Abnormality sensing system

Abstract

To provide an abnormality sensing system that can accurately sense an abnormality in grinding processing or in cutting processing.SOLUTION: The abnormality sensing system comprises a processing device that performs grinding processing to a work-piece and a PC that estimates a step of the grinding processing being performed by the processing device. When the grinding processing is started, the PC starts to obtain vibration data. When receiving a request for sending notification of a step of estimating, the PC inputs the vibration data to an estimation model, estimates a step of the grinding processing in which the vibration data can be generated (S23), and notifies a PLC of the estimated step of estimating (S25). The PLC compares an instruction step with the step of estimating (S13), and unless both steps match with each other (NO in S13), senses an abnormality (S14).SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present invention relates to an abnormality detection system that detects an abnormality in a grinding process or a cutting process. [Background technology]

[0002] There are known devices for detecting abnormalities in cutting and grinding processes. For example, JP 2008-132558 A (Patent Document 1) discloses an abnormality detection device for detecting abnormalities in cutting processes. For example, this abnormality detection device compares data on vibrations occurring when a workpiece is cut with a threshold value for determining abnormalities, and detects an abnormality when the data on vibrations exceeds the threshold value. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2008-132558 A Summary of the Invention [Problem to be solved by the invention]

[0004] The machining tools used in cutting and grinding processes may deteriorate with use. In cutting and grinding processes, multiple processes are generally performed in sequence. When the machining tools deteriorate, for example, in a grinding process, the amount of contact between the machining tool and the workpiece may become smaller than expected, and the process may not be performed properly. In such cases, the vibrations generated by the process become smaller, so that even if the vibration data is compared with a threshold value, the vibration data may not exceed the threshold value, and an abnormality may not be detected properly. There is a demand for improving the accuracy of detecting abnormalities in grinding and cutting processes, including the above cases.

[0005] The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide an abnormality detection system that can accurately detect abnormalities in grinding or cutting processes. [Means for solving the problem]

[0006] (1) The anomaly detection system according to the present disclosure includes a processing device that performs grinding or cutting on a processing object by sequentially executing a plurality of predetermined processes, and an estimation device that inputs data on vibrations generated in the processing device during the above processing into an estimation model trained by machine learning to estimate which process the processing device is executing. In this anomaly detection system, an anomaly is detected when an instructed process to be executed by the processing device differs from an estimated process estimated by the estimation device.

[0007] (2) Preferably, the estimation model is generated by a learning process using a training data set. The training data set includes a plurality of training data. Each of the plurality of training data includes information indicating a process and data on vibration when the process is performed.

[0008] (3) Preferably, the estimation model outputs a possibility indicating which of the multiple processes the input relates to for each process, and the estimation device determines the process with the highest possibility from the output possibilities as the estimated process.

[0009] (4) Preferably, even if the indication step and the estimation step coincide with each other, if the likelihood of the estimation step is less than a threshold value, an abnormality is detected.

[0010] (5) Preferably, the data relating to vibration is data obtained by accumulating vibrations occurring in the processing device for a predetermined period of time, or data obtained by performing FFT (Fast Fourier Transform) on the accumulated data. Effect of the Invention

[0011] According to the anomaly detection system of the present invention, an anomaly in a grinding process or a cutting process can be detected with high accuracy. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram showing an example of a configuration of an anomaly detection system according to an embodiment; [Diagram 2] 3A to 3C are diagrams for explaining each step of a grinding process performed by the processing device. [Diagram 3] FIG. 2 is a block diagram illustrating an example of a functional configuration of a PC. [Figure 4] FIG. 2 is a block diagram illustrating an example of a functional configuration of a PLC. [Diagram 5] FIG. 11 is a diagram for explaining a learning process of an estimation model. [Figure 6] FIG. 13 is a diagram for explaining a trained estimation model. [Figure 7] 5 is a flowchart showing a procedure of a process executed by a PLC and a PC according to the embodiment. [Figure 8] 13 is a diagram for explaining experimental results. [Figure 9] 1 is a diagram showing the experimental results when grinding is broadly divided into three steps. [Figure 10] 10 is a flowchart showing a procedure of a process executed by a PLC and a PC according to a first modification. [Figure 11] 11 is a flowchart showing a procedure of processes executed by a PLC and a PC according to a second modification. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same or corresponding parts are designated by the same reference characters and their description will not be repeated.

