Abnormality detection system, abnormality detection device, abnormality detection method, and computer program

Abstract

An abnormality detection system that detects an abnormality of a tool attached to a machine tool, the abnormality detection system including: a deformation sensor that measures a deformation generated by the tool during a period in which the machine tool processes an object to be processed by the tool; and an abnormality detection device that detects an abnormality of the tool on the basis of measurement data from the deformation sensor, the abnormality detection device comprising: a classification unit that is a machine-learned model that takes object data based on the measurement data from the deformation sensor as an input and takes the result of classification of the object data as an output; an abnormality detection unit that detects an abnormality of the tool on the basis of a result of classification of the object data by the classification unit; and an output unit that outputs a detection result obtained by the abnormality detection unit.

Description

Technical Field

[0001] The present invention relates to an anomaly detection system, an anomaly detection device, an anomaly detection method and a computer program. Background Art

[0002] Patent document 1 discloses an abnormality detection device for detecting abnormalities of a tool installed on a machine tool. The abnormality detection device disclosed in Patent document 1 uses a type of SVM method to learn the various measurement data of the tool's vibration information, cutting force information, sound information, spindle load, motor current, and power value, and creates a normal model. After the normal model is created, the measurement data is acquired during processing, and based on the normal model, the measurement data is diagnosed as normal or abnormal. The abnormality detection device also re-diagnoses the measurement data diagnosed as abnormal using a method different from the type of SVM method such as invariant analysis.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Publication No. 2018-024055 Summary of the invention

[0006] An abnormality detection system involved in one embodiment of the present invention is an abnormality detection system for detecting abnormalities of a tool installed on a machine tool, wherein the abnormality detection system comprises: a deformation sensor for measuring the deformation generated in the tool while the machine tool processes a processing object through the tool; and an abnormality detection device for detecting abnormalities of the tool based on measurement data of the deformation sensor, the abnormality detection device comprising: a machine learning completed model, i.e., a classification unit, which takes object data based on the measurement data of the deformation sensor as input and takes classification results of the object data as output; an abnormality detection unit, which detects abnormalities of the tool based on the classification results of the object data obtained by the classification unit; and an output unit, which outputs the detection results obtained by the abnormality detection unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a diagram showing an example of the overall configuration of the abnormality detection system according to the embodiment.

[0008] Figure 2A This is a diagram showing an example of the structure of a cutting tool according to an embodiment.

[0009] Figure 2B This is a diagram showing another example of the structure of the cutting tool according to the embodiment.

[0010] Figure 2CThis is a diagram showing another example of the structure of the cutting tool according to the embodiment.

[0011] Figure 2D This is a diagram showing another example of the structure of the cutting tool according to the embodiment.

[0012] Figure 3 This is a diagram showing an example of the structure of the sensor module according to the embodiment.

[0013] Figure 4 This is a block diagram showing an example of the hardware configuration of the abnormality detection device according to the embodiment.

[0014] Figure 5 A functional block diagram showing an example of the function of the abnormality detection device according to the embodiment.

[0015] Figure 6 This is a graph showing an example of a time-series waveform of the deformation of the cutting tool.

[0016] Figure 7 This is a diagram showing an example of a processing object used in an intermittent processing step.

[0017] Fig. 8A This is a graph showing an example of a time-series waveform of the deformation of the cutting tool in the intermittent machining process.

[0018] Figure 8B This is a graph showing another example of the time series waveform of the deformation of the cutting tool in the intermittent machining process.

[0019] Fig. 9 This is a diagram for explaining an example of the learning period of a one-class classifier.

[0020] Fig.10 It is a diagram schematically showing an overview of local outlier factors.

[0021] Fig.11 This is a diagram showing an example of an abnormality notification screen.

[0022] Fig.12 This is a diagram showing an example of an abnormality prediction screen.

[0023] Fig.13 This is a flowchart showing an example of an abnormality detection process by the abnormality detection device according to the embodiment.

[0024] Fig.14 This is a flowchart showing an example of classification processing of object data.

[0025] Fig.15 This is a diagram showing an example of the learning period of a multi-class classifier. DETAILED DESCRIPTION

[0026] <Problems to be Solved by the Invention>

[0027] The abnormality detection device disclosed in Patent Document 1 uses a vibration sensor that measures vibration of a tool. However, since the vibration sensor measures vibration of a spindle that is far from the tool, it is difficult to accurately obtain vibration information useful for detecting abnormality of the tool.

[0028] <Effects of the present invention>

[0029] According to the present invention, it is possible to detect abnormality of a tool using information that accurately reflects the state of a tool mounted on a machine tool.

[0030] <Overview of Embodiments of the Invention>

[0031] Hereinafter, the outline of the embodiments of the present invention will be listed and described.

[0032] (1) The abnormality detection system involved in this embodiment is an abnormality detection system for detecting abnormalities of a tool installed in a machine tool, the abnormality detection system comprising: a deformation sensor for measuring deformation of the tool while the machine tool processes a processing object through the tool; and an abnormality detection device for detecting abnormalities of the tool based on measurement data of the deformation sensor, the abnormality detection device comprising: a machine learning completed model, i.e., a classification unit, which takes object data based on the measurement data of the deformation sensor as input and takes classification results of the object data as output; an abnormality detection unit, which detects abnormalities of the tool based on the classification results of the object data obtained by the classification unit; and an output unit, which outputs the detection results obtained by the abnormality detection unit. Thus, it is possible to detect abnormalities of the tool using information that accurately reflects the state of the tool, i.e., the deformation of the tool.

[0033] (2) In the above (1), the strain sensor may be mounted on the tool. Thus, the strain of the tool can be directly measured by the strain sensor.

