JP7723926B2 - Anomaly detection device and anomaly detection method - Google Patents

Description

Article 30, paragraph 2 of the Patent Act applies. March 2, 2022: Published in the 2022 Japan Society for Precision Engineering Spring Meeting Academic Lecture Series. March 17, 2022: Published in the 2022 Japan Society for Precision Engineering Spring Meeting Academic Lecture Series.

The disclosed technology relates to an anomaly detection device and an anomaly detection method.

A method has been disclosed in the past for generating a learning model that predicts tool life by applying machine learning based on sound data during machining (Patent Document 1). This method improves accuracy by preparing multiple learning models and selecting one from among them depending on the condition.

A method has also been disclosed in which a normal model is created using machine learning of tool-related measurements (vibration, sound, motor current, etc.) during normal machining, and abnormalities are determined based on this normal model (Patent Document 2).

A method has also been disclosed for predicting tool wear by generating a learning model that correlates tool sound and temperature data and tool wear data with machining conditions (Patent Document 3).

Japanese Patent Application Laid-Open No. 2021-071818 Japanese Patent Application Laid-Open No. 2019-139755 Japanese Patent Application Laid-Open No. 2018-024055

Rule-based anomaly detection is one option, but as systems become more diverse and sensors become more multi-channel, the amount of data becomes vast and complex, making it impossible for workers to correctly grasp and understand the relationships between that data.

It is possible to perform deep learning using time-series data as input, but if features are extracted through preprocessing and used as input for a neural network model, offline preprocessing is required, making it impossible to perform real-time anomaly detection.

In addition, due to daily improvement efforts, there is almost no training data on abnormalities in the field, making it difficult to train a model that can accurately detect anomalies.

The disclosed technology was developed in light of the above points, and aims to provide an anomaly detection device and anomaly detection method that can accurately detect tool wear abnormalities in real time.

A first aspect of the present disclosure is an anomaly detection device that detects abnormal wear in a tool of a machine tool, and includes: an acquisition unit that acquires time-series data of measurement values related to the tool; an estimation unit that inputs the time-series data of measurement values acquired by the acquisition unit into a trained model that has been trained in advance based on the time-series data of measurement values acquired during normal machining and outputs the time-series data of measurement values, and acquires the output of the trained model; and a detection unit that compares the time-series data of measurement values acquired by the acquisition unit with the output of the trained model to detect abnormal wear in the tool.

A second aspect of the present disclosure is an anomaly detection device that detects abnormal wear in a tool of a machine tool, and includes: an acquisition unit that acquires time-series data of measurement values related to the tool; an estimation unit that inputs the time-series data of measurement values acquired by the acquisition unit into a trained model including an encoder and decoder that has been trained in advance based on the time-series data of measurement values acquired during normal machining and outputs the time-series data of measurement values, and acquires the output of the encoder of the trained model for each interval of the time-series data of measurement values; and a detection unit that detects abnormal wear in the tool by comparing the output of the encoder of the trained model acquired for each interval of the time-series data of measurement values.

A third aspect of the present disclosure is an anomaly detection method for detecting abnormal wear in a tool of a machine tool, in which an acquisition unit acquires time series data of measurement values related to the tool, an estimation unit inputs the time series data of measurement values acquired by the acquisition unit into a trained model that has been trained in advance based on the time series data of measurement values acquired during normal machining and outputs the time series data of measurement values, and acquires the output of the trained model, and a detection unit compares the time series data of measurement values acquired by the acquisition unit with the output of the trained model to detect abnormal wear in the tool.

A fourth aspect of the present disclosure is an anomaly detection method for detecting abnormal wear in a tool of a machine tool, in which an acquisition unit acquires time series data of measurement values related to the tool, an estimation unit inputs the time series data of measurement values acquired by the acquisition unit into a trained model including an encoder and decoder that has been trained in advance based on the time series data of measurement values acquired during normal machining and outputs the time series data of measurement values, and the output of the encoder of the trained model is acquired for each interval of the time series data of measurement values, and a detection unit compares the output of the encoder of the trained model acquired for each interval of the time series data of measurement values to detect abnormal wear in the tool.

The disclosed technology makes it possible to accurately detect abnormal tool wear in real time.

