FLUCTUATION AMOUNT ESTIMATION DEVICE IN MACHINE TOOL AND CORRECTION AMOUNT CALCULATION DEVICE

Abstract

Provided are a fluctuation amount estimation device (9) capable of evaluating reliability of an estimated value and a correction amount calculation device (1) including the fluctuation amount estimation device (9). The correction amount calculation device (1) includes the fluctuation amount calculation device (9), a correction amount calculation unit (5), and a correction amount output unit (7). The fluctuation amount estimation device (9) includes a parameter storage (3) storing parameters as constituent elements of a neural network obtained by machine learning, an estimation unit (2) estimating a fluctuation amount relevant to a position of an element arranged in a machine tool (11) or a fluctuation amount of a distance between elements arranged in the machine tool (11) for each physical condition information of the machine tool (11) by means of the neural network with a parameter freely selected from the parameters being omitted, and a reliability evaluation unit (4) evaluating reliability of estimated multiple fluctuation amounts based on the estimated fluctuation amounts. The correction amount calculation unit (5) calculates a correction amount for the estimated fluctuation amounts based on the fluctuation amounts, and the correction amount output unit (7) outputs the calculated correction amount to outside.

Claims

1. A fluctuation amount estimation device in a machine tool, comprising: a parameter storage storing therein a plurality of parameters defining a relation between a fluctuation amount relevant to a position of an element arranged in the machine tool or a fluctuation amount of a distance between elements arranged in the machine tool and physical condition information as information indicative of a physical condition of the machine tool, the plurality of parameters being constituent elements of a neural network and being obtained by machine learning using the neural network; an estimation unit estimating the fluctuation amount based on the parameters stored in the parameter storage and the physical condition information input successively from outside, the estimation unit repeating, for each of the input physical condition information, estimation of the fluctuation amount by means of the neural network multiple times with a parameter freely selected from the plurality of parameters being omitted; and a reliability evaluation unit evaluating reliability of multiple fluctuation amounts for each of the physical condition information estimated by the estimation unit, based on the estimated multiple fluctuation amounts.

2. The fluctuation amount estimation device of claim 1, wherein the fluctuation amount estimation device includes an evaluation information output unit outputting evaluation information regarding the reliability evaluated by the reliability evaluation unit to outside.

3. The fluctuation amount estimation device of claim 1, wherein the fluctuation amount of the distance between the elements is a fluctuation amount of a distance between a workpiece and a tool arranged in the machine tool.

4. The fluctuation amount estimation device of claim 1, wherein the physical condition information is temperature information of a predetermined portion of the machine tool and the fluctuation amount is a thermal displacement amount.

5. The fluctuation amount estimation device of claim 3, wherein the physical condition information is condition information relevant to tool wear and the fluctuation amount is a tool wear amount.

6. A correction amount calculation device in a machine tool, comprising: the fluctuation amount estimation device of claim 1; a correction amount calculation unit calculating a correction amount for the multiple fluctuation amounts for each of the physical condition information estimated by the estimation unit, based on the multiple fluctuation amounts; and a correction amount output unit outputting the correction amount calculated by the correction amount calculation unit to outside.

7. The correction amount calculation device of claim 6, wherein the correction amount calculation unit is configured to calculate a correction amount adjusted in accordance with an evaluation result evaluated by the reliability evaluation unit.

8. The correction amount calculation device of claim 6, wherein the correction amount calculation unit is configured to calculate a correction amount by moving averaging when an evaluation value for the reliability evaluated by the reliability evaluation unit is a value which is judged as having low reliability with respect to a predetermined reference value.

9. The correction amount calculation device of claim 6, wherein the correction amount calculation unit is configured to calculate a correction amount corresponding to a minimum one of the estimated multiple fluctuation amounts when an evaluation value for the reliability evaluated by the reliability evaluation unit is a value which is judged as having low reliability with respect to a predetermined reference value.

