a machine tool using a neural network. It is said.

In various lathes, the inventors of the present invention have arranged the relationship between the correction value calculated by the correction formula obtained by the test machining and the machining error of the workpiece actually generated based on the machining time of the day, It was recognized that there was a time delay between the calculated correction value and the machining error to be corrected. That is, taking the time of one day on the horizontal axis and taking the correction value on the vertical axis, in the machining at various times, the value to be corrected δ obtained from the calculated correction value δ and the error remaining in the machined workpiece _When the value of _t was plotted, as shown in FIG. 3, it was recognized that there was a time delay d between the two. The time delay d varies depending on the structure and size of the machine, but is almost constant at any time of the day.

  Furthermore, when time-dependent changes in machining errors when machining workpieces of the same shape are organized, there are several types of change patterns as shown in Fig. 4 depending on the machine structure, workpiece material, and type of cutting fluid used. It was recognized that there was. For example, in the pattern a of FIG. 4, the machining error rapidly increases to the minus side immediately after the start of the operation of the machine, and as the time elapses, the increase amount becomes gradual and eventually converges to a constant value. The change pattern is shown. The machining error occurs on the minus side when the amount of deformation of the workpiece due to machining heat is larger than the amount of deformation of the machine tool.

  Further, in the pattern b in the figure, the machining error slightly shifts to the minus side immediately after the start of machining, but then changes to the plus side, and the change pattern in which the machining error increases to the plus side as time elapses. Show. Such a change pattern is considered to appear when the thermal deformation amount of the workpiece is relatively small and the machine temperature rises in proportion to the operation time. When the machine temperature becomes high, the heat radiation to the surroundings increases. Therefore, it is considered that the error amount converges to a constant value on the plus side.

  Further, in the pattern c in the figure, after the error amount is once shifted to the plus side at the beginning of the operation, it tends to move to the minus side and converge to a constant value. Such a change pattern is considered to appear in a machining condition in which a local temperature rise of the machine in the headstock, the tool post or the like greatly affects the machining error of the workpiece.

  In an actual machining site, workpiece machining errors are measured by sampling inspection, etc., so that when a worker processes a workpiece with a certain machine, the workpiece machining error shows how it changes over time. I know empirically.

The present invention has been made on the basis of the above knowledge, and the method for correcting a thermal deformation error of a machine tool according to the invention of claim 1 of the present invention uses temperature data t _{1 of} each part of the machine at the time of machining as an input value, In a method for correcting a machining error due to thermal deformation of a machine tool using a neural network 1 that uses t ₂ , t ₃ ... to infer the machining error correction values δ _x , δ _y , δ _z of the workpiece at the time. The time-room temperature data obtained by measuring the room temperature at each time on the most recent days are registered, and the value t ₀ used as the room temperature at the time of processing is obtained from the time-room temperature data and used as input data for the input layer. It is characterized by this.

The invention of claim 2 of the present application is the method for correcting a thermal deformation error comprising the above means, and a delay time d for correcting an error in a correction amount based on a room temperature obtained in advance, and each time in a plurality of nearest days. The time-room temperature data obtained by measuring the room temperature is registered, and the input data of the input layer is obtained from the time-room temperature data obtained by shifting the value t ₀ used as the room temperature at the time of processing according to the delay time. It is characterized by using as.

Furthermore, the invention of claim 3 of the present application is a method for correcting a thermal deformation error comprising the means of claim 1 or 2, in which several patterns a of the change over time a that change depending on the elapsed time from the start of the machine a, .., b, c,... are registered, and the selection means 6 is provided, and a time-dependent change δ _f calculated from the selected pattern and the elapsed time at the time of processing is used as input data of the input layer. It is.

  According to the present invention, the correction value of the workpiece machining error caused by the thermal deformation of the machine tool can be inferred more accurately, and the workpiece machining accuracy can be increased. In addition, by specifying a thermal deformation pattern based on experience, it is possible to infer a highly accurate correction value with a small learning time.