FIG. 1 is a schematic block diagram of an example of a computer that functions as a model learning device and an anomaly detection device according to the first and second embodiments. FIG. 2 is a diagram for explaining a strain gauge, an accelerometer, a thermocouple, and an AE sensor provided on the tool. FIG. 10 is a diagram illustrating an example of time-series data of various measurement values. FIG. 10 is a diagram illustrating an example of time-series data of an AE signal. FIG. 1 is a block diagram showing a configuration of a model learning device according to a first embodiment and a second embodiment. FIG. 2 is a diagram illustrating an example of a model according to the first embodiment. 1 is a block diagram showing a configuration of an anomaly detection device according to a first embodiment and a second embodiment; 4 is a flowchart showing an abnormality detection routine of the abnormality detection device of the first embodiment. 10 is a graph showing an example of detecting abnormal wear from time-series data of an AE signal. 10 is a graph showing an example of detecting abnormal wear from time-series sound data. FIG. 10 is a diagram illustrating an example of a model according to a second embodiment. 10 is a flowchart showing an abnormality detection routine of the abnormality detection device of the second embodiment. 10 is a graph showing an example of detecting abnormal wear from time-series data of an AE signal. 10 is a graph showing an example of detecting abnormal wear from time-series sound data.

An example of an embodiment of the disclosed technology will be described below with reference to the drawings. Note that the same reference numerals are used to designate identical or equivalent components and parts in each drawing. Also, the dimensional proportions in the drawings have been exaggerated for the sake of explanation and may differ from the actual proportions.

[First embodiment]
<Configuration of model learning device according to first embodiment>
FIG. 1 is a block diagram showing the hardware configuration of a model learning device 10 according to this embodiment.

As shown in FIG. 1, the model learning device 10 has a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, storage 14, an input unit 15, a display unit 16, and a communication interface (I/F) 17. Each component is connected to each other via a bus 19 so that they can communicate with each other.

CPU 11 is a central processing unit that executes various programs and controls each component. That is, CPU 11 reads programs from ROM 12 or storage 14 and executes the programs using RAM 13 as a work area. CPU 11 controls the above components and performs various calculation processes in accordance with the programs stored in ROM 12 or storage 14. In this embodiment, ROM 12 or storage 14 stores a learning program for learning a model. The learning program may be a single program, or a group of programs consisting of multiple programs or modules.

ROM 12 stores various programs and data. RAM 13 serves as a working area for temporarily storing programs or data. Storage 14 is composed of an HDD (Hard Disk Drive) or SSD (Solid State Drive) and stores various programs, including the operating system, and various data.

The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used to enter various inputs.

The input unit 15 accepts as input learning data, which is time-series data of measurement values related to the machine tool's tool, acquired during machining when the machine tool's tool is in a normal state.

Specifically, as shown in FIG. 2, a strain gauge 31 attached to a tool 30 of the machine tool is used to measure the distortion of the tool 30, an accelerometer 34 attached to the tool 30 is used to measure the acceleration of the tool 30, a thermocouple 32 attached to the tool 30 is used to measure the temperature of the tool 30, a microphone (not shown) attached to the machine tool is used to measure the sound of the tool 30, and AE sensors 36 and 38 attached to the tool 30 are used to measure the AE (acoustic emission) signal of the tool 30 (see FIG. 3). FIG. 3 shows an example of time series data of distortion, acceleration, temperature, sound, and AE signal of the tool 30.

The time series data of the distortion, acceleration, temperature, sound, and AE signal of the tool 30 acquired during machining when the tool 30 is in a normal state, i.e., when there are no wear abnormalities, are used as learning data (see the normal time series data in Figure 4). Figure 4 shows an example of time series data of the AE signal when the tool 30 is normal, in a transitional period immediately before an abnormality, and when an abnormality occurs.

This normal learning data is prepared for each of the multiple tools 30, and the input unit 15 accepts the learning data for each of the multiple tools 30.

The display unit 16 is, for example, a liquid crystal display, and displays various information. The display unit 16 may also function as the input unit 15 by adopting a touch panel system.

The communication interface 17 is an interface for communicating with other devices, and uses standards such as Ethernet (registered trademark), FDDI (Fiber Distributed Data Interface), Wi-Fi (registered trademark), and USB (Universal Serial Bus).

Next, we will explain the functional configuration of the model learning device 10. Figure 5 is a block diagram showing an example of the functional configuration of the model learning device 10.