[0034] (3) In the above (2), the tool is a cutting tool including a turning tool or a milling tool, and the strain sensor and the wireless communication unit for wirelessly transmitting the measurement data of the strain sensor can be installed on the cutting tool. Thus, the strain sensor can be installed on a turning tool in which the tool itself does not rotate, or a milling tool in which the tool itself rotates, and the measurement data can be transmitted wirelessly.

[0035] (4) In any one of the above (1) to (3), the classifier is a one-class classifier that performs one-class classification, and the abnormality detection unit can detect the abnormality of the tool based on the abnormality degree generated in the one-class classification. Thus, the one-class classifier constructed by unsupervised learning can be used to detect the abnormality of the tool.

[0036] (5) In the above (4), the abnormality detection unit may perform multi-class classification of the object data by comparing the abnormality degree with a plurality of threshold values, and detect the abnormality of the tool based on the result of the multi-class classification. Thus, the abnormality of the tool can be accurately detected using the result of the multi-class classification of the object data.

[0037] (6) In the above (5), the abnormality detection unit may classify the target data into a normal state, a state close to abnormality, and an abnormal state by comparing the abnormality degree with a first threshold value and a second threshold value. Thus, the target data can be sub-classified using the abnormality degree.

[0038] (7) In the above (6), when the object data is classified as an abnormal state by the abnormality detection unit, the output unit may output information for notifying that the tool is abnormal. Thus, the user can be notified of the abnormality of the tool.

[0039] (8) In the above (6), when the object data is classified by the abnormality detection unit as being close to an abnormal state, the output unit may output information for notifying that an abnormality of the tool is predicted to occur. Thus, the user can be notified that an abnormality of the tool is predicted to occur.

[0040] (9) In any one of the above (6) to (8), the first threshold value for distinguishing the object data between a normal state and a state close to an abnormality may be determined based on the object data based on the measurement data measured by the deformation sensor of the tool close to an abnormality. Thus, it is possible to accurately distinguish whether the tool is in a normal state or in a state close to an abnormality by using the first threshold value determined based on the actually measured data.

[0041] (10) In any one of the above (1) to (9), the abnormality detection device can detect the abnormality of the tool based on the measurement data of the deformation of the tool in the intermittent processing step in which the cutting and non-cutting of the processing object are repeated. In the intermittent processing step, the state of the tool is more likely to be expressed as deformation than in the continuous processing step in which the cutting of the processing object is continuously performed. Therefore, the abnormality of the tool can be accurately detected.

[0042] (11) The abnormality detection device according to the present embodiment is an abnormality detection device for detecting abnormalities of a tool installed in a machine tool, and the abnormality detection device comprises: an acquisition unit for acquiring object data based on measurement data output from a deformation sensor for measuring deformation generated in the tool while the machine tool processes a processing object through the tool; a classification unit, which is a machine learning model, taking the object data acquired by the acquisition unit as input and taking the classification result of the object data as output; an abnormality detection unit for detecting abnormalities of the tool based on the classification result of the object data obtained by the classification unit; and an output unit for outputting the detection result obtained by the abnormality detection unit. Thus, it is possible to detect abnormalities of the tool using information accurately reflecting the state of the tool, namely, the deformation of the tool.

[0043] (12) The abnormality detection method involved in this embodiment is an abnormality detection method performed by an abnormality detection device for detecting an abnormality of a tool installed on a machine tool, and the abnormality detection method includes the following steps: while the machine tool processes a processing object through the tool, acquiring object data based on measurement data output from a deformation sensor that measures deformation generated in the tool; classifying the object data by a machine learning completed model, i.e., a classification unit, which takes the acquired object data as input and outputs a classification result of the object data; detecting an abnormality of the tool based on the classification result of the object data obtained by the classification unit; and outputting the detection result of the abnormality of the tool. Thus, it is possible to detect an abnormality of the tool using information that accurately reflects the state of the tool, i.e., the deformation of the tool.

[0044] (13) The computer program involved in this embodiment is a computer program for detecting an abnormality of a tool installed in a machine tool, and the computer program causes a computer to execute the following steps: while the machine tool processes a processing object using the tool, object data based on measurement data output from a deformation sensor that measures deformation generated in the tool is acquired; the object data is classified using a machine learning model, i.e., a classification unit, which takes the acquired object data as input and outputs a classification result of the object data; an abnormality of the tool is detected based on the classification result of the object data obtained by the classification unit; and the detection result of the abnormality of the tool is output. Thus, an abnormality of the tool can be detected using information that accurately reflects the state of the tool, i.e., the deformation of the tool.

[0045] <Details of Embodiments of the Invention>

[0046] Hereinafter, details of the embodiments of the present invention will be described with reference to the drawings. In addition, at least a part of the embodiments described below may be arbitrarily combined.

[0047] [1. Anomaly detection system]

[0048] Figure 1 1 is a diagram showing an example of the overall structure of the abnormality detection system involved in the present embodiment. The abnormality detection system 10 detects an abnormality of a cutting tool 30 of a machine tool 20. The cutting tool 30 is mounted on the machine tool 20. The machine tool 20 performs cutting processing on a processing object using the cutting tool 30. The "cutting processing" mentioned here includes "turning processing" in which a rotating processing object is cut by the cutting tool 30, and "milling processing" in which a fixed processing object is cut by the rotating cutting tool 30. The machine tool 20 can be a turning machine such as a lathe, or a milling machine such as a milling machine.

[0049] The abnormality detection system 10 includes a cutting tool 30, a wireless device 200, and an abnormality detection apparatus 300. The wireless device 200 is connected to the abnormality detection apparatus 300, for example, by wire. The wireless device 200 is, for example, an access point.

[0050] The cutting tool 30 includes a sensor module 100. As described later, the sensor module 100 includes a sensor.

