Fitting state detection method of tool holder and machine tool to which tool holder is to be attached

Abstract

To enable reliable detection of a fact that a tool holder is attached to a spindle of a machine tool in an appropriate state when a tool is automatically replaced by using an automatic tool changer (ATC).SOLUTION: A machine tool includes an automatic tool changer. In the machine tool, a displacement sensor that detects a distance between a tool holder replaceably attached to a spindle and the spindle is provided in the vicinity of the spindle. The machine tool includes: an operation part that performs conversion and creates a histogram for each axial direction of the spindle from data of a change in a distance with an outer peripheral part of the tool holder detected by the displacement sensor; a preprocessing part that creates a histogram group including a histogram of a normal attachment state and at least one abnormal attachment state with respect to the spindle and the tool holder and stores the histogram group as master data; and a determination part that determines normal or abnormal of an attachment state indicated by the histogram created by the operation part on the basis of the master data stored in the preprocessing part.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a method for detecting the mounting state of a tool holder in an NC machine tool or the like, and a machine tool to which a tool holder is mounted, and in particular to a method for detecting the mounting state of a tool holder suitable for a machine tool having an automatic tool changer (ATC), and a machine tool to which a tool holder is mounted. [Background technology]

[0002] In NC machine tools and machining centers equipped with an automatic tool changer (ATC), tools are automatically changed depending on the machining content of the workpiece. This tool change is programmed in advance in the NC machine tool or machining center, and a tool holder with a tool stored in a specified position in the ATC is attached to the spindle of the machine tool according to the program. Since this is performed automatically without human intervention, a detection means is provided near the spindle of the machine tool to detect whether the tool holder is properly attached to the spindle.

[0003] For example, in the machine tool described in Patent Document 1, when foreign matter such as chips is caught between the spindle and the tool holder, the inclination of the tool holder can be detected with high accuracy using an inexpensive sensor. Specifically, an eddy current sensor is disposed on the side of the flange of the tool holder, facing obliquely to the side of the eddy current sensor. The spindle is rotated, and the inclination of the tool holder is detected based on a change in the output of the eddy current sensor caused by a change in the relative position between the flange of the tool holder and the head of the eddy current sensor.

[0004] Another conventional example of a method for detecting the mounting state of a tool holder is described in Patent Document 2. The machine tool described in this publication is configured to automatically measure the amount of eccentricity when the tool holder is clamped and detect a clamping abnormality. That is, the headstock of the machine tool rotatably supports the spindle, and the tool holder is mounted on the spindle by an automatic tool changer. The tool holder has a flange, and a tool holder eccentricity detection device is disposed at a position facing the flange. The eccentricity detection device has an eddy current sensor head and measures the distance between the tool holder and the flange in a non-contact manner, and when there is a large change in the measured distance, it is determined that the tool holder is clamped eccentrically, and a clamping abnormality occurs.

[0005] Yet another conventional example of a method for detecting the mounting condition of a tool holder is described in Patent Document 3. In the machine tool described in this publication, a tool holder with a tool attached is attached to a spindle, and the spindle is rotated to machine a workpiece, during which a sensor measures the displacement of the outer peripheral surface of the flange of the tool holder, and a data processing device performs Fourier analysis on the measured data and judges whether the tool holder is mounted on the spindle abnormally from the power spectrum shape of the analysis results.

[0006] In addition, since a general-purpose tool holder has a notch for removal formed at a symmetrical position in the circumferential direction, the measurement result is interrupted at this notch regardless of which of the above methods is used. In order to solve this problem, Patent Document 4 describes a method of determining invalid data parts from the measurement results and interpolating the invalid parts. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] JP 2007-260885 A [Patent Document 2] JP 2002-331442 A [Patent Document 3] JP 2004-276145 A [Patent Document 4] JP 2008-93750 A Summary of the Invention [Problem to be solved by the invention]

[0008] As described in the above patent documents, when tools are changed using ATC without manual intervention, there is a risk that chips remaining in the environment in which the machine tool is placed may get caught between the tool holder and the spindle, causing the tool holder to be attached to the spindle tilted and / or eccentric. To eliminate this problem, it is necessary to provide some kind of sensor near the spindle to confirm that the tool holder is properly attached to the spindle of the machine tool.