10. The correction amount calculation device of claim 6, wherein the correction amount calculation unit is configured not to calculate a correction amount when an evaluation value for the reliability evaluated by the reliability evaluation unit is a value which is judged as having low reliability with respect to a predetermined reference value.

11. The fluctuation amount estimation device of claim 2, wherein the fluctuation amount of the distance between the elements is a fluctuation amount of a distance between a workpiece and a tool arranged in the machine tool.

12. The fluctuation amount estimation device of claim 2, wherein the physical condition information is temperature information of a predetermined portion of the machine tool and the fluctuation amount is a thermal displacement amount.

13. The fluctuation amount estimation device of claim 3, wherein the physical condition information is temperature information of a predetermined portion of the machine tool and the fluctuation amount is a thermal displacement amount.

14. The fluctuation amount estimation device of claim 11, wherein the physical condition information is temperature information of a predetermined portion of the machine tool and the fluctuation amount is a thermal displacement amount.

15. The fluctuation amount estimation device of claim 11, wherein the physical condition information is condition information relevant to tool wear and the fluctuation amount is a tool wear amount.

Description

BRIEF DESCRIPTION OF DRAWINGS

[0041] FIG. 1 is a block diagram illustrating a correction amount calculation device according to an embodiment of the present invention;

[0042] FIG. 2 is a block diagram illustrating a machine learning device;

[0043] FIG. 3 is a front view of a machine tool used in the embodiment;

[0044] FIG. 4 is a right side view of the machine tool illustrated in FIG. 3;

[0045] FIG. 5 is an illustrative diagram showing a neural network model as a manner of machine learning;

[0046] FIG. 6 is an illustrative diagram showing an arithmetic algorithm for estimation based on a neural network model;

[0047] FIG. 7 is an illustrative diagram showing a neural network model for deep learning;

[0048] FIG. 8 is an illustrative diagram for explaining a manner of evaluating reliability of estimated values;

[0049] FIG. 9 is a graph showing an example of temperature data as an input value for estimation based on a neural network model;

[0050] FIG. 10 is a graph showing an example of temperature data as an input value for estimation based on a neural network model;

[0051] FIG. 11 is a graph showing an example of temperature data as an input value for estimation based on a neural network model;

[0052] FIG. 12 is a graph showing an example of temperature data as an input value for estimation based on a neural network model;

[0053] FIG. 13 is an illustrative diagram for explaining evaluation of reliability of estimated values based on the input value shown in FIG. 9;

[0054] FIG. 14 is an illustrative diagram for explaining evaluation of reliability of estimated values based on the input value shown in FIG. 10;

[0055] FIG. 15 is an illustrative diagram for explaining evaluation of reliability of estimated values based on the input value shown in FIG. 11;

[0056] FIG. 16 is an illustrative diagram for explaining evaluation of reliability of estimated values based on the input value shown in FIG. 12; and

[0057] FIG. 17 is an illustrative diagram showing a result of evaluation of reliability.

DESCRIPTION OF EMBODIMENTS

[0058] Hereinafter, a specific embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram illustrating a correction amount calculation device according to the embodiment.

[0059] The correction amount calculation device 1 according to this embodiment estimates a fluctuation amount of a distance between a tool T and a workpiece W in a machine tool 11 as illustrated in FIGS. 3 and 4 that is caused by thermal displacement of components of the machine tool 11, and calculates a correction amount for the estimated fluctuation amount. Note that, needless to say, the machine tool 11 used in this embodiment is only an example and the invention may be applied to any other type of machine tool. Note further that the fluctuation amount of the distance between the tool T and the workpiece W estimated in this embodiment is only an example and the fluctuation amount to be estimated in the invention is not limited thereto.

[0060] [Machine Tool]

[0061] First, a schematic configuration of the machine tool 11 used in this embodiment is briefly described. As illustrated in FIGS. 3 and 4, the machine tool 11 is the so-called NC lathe, and includes a bed 12, a headstock 13 arranged on the bed 12, and a carriage 16 and a tailstock 19 arranged on the bed 12 to be movable in the direction of Z axis that is indicated by an arrow in the drawings. Motion mechanisms of the machine tool 11 are controlled by an NC device 10 shown in FIG. 1.