[0051] Furthermore, the abnormality detection system 10 is not limited to a configuration including one cutting tool 30 , and may include a plurality of cutting tools 30 .

[0052] The cutting tool 30 transmits the measurement results of the sensors in the sensor module 100 to the abnormality detection device 300 in a time series.

[0053] More specifically, the cutting tool 30 wirelessly transmits a wireless signal including a packet storing the measured value to the wireless device 200 .

[0054] The wireless device 200 acquires a packet included in the wireless signal received from the cutting tool 30 , and relays the packet to the abnormality detection apparatus 300 .

[0055] When receiving the sensor packet from the cutting tool 30 via the wireless device 200 , the abnormality detection device 300 acquires measurement information from the received sensor packet and processes the acquired measurement information.

[0056] The cutting tool 30 and the wireless device 200 perform wireless communication based on communication protocols such as ZigBee (registered trademark) in accordance with IEEE 802.15.4, Bluetooth (registered trademark) in accordance with IEEE 802.15.1, and UWB (Ultra Wide Band) in accordance with IEEE802.15.3a. In addition, communication protocols other than the above can also be used between the cutting tool 30 and the wireless device 200.

[0057] [2. Specific examples of cutting tools]

[0058] Figure 2A This is a diagram showing an example of the structure of the cutting tool according to the present embodiment.

[0059] A turning tool 30A, which is an example of the cutting tool 30, is a turning tool for machining a rotating object, and is mounted on a machine tool such as a lathe. The turning tool 30A includes a cutting portion 31A and a sensor module 100 provided on the cutting portion 31A.

[0060] For example, the cutting part 31A can be mounted with a cutting insert 32 having a cutting edge. Specifically, the cutting part 31A is a shank that holds the cutting insert 32. That is, the turning tool 30A is a so-called disposable turning tool.

[0061] More specifically, the cutting portion 31A includes a fixing member 33A and a fixing member 33B. The fixing members 33A and 33B hold the cutting insert 32 .

[0062] The cutting insert 32 is, for example, a polygonal shape such as a triangle, a square, a rhombus, or a pentagon in a plan view. The cutting insert 32 has, for example, a through hole formed in the center of the upper surface and is fixed to the cutting portion 31A by fixing members 33A and 33B.

[0063] Figure 2B This is a diagram showing another example of the structure of the cutting tool according to the present embodiment.

[0064] A turning tool 30B, which is an example of the cutting tool 30, is a tool for turning and is mounted on a machine tool such as a lathe. The turning tool 30B includes a cutting portion 31B and a sensor module 100 provided on the cutting portion 31B.

[0065] For example, the cutting portion 31B has a cutting edge 34. That is, the turning tool 30B is a peeling-type turning tool or a brazing-type turning tool.

[0066] Figure 2C This is a diagram showing another example of the structure of the cutting tool according to the present embodiment. Figure 2C A cross-sectional view showing a cutting tool.

[0067] A milling tool 30C, which is an example of the cutting tool 30, is a milling tool used for machining a fixed object, and is mounted on a machine tool such as a milling machine. The milling tool 30C includes a cutting portion 31C and a sensor module 100 provided on the cutting portion 31C.

[0068] For example, the cutting part 31C can be mounted with a cutting insert 32 having a cutting edge. Specifically, the cutting part 31C is a holder that holds the cutting insert 32. That is, the milling tool 30C is a so-called milling cutter.

[0069] More specifically, the cutting section 31C includes a plurality of fixing members 33C. The fixing members 33C hold the cutting inserts 32 .

[0070] The cutting insert 32 is fixed to the cutting portion 31C via a fixing member 33C.

[0071] Figure 2D This is a diagram showing another example of the structure of the cutting tool according to the present embodiment.

[0072] A milling tool 30D, which is an example of the cutting tool 30, is a tool for milling processing and is mounted on a machine tool such as a milling machine. The milling tool 30D includes a cutting portion 31D and a sensor module 100 provided on the cutting portion 31D.

[0073] For example, the cutting portion 31D has a cutting edge 35. That is, the milling tool 30D is an end mill.

[0074] [3.Sensor module]

[0075] Figure 3 This is a diagram showing an example of the structure of the sensor module according to the present embodiment.

[0076] The sensor module 100 includes a processor 101 , a nonvolatile memory 102 , a volatile memory 103 , a communication interface (I / F) 104 , a strain sensor 110A, and a strain sensor 110B.

[0077] The volatile memory 103 is, for example, a volatile memory such as SRAM (Static Random Access Memory) or DRAM (Dynamic Random Access Memory). The nonvolatile memory 102 is, for example, a nonvolatile memory such as a flash memory or ROM (Read Only Memory). The nonvolatile memory 102 stores, for example, a computer program (not shown) and data used in the execution of the computer program. The computer program is a program for transmitting the measurement values ​​of the strain sensor 110A and the strain sensor 110B in a time series.

[0078] The processor 101 is, for example, a CPU (Central Processing Unit). However, the processor 101 is not limited to a CPU. The processor 101 may also be a GPU (Graphics Processing Unit). In a specific example, the processor 101 is a multi-core GPU. The processor 101 may be, for example, an ASIC (Application Specific Integrated Circuit), or a programmable logic device such as a gate array or an FPGA (Field Programmable Gate Array).

[0079] The communication I / F 104 is a communication interface for a communication protocol such as ZigBee in accordance with IEEE 802.15.4, Bluetooth in accordance with IEEE 802.15.1, and UWB in accordance with IEEE802.15.3a. The communication I / F 104 is implemented by a communication circuit such as a communication IC (Integrated Circuit).