[0014] <Overall configuration of anomaly detection system> 1 is a block diagram showing an example of the configuration of an anomaly detection system 1 according to the present embodiment. The anomaly detection system 1 includes a processing device 2 and a PC (Personal Computer) 3. The processing device 2 and the PC 3 are connected to each other via a communication cable 4 so as to be able to communicate with each other.

[0015] The processing device 2 is a device for performing grinding or cutting on a processing object (hereinafter also referred to as a "workpiece") W. In this embodiment, an example will be described in which the processing device 2 is a device that performs grinding on the workpiece W, but it may also be a device that performs cutting on the workpiece W. The grinding process according to this embodiment includes a plurality of steps, as shown in Fig. 2 described later.

[0016] PC3 is a device that estimates which step of the grinding process is being performed when the processing device 2 is performing the grinding process. Specifically, PC3 acquires vibration data during the grinding process from a sensor provided in the processing device 2, and determines which step of the grinding process the vibration data corresponds to, thereby estimating the step being performed by the processing device 2. PC3 corresponds to an example of an "estimation device" according to the present disclosure.

[0017] The abnormality detection system 1 according to the present embodiment detects an abnormality when the process of grinding being performed on the workpiece W by the processing device 2 differs from the process estimated by the PC 3.

[0018] <<Processing equipment and PC configuration>> The processing device 2 includes a workpiece supporting device 10, a grindstone driving device 20, a PLC (Programmable Logic Controller) 30, and a display device 40. The workpiece supporting device 10, the grindstone driving device 20, and the display device 40 are controlled by the PLC30.

[0019] The workpiece supporting device 10 includes a stage 11, a fixing device 12, a rotating shaft 13, and a sensor 14.

[0020] The stage 11 is configured to be movable in the X-axis direction by a motor (not shown). The stage 11 may be configured to be movable in the Z-axis direction in addition to the X-axis direction. A fixing device 12 is attached to the stage 11.

[0021] The fixing device 12 is a device that fixes the workpiece W. By moving the stage 11 in the X-axis direction, the workpiece W fixed to the fixing device 12 can be moved in the X-axis direction. The fixing device 12 is provided with a rotation shaft 13 that rotates the fixed workpiece W. The rotation shaft 13 is connected to a motor (not shown) and is driven to rotate by the motor. This causes the workpiece W to rotate around the rotation shaft 13.

[0022] The sensor 14 is provided on the fixing device 12. The sensor 14 detects vibrations generated during grinding of the workpiece W, and outputs the detection result to the PC 3 as vibration data. For example, an acceleration sensor can be used as the sensor 14. Note that the sensor 14 is not limited to being provided on the fixing device 12 as long as it can detect vibrations generated in the workpiece W or a grindstone described below. For example, the sensor 14 may be provided on the stage 11, or on the grindstone driving device 20 side.

[0023] The grindstone driving device 20 includes a grindstone 21, a rotating shaft 22, a motor 23, and a feed mechanism 24.

[0024] The grindstone 21 is attached to a rotating shaft 22. The rotating shaft 22 is connected to a motor 23 and is rotated by the motor 23. This causes the grindstone 21 to rotate around the rotating shaft 22.

[0025] The feed mechanism 24 is configured to be movable in the Y-axis direction by a motor (not shown). The feed mechanism 24 moves the grindstone 21 in the Y-axis direction together with the rotating shaft 22 and the motor 23. That is, the feed mechanism 24 moves the grindstone 21 in the Y-axis direction. By moving the feed mechanism 24, the grindstone 21 can be brought into contact with the workpiece W. With the workpiece W and the grindstone 21 rotating, the workpiece W comes into contact with the grindstone 21, thereby grinding the workpiece W.

[0026] The PLC 30 controls each device of the processing device 2. The PLC 30 includes a CPU (Central Processing Unit) 30a and a memory 30b. The memory 30b includes, for example, a ROM (Read Only Memory) and a RAM (Random Access Memory). The memory 30b stores various programs to be executed by the CPU 30a.

[0027] The CPU 30a executes various programs stored in the memory 30b to control each device of the processing device 2. The control executed by the CPU 30a is not limited to software processing, but can also be processed by constructing a dedicated hardware (electronic circuit). The CPU 30a executes the programs stored in the memory 30b to sequentially execute each step of the grinding process.