[0062] A spindle 14 is arranged to have a center axis extending along the Z axis, and is supported by the headstock 13 to be rotatable about the center axis. The spindle 14 has a chuck 15 attached to a distal end thereof, and the chuck 15 clamps the workpiece W. Note that the spindle 14 rotates about the center axis by being driven by a spindle motor (not illustrated) incorporated in the headstock 13. The carriage 16 can be moved in the Z-axis direction by a Z-axis feed apparatus that is not illustrated in the drawings. The carriage 16 has a tool rest 17 arranged thereon. The tool rest 17 includes a turret 18 for attaching the tool T thereto, and can be moved in the direction of X axis that is indicated by an arrow in the drawings, by an X-axis feed apparatus that is not illustrated in the drawings.

[0063] Thus, in this machine tool 11, under control by the NC device 10, the spindle motor, the X-axis feed apparatus, and the Z-axis feed apparatus are driven as appropriate so that the tool T is moved relative to the workpiece W in an X-Z pane. By this relative movement of the tool T, the workpiece W is machined into a desired shape.

[0064] [Correction Amount Calculation Device]

[0065] Next, a configuration of the correction amount calculation device 1 according to this embodiment is described. As illustrated in FIG. 1, the correction amount calculation device 1 consists of processing units, namely, an estimation unit 2, a data storage 3, a reliability evaluation unit 4, a correction amount calculation unit 5, an input unit 6, and an output unit 7, and a display 8 connected to the output unit 7. In this correction amount calculation device 1, the estimation unit 2, the data storage 3, the reliability evaluation unit 4, the correction amount calculation unit 5, the input unit 6, and the output unit 7 are composed of a computer including a CPU, a RAM, and a ROM. The estimation unit 2, the reliability evaluation unit 4, the correction amount calculation unit 5, the input unit 6, and the output unit 7 are functionally implemented by a computer program to execute the processes described later. Further, the data storage 3 is composed of an appropriate storage medium such as a RAM. The estimation unit 2, the data storage 3, the reliability evaluation unit 4, the input unit 6, the output unit 7, and the display 8 together function as a fluctuation amount estimation device 9.

[0066] The data storage 3 is a functional unit that stores therein a plurality of parameters defining a relation between temperature data as information indicative of a physical condition of the machine tool 11 and the fluctuation amount of the distance between the tool T and the workpiece W. These parameters are obtained by backpropagation-based supervised machine learning using a neural network, which are calculated by a machine learning device 100 as illustrated in FIG. 2.

[0067] The temperature data (cause data) as training data is measured by temperature sensors S.sub.1 to S.sub.i that are respectively disposed at appropriate positions on the components of the machine tool 11, namely, the bed 12, the headstock 13, the carriage 16, and the tool rest 17. The fluctuation amount of the distance between the tool T and the workpiece W (effect data) is measured by a displacement sensor that is arranged appropriately so as to measure the distance between the tool T and the workpiece W. Note that i is a natural number from 1 to n.

[0068] Specifically, the temperature sensors S.sub.1 to S.sub.i disposed in the machine tool 11 respectively measure temperatures t.sub.1 to t.sub.i at predetermined sampling intervals, and simultaneously the displacement sensor measures a displacement amount (thermal displacement amount) d of the distance between the tool T and the workpiece W.

[0069] Note that the thermal displacement amount d is measured as a thermal displacement amount d.sub.X in the X-axis direction and a thermal displacement amount d.sub.Z in the Z-axis direction.

[0070] A typical neural network model used in the machine learning device 100 is illustrated in FIG. 5, and an arithmetic algorithm in this neural network model is illustrated in FIG. 6. In the arithmetic algorithm illustrated in FIG. 6, t.sub.1 to t.sub.i in the input layer respectively correspond to output values of the temperature sensors S.sub.1 to S.sub.i. Further, Hw.sub.i,j and Hw.sub.i are weighting factors, and V.sub.j and Kv are thresholds as reaction sensitivity. Further, output A.sub.j in the intermediate layer is calculated by Equation 1 below. Note that j is a natural number from 1 to m.