[0009] In the machine tool described in Patent Document 1, the eddy current sensor must be disposed at an angle to the spindle or tool holder so as not to interfere with tool changing operations by the ATC, which limits the setting position of the eddy current sensor. Also, to increase the sensitivity of the eddy current sensor, the sensor must be placed as close as possible to the tool holder, but since tool holders generally have notches, the closer the sensor is to the tool holder, the more the effect of the notch will be included in the measurement data, and this effect must be removed.

[0010] The machine tools described in Patent Documents 2 and 3 differ from Patent Document 1 in that eddy current sensors are arranged substantially horizontally facing each other on the side surface of the tool holder. However, other configurations are similar, and it is considered necessary to eliminate the effect of the notch formed in the tool holder, but Patent Document 2 does not disclose how to eliminate the effect of the notch.

[0011] In addition, in Patent Document 3, the measurement data is compared with the basic data measured when the tool is normally installed, so even if there is a notch, the effect of the notch can be eliminated from the measurement data. However, in the machine tool described in this publication, the data of the notch is eliminated, so there is an issue that needs to be improved, that is, data loss.

[0012] In Patent Document 4, which takes into consideration the handling of the notched portion of the tool holder, a data processing device determines the notched portion and interpolates the data of that portion to cover the missing data amount. This allows for more accurate detection of tool holder mounting abnormalities. In the machine tool described in Patent Document 4, in order to use the interpolated data as data for mounting abnormality judgment, registered data, which is data when the tool is normally mounted, is obtained and compared with the measured data. When making this comparison, it is necessary to align the circumferential positions (phase) of the registered data and the measured data, but there are rare cases where a correct judgment is not made due to differences in circumferential positions.

[0013] The present invention has been made in consideration of the above-mentioned problems in the conventional technology, and has an object to reliably detect whether a tool holder is properly attached to a spindle of a machine tool when a tool is automatically changed using an ATC. In addition to this object, another object is to detect an abnormality in the attachment of the tool holder to the spindle as quickly as possible. [Means for solving the problem]

[0014] The present invention, which achieves the above-mentioned object, is characterized in that, in a method for detecting a mounting state of a tool holder using a displacement sensor provided in the vicinity of a spindle when automatically mounting the tool holder on a spindle of a machine tool, the method includes the steps of: measuring, by the displacement sensor, the distance between the displacement sensor and the spindle for one or more revolutions of the spindle; converting data of the measured distance into a histogram for each axial position of the spindle; and determining whether or not the tool holder is correctly mounted on the spindle within a tolerance range by using an HBOS (histogram based outlier detection) method, based on the obtained histogram and a plurality of histograms stored in advance.

[0015] In this feature, the tool holder may have cutouts at positions substantially symmetrical in the circumferential direction, and the histogram may be created based on a change in output of the displacement sensor at a circumferential position of the tool holder excluding the cutouts. Also, it is preferable to include a step of obtaining, before automatically mounting the tool holder on the spindle, at least one histogram when the tool holder is correctly mounted on the spindle within an allowable error range and at least one histogram when the tool holder is abnormally mounted on the spindle outside the allowable error range. It is further preferable to include a step of adding a histogram when it is determined in the determination step that the mounting is abnormal to the stored group of histograms.

[0016] Another feature of the present invention for achieving the above-mentioned object is that in a machine tool equipped with an automatic tool changer and having a displacement sensor disposed near a spindle to detect the distance between the spindle and a tool holder replaceably attached to the spindle, a calculation device is provided having: a calculation unit that converts data of changes in the distance to the outer periphery of the tool holder detected by the displacement sensor into a histogram for each axial direction of the spindle; a pre-processing unit that creates a histogram group having histograms of a normal mounting state and at least one abnormal mounting state regarding the spindle and the tool holder, and stores the histogram group as master data; and a judgment unit that judges whether the mounting state indicated by the histogram created by the calculation unit is normal or abnormal, based on the master data stored in the pre-processing unit.