As shown in Figure 5, the model learning device 10 functionally comprises a learning database (DB) 20, a learning unit 24, and a model storage unit 26.

The learning database 20 stores learning data for each of the multiple input tools 30.

Based on the received learning data for the multiple tools 30, the learning unit 24 learns a model that takes time series data of measurement values as input and outputs time series data of measurement values.

Specifically, when time series data of measurement values, which is training data, is input to the model, the model is trained so that it outputs time series data of measurement values corresponding to the training data.

More specifically, as shown in Figure 6, model 70 is a model based on LSTM (Long Short-Term Memory) and an autoencoder, and includes an encoder 72 and a decoder 74. The encoder 72 receives, for each type of measurement value, the measurement value at each time of one interval of time series data, and outputs encoded data. The decoder 74 receives, for each type of measurement value, the encoded data and decodes it into the measurement value at each time of one interval of time series data.

The model storage unit 26 stores the trained model.

<Configuration of anomaly detection device according to the first embodiment>
FIG. 1 is a block diagram showing the hardware configuration of an abnormality detection device 50 according to this embodiment.

As shown in Figure 1 above, the anomaly detection device 50 has the same configuration as the model learning device 10, and the ROM 12 or storage 14 stores an anomaly detection program for detecting tool wear abnormalities.

The input unit 15 accepts time-series data of measurement values related to the tool 30 as input. Specifically, the strain of the tool 30 is measured using a strain gauge 31 provided on the tool 30 of the machine tool, the acceleration of the tool 30 is measured using an accelerometer 34 provided on the tool 30, the temperature of the tool 30 is measured using a thermocouple 32 provided on the tool 30, the sound of the tool 30 is measured using a microphone (not shown) provided on the machine tool, and the AE (acoustic emission) signal of the tool 30 is measured using AE sensors 36 and 38 provided on the tool 30.

The input unit 15 accepts as input time series data of tool distortion, time series data of acceleration, time series data of temperature, time series data of sound, and time series data of AE signals acquired during tool machining.

Next, we will explain the functional configuration of the anomaly detection device 50. Figure 7 is a block diagram showing an example of the functional configuration of the anomaly detection device 50.

As shown in FIG. 7, the anomaly detection device 50 functionally comprises a model storage unit 60, an acquisition unit 62, an estimation unit 64, and a detection unit 66.

The model storage unit 60 stores trained models trained by the model learning device 10.

The acquisition unit 62 acquires time series data of measurement values related to the input tool 30.

The estimation unit 64 inputs the time series data of the measurement values acquired by the acquisition unit 62 into the trained model and acquires the output of the trained model.

Specifically, for each type of measurement value, the measurement value at each time of one section of time series data is input to the encoder 72, and the measurement value at each time of one section of time series data is decoded by the decoder 74 to obtain the measurement value.

The detection unit 66 compares the time series data of the measurement values acquired by the acquisition unit 62 with the output of the trained model to detect abnormal wear of the tool 30.

Specifically, the detection unit 66 calculates an abnormality score for each interval of the time series data using an error value that is the sum of the differences between the measured values and the output of the trained model, and detects wear abnormalities in the tool 30 based on the abnormality score for each interval.

More specifically, for each time t in one interval, the difference a(t) between the measurement value x(t) and the output ^y(t) of the trained model is calculated according to the following formula, and the error value ε(t) is calculated by summing up the differences a(t) for one interval.

Then, using the error value ε(t) for that section, the anomaly score S(t) is calculated according to the following formula:

However, α and β are predetermined values.

For each section, the detection unit 66 detects abnormal wear of the tool 30 when the abnormality score S(t) for that section is equal to or greater than the threshold value.

<Operation of the model learning device according to the first embodiment>
Next, the operation of the model learning device 10 according to this embodiment will be described.

First, the CPU 11 of the model learning device 10 reads a learning program from the ROM 12 or storage 14, expands it into the RAM 13, and executes it, thereby performing the learning process. Furthermore, learning data for each of the multiple tools 30 is input into the model learning device 10 and stored in the learning database 20.

Then, the CPU 11, as the learning unit 24, learns a model that takes time-series data of measurement values as input and outputs time-series data of measurement values based on the received learning data for the multiple tools 30, stores the learned model in the model storage unit 26, and terminates the learning process.