[0080] The sensor module 100 includes a battery 105. The battery 105 supplies power to the processor 101, the nonvolatile memory 102, the volatile memory 103, the communication I / F 104, and the strain sensor 110A and the strain sensor 110B.

[0081] The deformation sensor 110A and the deformation sensor 110B are, for example, provided near the cutting edge of the cutting tool 30. The deformation sensor 110A and the deformation sensor 110B are mounted on the cutting tool 30 in such a manner as to measure deformations in different directions from each other. For example, the deformation sensor 110A measures the deformation in the length direction of the cutting tool 30, and the deformation sensor 110B measures the deformation in the width direction of the cutting tool 30. Hereinafter, the deformation sensor 110A and the deformation sensor 110B are also collectively referred to as the deformation sensor 110. The deformation sensor 110 is an example of a sensor. The sensor is driven by the power supplied from the battery 105.

[0082] In addition, the sensor module 100 is not limited to a structure including two strain sensors 110, and may include one or more than three strain sensors 110. In addition, instead of or in addition to the strain sensor 110, the sensor module 100 may also include other sensors such as an acceleration sensor, a pressure sensor, a sound sensor, and a temperature sensor.

[0083] [4. Hardware structure of abnormality detection device]

[0084] Figure 4This is a block diagram showing an example of the hardware configuration of the abnormality detection device according to the present embodiment.

[0085] The abnormality detection device 300 includes a processor 301 , a non-volatile memory 302 , a volatile memory 303 , an input / output interface (I / O) 304 , a graphic controller 305 , and a display device 306 .

[0086] The volatile memory 303 is, for example, a volatile memory such as SRAM or DRAM. The nonvolatile memory 302 is, for example, a nonvolatile memory such as a flash memory, a hard disk, or a ROM. The nonvolatile memory 302 stores the abnormality detection program 307, which is a computer program, and data used in the execution of the abnormality detection program 307. Each function of the abnormality detection device 300 is realized by the processor 301 executing the abnormality detection program 307. The abnormality detection program 307 can be stored in a recording medium such as a flash memory, a ROM, or a CD-ROM.

[0087] A measurement result database (DB) 309 is provided in the nonvolatile memory 302. The measurement result DB 309 stores the measurement results obtained by the strain sensor 110. More specifically, in the measurement result DB 309, the measurement values ​​of the strain output from the strain sensor 110 are stored in a time series at every constant sampling period.

[0088] The nonvolatile memory 302 stores a classifier 310. The classifier 310 is a machine-learned model and is an example of a classifier. The classifier 310 will be described later.

[0089] The processor 301 is, for example, a CPU. The processor 301 may also be one or more CPUs. However, the processor 301 is not limited to a CPU. The processor 301 may also be a GPU. In a specific example, the processor 301 is a multi-core GPU. The processor 301 may be, for example, a programmable logic device such as an ASIC, a gate array, or an FPGA. In this case, the ASIC or the programmable logic device is configured to be able to perform the same processing as the anomaly detection program 307.

[0090] The I / O 304 is connected to the wireless device 200. The I / O 304 is, for example, a communication I / F, and is capable of communicating based on a specific communication protocol. The I / O 304 includes, for example, an Ethernet interface (“Ethernet” is a registered trademark). The I / O 304 is capable of receiving the measurement result of the deformation sensor 110 from the cutting tool 30 via the wireless device 200.

[0091] The graphics controller 305 is connected to the display device 306 and controls the display on the display device 306. The graphics controller 305 includes, for example, a GPU and a VRAM (Video RAM), stores data displayed on the display device 306 in the VRAM, periodically reads out one frame of video data from the VRAM, and generates a video signal. The generated video signal is output to the display device 306, and the video is displayed on the display device 306. The function of the graphics controller 305 may be included in the processor 301. A part of the area of ​​the volatile memory 303 may also be used as the VRAM.

[0092] The display device 306 includes, for example, a liquid crystal panel or an OEL (organic electroluminescence) panel. The display device 306 can display text or graphic information.

[0093] [5. Function of abnormality detection device]

[0094] Figure 5 3 is a functional block diagram showing an example of the function of the abnormality detection device according to the embodiment. The abnormality detection device 300 has the functions of a classifier 310 , an input unit 311 , an abnormality detection unit 313 , and an output unit 314 .

[0095] While the machine tool 20 processes the object to be processed by the cutting tool 30, the input unit 311 acquires measurement data of strain generated in the cutting tool 30. The input unit 311 receives the measurement data of strain from the measurement result DB 309. The measurement data is data of the measurement value of the strain sensor 110, for example, time series data of the measurement value of strain.

[0096] The classifier 310 is a machine-learned model that receives object data based on measurement data of deformation of the cutting tool 30 while the machine tool 20 is machining an object to be machined with the cutting tool 30 , and outputs a classification result of the object data.

[0097] Figure 6 This is a graph showing an example of a time series waveform of the deformation of a cutting tool. Figure 6 In the figure, the vertical axis represents deformation and the horizontal axis represents time. The cutting process of the object to be processed includes multiple processes with different processing conditions. For example, the turning process based on a lathe for one object to be processed includes multiple outer diameter processing processes and multiple inner diameter processing processes. In a series of cutting processes, the waveform pattern of the measured value of the deformation of the cutting tool 30 changes in each process. Figure 6In the example of , the cutting process of one object includes process A, process B, process C, process D, process E, and process F. Process A is performed in period P1, process B is performed in period P2, process C is performed in period P3, process D is performed in period P4, process E is performed in period P5, and process F is performed in period P6. The processing conditions of process A, process B, process C, process D, process E, and process F are different from each other. Therefore, the waveform patterns of deformation in process A, process B, process C, process D, process E, and process F are different from each other.

