Abnormality detection device and abnormality detection method of cutting tool

Abstract

To enable stable abnormality detection of a cutting tool regardless of a progress state of tool wear.SOLUTION: Vibration in components of a machine tool during machining is detected. A frequency spectrum of the vibration is calculated. Amplitude for all frequency components defined by a frequency corresponding to a spindle rotational number of the machine tool × n (1≤n≤tool blade number, n is an integral number) is extracted from the frequency spectrum. The presence or absence of blade edge chipping of a cutting tool is determined on the basis of a Mahalanobis distance calculated by a MT (Mahalanobis Taguchi) method by defining extracted amplitude data as a feature amount and a prescribed threshold value.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an abnormality detection device and an abnormality detection method for a cutting tool. [Background technology]

[0002] Patent Document 1 describes an anomaly detection device and method for a cutting tool in a machine tool equipped with a cutting tool having multiple blades, which detects the load on the spindle during cutting as a current value or power value of the spindle motor, calculates the frequency spectrum of the time-series data of the detected current value or power value, and determines whether the cutting tool is damaged based on the amplitude of a specified frequency component and a specified threshold value. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2020-104257 A Summary of the Invention [Problem to be solved by the invention]

[0004] The cutting tool anomaly detection device and anomaly detection method described in Patent Document 1 are particularly effective in detecting wear on multi-blade tools with a large number of blades. As wear progresses, the progression of wear on each blade is often not uniform, resulting in greater variation in the cutting load on each blade, and the frequency spectrum reflects periodic fluctuations caused by the variation in cutting load between each blade. For this reason, when a single variable such as the amplitude at a specified frequency is used to determine the presence or absence of cutting edge chipping, as in the cutting tool abnormality detection device and method described in Patent Document 1, it may become difficult to detect cutting tool abnormalities when cutting edge chipping occurs at different stages of tool wear progression. An object of the present invention is to provide an abnormality detection device and an abnormality detection method for a cutting tool that enable stable abnormality detection of the cutting tool regardless of the progress of tool wear. [Means for solving the problem]

[0005] The method for detecting an abnormality according to one embodiment of the present invention includes detecting vibrations in components of a machine tool during machining, calculating a frequency spectrum of the vibrations, and calculating a frequency x n (where The amplitudes are extracted for all frequency components defined within the range of 1≦n≦number of tool blades, n is an integer), and the presence or absence of chipping of the cutting tool blade edge is judged based on the Mahalanobis distance calculated by the MT (Mahalanobis-Taguchi) method using the extracted amplitude data as a feature quantity and a predetermined threshold value. Furthermore, as unit spaces (normal data groups) for calculating the Mahalanobis distance in the MT method, (1) an amplitude data group at the beginning of use of the cutting tool, (2) an amplitude data group for multiple cutting tools of the same type, for each tool for machining an arbitrary number of workpieces, and (3) an amplitude data group from the machining cycle immediately preceding the current machining cycle up to an arbitrary number of machining cycles back, for the machining cycles currently being executed, since it is impossible to judge when the reference data group is acquired using (1) and (3), tool damage is judged and detected using (2), and the absolute values ​​of the differences in the Mahalanobis distances between (1) and (2) and between (2) and (3) for the state of the tool before use are evaluated, thereby making it possible to detect abnormalities in the tool itself and evaluate the performance. Effect of the Invention

[0006] Therefore, in the present invention, it is possible to detect the occurrence of cutting tool tip fractures and the number of cutting tool tip fractures regardless of the progress of tool wear due to the cumulative number of pieces processed, and it is possible to suppress the occurrence of continuous machining defects due to deterioration of machining quality and machining accuracy as the cutting tool tip fractures increase, preventing the occurrence of a large number of defective products. It is also possible to detect cutting tool tip fractures when acquiring a reference data group in an inline process, preventing defective products from being overlooked in an inline process, and it is possible to detect whether the tool used is normal or abnormal, preventing the occurrence of a large number of defective products due to the use of abnormal tools. [Brief description of the drawings]