A.sub.j=f((ΣHw.sub.i,j.Math.t)+V.sub.j)   (Equation 1)

Further, output D as an estimated value in the output layer is calculated by Equation 2 below.

D=f((ΣKw.sub.jA.sub.j)+Kv)   (Equation 2)

Note that the outputs A.sub.j and D are transformed by a sigmoid function represented by Equation 3 below or the like.

f(u)=1/(1+exp(−u))   (Equation 3)

[0071] By using data regarding the temperatures t.sub.1 to t.sub.i and thermal displacement amounts d.sub.X, d.sub.Z obtained as described above as training data and using the above-described arithmetic algorithm, after setting the number of intermediate layer neurons and the number of intermediate layers appropriately, the machine learning device 100 calculates parameters A.sub.Xj, Kw.sub.Xj, Kv.sub.X, Hw.sub.Xi,j, and V.sub.Xj for the thermal displacement amount D.sub.X in the X-axis direction and parameters A.sub.Zj, Kw.sub.Zj, Kv.sub.Z, Hw.sub.Zi,j, and V.sub.Zj for the thermal displacement amount D.sub.Z in the Z-axis direction by the above-described backpropagation-based supervised machine learning. Note that the number of intermediate layer neurons j is freely selected; generally, a larger number of intermediate layer neurons j leads to higher accuracy.

[0072] The backpropagation-based supervised machine learning in this embodiment is such that: output values of the temperature sensors S.sub.1 to S.sub.i that are previously obtained are each input as an input value t for the arithmetic algorithm illustrated in FIG. 6; an output value (estimated thermal displacement amount D (D.sub.X, D.sub.Z)) obtained by setting the weighting factors Hw.sub.i,j and Kw.sub.j and the thresholds V.sub.j and Kv appropriately is compared with the true value (measured thermal displacement amount=d (d.sub.X, d.sub.Z)); and the operation of changing the weighting factors Hw.sub.i,j and Kw.sub.j and the thresholds V.sub.j and Kv is repeatedly carried out so that the difference between the obtained output value and the true value is reduced, that is to say, so that the obtained output value converges to the true value, whereby optimal values are set for the weighting factors Hw.sub.i,j and Hw.sub.i and thresholds V.sub.j and Kv as parameters. Note that the input value t in the input layer and the thresholds V.sub.j, Kv in the intermediate layer in FIG. 6 are each called “node”, while the weighting factors Hw.sub.i,j and Kw.sub.j are each called “edge”. Note further that the thermal displacement amounts D.sub.X and d.sub.X in the X-axis direction in this embodiment are radius values.

[0073] By performing this backpropagation-based machine learning based on data regarding the temperatures t.sub.1 to t.sub.i and thermal displacement amounts d.sub.X, d.sub.Z obtained as training data, the machine learning device 100 sets the parameters A.sub.Xj, A.sub.Zj, Kw.sub.Xj, Kw.sub.Zj, Kv.sub.X, Kv.sub.Z, V.sub.Xj, and V.sub.Zj for estimating (calculating) the thermal displacement amounts D.sub.X, D.sub.Z based on the temperatures t.sub.1 to t.sub.i measured by the temperature sensors Si to Si. The calculated parameters A.sub.Xj, A.sub.Zj, Kw.sub.Xj, Kw.sub.Zj, Kv.sub.X, Kv.sub.Z, V.sub.Xj, and V.sub.Zj are stored into the data storage 3 through the input unit 6.

[0074] Note that, although the neural network model illustrated in FIG. 5 that has a relatively simple structure is only an example for easy understanding and the neural network model to be used is not limited thereto. A neural network model with multiple intermediate layers as illustrated in FIG. 7, which is generally called “deep learning”, may be used. In such a case, the arithmetic algorithm illustrated in FIG. 6 is also to be adapted for such a deep-learning neural network model.