<Action of the abnormality detection device according to the first embodiment>
Next, the operation of the abnormality detection device 50 according to this embodiment will be described.

Figure 8 is a flowchart showing the flow of the anomaly detection process performed by the anomaly detection device 50. The CPU 11 reads the anomaly detection program from the ROM 12 or storage 14, deploys it in the RAM 13, and executes it to perform the anomaly detection process. Time-series data of measurement values related to the tool 30 is also input to the anomaly detection device 50. Note that the anomaly detection process is one example of an anomaly detection method.

In step S100, the CPU 11, as the acquisition unit 62, acquires one section of time series data of measurement values related to the input tool 30.

In step S102, the CPU 11, functioning as the estimation unit 64, inputs one section of time series data acquired by the acquisition unit 62 into the trained model and acquires the output of the trained model.

Specifically, the measurement values for each time of one section of time series data are input to the encoder 72, and the measurement values for each time of one section of time series data are decoded by the decoder 74 to obtain them.

In step S104, the CPU 11, as the detection unit 66, calculates an error value by summing up the differences between the measured values and the output of the trained model for that one interval.

In step S106, the CPU 11, as the detection unit 66, calculates an anomaly score for that one section using the error value.

In step S108, the CPU 11, functioning as the detection unit 66, determines whether the abnormality score is equal to or greater than the threshold value. If the abnormality score is less than the threshold value, the process returns to step S100, and the processes of steps S100 to S108 are repeated for the next section of time-series data. On the other hand, if the abnormality score is equal to or greater than the threshold value, the process proceeds to step S110.

In step S110, the CPU 11 outputs a message on the display unit 16 indicating that the tool 30 is experiencing abnormal wear, and then ends the abnormality detection process.

Example 1
An example of detecting wear abnormalities using the model learning device 10 and the abnormality detection device 50 of the first embodiment described above will be described.

Figures 9(A) and (B) show examples of the measurement value x(t), the trained model output ^y(t), the difference a(t), the error value ε(t), and the anomaly score S(t) when time series data of an AE signal is input.

As shown in Figure 9(A), an anomaly is detected during the transition period before the anomaly occurs. Figure 9(B) shows that the anomaly score S(t) is above the threshold for unpredictable events. For such unpredictable events, anomalies can be detected with high accuracy by performing processing such as noise removal on the anomaly score S(t).

Figures 10(A) and (B) show examples of the measurement value x(t), the trained model output ^y(t), the difference a(t), the error value ε(t), and the anomaly score S(t) when sound time series data is input.

As shown in Figures 10(A) and (B), it can be seen that an abnormality was detected during the transient period before the abnormality occurred.

As described above, the anomaly detection device according to the first embodiment inputs the time series data of acquired measurement values into a trained model that has been trained in advance based on the time series data of measurement values acquired during normal machining conditions, obtains the output of the trained model, and compares the time series data of acquired measurement values with the output of the trained model to detect abnormal tool wear. This enables accurate detection of abnormal tool wear in real time.

Second Embodiment
Next, a second embodiment will be described. Note that the configurations of the learning device and the anomaly detection device according to the second embodiment are the same as those of the first embodiment, so the same reference numerals are used and the description thereof will be omitted.

In the second embodiment, the model configuration differs from that of the first embodiment.

<Configuration of model learning device according to second embodiment>
The learning unit 24 of the model learning device 10 according to the second embodiment learns a model that takes time series data of measurement values as input and outputs time series data of measurement values based on the received learning data for a plurality of tools 30.

Specifically, when time series data of measurement values, which is training data, is input to the model, the model is trained so that it outputs time series data of measurement values corresponding to the training data.

More specifically, as shown in FIG. 11(A), model 270 is a model based on an autoencoder and includes an encoder 272 and a decoder 274. The encoder 272 receives, for each type of measurement value, the measurement value at each time of one interval of time series data, and outputs encoded data. The decoder 274 receives, for each type of measurement value, the encoded data and decodes it into the measurement value at each time of one interval of time series data.

<Configuration of an abnormality detection device according to the second embodiment>
As shown in FIG. 7, the anomaly detection device 50 according to the second embodiment functionally comprises a model storage unit 60, an acquisition unit 62, an estimation unit 64, and a detection unit 66.

The model storage unit 60 stores trained models trained by the model learning device 10.

The acquisition unit 62 acquires time series data of measurement values related to the input tool 30.