[0007] [Figure 1] 1 is an overall view of an abnormality detection device for a cutting tool according to a first embodiment. [Diagram 2] 1 is a flowchart 1 showing the flow of an anomaly detection method in the first embodiment. [Diagram 3] 2 is a flowchart 2 showing the flow of the anomaly detection method in the first embodiment. [Figure 4] FIG. 13 is a diagram showing the amplitude intensity distribution under normal conditions when the cutting tool of the first embodiment has machined 300 workpieces in total from the beginning of use. [Diagram 5] FIG. 5 is a diagram showing the amplitude intensity distribution when machining is performed using a cutting tool having 42 blades, one of which has an artificial defect on its cutting edge, that has been used to machine up to 300 workpieces as shown in FIG. 4 of the first embodiment. [Figure 6] FIG. 13 is a diagram showing the amplitude intensity distribution under normal conditions when the cutting tool of the first embodiment has been used to machine 1,500 workpieces in total from the beginning of use. [Figure 7] FIG. 7 is a diagram showing the amplitude intensity distribution when machining is performed using a cutting tool having 42 blades, one of which has an artificial defect in the tip, that has been used to machine up to 1,500 workpieces shown in FIG. 6 according to the first embodiment. [Figure 8] FIG. 13 is a diagram showing the amplitude intensity distribution when the cutting tool of the first embodiment has been used to machine 3,300 workpieces in total from the beginning of use. [Figure 9] FIG. 11 is a diagram showing the amplitude intensity distribution when machining is performed using a cutting tool having 42 blades, one of which has an artificial defect in the tip, that has been used to machine up to 3,300 workpieces shown in FIG. 8 according to the first embodiment. [Figure 10] 11 is a graph showing the relationship between the cumulative number of processes and the Mahalanobis distance in the first embodiment. [Figure 11] 11 is a graph showing the relationship between a first cumulative number of processes and a Mahalanobis distance in the first embodiment. [Figure 12] 13 is a graph showing the relationship between the second cumulative number of processes and the Mahalanobis distance in the first embodiment. [Figure 13] 13 is a graph showing the relationship between a third cumulative processing number and a Mahalanobis distance in the first embodiment. [Figure 14] 10 is a flowchart showing a flow of defect detection in real time when an amplitude data group at the beginning of use of the cutting tool of the first embodiment is set as a unit space (normal data group). [Figure 15] FIG. 11 is a flowchart showing the flow of defect detection in real time when amplitude data for machining 20 (any number) workpieces of each of multiple cutting tools of the same type in embodiment 1 is collected and the amplitude data group is set as a unit space (normal data group). [Figure 16] 10 is a flowchart showing a flow of defect detection in real time when a group of amplitude data from a machining cycle immediately before an arbitrary number of machining cycles back is set as a unit space for a machining cycle being executed in the first embodiment. [Figure 17] 11 is a graph showing the relationship between the cumulative number of machining operations and the Mahalanobis distance when a group of 100 amplitude data at the beginning of use of a cutting tool in an in-line process of the first embodiment is set as a unit space (normal data group). DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] [Embodiment 1] FIG. 1 is an overall view of a cutting tool abnormality detection device according to a first embodiment.

[0009] (Machine Tool Configuration) The machine tool 1 has a cutting tool 5 equipped with a plurality of blades 6, a spindle 9 driven by a drive motor (not shown) for axially moving and rotating the cutting tool 5, a spindle holding member 10 for holding the spindle 9, a work chuck 8 for gripping and synchronously rotating a workpiece 7 to be machined into an internal gear by skiving with the cutting tool 5, an NC control unit 2 for controlling the axial movement and rotation of the spindle 9 and the synchronous rotation of the work chuck 8, and an alarm unit 4.

[0010] (Configuration of anomaly detection device) The abnormality detection device 3 includes a vibration sensor 11 attached to a spindle holding member 10, a signal amplifier 12 that amplifies a vibration signal measured by the vibration sensor 11, a signal memory unit 13 that stores the vibration signal amplified by the signal amplifier 12, a frequency analysis unit 14 that calculates a frequency spectrum of the vibration signal stored in the signal memory unit 13, a multivariate analysis unit 15 that extracts amplitudes for all frequency components defined as a frequency corresponding to the rotation speed of the spindle 9 × n (where 1≦n≦the number of tool blades, n is an integer) from the frequency spectrum calculated by the frequency analysis unit 14 and calculates a Mahalanobis distance by the MT (Mahalanobis-Taguchi) method using the extracted amplitude data as a feature, a frequency amplitude data memory unit 16 that stores the extracted amplitude data for each machining cycle, and a judgment unit 17 that judges whether or not the cutting edge of the cutting tool is chipped by comparing the Mahalanobis distance calculated by the multivariate analysis unit 15 with a predetermined threshold value.