[0075] The estimation unit 2 inputs the temperatures t.sub.1 to t.sub.i measured by the temperature sensors S.sub.1 to S.sub.i arranged in the machine tool 11 at predetermined sampling intervals through the input unit 6, and calculates, for each set of temperatures t.sub.1 to t.sub.i, the thermal displacement amount D.sub.X in the X-axis direction and the terminal displacement amount D.sub.Z in the Z-axis direction of the distance between the tool T and the workpiece W in accordance with Equation 4 and Equation 5 below using the parameters stored in the data storage 3, respectively.

D.sub.X=1/(1+exp(−((ΣA.sub.Xj.Math.Kw.sub.Xj)+Kv.sub.X))

A.sub.Xj=1/(1+exp(−((ΣHw.sup.Xi,j.Math.t.sub.i)+V.sub.Xj))   (Equation 4)

D.sub.Z=1/(1+exp(−((ΣA.sub.Zj.Math.Kw.sub.Zj)+Kv.sub.Z))

A.sub.Zj=1/(1+exp(−((ΣHw.sub.Zi,j.Math.t.sub.i)+V.sub.Zj))   (Equation 5)

[0076] In this process, each time when a set of temperature data t.sub.1 to t.sub.i is input, the estimation unit 2 repeats estimation of the thermal displacement amounts D.sub.X, D.sub.Z using the neural network model, i.e., in accordance with Equation 4 and Equation 5 above, multiple times with one or more parameters freely selected from the parameters being omitted. For example, the estimation unit 2 carries out the estimation of the thermal displacement amounts D.sub.X, D.sub.Z with one or more parameters (nodes) freely selected from the parameters V.sub.Xj and V.sub.Zj (intermediate layer nodes) of the neural network model illustrated in FIG. 5 being omitted and the parameters Hw.sub.Xi,j, Hw.sub.Zi,j, Kw.sub.Xj, Kw.sub.Zj (intermediate layer edges) relevant (connected) to the selected parameter or parameters being omitted. Each time when a set of temperature data t.sub.1 to t.sub.i is input, the estimation unit 2 repeats the estimation multiple times while changing the parameters to be omitted. For the purpose of easier understanding, a conceptual diagram in which freely selected nodes and edges relevant to the nodes are omitted from the neural network model illustrated in FIG. 7 is shown in FIG. 8. In FIG. 8, the omitted nodes are shown by broken lines and the lines indicative of the relevant edges are deleted.

[0077] The reliability evaluation unit 4 is a functional unit that evaluates, for each input set of temperature data t.sub.1 to t.sub.i, reliability of the multiple thermal displacement amounts D.sub.Xk, D.sub.Zk estimated by the estimation unit 2, based on the estimated multiple thermal displacement amounts D.sub.Xk, D.sub.Zk. Note that k is a natural number from 1 to p, and p means the number of times of repeat.

[0078] As already mentioned above, the inventors of the present invention have obtained the following knowledge as a result of intensive studies: in the process of estimating the thermal displacement amounts D.sub.X, D.sub.Z (fluctuation amount) as an output value by using the neural network, for example, with a set of temperature data t.sub.1 to t.sub.i (physical condition information) obtained successively as an input value, in the case where estimation of the thermal displacement amounts D.sub.X, D.sub.Z is repeated multiple times with one or more parameters freely selected from the plurality of parameters being omitted, there is small variation in the estimated values when the input value is a normal value that has been machine-learned, whereas there is large variation in the estimated values when the input value is a value which has not been machine-learned.

[0079] Therefore, when the degree of variation in the thermal displacement amounts D.sub.Xk, D.sub.Zk estimated through the repeated estimation exceeds a predetermined reference value (threshold), the estimated values can be judged as having low reliability. On the other hand, when the degree of variation in the estimated fluctuation amounts does not exceed the reference value (threshold), the estimated values can be judged as having high reliability.