1 and 2 show an embodiment of the present invention, where 1 is a three-layer neural network, 2 is first computing means for predicting a virtual room temperature T ₀ from a time T _d in one day, and 3 is a machine This is a second calculation means for inferring a virtual error δ _f from the elapsed time T _s after activation. Machine temperature t ₁ , t ₂ ... Detected by thermocouples s ₁ , s ₂ ... Arranged in each part of the machine is input to the neural network 1 via the communication unit 4 of the temperature measuring device. The auxiliary storage device of the CNC device 5 stores a data set of room temperature data t _{0 to be} described later, a delay time d, and a time-varying pattern, and is provided with means for selecting this time-varying pattern. The correction value output from the neural network 1 is sent to the NC device 7 of the CNC device, and the corrected position command is given to the machine tool 8 from the NC device 7.

The input layer of the neural network 1 has an estimated value t ₀ of the virtual room temperature, detected temperatures t ₁ , t ₂ ... From the temperature sensors provided in each part of the machine, and an expected correction value δ estimated from the selected thermal deviation pattern. _f is given as input data, and correction values δ _x , δ _y , δ _z in the X, Y, and Z directions inferred from the output layer are output.

For example, the first computing means 2 may measure the room temperature data measured at the time of 0:00, 0:05, 0:10,. Accumulated for 7 days, for example, average room temperature at each measurement time {t _0:00 , t _0:05 , t _0:10 ... T ₂₃ ; ₅₅ }
Is calculated and stored. On the other hand, a delay time d (see FIG. 3) from room temperature, which is obtained in advance for each machine, is set in the first calculation means 2.

When the time T _d at the processing time is given, the first calculation means 2 performs a complementary calculation of the room temperature t ₀ at the time before the delay time d from the calculated data indicating the time-dependent change in the average room temperature. As an input value.

  In the second calculation means 3, a time-dependent change pattern of the workpiece dimensional error from the time of starting the machine as shown in FIG. 4 is registered as an equation using an exponential function, for example.

Prior to machining the workpiece, the operator considers the workpiece shape and material, the type of machine, the room temperature, the cutting fluid used, etc., and the change pattern used from the registered patterns a, b, c. And the maximum amplitude δ _max is also input as an input term. The second calculation means 3 determines a coefficient to be multiplied by the change pattern calculation formula based on the input maximum amplitude δ _max , and calculates the predicted correction amount δ _f by substituting the elapsed time T _s from the start of the machine. The input data of the neural network 1 is used.

The neural network 1 infers correction values δ _x , δ _y , δ _z with input values given from the first calculation means and the second calculation means and temperatures t ₁ , t _2. The correction values δ _x , δ _y are adjusted so that the machining accuracy of the workpiece is less than or equal to a desired value by correcting the inference value and the error remaining in the machined workpiece and correcting the weighting coefficient multiplied by each input value. , Δ _z is determined. Even after entering the actual workpiece machining process, the operator appropriately measures the workpiece and inputs the measured value to the computer. The computer improves the accuracy of the correction value by correcting the inference process of the neural network 1 using the input data.

Block diagram showing an embodiment of the present invention Explanatory diagram showing neural network and its input / output terms Diagram showing the time delay given to the correction value for room temperature change Diagram showing some of the time-dependent change patterns of machining errors due to the elapsed time after machine startup

Explanation of symbols

The value t ₁ of inputting a first neural network 6 selection means t ₀ rt, t _2, t ₃ ··· machine each part of the temperature d delay time δ _x, δ _y, δ _z machining error correction value a, b, c ·・ ・ Time-dependent change pattern δ _f Time-dependent change calculated from time-dependent change pattern

Claims (3)

In the correction method of the machining error due to the thermal deformation of the machine tool using the neural network that infers the correction value of the machining error of the workpiece at the time using the temperature data of each part of the machine at the machining time as an input value,
Registering the time-room temperature data obtained by measuring the room temperature at each time on the most recent days, and obtaining the room temperature at the time of processing from this time-room temperature data and using it as input data of the input layer, A method for correcting thermal deformation errors of machine tools.   Register the delay time for correcting the error of the correction amount based on the room temperature obtained in advance and the time-room temperature data obtained by measuring the room temperature at each time on the most recent days, and the room temperature at the time of processing is 2. The thermal deformation error correction method according to claim 1, wherein the method is used as input data of an input layer obtained from the time-room temperature data shifted according to a delay time.   Registers several patterns of change over time that change with the elapsed time from the start of the machine, and provides a selection means for the input pattern to input the change over time calculated from the selected pattern and the elapsed time during processing. The method of correcting a thermal deformation error according to claim 1 or 2, wherein

Images