[0011] FIG. 2 is a flowchart 1 showing the flow of the anomaly detection method in the first embodiment, and FIG. 3 is a flowchart 2 showing the flow of the anomaly detection method in the first embodiment.

[0012] (Operation of the anomaly detection device) In step S1, when machining of the workpiece 7 is started by the machine tool 1, a machining start signal is generated from the NC control unit 2 as a trigger to start an abnormality detection process. In step S2, the vibration sensor 11 starts measuring the vibration caused by the cutting process, and the signal amplifier 12 amplifies the vibration signal measured by the vibration sensor 11, outputting the amplified vibration signal to the signal memory unit 13 (vibration detection step). In step S3, the signal storage unit 13 stores the input amplified vibration signal as time-series data up to a predetermined number of samples (signal storage step). In step S4, the frequency analysis unit 14 calculates the frequency spectrum of the amplified vibration signal from the time series data of the amplified vibration signal stored in the signal storage unit 13 (frequency analysis step). In step S5, the multivariate analysis unit 15 extracts amplitude data for all frequency components defined as the frequency corresponding to the spindle speed × n (where 1≦n≦number of tool blades, n is an integer) from the frequency spectrum calculated by the frequency analysis unit 14 (amplitude extraction step). In step S6, the extracted amplitude data is stored in the frequency amplitude data storage unit 16 for each machining cycle (amplitude data storage step). The amplitude data stored for each machining cycle is used to construct the unit space (normal population) of the MT (Mahalanobis-Taguchi) method. In step S7, the multivariate analysis unit 15 calculates the Mahalanobis distance by the MT (Mahalanobis-Taguchi) method using the amplitude data stored in the frequency amplitude data storage unit 16 as a feature amount (analysis step). In calculating the Mahalanobis distance, a group of amplitude data at the beginning of use of the cutting tool is used to set the unit space (normal group) of the MT (Mahalanobis-Taguchi) method. The reason for this is that when the cutting tool is first used, the cutting tool is in a brand new state immediately after replacement, and therefore a group of amplitude data in a normal cutting state can be obtained. In addition, in setting the unit space (normal data group) in embodiment 1, amplitude data for machining 20 workpieces (any number) of workpieces for each of multiple cutting tools of the same type other than the machining examples in each of Figures 11, 12, and 13 described later was collected and used as the amplitude data group. This makes it possible to average out the variation in the performance of the cutting tools by using a group of amplitude data across multiple cutting tools to set a unit space (group of normal data), and by setting a unit space (group of normal data) that serves as a common standard, it is possible to avoid the hassle of setting a unit space (group of normal data) each time a tool is changed. In addition, the sensitivity of detecting chipping of cutting edges can be improved by dynamically setting the unit space (normal data group) of the MT (Mahalanobis-Taguchi) method using the amplitude data group from the previous machining cycle to an arbitrary number of machining cycles back for the currently executing machining cycle. In step S8, the determination unit 17 compares the calculated Mahalanobis distance with a predetermined threshold value (comparison step). If the calculated Mahalanobis distance is greater than the predetermined threshold value, the process proceeds to step S9, and if the calculated Mahalanobis distance is less than the predetermined threshold value, the process proceeds to step S10, where the abnormality detection process ends. In step S9, the judgment unit 17 outputs an abnormality signal to the NC control unit 2 and the alarm unit 4, the NC control unit 2 stops the machine tool 1, the alarm unit 4 issues an alarm indicating that an abnormality has occurred, and the process proceeds to step S10, where the abnormality detection process is terminated (output step).

[0013] (Amplitude extraction step) FIG. 4 is a diagram showing the amplitude intensity distribution under normal conditions when the cutting tool of embodiment 1 has been used to machine up to a cumulative number of 300 workpieces from the start of use, and FIG. 5 is a diagram showing the amplitude intensity distribution when machining is performed using a cutting tool of embodiment 1 in FIG. 4 with one of its 42 blades artificially damaged. That is, Figures 4 and 5 show the amplitude intensity distribution extracted for all frequency components defined as the frequency corresponding to the rotation speed of the spindle 9 × n (where 1 ≦ n ≦ number of tool blades, n is an integer) from the frequency spectrum of vibration detected by the vibration sensor 11 when machining was performed using a 42-blade cutting tool under the machining conditions of a spindle rotation speed of 3676 min-1, a tool feed amount per spindle revolution of 0.1 mm / rev, and a machining allowance of 0.2 mm. Under the above conditions, the frequency when n = 1 is 61.3 Hz, and the maximum frequency is 2573 Hz when n = 42 (hereinafter, the frequency corresponding to the rotation speed of the spindle 9 x n is referred to as the nth frequency, and the amplitude of the nth frequency is referred to as the nth frequency component).