[0080] FIGS. 9 to 12 show examples of output data which is output in the event of failure of a temperature sensor used in measurement. FIGS. 9 and 11 each show output data which is output in the event of disconnection of a temperature sensor of a resistance-value output type. In each case, an extremely abnormal value (as low temperature data as −128° C.) is output. FIGS. 10 and 12 each show output data which is output in the event of disconnection of a temperature sensor of voltage output type. In such a case, the output value of 0° C. is obtained.

[0081] FIGS. 13 to 16 show results obtained in the case where the thermal displacement amounts D.sub.Xk are estimated through the above-described repeated estimation when the temperature data shown in FIGS. 9 to 12 are input. FIG. 13 shows the estimated thermal displacement amounts D.sub.Xk obtained when the temperature data shown in FIG. 9 is input. FIG. 14 shows the estimated thermal displacement amounts D.sub.Xk obtained when the temperature data shown in FIG. 10 is input. FIG. 15 shows the estimated thermal displacement amounts D.sub.Xk obtained when the temperature data shown in FIG. 11 is input. FIG. 16 shows the estimated thermal displacement amounts D.sub.Xk obtained when the temperature data shown in FIG. 12 is input. In FIGS. 13 to 16, the broken-line curve represents the measured value (d.sub.X) of the thermal displacement amount. Further, the upper thin solid line represents a curve connecting values obtained by adding a standard deviation to an average of the thermal displacement amounts D.sub.Xk estimated through the repeated estimation, while the lower thin solid line represents a curve connecting values obtained by subtracting the standard deviation from the average of the estimated thermal displacement amounts D.sub.Xk. The bold solid line represents a curve connecting the averages. Further, the dashed and dotted line represents a curve that indicates the standard deviation of the estimated thermal displacement amounts D.sub.Xk.

[0082] As shown in FIGS. 13 and 15, in the case where an extremely abnormal value is output because of disconnection of a temperature sensor of resistance-value output type, there is a large degree of variation in the values of the thermal displacement amounts D.sub.Xk estimated through the repeated estimation. In FIGS. 13 and 15, input of an abnormal value from the temperature sensor begins after the elapsed time exceeds 10 hours. On the other hand, extremely abnormal data is not output in the event of failure of a temperature sensor of voltage output type. Therefore, as shown in FIGS. 14 and 16, the variation in the values of the thermal displacement amounts D.sub.Xk estimated through the repeated estimation is not much different from that seen when normal data is output.

[0083] Therefore, when the degree of variation in the thermal displacement amounts D.sub.Xk, D.sub.Zk estimated through the repeated estimation exceeds, for example, a predetermined reference value (threshold), the reliability evaluation unit 4 judges the estimated values as having low reliability. On the other hand, when the degree of variation in the thermal displacement amounts D.sub.Xk, D.sub.Zk does not exceed the reference value (threshold), the reliability evaluation unit 4 judges the estimated values as having high reliability. Examples of the index for the degree of variation, i.e., the evaluation value for reliability, include range of variation (difference between maximum and minimum values), standard deviation σ obtained by statistical processing of the variation, as well as 2σ, 3σ, 4σ, 6σ obtained from standard deviation σ. By using any one of these indexes as the evaluation value, the reliability evaluation unit 4 judges the estimated values as having low reliability when the value of the index exceeds a predetermined reference value (threshold), and judges the estimated values as having high reliability when the value of the index does not exceed the reference value. The reliability evaluation unit 4 displays the evaluation result on the display 8 through the output unit 7. Alternatively, the reliability evaluation unit 4 may be configured to display an error indication on the display 8 when the estimated values are judged as having low reliability. In this case, the output unit 7 functions as an evaluation information output unit. Note that the reliability judgement is not limited to using a single reference value. The reliability judgment may use multiple reference values to rate reliability of the estimated values on a scale of grades such as Reliability 1, Reliability 2, Reliability 3, and so on.