[0028] Fig. 2 is a diagram for explaining each step of the grinding process executed by the processing device 2. Referring to Fig. 2, the grinding process includes a first step, a second step, a third step, a fourth step, a fifth step, a sixth step, and a seventh step. The PLC 30 of the processing device 2 sequentially executes steps (hereinafter, steps are abbreviated as "S") 1 to S7 in the grinding process. The process of the flowchart shown in Fig. 2 is started, for example, by operating a start button (not shown) of the processing device 2.

[0029] When grinding is started, the grindstone 21 and the workpiece W start to be rotated. The PLC 30 manages the coordinates of the grindstone 21 in each process. Specifically, the coordinates of the grindstone 21 on the Y axis are determined in advance in each process, and the PLC 30 moves (sends) the workpiece W to the coordinates determined in the process to be executed. When one process is completed, the PLC 30 moves the grindstone 21 to the coordinates of the next process.

[0030] The first step of S1 is a preparation step. That is, in the first step, the grindstone 21 is in an initial position and is in a stage before the grindstone 21 is sent to the workpiece W.

[0031] The second step of S2 is a step of moving (sending) the grindstone 21 closer to the workpiece W. In the second step, the feed mechanism 24 is moved in the Y-axis direction to move the grindstone 21 closer to the workpiece W.

[0032] The third step of S3 is a step of bringing the grindstone 21 into contact with the workpiece W. In the third step as well, the feed mechanism 24 is moved in the Y-axis direction to bring the grindstone 21 into contact with the workpiece W.

[0033] The fourth step of S4 is a step of roughly grinding the workpiece W. In the fourth step, the stage 11 is moved in the X-axis direction at a relatively high speed (specifically, a speed faster than that in the sixth step), and the workpiece W is reciprocated in the X-axis direction relative to the grindstone 21.

[0034] The fifth step of S5 is a step of rotating the grindstone 21 while stopping the feeding of the grindstone 21 (stopping the movement of the feed mechanism 24 and the stage 11). In the fifth step, the grindstone 21 is rotated and the process waits for the deflection in the direction of the rotation axis 22 to be eliminated. That is, in the fifth step, the grindstone 21 is rotated until no processing resistance is generated (i.e., until the grindstone starts to rotate).

[0035] The sixth step of S6 is a step of performing finish grinding of the workpiece W. In the sixth step, the stage 11 is moved in the X-axis direction at a slower speed than in the fourth step, and the workpiece W is reciprocated in the X-axis direction relative to the grindstone 21.

[0036] The seventh step of S7 is a step of rotating the grindstone 21 while stopping the feeding of the grindstone 21 (stopping the movement of the feeding mechanism 24 and the stage 11). In the seventh step, similar to the fifth step, the grindstone 21 is rotated and the process waits for the deflection in the direction of the rotation axis 22 to be eliminated.

[0037] 1 again, the display device 40 is, for example, a liquid crystal display. The display device 40 displays various information in accordance with a control signal from the PLC 30. For example, the PLC 30 outputs a control signal to the display device 40 so as to display the process currently being executed during the execution of the grinding process. The display device 40 displays the process currently being executed by the processing device 2 in accordance with the control signal from the PLC 30, so that, for example, an operator can recognize the progress of the grinding process.

[0038] Here, the grindstone 21 may deteriorate with use. When the grindstone 21 deteriorates, for example, in a certain step of the grinding process (for example, the fourth step), the amount of contact between the grindstone 21 and the workpiece W becomes smaller than expected, and the step may not be performed appropriately. Generally, an abnormality may be detected when the vibration detected by the sensor 14 exceeds a predetermined vibration amount, but such a detection method may not be able to detect the above-mentioned case where the vibration becomes smaller as an abnormality.

[0039] Therefore, the abnormality detection system 1 according to the present embodiment includes a PC 3 configured to acquire vibration data during grinding from a sensor 14 of the processing device 2 and estimate which process of the grinding process the vibration data corresponds to. The PLC 30 detects an abnormality when a process that it instructs each device (workpiece support device 10, grindstone driving device 20) of the processing device 2 to execute (hereinafter also referred to as an "instructed process") differs from a process estimated by the PC 3 (hereinafter also referred to as an "estimated process").