[0014] 4 and 5, the primary frequency component 22 after the cutting edge is chipped is significantly increased compared to the primary frequency component 21 before the cutting edge is chipped. In this case, by setting the threshold value 23 so that the primary frequency component 22 after the cutting edge is chipped exceeds it, it is possible to detect an abnormality in which one cutting edge is chipped.

[0015] FIG. 6 is a diagram showing the amplitude intensity distribution under normal conditions when the cutting tool of embodiment 1 has been used to machine 1,500 workpieces in total from the beginning of use, and FIG. 7 is a diagram showing the amplitude intensity distribution when machining is performed using the cutting tool of embodiment 1, FIG. 6 , in which an artificial defect has been generated in the tip of one of the 42 blades used to machine 1,500 workpieces. The processing conditions are the same as those in Figs.

[0016] 6 and 7, the 10th frequency component 25 after the cutting edge is chipped is significantly increased compared to the 10th frequency component 24 before the cutting edge is chipped. In this case, by setting a threshold value 26 so that the 10th frequency component 25 after the cutting edge is chipped exceeds, it is possible to detect an abnormality in which one cutting edge is chipped.The 10th frequency component 25 after the cutting edge is chipped is significantly increased compared to the 10th frequency component 24 before the cutting edge is chipped. In this case, by setting a threshold value 26 so that the 10th frequency component 25 after the cutting edge is chipped exceeds, it is possible to detect an abnormality in which one cutting edge is chipped.

[0017] FIG. 8 is a diagram showing the amplitude intensity distribution when the cutting tool of embodiment 1 has been used to machine 3,300 workpieces in total from the beginning of use, and FIG. 9 is a diagram showing the amplitude intensity distribution when machining is performed using the cutting tool of embodiment 1 in FIG. 7 with one of the 42 blades artificially damaged. The processing conditions are the same as those in Figs.

[0018] 8 and 9, the fourth-order frequency component 28 after the cutting edge is chipped is significantly increased compared to the fourth-order frequency component 27 before the cutting edge is chipped. In this case, by setting a threshold value 29 so that the fourth-order frequency component 28 after the cutting edge is chipped exceeds it, it is possible to detect an abnormality in which one cutting edge is chipped.

[0019] As described above, it is clear that it is possible to detect chipping of the cutting tool's cutting edge by comparing a specified frequency component with a threshold value for each cumulative machining count. However, since the frequency components suitable for determining an abnormality differ depending on the cumulative machining count, it is difficult to determine chipping of the cutting edge using a single variable such as the amplitude of a specified frequency component. The reason for this is that vibration during cutting is caused by fluctuations in the cutting load. With a cutting tool with multiple blades, the cutting load from each blade is repeated with each rotation of the tool when cutting the workpiece, generating vibration with a frequency component of the product of the spindle speed x the number of tool teeth. If there is no difference in cutting load between each blade, then in principle, vibration with a frequency lower than the product of the spindle speed x the number of tool teeth will not occur. On the other hand, if there are differences in cutting loads between the teeth, a repeating pattern of cutting load fluctuations based on the difference in cutting loads for each tooth occurs with each rotation of the tool, and vibrations with a frequency spectrum that reflects this repeating pattern of cutting load fluctuations are generated.For example, if the cutting load of one tooth is significantly larger than the other teeth, a large cutting load fluctuation occurs once with each rotation of the tool, and vibrations with a frequency equivalent to 1 x spindle rotation speed, i.e., a first-order frequency component, appear in the frequency spectrum. The causes of differences in cutting load between the blades are differences in the wear state of the blades and damage to the cutting edge. In the present invention, the change in the repeating pattern of the cutting load fluctuation before and after the blade edge damage is detected as a change in the frequency spectrum of the vibration during cutting, and abnormality detection is performed. The repetitive pattern of the cutting load fluctuation when the cutting edge is chipped is a pattern in which the cutting load fluctuation due to the chipped cutting edge is superimposed on the repetitive pattern of the cutting load fluctuation before the chipped cutting edge. Therefore, the change in the frequency spectrum of the vibration before and after the chipped cutting edge depends on the frequency spectrum of the vibration before the chipped cutting edge. In a cutting tool with multiple blades, if the wear state of each blade changes differently as the cumulative number of cuts increases, the repeating pattern of the cutting load fluctuation changes depending on the cumulative number of cuts. In other words, the frequency spectrum before the blade breakage differs depending on the cumulative number of cuts when the blade breakage occurs, so the change in the frequency spectrum of the vibration after the blade breakage depends on the cumulative number of cuts. Therefore, the frequency components suitable for determining an abnormality also change under the influence of the cumulative number of cuts.