The estimation unit 64 inputs the time series data of the measurement values acquired by the acquisition unit 62 into the trained model, and acquires the output of the encoder 272 of the trained model as a feature for each section of the time series data of the measurement values.

Specifically, for each type of measurement value, the measurement value at each time of one interval of time-series data is input to the encoder 72, and the encoded data output by the encoder 72 is obtained as a feature.

The detection unit 66 compares the output of the encoder 272 of the trained model obtained for each section of the time series data of the measurement values to detect abnormal wear of the tool 30.

Specifically, for each type of measurement value, the detection unit 66 calculates an anomaly score for each interval of the time-series data using the local density of the output of the trained model encoder 272 acquired for that interval and the output of the trained model encoder 272 acquired during normal machining, and detects wear abnormalities in the tool 30 based on the anomaly score for each interval. Here, the output of the trained model encoder 272 acquired during normal machining is the output of the trained model encoder 272 acquired for an interval prior to the current interval in which wear abnormalities in the tool 30 were not detected.

11B , for each section, a local density ratio is calculated between the output of the encoder 272 of the trained model acquired for that section and the output of the encoder 272 of the trained model acquired during processing in a normal state using a local outlier factor (LOF).The local density ratio for that section is then normalized to calculate an anomaly score.

where N _k(P) is a set of outputs from the encoder 272 obtained for the section prior to the current section, reachability_distance _{k(P, A)} is the distance from the output A of the kth neighbor, lrd _k (P) is the local density, and lof(P) is the local density ratio.

For each section, the detection unit 66 detects abnormal wear of the tool 30 if the abnormality score for that section is equal to or greater than the threshold value.

<Operation of the model learning device according to the second embodiment>
Next, the operation of the model learning device 10 according to this embodiment will be described.

First, the CPU 11 of the model learning device 10 reads a learning program from the ROM 12 or storage 14, expands it into the RAM 13, and executes it, thereby performing the learning process. Furthermore, learning data for each of the multiple tools 30 is input into the model learning device 10 and stored in the learning database 20.

Then, the CPU 11, as the learning unit 24, learns a model that takes time-series data of measurement values as input and outputs time-series data of measurement values based on the received learning data for the multiple tools 30, stores the learned model in the model storage unit 26, and terminates the learning process.

<Action of the abnormality detection device according to the second embodiment>
Next, the operation of the abnormality detection device 50 according to this embodiment will be described.

Figure 12 is a flowchart showing the flow of the anomaly detection process performed by the anomaly detection device 50. The CPU 11 reads the anomaly detection program from the ROM 12 or storage 14, deploys it in the RAM 13, and executes it to perform the anomaly detection process. In addition, time-series data of measurement values related to the tool 30 is input to the anomaly detection device 50.

In step S200, the CPU 11, as the acquisition unit 62, acquires one section of time series data of measurement values related to the input tool 30.

In step S202, the CPU 11, functioning as the estimation unit 64, inputs one section of time series data acquired by the acquisition unit 62 into the trained model and acquires the output of the encoder 272 of the trained model as a feature.

Specifically, the measured values at each time of one interval of time series data are input to the encoder 272, and the encoded data output by the encoder 272 is obtained as a feature.

In step S204, the CPU 11, as the detection unit 66, calculates the local density ratio between the feature value that is the output of the encoder 272 of the trained model acquired for that one section and the feature value that is the output of the encoder 272 of the trained model acquired during processing in a normal state.

In step S206, the CPU 11, as the detection unit 66, calculates an anomaly score for the relevant section using the local density ratio.

In step S208, the CPU 11, functioning as the detection unit 66, determines whether the abnormality score is equal to or greater than the threshold value. If the abnormality score is less than the threshold value, the process returns to step S200, and the processes of steps S200 to S208 are repeated for the next section of time-series data. On the other hand, if the abnormality score is equal to or greater than the threshold value, the process proceeds to step S210.

In step S210, the CPU 11 outputs a message to the display unit 16 indicating that the tool 30 is experiencing abnormal wear, and then ends the abnormality detection process.

Example 2
An example of detecting wear abnormalities using the model learning device 10 and the abnormality detection device 50 of the second embodiment described above will be described.