[0084] Based on the thermal displacement amounts D.sub.Xk, D.sub.Zk estimated by the estimation unit 2 and the evaluation result evaluated by the reliability evaluation unit 4, the correction amount calculation unit 5 calculates correction amounts C.sub.X, C.sub.Z that are adjusted in accordance with the evaluation result. For example, when evaluation result evaluated by the reliability evaluation unit 4 is “high reliability”, the correction amount calculation unit 5 calculates thermal displacement amounts D.sub.Xa, D.sub.Za as representative values that are obtained by simple averaging of the thermal displacement amounts D.sub.Xk, D.sub.Zk, and then calculates a correction amount C.sub.X for the thermal displacement amount D.sub.Xa in the X-axis direction and a correction amount C.sub.Z for the thermal displacement amount D.sub.Za in the Z-axis direction in accordance with Equation 6 below. On the other hand, when the evaluation result evaluated by the reliability evaluation unit 4 is “low reliability”, the calculation amount calculation unit 5 calculates thermal displacement amounts D.sub.Xa, D.sub.Za as representative values by performing simple averaging of the thermal displacement amounts D.sub.Xk, D.sub.Zk and moving averaging of the simple average with previous ones, and then calculates correction amounts C.sub.X, C.sub.Z in accordance with Equation 6 below.

C.sub.X=−D.sub.Xa

C.sub.Z=−D.sub.Za   (Equation 6)

[0085] The correction amount calculation unit 5 outputs the calculated correction amounts C.sub.X, C.sub.Z to the NC device 10 through the output unit 7. The NC device 10 corrects a relative positioning position for the workpiece W and the tool T in the machine tool 100 based on the correction amounts C.sub.X, C.sub.Z. In this process, the output unit 7 functions as a correction amount output unit.

[0086] In the correction amount calculation device 1 having the above-described configuration, the estimation unit 2 estimates, based on temperature data t.sub.1 to t.sub.i input successively from the machine tool 11 and the parameters stored in the data storage 3, thermal displacement amounts D.sub.X, D.sub.Z between the tool T and the workpiece W for each input set of temperature data t.sub.1 to t.sub.i.

[0087] In this process, the estimation unit 2 carries out the estimation using the Equation 4 and Equation 5 above with one or more parameters freely selected from the parameters being omitted, and repeats the estimation multiple times while changing the parameters to be omitted, thereby calculating thermal displacement amounts D.sub.Xk, D.sub.Zk. The estimation unit 2 outputs data regarding the calculated thermal displacement amounts D.sub.Xk, D.sub.Zk to the reliability evaluation unit 4. An example of the thus-estimated thermal displacement amounts in the X-axis direction is shown in FIG. 17. In FIG. 17, the upper thin solid line represents a curve connecting values obtained by adding a standard deviation to an average of the thermal displacement amounts D.sub.Xk estimated through the repeated estimation, the lower thin solid line represents a curve connecting values obtained by subtracting the standard deviation from the average of the estimated thermal displacement amounts D.sub.Xk, and the bold solid line represents a curve connecting the averages.

[0088] The reliability evaluation unit 4 evaluates reliability of the estimated thermal displacement amounts D.sub.Xk, D.sub.Zk input from the estimation unit 2 on the basis of an index (evaluation value) for the degree of variation in the estimated thermal displacement amounts D.sub.Xk, D.sub.Zk, and outputs the evaluation result to the correction amount calculation unit 5 and to the display 8 through the output unit 7 so that the evaluation result is displayed on the display 8. Since the reliability of the thermal displacement amounts D.sub.Xk, D.sub.Zk estimated by the estimation unit 2 is evaluated by the evaluation unit 4 and the evaluation result is displayed on the display 8, it is possible to make an objective judgement on reliability of the estimated thermal displacement amounts D.sub.Xk, D.sub.Zk.