[0020] For this reason, the anomaly detection method in one embodiment of the present invention further determines whether or not the cutting tool tip is chipped based on the Mahalanobis distance calculated by the MT (Mahalanobis-Taguchi) method using the extracted amplitude data as a feature and a predetermined threshold value.

[0021] (Analysis step) FIG. 10 is a graph showing the relationship between the cumulative number of processes and the Mahalanobis distance in the first embodiment. In other words, this graph shows the transition of the Mahalanobis distance against the cumulative number of machining operations when machining 300 cumulative pieces under the machining conditions (spindle speed 3676 min-1, tool feed per spindle revolution 0.1 mm / rev, machining allowance 0.1 mm), and then machining one workpiece with a cutting tool in which a defect was artificially created in the tip of one of the 42 blades.

[0022] The unit space (normal data group) S of the Mahalanobis-Taguchi (MT) method was set using the amplitude data of the machining cycles from 1 to 10 cycles before the current machining cycle. As the cumulative number of machining operations increases, the unit space (normal data group) S, which is the basis for calculating the Mahalanobis distance, also changes dynamically, so the Mahalanobis distance becomes smaller unless a sudden phenomenon such as cutting edge chipping occurs. For this reason, while the Mahalanobis distance of the unit space (normal data group) S before the cutting edge is chipped is approximately 1, the Mahalanobis distance 46 after the cutting edge is chipped is approximately 400. This shows that the detection sensitivity for cutting edge chipping is higher than the average difference of approximately 20 in the Mahalanobis distance between the case where there is no cutting edge chipping and the case where there is one cutting edge chipping, as shown in Figures 11, 12, and 13 described below.

[0023] FIG. 11 is a graph showing the relationship between the first cumulative number of processing operations and the Mahalanobis distance in the first embodiment. In other words, this graph shows the change in Mahalanobis distance versus the cumulative number of machining operations when a cutting tool, which has been used to machine up to 300 workpieces, artificially damages one of the cutting edges of its 42 blades, machined one workpiece with the tool, then continued to machine a total of 10 workpieces, and then machined another 10 workpieces with a blade edge other than the one that initially caused the damage, and so on, machining 10 workpieces in the same manner each time an additional blade edge damage was added, until a total of four blade edges were damaged.

[0024] The Mahalanobis distance in 10 workpiece machining operations with one chipped cutting edge forms a data group 30 of Mahalanobis distances having large values ​​compared to the unit space (normal data group) S of the Mahalanobis distance before the chipped cutting edge, that is, in machining up to a cumulative number of 300 pieces. For this reason, the chipped cutting edge in one place can be detected by setting the first threshold value 42. In addition, data group 31 of Mahalanobis distances when there are two missing edges is composed of Mahalanobis distances larger than data group 30 of Mahalanobis distances, and data group 32 of Mahalanobis distances when there are three missing edges and data group 33 of Mahalanobis distances when there are four missing edges similarly form groups composed of Mahalanobis distances whose size corresponds to an increase in the number of missing edges. Therefore, the number of cutting edge chips can be detected by setting a second threshold value 43 between the Mahalanobis distance data group 30 and the Mahalanobis distance data group 31, setting a third threshold value 44 between the Mahalanobis distance data group 31 and the Mahalanobis distance data group 32, and further setting a fourth threshold value 45 between the Mahalanobis distance data group 32 and the Mahalanobis distance data group 33.

[0025] FIG. 12 is a graph showing the relationship between the second cumulative number of processing operations and the Mahalanobis distance in the first embodiment. In other words, after machining a cumulative number of 1,500 workpieces, a total of 10 workpieces are machined continuously, and then the process is repeated to machine 10 workpieces each time a chip is added to the cutting edge, until a total of four chips are added to the cutting edge. This shows the change in Mahalanobis distance versus the cumulative number of workpieces machined.