[0040] The PC 3 includes a CPU 3a and a memory 3b. The memory 3b stores an estimation model learned by machine learning. The estimation model receives vibration data from the sensor 14 (or data obtained by performing FFT on the vibration data as described later) as input, and estimates which process of the grinding process the input data indicates vibration from. In other words, the estimation model estimates the process being executed by the processing device 2 based on the input data. The details of the estimation model will be described later.

[0041] The PC 3 is communicatively connected to the PLC 30 of the processing device 2 via a communication cable 4. The CPU 3a of the PC 3 estimates the process being executed by the processing device 2 by inputting the vibration data acquired from the sensor 14 to an estimation model. The PC 3 outputs the estimated process (estimated process) to the processing device 2 via the communication cable 4.

[0042] 3 is a block diagram for explaining an example of the functional configuration of the PC 3. A CPU 3a of the PC 3 functions as an acquisition unit 301, a processing unit 303, and an output unit 305 by executing a program stored in a memory 3b.

[0043] The acquiring unit 301 acquires vibration data from the sensor 14. For example, the acquiring unit 301 starts acquiring the vibration data from the sensor 14 on the condition that a start signal output from the processing device 2 at the start of grinding is received. The start signal may be output in conjunction with the above-mentioned start button provided on the processing device 2, for example.

[0044] The processing unit 303 inputs the vibration data acquired by the acquisition unit 301 into the estimation model stored in the memory 3b, and obtains an estimation result. Since the contact state between the workpiece W and the grinding wheel 21 is different in each process of the grinding process, the process that may generate the vibration data can be estimated by using the vibration data. Note that data obtained by applying FFT to the vibration data can also be used as an input to the estimation model. When data obtained by applying FFT to the vibration data is used, the processing unit 303 executes FFT on the vibration data acquired by the acquisition unit 301, and frequency-resolves the vibration data. This makes it possible to obtain spectrum data of the vibration (amplitude) for each frequency. The processing unit 303 then inputs the frequency-resolved data into the estimation model to obtain an estimation result.

[0045] The output unit 305 outputs the estimation result to the PLC 30. Specifically, the output unit 305 outputs to the PLC 30 an estimated process that is estimated to be being executed by the processing device 2, based on the vibration data or data obtained by performing FFT on the vibration data.

[0046] 4 is a block diagram for explaining an example of a functional configuration of the PLC 30. A CPU 30a of the PLC 30 functions as an acquisition unit 31, a process management unit 33, a comparison unit 35, and an output unit 37 by executing a program stored in a memory 30b.

[0047] The acquisition unit 31 requests the PC 3 to notify the estimated process and acquires the estimated process from the PC 3. The acquisition unit 31 outputs the estimated process acquired from the PC 3 to the comparison unit 35. The acquisition unit 31 outputs a notification to the process management unit 33 at the timing when the acquisition unit 31 requests the PC 3 for the estimated process.

[0048] The process management unit 33 manages the process being executed by the processing device 2. Specifically, the process management unit 33 manages the instructed process that the PLC 30 instructs each device (the workpiece supporting device 10, the grindstone driving device 20) of the processing device 2. Upon receiving the notification from the acquisition unit 31, the process management unit 33 outputs the instructed process at that time to the comparison unit 35.

[0049] The comparison unit 35 compares the estimated process step output from the acquisition unit 31 with the instructed process step output from the process management unit 33, and determines whether or not they match. The comparison unit 35 outputs the comparison result to the output unit 37.

[0050] The output unit 37 outputs a control signal to the display device 40 based on the comparison result received from the comparison unit 35. For example, if the estimated process and the instructed process are different, it is determined that some abnormality has occurred in the processing device 2, and the output unit 37 outputs a control signal to the display device 40 for displaying the abnormality. If the estimated process and the instructed process match, it is determined that the grinding process is being performed normally, and the output unit 37 outputs a control signal to the display device 40 for displaying the name of the matched process.

[0051] <Estimation model> FIG. 5 is a diagram for explaining the learning process of the estimation model. The estimation model stored in the memory 3b of the PC 3 according to the present embodiment is a trained model trained by machine learning. More specifically, the estimation model according to the present embodiment is a trained model trained by deep learning. The learning process of the estimation model will be explained with reference to FIG. 5.

[0052] Machine learning is a method of iteratively learning based on given information (for example, a learning data set) and independently establishing rules or judgment criteria. Deep learning is machine learning that uses a multi-layered neural network. It is also possible to apply support vector machines, etc. as machine learning.