[0017] In this feature, it is preferable that the master data stores a histogram of a normal mounting state, the number of which corresponds to the number of tool holders to be replaced, and a histogram of an abnormal mounting state equal to or greater than the number of tool holders, and it is preferable that the judgment unit judges a mounting abnormality between the spindle and the tool holder using an HBOS (histogram based outlier detection) method. Also, it is preferable that the machine tool is an NC machine tool or a machining center. Effect of the Invention

[0018] According to the present invention, the occurrence frequency of displacement at positions divided in the axial direction is calculated from the data of the distance between the displacement sensor and the tool holder detected by the displacement sensor, and the occurrence frequency is compared with the occurrence frequency in the reference data, so that it is possible to eliminate the influence of the circumferential direction (phase) and to reliably detect that the tool holder is properly attached to the spindle of the machine tool. Also, since phase matching is no longer necessary simply by converting the measurement data into a frequency distribution, it is possible to detect an abnormal attachment of the tool holder to the spindle as quickly as possible. [Brief description of the drawings]

[0019] [Figure 1] 3A and 3B are a front view and a bottom view of a displacement sensor mounting portion in the embodiment of the present invention. [Diagram 2] 2 is a graph showing an example of an output from the displacement sensor shown in FIG. 1. [Diagram 3] FIG. 13 is a diagram illustrating a method for creating a histogram from displacement data. [Figure 4] FIG. 13 is a diagram showing an example of comparing displacement data with master data. [Diagram 5] FIG. 11 is a diagram showing an example of displacement data used in data processing; [Figure 6] 2 is a block diagram of an embodiment of the data calculation processing unit shown in FIG. 1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] Hereinafter, a method for detecting the presence or absence of an abnormality when mounting a tool holder on a spindle of a machine tool according to the present invention will be described with reference to the drawings. Note that in the following embodiments, the machine tool will be described as an NC machine tool or a machining center equipped with an automatic tool changer (ATC), but the machine tool is not limited to these and includes any machine that automatically changes tools.

[0021] Fig. 1 shows only the mounting portion of a tool holder 110 of a machine tool 100. The tool holder 110, to which a drill, which is a tool 112, is attached using an ATC (not shown), is shown mounted on a spindle 120 of the machine tool 100. Fig. 1(a) is a front view of the spindle 120, and Fig. 1(b) is a view from below. Fig. 1 also shows a mounting state detection device 200 that detects whether there is an abnormality in the mounting state.

[0022] In machine tool 100, tool holder 110 with tool 112 attached thereto is replaceably mounted on spindle 120. In this mounting, a conical fitting portion 114 provided on the upper end side of tool holder 110 is fitted into a complementary conical fitting portion 116 formed on spindle 120 in order to fit fitting portion 114 onto spindle 120. A mounting state detection device 200 is provided to monitor for mounting abnormalities caused by chips 102 or the like adhering to the fitting portion when the two are fitted together.

[0023] The mounting state detection device 200 of this embodiment automatically detects mounting abnormalities of the tool holder 110 mounted on the spindle 120, particularly by using the ATC, and is mainly composed of a sensor 210 and a processor 220, as shown in FIG. 1(a). The sensor 210 is a displacement sensor that detects the displacement of the cylindrical outer circumferential surface 110a of the flange portion of the tool holder 110 in a non-contact manner. For this reason, the sensor 210 is attached to the spindle 120 via a bracket 202 at a position at the lower end of the spindle 120 facing the outer circumferential surface 110a of the tool holder 110, and is disposed at a small distance from the tool holder 110. The distance from the outer circumferential surface 110a of the tool holder 110 detected by the sensor 210 is input to the processor 220 as a displacement signal Sg.