[0026] Using the same first threshold value 42, second threshold value 43, third threshold value 44, and fourth threshold value 45 as in Figure 11, it is possible to distinguish between the unit space (normal data group) S of Mahalanobis distance in machining up to a cumulative machining number of 1,500 pieces with no cutting edge chipping, the data group 34 of Mahalanobis distance in a state where there is one cutting edge chipping, the data group 35 of Mahalanobis distance in a state where there are two cutting edge chippings, the data group 36 of Mahalanobis distance in a state where there are three cutting edge chippings, and the data group 37 of Mahalanobis distance in a state where there are four cutting edge chippings, and it is clear that it is possible to detect the occurrence of cutting edge chipping and the number of cutting edge chippings.

[0027] FIG. 13 is a graph showing the relationship between the third cumulative processing number and the Mahalanobis distance in the first embodiment. In other words, after machining a cumulative number of 3,300 pieces, a total of 10 workpieces are machined continuously, and then the process is repeated to machine 10 workpieces in the same manner each time a chip is added to the cutting edge, until a total of four chips are added to the cutting edge. This shows the change in Mahalanobis distance versus the cumulative number of pieces machined.

[0028] Using the first threshold value 42, second threshold value 43, third threshold value 44, and fourth threshold value 45, which are the same as those in Figure 10, it is possible to distinguish between the unit space (normal data group) S of Mahalanobis distance in machining up to a cumulative machining number of 3,300 pieces with no cutting edge chipping, the Mahalanobis distance data group 38 in a state where there is one cutting edge chipping, the Mahalanobis distance data group 39 in a state where there are two cutting edge chippings, the Mahalanobis distance data group 40 in a state where there are three cutting edge chippings, and the Mahalanobis distance data group 41 in a state where there are four cutting edge chippings, and it can be seen that it is similarly possible to detect the occurrence of cutting edge chipping and the number of cutting edge chippings.

[0029] Next, the effects of the first embodiment will be described.

[0030] (1) Vibrations in the components of the machine tool during machining are detected, the frequency spectrum of the vibrations is calculated, and from the frequency spectrum, the amplitudes are extracted for all frequency components defined as the frequency equivalent to the spindle speed of the machine tool x n (where 1≦n≦number of tool teeth, n is an integer). The extracted amplitude data is used as a feature to calculate the Mahalanobis distance using the MT (Mahalanobis-Taguchi) method, and the presence or absence of chipping on the cutting tool tip is determined based on a specified threshold value and the Mahalanobis distance. Therefore, it becomes possible to detect the occurrence of chipping of the cutting tool's cutting edge and the number of chipped edges regardless of the progress of tool wear due to the cumulative number of processed items. This makes it possible to suppress the occurrence of continuous machining defects due to deterioration of machining quality and machining accuracy as the chipping of the cutting edge increases, and prevents the production of a large number of defective products.

[0031] (2) In calculating the Mahalanobis distance, the amplitude data set at the beginning of use of the cutting tool was used to set the unit space (normal data set) of the MT (Mahalanobis-Taguchi) method. Therefore, when the cutting tool is first used, the cutting tool is in a brand new state immediately after replacement, so a group of amplitude data in a normal cutting state can be obtained.

[0032] (3) In calculating the Mahalanobis distance, amplitude data for 20 (any number) workpieces processed by each of the same type of cutting tools was collected and used as an amplitude data group to set the unit space (normal data group) of the MT (Mahalanobis-Taguchi) method. Therefore, by using a group of amplitude data across multiple cutting tools to set a unit space (a group of normal data), the variation in the performance of the cutting tools can be averaged out, and by setting a common standard unit space, the complication of setting a unit space each time a tool is changed can be avoided.

[0033] (4) In calculating the Mahalanobis distance, the unit space (normal data group) of the MT (Mahalanobis-Taguchi) method is dynamically set using the amplitude data group from the previous machining cycle to an arbitrary number of machining cycles going back for the currently executing machining cycle. Therefore, the detection sensitivity for chipping of the cutting edge can be improved.

[0034] Next, the case where the first embodiment is applied to an in-line process will be described.