[0053] The estimation model is generated by repeatedly executing a learning process using a training data set that includes, for example, a plurality of pieces of training data.

[0054] Each of the learning data includes information indicating the process name (any of the first to seventh processes) of the grinding process as the correct answer data, and vibration data when the process with that process name is performed. In other words, the vibration data when a certain process is performed is labeled with information indicating the process name of the certain process, and the vibration data is prepared as the learning data.

[0055] The learning data can be prepared, for example, by consistently acquiring vibration data during grinding processing, dividing the acquired vibration data into vibration data for each process (process 1 to process 7), and labeling each divided vibration data with the process name (information indicating the process).

[0056] FIG. 5 shows, as an example, an example in which the learning process is performed using learning data in which the fourth step is labeled as correct answer data. Vibration data or data obtained by performing FFT on vibration data is input to the input layer as an input (feature). In response to this input, the output layer outputs the possibility (score) that the input data represents each step. The score is expressed, for example, as a probability. In FIG. 5, scores a1 to a7 are output as the scores for the first to seventh steps, respectively.

[0057] Once the scores are calculated, the error between the scores and the labels (correct answers) is calculated, and the internal parameters of the estimation model are updated so that the calculated error is reduced. Specifically, the internal parameters are updated so that each calculated score approaches its corresponding correct answer.

[0058] The internal parameters are updated, for example, by updating the weighting coefficients and biases in each layer (input layer, intermediate layer, and output layer). The output value y of each node in each layer is expressed by the total product of the input X (X1 to Xn) from each node in the previous layer and the weighting coefficient K (K1 to Kn), plus a bias B, and further applying the activation function h(x), as shown in the following formula (1).

[0059] y=h[{(K1×X1+K2×X2+…+Kn×Xn)+B}]…(1) Specifically, in the output layer, for example, focusing on the fourth node N4 in the output layer, the output value (score) a4 of node N4 is expressed by the following equation (2) by substituting the inputs x (x1 to xn) from each node P2-1 to P2-n in the previous layer (the second layer of the intermediate layer), the weighting coefficients k (k1 to kn), and the bias b4 into equation (1).

[0060] a4=h[{(k1×x1+k2×x2+…+kn×xn)+b4}]…(2) The weighting coefficient K and bias B in each layer (input layer, intermediate layer, and output layer) are updated so that the error between this output value a4 and the label becomes smaller. The same is true for other nodes in the output layer.

[0061] By performing the learning process on the estimation model as described above, the estimation model learns the vibrations (vibration pattern and amplitude) of each process of the grinding process and becomes able to estimate, from the input vibration data (or data obtained by applying FFT to the vibration data), the process that may generate the vibration data (or data obtained by applying FFT to the vibration data).

[0062] Fig. 6 is a diagram for explaining the trained estimation model. When vibration data or data obtained by performing FFT on vibration data is input, the trained estimation model calculates the possibility (score) that the input data represents each process. The score is expressed, for example, as a probability.

[0063] In Fig. 6, scores b1 to b7 are calculated as the scores for the first to seventh steps, respectively. The estimation model estimates the step showing the highest score among the calculated scores as the step indicated by the input data (vibration data or data obtained by performing FFT on the vibration data). In other words, if it is assumed that score b4 is the highest score among scores b1 to b7, the estimation model outputs the fourth step as the step indicated by the input data.

[0064] The estimated process estimated as described above is output to the PLC 30. Then, in the PLC 30, this estimated process is compared with the instruction process that the PLC 30 has instructed each device (work support device 10, grinding wheel drive device 20) of the processing device 2, and an abnormality is detected based on the comparison result (whether or not they match).

[0065] <Processing Executed by PLC and PC> FIG. 7 is a flowchart showing the procedure of the processing executed by the PLC 30 and the PC 3 according to the embodiment. The processing shown in this flowchart starts along with the start of the grinding process.

[0066] When the grinding process is started, the PLC 30 transmits a start signal indicating that the grinding process has started to the PC 3 (S11).

[0067] When the PC 3 receives the start signal from the PLC 30, it starts acquiring vibration data from the sensor 14 (S21). The PC 3 accumulates the vibration data acquired from the sensor 14.

[0068] Next, the PLC 30 transmits a request signal requesting notification of the estimation process to the PC 3 (S12). Note that in S12, the PLC 30 stores the instruction process at this time.