[0024] The arithmetic processing device 220 detects an abnormality in mounting of the tool holder 110 based on the displacement signal Sg detected by the sensor 210. The arithmetic processing device 220 includes an A / D converter 222 that converts the detected displacement signal Sg into a digital signal, an arithmetic processing unit 224 that performs various arithmetic processing, the details of which will be described later, on the displacement signal Sg converted into a digital signal, a storage unit 226 that stores master data and the like, and an input / output unit 228 that manages input / output processing such as outputting arithmetic results to the control device 150 of the machine tool 100.

[0025] 1(b), two notches 110c are formed at approximately symmetrical positions in the circumferential direction on an outer circumferential surface 110a of a flange portion 110b of a tool holder 110 in order to mount and remove the tool holder 110 to and from a spindle 120. As a result, the displacement signal detected by the sensor 210 has drastic changes at two positions in the circumferential direction (see FIG. 2).

[0026] FIG. 2 shows an example of a displacement signal output from the sensor 210 when the sensor 210 is arranged as shown in the embodiment of FIG. 1. The horizontal axis of FIG. 2 indicates the circumferential position of the tool holder 110, and is shown as a percentage (%) of the entire circumference. The vertical axis of FIG. 2 indicates the output value of the sensor 210, which can be set arbitrarily. If the output range that guarantees the linearity of the sensor 210 is, for example, ±1V, FIG. 2 shows 0 to -500mV. Since the sensor 210 is arranged facing the outer peripheral surface 110a of the tool holder 110, it can be seen that the output signal of the sensor 210 drops sharply at positions of about 15% and about 65% of the entire circumference. As described above, this corresponds to the notch 110c of the tool holder 110 shown in FIG. 1(b).

[0027] Conventionally, the displacement signal Sg detected by the sensor 210 for one or more revolutions in the circumferential direction of the tool holder 110 is compared with a master, which is a reference displacement signal, and a judgment is made as to whether the signal is normal or abnormal, depending on the degree of deviation from the master. In this case, many conventional methods use a method for eliminating the influence of the notch 110c included in the tool holder 110. However, what is required to judge whether the signal is normal or abnormal is not data on the notch 110c, but data on the outer peripheral surface 110a, which is a cylindrical surface, so it is not a good idea to spend time processing the notch 110c.

[0028] Another method is to recognize the presence of the notches 110c, treat the entire data including the data of the notches 110c as data for one or more revolutions in the circumferential direction of the tool holder 110 as a displacement signal Sg, and compare it with the displacement signal stored as a master in the memory unit 226 when the tool holder 110 is normally attached. In this case, the positional deviation of a pair of notches 110c has a significant effect on the determination of abnormality / normality. For example, it is necessary to distinguish between an abnormal state in which the chips 102 are included in the fitting portion of the tool holder 110 and the spindle 120 and the tool holder 110 is attached eccentrically or tilted, and a state in which the attachment state is normal but the alignment between the master performed to detect an abnormality and the data of the displacement signal Sg detected by the sensor 210 is not good. A typical example of the latter is a case in which the positions of two notches 110c formed almost symmetrically are recognized as opposite positions in the master and the detected displacement signal.

[0029] To correctly perform this distinction, it is necessary to align the master and the data, taking into consideration that the data detected by the displacement sensor 210 for one or more revolutions in the circumferential direction of the tool holder 110 varies within the assembly tolerance. To perform this alignment, it is necessary to provide an additional device, such as a camera for capturing images, or to provide a new reference point on the tool holder, which makes it less versatile and difficult to automate.

[0030] In the present invention, an outlier detection method using a histogram as shown below is adopted to judge whether the mounting state of the spindle 120 and the tool holder 110 is abnormal / normal. Fig. 3 shows a diagram for explaining a method for creating a histogram. Fig. 3(a) is a diagram showing the displacement signal Sg shown in Fig. 2 with the vertical and horizontal axes interchanged, and Fig. 3(b) is a histogram obtained as a result of arithmetic processing of the displacement signal Sg in Fig. 3(a) by the arithmetic processing unit 224.