[0098] The abnormality detection device 300 analyzes the measurement data of a process (hereinafter also referred to as "target process") selected from the above-mentioned multiple processes, and detects abnormality of the cutting tool 30. In a specific example, the target process is an intermittent processing process that repeats cutting and non-cutting of the processing object.

[0099] Figure 7 1 is a diagram showing an example of a processing object used in an intermittent processing step. The processing object 400 has a plurality of grooves 401 extending in the axial direction on the outer circumference of a cylinder. That is, in the processing object 400, arc surfaces 402 and grooves 401 are arranged alternately in the circumferential direction. In the outer diameter processing step of the processing object 400 as described above, the processing object 400 rotates in the circumferential direction (direction of the arrow), and the cutting tool 30 processes the outer circumferential surface of the processing object 400. The cutting edge of the cutting tool 30 repeatedly contacts (cuts) the arc surface 402 and passes (non-cuts) the groove 401.

[0100] Fig. 8A This is a graph showing an example of a time series waveform of the deformation of the cutting tool in the intermittent machining process. Figure 8B This is a graph showing another example of the time series waveform of the deformation of the cutting tool in the intermittent machining process. Fig. 8A This is a graph with a measurement time of 14 seconds. Figure 8B This is a graph with a measurement time of 1.5 seconds. Fig. 8A and Figure 8B In the figure, the vertical axis represents deformation and the horizontal axis represents time. Figure 8B The scale of the time axis (the length corresponding to a unit of time) is greater than Fig. 8A In a continuous machining process in which the cutting of the object is continuous, even if an abnormality occurs in the cutting tool 30, a constant deformation is stably generated in the cutting tool 30. In contrast, in an intermittent machining process, no deformation is generated in the cutting tool 30 during non-cutting, and deformation is generated in the cutting tool 30 during cutting. That is, the state in which deformation is generated in the cutting tool 30 and the state in which no deformation is generated are repeated. In particular, Figure 8BAs shown in FIG. 1 , in the measurement result of the deformation of the cutting tool in the intermittent machining process, high values ​​(cutting) and low values ​​(non-cutting) are repeated alternately. Therefore, the abnormality generated in the cutting tool 30 is manifested as, for example, a change in the deformation of the cutting tool 30. Therefore, the abnormality of the cutting tool 30 can be accurately predicted using the measurement result of the deformation in the intermittent machining process.

[0101] Back to Figure 5 The object data is, for example, feature data including feature quantities of measured values ​​of deformation of the cutting tool 30. The abnormality detection device 300 further has a function of a conversion unit 312. The conversion unit 312 converts the measured values ​​of deformation into feature quantities and generates object data.

[0102] The characteristic quantity of the deformation is, for example, a moving average of the measured values ​​of the deformation. Another example of the characteristic quantity of the deformation is the moving standard deviation of the measured values ​​of the deformation. Another example of the characteristic quantity of the deformation is the difference between the percentiles of the measured values ​​of the deformation, in a specific example, the difference between the 95th percentile and the 5th percentile. Another example of the characteristic quantity of the deformation is the RMS (root mean square) of the measured value of the deformation filtered with a bandpass filter.

[0103] The object data may be measurement data (so-called raw data) of the deformation of the cutting tool 30. In this case, the conversion section 312 may be omitted.

[0104] The classifier 310 is a one-class classifier (one-class classification unit). The one-class classifier, that is, the classifier 310 is constructed by unsupervised learning.

[0105] Fig. 9 is a diagram used to illustrate an example of the learning process of a one-class classifier. Fig. 9 In the figure, the vertical axis represents deformation and the horizontal axis represents time. Fig. 9 In the example, an abnormality of the cutting tool 30 occurs at a time point indicated as "abnormality" (hereinafter referred to as "abnormality occurrence point"). Fig. 9 In the graph of , the cutting tool 30 is in normal condition from the initial stage of processing to the point where the abnormality occurs. The learning period is set as the period when the cutting tool 30 is in normal condition. That is, in the learning of the classifier 310, the measurement data in the learning period stored in the measurement result DB 309 is used as the learning data. Thus, the classifier 310 captures the characteristics of the measurement data when the cutting tool 30 is in normal condition (hereinafter also referred to as "normal data"), has characteristics different from those of the normal data, and can distinguish the measurement data when the cutting tool 30 is in abnormal condition (hereinafter also referred to as "abnormal data") from the normal data.

[0106] For example, the classifier 310 is a local outlier factor (LOF). Hereinafter, the classifier 310 of the local outlier factor will be described.

[0107] Fig.10 is a diagram schematically showing the outline of the local outlier factor. LOF is an outlier detection algorithm that detects outliers based on a set of data. In LOF, for each point, the local density (represented by the dotted circle in the figure) is calculated based on the distance to k nearby points, and the local density of the focus point A is compared with the local density of the nearby group. If the local density of the focus point is low, the focus point is judged to be an abnormal value (outlier).

[0108] LOF is described in detail. Let k-distance(A) be the distance from point A to the kth nearby point, and let Nk(A) be the set of k nearest points. The reachability distance k(A, B) from point B to point A is defined by equation (1).

[0109] [Number 1]

[0110] reachability-distancek(A,B)=max{k-distance(B),d(A,B)}…(1)

[0111] The local reachability density lrd k(A) of point A is defined by equation (2).

[0112] [Number 2]

[0113]

[0114] The LOF of point A is also defined by equation (3).

[0115] [Number 3]

[0116]

[0117] The calculated LOF indicates the degree to which the local density of the focus point A is equal to the local density of the nearby points. The lower the LOF, the higher the local density of the focus point A, and the higher the possibility that the focus point A is normal data. The higher the LOF, the lower the local density of the focus point A, and the higher the possibility that the focus point A is abnormal data. That is, LOF is an outlier score, which is an example of abnormality.