[0035] FIG. 14 is a flow chart showing the flow up to the determination of chipping of the cutting edge in an in-line process in the method (2) for setting the unit space (normal data group). Anomaly detection is started in step S21, and line operation is started in step S22. After the tool change is performed in step S23, an arbitrary number of amplitude data groups are acquired in step S24. In step S25, a unit space (normal data group) is set from the acquired amplitude data group, in step S26, the Mahalanobis distance is calculated from the feature amount of the next acquired amplitude data, and in step S27, it is determined whether the tool is damaged. Steps S26 and S27 are repeated while the line is in operation.

[0036] FIG. 15 is a flow chart showing the flow up to the determination of chipping of the cutting edge in an in-line process in the method (3) for setting the unit space (normal data group) described above. In step S31, for a plurality of cutting tools of the same type, amplitude data for machining 20 workpieces (any number) of the tools is collected. Anomaly detection is started in step S32, a unit space (normal data group) is set in step S33 using the amplitude data group collected in step S31, and then line operation is started in step S . In step S35, the Mahalanobis distance is calculated from the feature amount of the acquired amplitude data, and in step S36, it is determined whether the tool is broken. Steps S35 and S36 are repeated while the line is in operation.

[0037] FIG. 16 is a flow chart showing the flow up to the determination of chipping of the cutting edge in an in-line process in the method (4) for setting the unit space (normal data group). Anomaly detection is started in step S41, and line operation is started in step S42. After acquiring an arbitrary number of amplitude data groups in step S43, a unit space (normal data group) is set from the acquired amplitude data groups in step S44, the Mahalanobis distance is calculated from the features of the next acquired amplitude data in step S45, and it is determined whether the tool is damaged or not in step S46. In step S47, the oldest data is deleted from the unit space (normal data group), and newly determined amplitude data is added, and the process returns to step S44. While the line is in operation, steps S44 to S47 are repeated.

[0038] FIG. 17 is a graph showing the relationship between the cumulative number of processing operations and the Mahalanobis distance when the method for setting the unit space (normal data group) described in (2) above is used in an in-line process of the first embodiment. When detecting chipped cutting edges in an in-line process, the methods of setting the unit space (normal data group) described above in (2) and (4) have the problem that it is impossible to detect or determine whether chipped cutting edges are present in the data acquired in steps S24 and S43, and therefore it is not possible to determine the total number. Therefore, in steps S24 and S43, the determination is made using the method of setting the unit space (normal data group) described in (3) above, thereby making it possible to perform a total determination in an inline process.

[0039] In addition, the method (2) has the advantage that it is possible to capture changes in amplitude data for the same cutting tool, and the method (4) has the advantage that it is highly sensitive to changes in amplitude data, but it has a disadvantage that it is not possible to compare with other cutting tools. Therefore, there is a problem in that it is not possible to detect abnormalities in the tool itself or the workmanship, such as chipped cutting edges, before the tool is replaced.

[0040] To address the above-mentioned issues, an evaluation is made of tool abnormalities and workmanship by taking the difference between the Mahalanobis distance calculated in step S26 of (2) and step S45 of (4) above and the Mahalanobis distance calculated in step S35 of (3) above.

[0041] Regarding the method of calculating the difference in Mahalanobis distance when evaluating tool abnormalities and workmanship, the difference in cutting edge condition for each amplitude data can be grasped by comparing the absolute value of the difference between the Mahalanobis distance in the setting method of the unit space (normal data group) in (3) above and the Mahalanobis distance in the setting method of the unit space (normal data group) in (2) or (4) above. Alternatively, by comparing the absolute value of the difference between the average of an arbitrary number of Mahalanobis distances in the setting method of the unit space (normal data group) in (3) above and the average of an arbitrary number of Mahalanobis distances in the setting method of the unit space (normal data group) in (2) or (4) above, the rough difference in cutting edge condition can be grasped without being bound by the size of small differences.

[0042] Other Embodiments The above describes an embodiment for carrying out the present invention, but the specific configuration of the present invention is not limited to the configuration of the embodiment, and design changes and the like that do not deviate from the gist of the invention are also included in the present invention. [Explanation of symbols]

[0043] 1 machine tool, 2 NC control unit, 3 abnormality detection device, 4 alarm unit, 5 cutting tool, 11 vibration sensor, 12 signal amplifier, 13 signal storage unit, 14 frequency analysis unit, 15 multivariate analysis unit, 16 frequency amplitude data storage unit, 17 judgment unit, 41 first threshold value (predetermined threshold value), 42 second threshold value (predetermined threshold value), 43 third threshold value (predetermined threshold value), 44 fourth threshold value (predetermined threshold value), 50 detection and judgment impossible area