[0031] First, in FIG. 3(a), the number of intersections 16, 18 between the straight lines 12, 14 extending in the vertical direction, which have a constant displacement, and the displacement data 10 detected by the displacement sensor 210 is counted. Here, the displacement data 10 includes data for the entire circumference of the tool holder 110. In the case of the straight line 12 intersecting with the data of the notch 110c, the number of intersections 16 is four, which correspond to both edges of the notch 110c, since the outer peripheral surface of the notch 110c is not a cylindrical surface but a flat surface 110d (see FIG. 1(b)). If the straight line with a constant displacement is moved to the right in FIG. 3(a), the number of intersections with the displacement data 10 from the edge of the notch 110c increases. In an ideal state in which the outer peripheral portion of the tool holder 110 is a perfect circle except for the notch 110c, and is attached completely concentrically and straight to the spindle 120, the number of intersections between the displacement data 10 and the straight line with constant displacement becomes infinite. However, in actual mounting, the tool holder 110 is mounted with variations within an assembly tolerance, and the outer peripheral surface 110a of the tool holder 110 is not a perfect circle when viewed microscopically, so a finite number of intersections 18 are obtained, as in the case of the straight line 14 and the displacement data 10. The number of intersections obtained in this manner is shown on the horizontal axis in Figure 3(b).

[0032] As is clear from the displacement data in Fig. 3(a), the number of intersections 16 is constant in the cutout 110c, and the number of intersections does not change in this portion even if the tool holder 110 is mounted at an angle or abnormally eccentric (portion of range B). On the other hand, near the position where the output of the displacement sensor 210 is at its maximum, that is, where the displacement sensor 210 and the outer periphery of the tool holder 110 are closest to each other (portion of range A), the number of intersections 18 is significantly increased compared to other portions. Therefore, it is clear that only the portion of range A needs to be considered in the subsequent judgments.

[0033] FIG. 4 shows an enlarged view of range A. FIG. 4(a) shows an example of the case where the tool holder 110 is in a normally attached state detected in advance, and this is hereinafter also referred to as a master 240 or master data 240. The master 240 is stored in the storage unit 226 of the arithmetic processing device 220 shown in FIG. 1. When creating the master, the peaks of the histogram as shown in FIG. 3(b) are connected by broken lines to make it convenient for subsequent processing. As explained in the explanation of FIG. 3, a sudden increase is observed near the output value of 0.

[0034] Fig. 4(b) is a diagram showing a comparison between the master data 240 shown in Fig. 4(a) and the displacement data 250 obtained by the sensor 210 when the tool holder 110 was actually attached to the spindle 120 this time. In Fig. 4(b), the solid line graph is a line connecting the vertices of the histogram of the displacement data 250 detected by the sensor 210, and the dashed dotted line graph is a line connecting the vertices of the histogram of the master 240. The memory unit 226 stores in advance the displacement data when the tool holder 110 is attached to the spindle 120 in an appropriate or normal state, and, if possible, the displacement data in an abnormal state.

[0035] Since the tool holder 110 is replaced by the ATC in accordance with the machining content of the workpiece, the number of tool holders 110 corresponding to the machining content of the workpiece or the type of the workpiece are used in the same machine tool 100. Therefore, the storage unit 226 stores master data 240 for at least the number of tool holders 110 that are replaced and used using the ATC. When using the same tool holder 110 and changing only the tool 112, it is sufficient to have master data 240 in the number corresponding to the number of tool holders 110 rather than the number of tools 112. However, as will be described later, if multiple master data for the same tool holder 110 are stored in the storage unit 226, mounting abnormalities can be determined with even higher accuracy.

[0036] Hereinafter, abnormal mounting of the tool holder 110 is detected by the HBOS (histogram-based outlier detection) method. In FIG. 4(b), there are discrepant parts in the line graphs of the two data 240, 250. The hatched part 20 is the common part of the line graphs of the two data 240, 250, the white part 22 is the part that exists only in the master 240, and the cross-hatched part 24 is the part that exists only in the detected displacement data 250. However, the shapes as a whole are considered to be almost the same, and the changes in the output value are also considered to be almost the same. Whether the two data 240, 250 can be considered to be the same, that is, whether the mounting can be considered to be normal, depends on whether the difference parts 22, 24 are within the allowable range.