[0118] The classifier 310 is not limited to LOF. The classifier 310 may be a class classifier other than LOF. For example, the classifier 310 may be an Isolation Forest. Another example of the classifier 310 is a class SVM.

[0119] Back to Figure 5The abnormality detection unit 313 detects abnormality of the cutting tool 30 based on the classification result of the object data obtained by the classifier 310. Specifically, the abnormality detection unit 313 detects abnormality of the cutting tool 30 based on the abnormality degree generated in one-class classification obtained by the classifier 310.

[0120] More specifically, the abnormality detection unit 313 performs multi-class classification of the object data by comparing the abnormality degree with a plurality of threshold values, and detects abnormality of the cutting tool 30 based on the result of the multi-class classification.

[0121] In a specific example, the abnormality detection unit 313 classifies the object data into a normal state, an adjacent abnormal state, and an abnormal state by comparing the abnormality degree with a first threshold value and a second threshold value. The first threshold value is used to distinguish the object data into a normal category, an adjacent abnormal category, and a set of abnormal categories. The normal category is a category of object data in which the cutting tool 30 is in a normal state, the abnormal category is a category of object data in which the cutting tool 30 is in an abnormal state, and the adjacent abnormal category is a category of object data in which the cutting tool 30 is in an adjacent abnormal state. The normal state mentioned here refers to a state in which the object to be processed can be processed normally by the cutting tool 30. In this state, even if the cutting tool 30 is continued to be used, it will temporarily be in a state without abnormality. The abnormal state refers to a state in which the object to be processed cannot be processed normally by the cutting tool 30. The adjacent abnormal state refers to an intermediate state between the normal state and the abnormal state of the cutting tool 30. The object to be processed can be processed normally by the cutting tool 30, but in this state, if the cutting tool 30 is continued to be used, it will be in a state that will become abnormal.

[0122] The first threshold is a value less than the second threshold. When the degree of abnormality (LOF) is below the first threshold, the object data belongs to the normal category. When the degree of abnormality is greater than the first threshold, the object data belongs to the adjacent abnormal category or the abnormal category. When the degree of abnormality is below the second threshold, the object data belongs to the normal category or the adjacent abnormal category. When the degree of abnormality is greater than the second threshold, the object data belongs to the abnormal category. That is, when the degree of abnormality is greater than the first threshold and is below the second threshold, the object data belongs to the adjacent abnormal category.

[0123] When the classifier 310 is LOF, for example, the first threshold is 1, and the second threshold is a value greater than 1, for example, 1.25.

[0124] When the abnormality is greater than the second threshold, the abnormality detection unit 313 detects an abnormality of the cutting tool 30. When the abnormality is greater than the first threshold and less than the second threshold, the abnormality detection unit 313 can determine that the cutting tool 30 is in an abnormal state. That is, the abnormality detection unit 313 can predict the occurrence of an abnormality of the cutting tool 30.

[0125] The output unit 314 inputs the detection result of the abnormality of the cutting tool 30 obtained by the abnormality detection unit 313. In a specific example, when the abnormality detection unit 313 detects the abnormality of the cutting tool 30, the output unit 314 outputs information for notifying the user of the occurrence of the abnormality. For example, the output unit 314 displays a screen for notifying the occurrence of the abnormality of the cutting tool 30 (hereinafter also referred to as an "abnormality notification screen") on the display device 306. Fig.11 The abnormality notification screen 500 includes text, graphics, colors, or a combination thereof for notifying the occurrence of an abnormality in the cutting tool 30. Fig.11 In the example of FIG. 5 , the abnormality notification screen 500 includes a text such as “An abnormality has occurred in the tool.” This allows the user to be notified of the abnormality of the cutting tool 30 .

[0126] In another example, when the abnormality detection unit 313 predicts that the cutting tool 30 has an abnormality, the output unit 314 outputs information for notifying the user that the abnormality is predicted. For example, the output unit 314 displays a screen on the display device 306 for notifying that the cutting tool 30 has an abnormality. Fig.12 510 is a diagram showing an example of an abnormality prediction screen. The abnormality prediction screen 510 includes text, graphics, colors, or a combination thereof for notifying that an abnormality of the cutting tool 30 is predicted to occur. Fig.12 In the example of FIG. 5 , the abnormality prediction screen 510 includes a text such as “abnormality is about to occur in the tool.” Thus, the user can be notified that the abnormality of the cutting tool 30 is predicted to occur.

[0127] [6. Operation of abnormality detection device]

[0128] The processor 301 of the abnormality detection device 300 executes the abnormality detection program 307 , whereby the processor 301 performs the abnormality detection process described below. Fig.13 This is a flowchart showing an example of an abnormality detection process by the abnormality detection device according to the embodiment.

[0129] While the cutting tool 30 is used to cut the object to be processed by the machine tool 20, the strain sensor 110 measures the deformation of the cutting tool 30. The sensor module 100 wirelessly transmits the measurement result obtained by the strain sensor 110 to the abnormality detection device 300. The abnormality detection device 300 receives the measurement result of the deformation and saves the measurement result in the measurement result DB 309. In this way, the measurement data is stored in the measurement result DB 309.

[0130] The processor 301 acquires measurement data of deformation of the cutting tool 30 according to the measurement result DB (step S101 ).

[0131] When the object data is feature data, the processor 301 calculates feature values ​​based on the measurement values ​​of the acquired measurement data and generates feature data (step S102). However, when the object data is measurement data, step S102 can be omitted.

[0132] Next, the processor 301 performs classification processing (one-class classification) of the object data by the classifier 310 (step S103).

[0133] Fig.14 This is a flowchart showing an example of classification processing of object data. Fig.14 In the figure, it represents the classification processing based on LOF.