Claims

1. An anomaly detection method executed by an anomaly detection device, comprising: The anomaly detection device, A vibration detection step of detecting vibration during machining of a component of the machine tool in cutting processing using a cutting tool having a plurality of blades; a signal storage step of recording the detected vibration signal as time-series data up to a predetermined number of samples; a frequency analysis step of calculating a frequency spectrum of the time series data of the vibration signal recorded in the signal storage step; an amplitude extraction step of extracting an amplitude for a frequency component defined as a frequency corresponding to a spindle speed of the machine tool×n (where 1≦n≦number of tool teeth, n is an integer) from the frequency spectrum; an amplitude data storage step of storing the amplitude data extracted in the amplitude extraction step for each machining cycle; An analysis step of calculating a Mahalanobis distance by multivariate analysis using the extracted amplitude data as a feature value by a Mahalanobis-Taguchi (MT) method, In the calculation of the Mahalanobis distance, a first unit space that is set by a data group in which an arbitrary number of amplitude data of normal cutting tools of the same type as the cutting tool are selected in advance; a first Mahalanobis distance calculated using the amplitude data of the cutting tool with respect to the first unit space as a feature amount; a second Mahalanobis distance calculated from a second unit space set by an arbitrary number of amplitude data of the cutting tool; a comparison step of comparing a difference between the first Mahalanobis distance and the second Mahalanobis distance with a predetermined threshold value; an output step of outputting an abnormality signal indicating that the cutting tool has an initial defect when the difference exceeds the predetermined threshold value; A method for detecting an abnormality in a cutting tool, comprising:

2. The method for detecting an abnormality in a cutting tool according to claim 1, the difference is a difference between an average of the plurality of first Mahalanobis distances and an average of the plurality of second Mahalanobis distances; A method for detecting an abnormality in a cutting tool, comprising:

3. The method for detecting an abnormality in a cutting tool according to claim 1, the difference is a difference between one of the plurality of first Mahalanobis distances and one of the plurality of second Mahalanobis distances corresponding to the first Mahalanobis distance; A method for detecting an abnormality in a cutting tool, comprising:

4. The method for detecting an abnormality in a cutting tool according to claim 1, If the difference is equal to or smaller than the predetermined threshold value, the cutting tool is determined to be normal, the calculation of the first Mahalanobis distance is continued, and occurrence of a defect during use of the cutting tool is detected based on whether or not the first Mahalanobis distance exceeds a predetermined value. A method for detecting an abnormality in a cutting tool, comprising:

5. The method for detecting an abnormality in a cutting tool according to claim 1, When the difference is equal to or smaller than the predetermined threshold value, the cutting tool is determined to be normal, the calculation of the second Mahalanobis distance is continued, and occurrence of a defect during use of the cutting tool is detected based on whether or not the second Mahalanobis distance exceeds a predetermined value. A method for detecting an abnormality in a cutting tool, comprising:

6. An abnormality detection device for detecting a defect in a cutting tool having a plurality of blades, a vibration sensor that detects vibration of the cutting tool during machining; a signal amplifier that amplifies the vibration signal detected by the vibration sensor; a signal storage unit that stores the vibration signal amplified by the signal amplifier; a frequency analysis unit that calculates a frequency spectrum of the vibration signal stored in the signal storage unit; a multivariate analysis unit which extracts amplitude data for all frequency components defined by a frequency corresponding to a spindle rotation speed of the cutting tool×n (where 1≦n≦number of tool teeth, n is an integer) from the frequency spectrum calculated by the frequency analysis unit, and calculates a Mahalanobis distance by an MT (Mahalanobis-Taguchi) method using the extracted amplitude data as a feature value, The multivariate analysis unit a first unit space that is set in advance by amplitude data of an arbitrary number of other cutting tools of the same type as the cutting tool; A first Mahalanobis distance is calculated using the amplitude data of the cutting tool with respect to the first unit space as a feature amount; A second Mahalanobis distance is calculated from a second unit space set by an arbitrary number of amplitude data of the cutting tool; The determination unit is a determination unit that compares a difference between the first Mahalanobis distance and the second Mahalanobis distance with a predetermined threshold value, and determines that the cutting tool has an initial defect when the difference exceeds the predetermined threshold value; An abnormality detection device for a cutting tool having the above structure.

Images