[0037] For example, supervised machine learning with normal / abnormal labels can be used to determine the difference. Specifically, for example, when a new tool holder is introduced, the operation of mounting the tool holder 110 on the spindle 120 is repeated, and the displacement signal Sg is detected for each of them. Then, a plurality of patterns in which a histogram similar to that shown in FIG. 4(a) is plotted as a line graph are obtained. The operator judges whether the tool holder is normal or abnormal one by one from the results obtained by visually inspecting the tool holder or by performing simulated machining. The obtained large number of data are classified into a normal group and an abnormal group, and the classified data are trained in a binary classification machine learning model, and the data and the binary classification machine learning model are stored in the storage unit 226. Here, the normal group includes at least one data for each tool holder 110, and the abnormal group includes one or more data for each tool holder 110. Thereafter, the binary classification machine learning model is used to determine whether the tool holder is normal or abnormal. It is preferable that the determined result is further added as belonging to the normal group or the abnormal group, and the binary classification machine learning model is re-trained. This allows for the acquisition of displacement data for various wearing conditions, and the more data there is, the more stored wearing conditions there are to compare with, making it possible to obtain a more similar wearing condition from among them, and thus enabling highly accurate judgment of the wearing condition. Furthermore, the data that the binary classification machine learning model learns as normal and abnormal groups can be data actually acquired from the device, or data generated by simulation.

[0038] Fig. 5 shows an example of a comparison of each displacement data 250 obtained when the same tool holder 110 is mounted on the same spindle 120 multiple times with one master data 240. In Fig. 5, a line graph is created by connecting the peaks of the histograms, as in Fig. 4(a). The data located in the center (Fig. 5(a)) is the master data 240, and the data located around it (Figs. 5(b) to (g)) are the detected displacement data 250. In the calculation process, these two types of data 240, 250 are superimposed to determine the difference, but each data is shown as individual data to make the figure easier to see.

[0039] The multiple detected displacement data are classified into a first group G1 shown in Fig. 5(b)-(d) and a second group G2 shown in Fig. 5(e)-(g). Each data belonging to the first group G1 has multiple peaks near the center and also has a peak on the negative side of the horizontal axis. This characteristic is similar to the master data 240. On the other hand, each data belonging to the second group G2 has a small number of peaks on the positive side of the horizontal axis or no peaks on the negative side of the horizontal axis, and the gain (vertical axis scale) is low overall. Since the difference can be visually determined in this way, this is automatically determined using the above-mentioned labeled supervised machine learning.

[0040] The above-mentioned processes are executed by the calculation processing unit 224 included in the calculation processing device 220 of the wearing state detection device 200. Details of the calculation processing unit 224 will be described with reference to the block diagram shown in Fig. 6. The calculation processing unit 224 is broadly divided into a data pre-processing unit 310, a learning model generation unit 330, and a data determination unit 320. The data pre-processing unit 310 is responsible for processing up to obtaining the image (HBOS image) shown in Fig. 4(b).

[0041] First, distance (displacement) data between the tool holder 110 and the sensor 210 when each tool holder 110 to be used is properly attached to the spindle 120 is acquired before machining, and the data registration unit 312 registers the acquired data in the memory unit 226 as master data. Next, the learning data generation unit 314 generates a plurality of displacement data by simulation randomly or according to a certain criterion. The generated plurality of displacement data are defined as normal data and abnormal data based on the master data, and then grouped into normal data and abnormal data groups and registered in the memory unit 226.

[0042] The normal data and abnormal data groups may include displacement data detected by the sensor 210 obtained when machining was performed using the same spindle 120 and the same tool holder 110, including the previous time and previous times. In this case, data detected by the sensor 210 when the attachment of the tool holder 110 to the spindle 120 was abnormal is grouped into an abnormal data group, and data detected when the attachment was normal is grouped into a normal data group.