[0134] The processor 301 calculates the reachability distance using the above equation (1) (step S201 ).

[0135] The processor 301 calculates the local reachability density using the above equation (2) (step S202 ).

[0136] The processor 301 calculates LOF (abnormality level) using the above equation (3) (step S203).

[0137] The processor 301 classifies the object data by LOF (step S203). That is, it determines whether the object data is an outlier (abnormal value). However, in the case of multi-class classification based on abnormality, the result of step S204 is not used. In this case, step S204 can be omitted. The above ends the classification process.

[0138] Back to Fig.13 The processor 301 classifies the object data into multiple categories using the abnormality calculated by the classifier 310 (step S104). That is, the processor 301 compares the abnormality with the first threshold and the second expected position to classify the object data into a normal category, a proximate abnormal category, and an abnormal category.

[0139] Processor 301 determines whether the classification result is an abnormality category (step S105). If the classification result is an abnormality category ("Yes" in step S105), that is, when an abnormality of cutting tool 30 is detected, processor 301 displays abnormality notification screen 500 on display device 306 (step S107).

[0140] When the classification result is not an abnormal category ("No" in step S105), the processor 301 determines that the classification result is a near abnormal category (step S106). When the classification result is a near abnormal category ("Yes" in step S106), that is, when the occurrence of an abnormality of the cutting tool 30 is predicted, the processor 301 displays the abnormality prediction screen 510 on the display device 306 (step S107).

[0141] The abnormality detection process ends when the abnormality notification screen 500 or the abnormality prediction screen 510 is displayed on the display device 306. When the classification result is a normal class (No in step S106), the processor 301 ends the abnormality detection process.

[0142] [7. Deformation example]

[0143] The object data may include not only data based on the measurement data of the deformation of the cutting tool 30 (characteristic data or measurement data. Hereinafter, also referred to as "deformation data"), but also data based on the measurement data of the acceleration of the cutting tool 30 (characteristic data or measurement data. Hereinafter, also referred to as "acceleration data"). That is, the object data may be a combination of deformation data and acceleration data at the same time. The acceleration of the cutting tool 30 is measured by, for example, an acceleration sensor (not shown) mounted on the spindle of the machine tool 20.

[0144] The conversion unit 312 can generate not only characteristic data of deformation, but also characteristic data of acceleration. The characteristic quantity of acceleration is, for example, the RMS of the measured value of acceleration. Another example of the characteristic quantity of acceleration is the moving standard deviation of the measured value of acceleration. Another other example of the characteristic quantity of acceleration is the percentile of the measured value of acceleration, in a specific example, the 95th percentile or the 5th percentile. Another other example of the characteristic quantity of acceleration is the difference between the percentiles of the measured value of acceleration, in a specific example, the difference between the 95th percentile and the 5th percentile. Another other example of the characteristic quantity of acceleration is the RMS of the value after filtering the measured value of acceleration with a bandpass filter. Another other example of the characteristic quantity of acceleration is the kurtosis of the time series measured value of acceleration. Another other example of the characteristic quantity of acceleration is the skewness of the time series measured value of acceleration.

[0145] In the above embodiment, the classifier 310 is set as a one-class classifier, but the present invention is not limited thereto. The classifier 310 may be a multi-class classifier.

[0146] The multi-class classifier, i.e., the classifier 310, performs multi-class classification of the object data. In a specific example, the classifier 310 classifies the object data into a normal class, a proximal abnormal class, and an abnormal class. The multi-class classifier, i.e., the classifier 310, is constructed by unsupervised learning.

[0147] Fig.15 is a diagram for explaining an example during the learning process of a multi-class classifier. Fig.15 In the figure, the vertical axis represents deformation and the horizontal axis represents time. The measurement data when the cutting tool 30 is in a normal state, the measurement data when the cutting tool 30 is in an abnormal state, and the measurement data when the cutting tool 30 is in an abnormal state are used for learning the multi-class classifier. The learning period is set to the period when the cutting tool 30 is in a normal state (normal period), the period when the cutting tool 30 is in an abnormal state (near abnormal period), and the period when the cutting tool 30 is in an abnormal state (abnormal period). That is, in the learning of the classifier 310, the measurement data (or feature data) in the normal period, the near abnormal period, and the abnormal period stored in the measurement data of the measurement result DB309 are used as learning data. Thus, the classifier 310 can capture the features of the measurement data (normal data) when the cutting tool 30 is in a normal state, the features of the measurement data (hereinafter, also referred to as "near abnormal data") when the cutting tool 30 is in an abnormal state, and the features of the measurement data (abnormal data) when the cutting tool 30 is in an abnormal state, and distinguish between normal data, near abnormal data, and abnormal data.

[0148] In the above embodiment, the multi-class classification of the object data is performed using the abnormality calculated by the one-class classifier, but the present invention is not limited thereto. The one-class classifier, namely the classifier 310 , may detect an outlier (abnormal data) to detect an abnormality of the cutting tool 30 .

[0149] In the above embodiment, the interception period is set in the intermittent processing step, and the abnormality of the cutting tool in the intermittent processing step is detected, but the present invention is not limited thereto. The interception period may also be set in the continuous processing step, and the abnormality of the cutting tool in the continuous processing step may be detected. Furthermore, a period including one or more intermittent processing steps and one or more continuous processing steps may be set as the interception period, and the abnormality of the cutting tool in the interception period may be detected.

[0150] [8.Supplementary Note]

[0151] The embodiments disclosed this time are illustrative in all aspects and are not restrictive. The scope of rights of the present invention is indicated not only by the above-mentioned embodiments but also by the claims, and includes all changes within the meaning equivalent to the claims and the scope thereof.