[0089] The correction amount calculation unit 5 calculates, based on the thermal displacement amounts D.sub.Xk, D.sub.Zk estimated by the estimation unit 2 and the evaluation result provided by the reliability evaluation unit 4, correction amounts C.sub.X, C.sub.Z adjusted in accordance with the evaluation result, and outputs the calculated correction amounts C.sub.X, C.sub.Z to the NC device 10 through the output unit 7. Thus, with the correction amount calculation device 1, since correction amounts subjected to reliability-based adjustment are calculated even when the estimated thermal displacement amounts D.sub.Xk, D.sub.Zk have low reliability because of a large degree of variation in the thermal displacement amounts D.sub.Xk, D.sub.Zk, it is possible to prevent the occurrence of large variation in successively calculated correction amounts, so that it is possible to carry out correction smoothly to some degree.

[0090] Hereinbefore, an embodiment of the present invention has been described. However, it should be understood that the present invention is not limited to the above-described embodiment and may be implemented in other manners.

[0091] For example, the correction amount calculation unit 5 may be configured to, when the evaluation value for the reliability evaluated by the reliability evaluation unit 4 is a value which is judged as having low reliability with respect to a predetermined reference value, designate thermal displacement amount D.sub.Xmin, D.sub.Zmin respectively having the minimum absolute value among the estimated thermal displacement amounts D.sub.Xk, D.sub.Zk as representative values, that is to say, D.sub.Xa=D.sub.Xmin, D.sub.Za=D.sub.Zmin, and calculate correction amounts C.sub.X, C.sub.Z for the designated thermal displacement amounts. With this configuration, even when the estimated thermal displacement amounts D.sub.Xk, D.sub.Zk have low reliability, minimum correction amounts are calculated, so that it is possible to carry out correction and it is possible to prevent extreme and excessive correction.

[0092] Alternatively, the correction amount calculation unit 5 may be configured not to perform the calculation of correction amounts C.sub.X, C.sub.Z when the evaluation value for the reliability evaluated by the reliability evaluation unit 4 is a value which is judged as having low reliability with respect to a predetermined reference value. With this configuration, correction amounts C.sub.X, C.sub.Z are calculated only when the estimated thermal displacement amounts D.sub.Xk, D.sub.Zk have high reliability, whereas correction amounts C.sub.X, C.sub.Z are not calculated when the estimated thermal displacement amounts D.sub.Xk, D.sub.Zk have low reliability; therefore, it is possible to prevent correction based on an inappropriate correction amount.

[0093] Further, the fluctuation amount to be estimated in the above embodiment is a thermal displacement amount between the tool T and the workpiece W; however, the present invention is not limited thereto. The fluctuation amount to be estimated may be position fluctuation (thermal displacement) by temperature change of a predetermined portion of an element arranged in the machine tool 11, e.g., position fluctuation (thermal displacement) by temperature change of portions of ball screws used in the X-axis feed apparatus (not illustrated) and Z-axis feed apparatus (not illustrated) or position fluctuation (thermal displacement) by temperature change of a distal end of a spindle holding and rotating a tool or a workpiece.

[0094] Alternatively, the physical condition information of the machine tool 11 may be condition information relevant to tool wear, e.g., information on load (supplied power) of the spindle motor and/or a feed motor in machining using the tool or information on the temperature of the tool or chips in machining, and the fluctuation amount to be estimated may be a tool wear amount. Note that the information on the temperature of the tool or chips in machining can be measured, for example, by an infrared camera. Further, the tool wear amount can be calculated by measurement using a displacement sensor or by processing an image of a tool tip captured by a camera.

[0095] As already mentioned above, the foregoing description of the embodiments is not limitative but illustrative in all aspects. One skilled in the art would be able to make variations and modifications as appropriate. The scope of the invention is not defined by the above-described embodiments, but is defined by the appended claims. Further, the scope of the invention encompasses all modifications made within the scope equivalent to the scope of the claims.

REFERENCE SIGNS LIST

[0096] 1 Correction amount calculation device

[0097] 2 Estimation unit

[0098] 3 Data storage

[0099] 4 Reliability evaluation unit

[0100] 5 Correction amount calculation unit

[0101] 6 Input unit

[0102] 7 Output unit

[0103] 8 Display

[0104] 9 Fluctuation amount estimation device

[0105] 10 NC device

[0106] 11 Machine tool

[0107] T Tool

[0108] W Workpiece