[0043] Next, the pre-processing unit (HBOS image generating unit) 316 converts the master data and the normal group data stored in the storage unit 226 into histograms (binarization). Then, the master histogram and one histogram of the normal group data are superimposed to generate an HBOS image as shown in FIG. 4(b) or FIG. 5. This operation is performed the number of times corresponding to the number of normal group data included in the normal group. Similarly, the abnormal data in the abnormal group stored in the storage unit 226 is converted into a histogram, and the master histogram is superimposed on one data in the abnormal group to generate an HBOS image as shown in FIG. 4(b) or FIG. 5. After the above processing is completed, the storage unit 226 may store only the HBOS image and the master histogram.

[0044] The above is the preparation before mounting the workpiece (work) on the processing device. After mounting the workpiece on the processing device, the sensor 210 detects the mounting state of the tool holder 110 on the spindle 120. The detected displacement signal (Sg) 10 is converted into a digital signal via the A / D converter 222 and transmitted to a data acquisition unit 322 provided in the data determination unit 320. The displacement signal 10 acquired by the data acquisition unit 322 is histogrammed in the pre-processing unit (HBOS image generation unit) 316 and superimposed on the histogram of the master to generate an HBOS image as shown in FIG. 4(b) or FIG. 5.

[0045] The learning model generation unit 330 learns the HBOS images created by the data preprocessing unit 310 in a "supervised manner with normal / abnormal labels". As a result of the learning, a binary classification machine learning model 332 is generated. This determines the criteria for determining abnormality / normality. The judgment unit 326 judges whether the HBOS image of the displacement signal 10 newly acquired this time is abnormal / normal through learning by referring to the generated binary classification machine learning model 332. The HBOS image for which a judgment has been made is added to the histogram of the normal mounting group or the abnormal mounting group.

[0046] In addition, after the HBOS image is generated, other conventional methods can be used to judge whether the image is normal or abnormal. For example, outlier detection methods such as the 3σ method and the quartile deviation method can be used.

[0047] As described above, according to the embodiment of the present invention, when detecting the mounting state of the tool holder on the spindle, it is possible to judge whether the mounting is abnormal or normal without considering the relative circumferential position of the tool holder and the spindle. Therefore, when the tool holder is replaced by the ATC depending on the machining in an unmanned NC machine tool or machining center, it is possible to reliably detect the mounting abnormality of the tool holder on the spindle without taking much time. As a result, the reliability of machining in the machine tool is improved, and the machining throughput is also improved. [Explanation of symbols]

[0048] 10...displacement signal, 12, 14...straight line (with constant output value), 16, 18...intersection, 20...hatched portion (common portion), 22, 24...difference portion, 100...machine tool, 102...chip, 110...tool holder, 110a...outer periphery, 110b...flange portion, 110c...notch, 110d...plane, 112...tool, 114, 116...fitting portion, 120...spindle, 150...control device, 200...mounting state detection device, 202...bracket, 210...(displacement ) Sensor, 220... arithmetic processing device, 222... A / D converter, 224... arithmetic processing unit, 226... memory unit, 228... input / output unit, 240... master (data), 250... displacement data, 310... data pre-processing unit, 312... data registration unit, 314... learning data generation unit, 316... pre-processing unit (HBOS image generation unit), 320... data judgment unit, 322... data acquisition unit, 326... judgment unit, 330... learning model generation unit, 332... binary classification machine learning model

Claims

1. A determination unit is provided for determining a mounting state of a tool holder mounted on a spindle, the determination unit is provided with a machine learning model that determines the mounting state based on a histogram that represents a frequency of occurrence of an output value, the histogram being generated from data composed of a circumferential position of the tool holder measured by a displacement sensor and an output value of the displacement sensor at the circumferential position.

2. The machine tool according to claim 1 , wherein the machine learning model is generated by supervised learning based on data in which a label of the mounting state is attached to a group of histograms acquired in advance.

3. Further, a data registration unit is provided, The machine tool according to claim 1 , wherein the data registration unit adds the histogram when the determination unit determines that the mounting is abnormal to the histogram group.

Images