Figures 13 (A) and (B) show examples of measurement values when time-series data of an AE signal is input, the results of applying envelope processing to the measurement values, the average value of the results, and the anomaly score when the average value is input to the trained model. The closer the anomaly score value is to 1, the more abnormal it is. As shown in Figures 13 (A) and (B), it can be seen that an anomaly is detected in the transient period before the anomaly occurs.

Figures 14(A) and (B) show examples of measurement values and anomaly scores when sound time-series data is input. As shown in Figures 14(A) and (B), it can be seen that an anomaly is detected during the transition period before the anomaly occurs.

As described above, the anomaly detection device according to the second embodiment inputs the time series data of acquired measurement values into a trained model that has been trained in advance based on the time series data of measurement values acquired during normal machining, obtains the output of the trained model's encoder for each interval of the time series data of measurement values, and compares the output of the trained model's encoder obtained for each interval of the time series data of measurement values to detect abnormal tool wear. This enables accurate detection of abnormal tool wear in real time.

The normal, transitional, and abnormal periods in Figures 13 and 14 above are determined by the on-site worker, and experienced on-site workers generally tend to change tools with ample time to spare, so they change tools towards the end of normal times.

In contrast, in this embodiment, an abnormality is detected once the transition period has begun, and by replacing the tool when the abnormality detection device detects an abnormality, the tool can be used longer than the operator. Specifically, this can be extended by approximately 40 seconds, which is equivalent to 15-20% of the tool's lifespan, and when the reduction in downtime of the machine tool is taken into account, the utilization rate of the machine and tool can be improved by nearly 30%.

Furthermore, as can be seen from the anomaly detection results obtained by the anomaly detection device, even when new tools of the same model number are used, the tool lifespans vary, and the timing of anomaly detection also differs, and the anomaly detection device of this embodiment can handle these situations.

<Modification>
The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the spirit and scope of the present invention.

For example, while the example has been described in which the model learning device and the anomaly detection device are configured as separate devices, this is not limited to this, and the model learning device and the anomaly detection device may also be configured as a single device.

Furthermore, the various processes executed by the CPU after reading software (programs) in each of the above embodiments may be executed by various processors other than the CPU. Examples of processors in this case include dedicated electrical circuits, such as GPUs (Graphics Processing Units), FPGAs (Field-Programmable Gate Arrays), and other PLDs (Programmable Logic Devices) whose circuit configuration can be changed after manufacture, and ASICs (Application Specific Integrated Circuits), which are processors with circuit configurations specifically designed to execute specific processes. Furthermore, the learning process and anomaly detection process may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, or a combination of a CPU and an FPGA, etc.). Furthermore, the hardware structure of these various processors is, more specifically, an electrical circuit that combines circuit elements such as semiconductor devices.

In addition, while the above embodiments have described a configuration in which the learning program and anomaly detection program are pre-stored (installed) in storage 14, this is not limiting. The programs may also be provided in a form stored on a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The programs may also be downloaded from an external device via a network.

10 Model learning device 11 CPU
14 Storage 15 Input unit 16 Display unit 20 Learning database 24 Learning unit 26 Model storage unit 30 Tool 31 Strain gauge 34 Accelerometer 32 Thermocouple 36, 38 AE sensor 50 Anomaly detection device 60 Model storage unit 62 Acquisition unit 64 Estimation unit 66 Detection unit 70, 270 Model 72, 272 Encoder 74, 274 Decoder

Claims (8)