[0152] Description of Reference Numerals

[0153] 10: Anomaly detection system;

[0154] 20: Machine tools;

[0155] 30: cutting tools;

[0156] 30A, 30B: turning tools;

[0157] 30C, 30D: milling tools;

[0158] 31A, 31B, 31C, 31D: cutting part;

[0159] 32: cutting blade;

[0160] 33A, 33B, 33C: fixing components;

[0161] 34, 35: cutting edge;

[0162] 100: sensor module;

[0163] 101: Processor;

[0164] 102: non-volatile memory;

[0165] 103: Volatile memory;

[0166] 104: communication interface (communication I / F);

[0167] 105: Battery;

[0168] 110, 110A, 110B: deformation sensors;

[0169] 200: wireless devices;

[0170] 300: anomaly detection device;

[0171] 301: processor;

[0172] 302: non-volatile memory;

[0173] 303: Volatile memory;

[0174] 304: Input and output interface (I / O);

[0175] 305: Graphics controller;

[0176] 306: display device;

[0177] 307: Anomaly detection program;

[0178] 308: Benchmark data;

[0179] 309: measurement result DB;

[0180] 310: classifier (classification unit);

[0181] 311: input unit;

[0182] 312: Conversion department;

[0183] 313: Abnormality detection unit;

[0184] 314: output unit;

[0185] 400: Processing object;

[0186] 401: slot;

[0187] 402: arc surface;

[0188] 500: abnormal notification screen;

[0189] 510: abnormal prediction screen;

[0190] A, B, C, D, E, F: process;

[0191] P1, P2, P3, P4, P5, P6: Period.

Claims

1. An abnormality detection system for detecting an abnormality of a tool installed on a machine tool, wherein: The anomaly detection system comprises: a deformation sensor for measuring deformation of the tool while the machine tool processes the object to be processed by the tool; as well as an abnormality detection device for detecting an abnormality of the tool based on measurement data of the deformation sensor, The abnormality detection device comprises: The machine learning completed model, i.e., the classification unit, takes object data based on the measurement data of the deformation sensor as input and takes the classification result of the object data as output; an abnormality detection unit that detects an abnormality of the tool based on the classification result of the object data obtained by the classification unit; and An output unit outputs the detection result obtained by the abnormality detection unit.

2. The abnormality detection system according to claim 1, wherein: The deformation sensor is mounted on the tool.

3. The abnormality detection system according to claim 2, wherein: The tool is a cutting tool including a turning tool or a milling tool, The cutting tool includes the strain sensor and a wireless communication unit that wirelessly transmits measurement data of the strain sensor.

4. The abnormality detection system according to any one of claims 1 to 3, wherein: The classification unit is a one-category classification unit that performs one-category classification. The abnormality detection unit detects abnormality of the tool based on the abnormality degree generated in the one-class classification.

5. The abnormality detection system according to claim 4, wherein: The abnormality detection unit performs multi-class classification of the object data by comparing the abnormality degree with a plurality of threshold values, and detects abnormality of the tool based on a result of the multi-class classification.

6. The abnormality detection system according to claim 5, wherein: The abnormality detection unit classifies the object data into a normal state, a state close to abnormality, and an abnormal state by comparing the abnormality degree with a first threshold value and a second threshold value.

7. The abnormality detection system according to claim 6, wherein: When the object data is classified as being in an abnormal state by the abnormality detection unit, the output unit outputs information for notifying that the tool is in an abnormal state.

8. The abnormality detection system according to claim 6, wherein: When the object data is classified by the abnormality detection unit as being in a state close to abnormality, the output unit outputs information for notifying that the abnormality of the tool is predicted to occur.

9. The abnormality detection system according to any one of claims 6 to 8, wherein: A first threshold value for distinguishing the object data between a normal state and a state approaching an abnormality is determined based on the object data based on measurement data measured by the deformation sensor of the tool approaching an abnormality.

10. The abnormality detection system according to any one of claims 1 to 9, wherein: The abnormality detection device detects abnormality of the tool based on measurement data of deformation of the tool in an intermittent machining process in which cutting and non-cutting of the object to be machined are repeated.

11. An abnormality detection device for detecting an abnormality of a tool installed on a machine tool, wherein: The abnormality detection device comprises: an acquisition unit that acquires object data based on measurement data output from a strain sensor that measures strain generated in the tool while the machine tool is processing the object to be processed using the tool; The machine learning completed model, i.e., the classification unit, takes the object data acquired by the acquisition unit as input and takes the classification result of the object data as output; an abnormality detection unit that detects an abnormality of the tool based on the classification result of the object data obtained by the classification unit; as well as An output unit outputs the detection result obtained by the abnormality detection unit.

12. An abnormality detection method, performed by an abnormality detection device for detecting an abnormality of a tool installed in a machine tool, wherein: The anomaly detection method comprises the following steps: acquiring object data based on measurement data output from a strain sensor that measures strain generated in the tool while the machine tool is processing the object to be processed using the tool; Classifying the object data by a classification unit that is a machine learning model that takes the acquired object data as input and outputs a classification result of the object data; detecting an abnormality of the tool based on a classification result of the object data obtained by the classification unit; as well as The abnormality detection result of the tool is output.

13. A computer program for detecting an abnormality of a tool installed in a machine tool, wherein: The computer program causes the computer to execute the following steps: acquiring object data based on measurement data output from a strain sensor that measures strain generated in the tool while the machine tool is processing the object to be processed using the tool; Classify the object data using a classification unit that is a machine learning model that takes the acquired object data as input and outputs a classification result of the object data; detecting an abnormality of the tool based on a classification result of the object data obtained by the classification unit; as well as The abnormality detection result of the tool is output.

Images