An abnormality detection device that detects abnormal wear of a tool of a machine tool,
an acquisition unit that acquires time series data of measurement values related to the tool;
an estimation unit that inputs the time series data of the measurement values acquired by the acquisition unit into a trained model that has been trained in advance based on the time series data of the measurement values acquired during processing in a normal state and outputs the time series data of the measurement values, and acquires an output of the trained model;
a detection unit that compares the time series data of the measurement values acquired by the acquisition unit with an output of the trained model to detect wear abnormalities of the tool;
Including,
The detection unit calculates an abnormality score for each interval of the time series data using an error value that is the sum of the differences between the measurement values at each time of the interval and the output of the trained model, and detects abnormal wear of the tool when the abnormality score for each interval is equal to or greater than a threshold . the acquisition unit acquires time-series data of measurement values for each type of measurement value related to the tool, which is at least two types of measurement values selected from distortion, acceleration, temperature, sound, and acoustic emission (AE) of the tool;
the estimation unit inputs the time series data of the measurement values acquired by the acquisition unit into a trained model that has been trained in advance for each type of measurement value based on the time series data of the measurement values acquired during processing in a normal state, and outputs the time series data of the measurement values, and acquires an output of the trained model;
2. The anomaly detection device according to claim 1, wherein the detection unit calculates, for each type of measurement value, an anomaly score for each interval of the time series data using an error value that is the sum of the differences between the measurement value at each time point in the interval and the output of the trained model, and detects wear abnormality in the tool when the anomaly score for each interval is equal to or greater than a threshold value . The anomaly detection device of claim 1, wherein the trained model includes an encoder and a decoder. An abnormality detection device that detects abnormal wear of a tool of a machine tool,
an acquisition unit that acquires time series data of measurement values related to the tool;
an estimation unit that inputs the time series data of the measurement values acquired by the acquisition unit into a trained model including an encoder and a decoder that has been trained in advance based on the time series data of the measurement values acquired during processing in a normal state and that outputs the time series data of the measurement values, and acquires the output of the encoder of the trained model for each section of the time series data of the measurement values;
a detection unit that detects abnormal wear of the tool by comparing , for each section of the time-series data of the measurement values, the output of the encoder of the trained model acquired for that section with the output of the encoder of the trained model acquired during normal machining;
Including,
The detection unit calculates an abnormality score for each interval of the time series data using a local density ratio between the output of the encoder of the trained model obtained for that interval and the output of the encoder of the trained model obtained during normal processing, and detects abnormal wear of the tool when the abnormality score for each interval is equal to or greater than a threshold. the acquisition unit acquires time-series data of measurement values for each type of measurement value related to the tool, which is at least two types of measurement values selected from distortion, acceleration, temperature, sound, and acoustic emission (AE) of the tool;
the estimation unit inputs the time series data of the measurement values acquired by the acquisition unit into a trained model including an encoder and a decoder that has been trained in advance based on the time series data of the measurement values acquired during processing in a normal state for each type of measurement value and outputs the time series data of the measurement values, and acquires an output of the encoder of the trained model for each section of the time series data of the measurement values;
5. The anomaly detection device according to claim 4, wherein the detection unit calculates an anomaly score for each type of measurement value in each interval of the time series data using a local density ratio between the output of the encoder of the trained model acquired for that interval and the output of the encoder of the trained model acquired during machining in a normal state, and detects wear abnormality in the tool when the anomaly score for that interval is equal to or greater than a threshold . 5. The abnormality detection device according to claim 1, wherein the measured value is distortion, acceleration, temperature, sound, or acoustic emission (AE) of the tool. An abnormality detection method for detecting abnormal wear of a tool of a machine tool, comprising:
an acquisition unit acquires time series data of measurement values related to the tool;
an estimation unit inputs the time series data of the measurement values acquired by the acquisition unit into a trained model that has been trained in advance based on the time series data of the measurement values acquired during processing in a normal state and that outputs the time series data of the measurement values, and acquires an output of the trained model;
A detection unit compares the time-series data of the measurement values acquired by the acquisition unit with an output of the trained model to detect abnormal wear of the tool.
This includes:
The detection unit calculates an abnormality score for each interval of the time series data using an error value obtained by summing up the differences between the measurement values at each time of the interval and the output of the trained model, and detects wear abnormality of the tool when the abnormality score for each interval is equal to or greater than a threshold value.
Anomaly detection methods. An abnormality detection method for detecting abnormal wear of a tool of a machine tool, comprising:
an acquisition unit acquires time series data of measurement values related to the tool;
an estimation unit inputs the time series data of the measurement values acquired by the acquisition unit into a trained model including an encoder and a decoder, the trained model having been trained in advance based on the time series data of the measurement values acquired during processing in a normal state, and outputs the time series data of the measurement values, and acquires an output of the encoder of the trained model for each section of the time series data of the measurement values;
The detection unit compares the output of the encoder of the trained model acquired for each section of the time-series data of the measurement values with the output of the encoder of the trained model acquired during machining in a normal state , and detects abnormal wear of the tool.
This includes:
The detection unit calculates an abnormality score for each interval of the time series data using a local density ratio between the output of the encoder of the trained model obtained for that interval and the output of the encoder of the trained model obtained during normal processing, and detects abnormal wear of the tool for each interval if the abnormality score for that interval is equal to or greater than a threshold .