[0069] Upon receiving the request signal, the PC3 estimates the process currently being performed by the processing device 2 (S23). Specifically, the PC3 extracts vibration data for a predetermined period from the accumulated vibration data, counting from the time when the request signal was received. More specifically, for example, it is assumed that the accumulation of vibration data is started at time t1 (S21) and the request signal is received at time t2. In this case, if the predetermined period is Lx, the vibration data from time t2-Lx to time t2 is extracted from the accumulated vibration data. Note that, if the total time for accumulating the vibration data does not reach the predetermined period, all the accumulated vibration data may be extracted. Then, the PC3 inputs the extracted vibration data into an estimation model and estimates which process of the grinding process the extracted vibration data indicates (obtains an estimated process). Note that the PC may perform FFT on the extracted vibration data and input the FFT-applied data into the estimation model to obtain an estimated process.

[0070] When the estimated process step is obtained, the PC 3 notifies the PLC 30 of the estimated process step (S25). When the PLC 30 receives the estimated process step, the PLC 30 compares the instructed process step with the estimated process step and determines whether or not the two steps match (S13).

[0071] If the instructed process and the estimated process do not match (NO in S13), the PLC 30 determines that some abnormality has occurred in the grinding process, and outputs the abnormality (S14). For example, the PLC 30 causes the display device 40 to display that an abnormality has occurred. Note that the output of the abnormality is not limited to displaying the abnormality on the display device 40, and may be any other method that can notify the operator. For example, a buzzer provided in the processing device 2 may sound, or a warning light may be turned on.

[0072] On the other hand, if the instructed process and the estimated process match (YES in S13), the PLC 30 skips S14 and advances the process to S15.

[0073] In S15, the PLC 30 judges whether or not all the grinding processes have been completed. If all the grinding processes have not been completed (NO in S15), the PLC 30 returns the process to S12. On the other hand, if all the grinding processes have been completed (YES in S15), the PLC 30 transmits an end signal indicating that all the grinding processes have been completed to the PC 3 (S16).

[0074] When the PC 3 receives the end signal, it stops obtaining the vibration data from the sensor 14 and ends the process (S27).

[0075] As described above, the anomaly detection system 1 according to this embodiment includes the processing device 2 that performs grinding (first to seventh steps) on the workpiece W, and the PC 3 that estimates which step of the grinding process the processing device 2 is performing. The PC 3 acquires vibration data from the sensor 14 provided on the processing device 2, and inputs the acquired vibration data into an estimation model learned by deep learning, thereby estimating which step the vibration data belongs to. This makes it possible to estimate the step being performed on the workpiece W based on the vibration data.

[0076] Then, the PLC 30 compares the instructed process with the estimated process estimated by the PC 3 to determine whether there is a difference between the two. If there is a difference between the two, it is determined that some abnormality has occurred in the processing device 2, and the abnormality is detected. For example, it is possible to detect a case where the amount of contact between the grinding wheel 21 and the workpiece W is smaller than expected due to deterioration of the grinding wheel 21, or a case where the amount of contact between the grinding wheel 21 and the workpiece W is larger or smaller than expected due to a control error and / or a manufacturing error, etc.

[0077] In addition, by operating the abnormality detection system 1 from the start to the end of the grinding process, even if a process skip occurs in the grinding process, the instructed process and the estimated process will not match, so this can be detected as an abnormality.

[0078] The abnormality detection system 1 according to the present embodiment can detect an abnormality in the processing device 2 by using the estimated process estimated by the PC3.

[0079] <Experimental Results> The inventors conducted an experiment by inputting vibration data for each grinding process into PC3 equipped with a trained estimation model to see whether PC3 can accurately estimate the process. Figure 8 is a chart for explaining the experimental results. Figure 8 shows the probability of the estimated process output by PC3 for the vibration data for each process input into PC3 (estimation model).

[0080] For example, when the vibration data of the first process is input to PC3, the probability that the input vibration data is estimated to be from the first process is 98%, the probability that the input vibration data is estimated to be from the third process is 1%, and the probability that the input vibration data is estimated to be from the seventh process is 1%. In other words, the input vibration data is estimated to be from the first process with a probability of 98%. Also, when the vibration data of the second process is input to PC3, the input vibration data is estimated to be from the second process with a probability of 82%.

[0081] The average accuracy rate of the overall output (estimated process) for the input vibration data was over 83%, indicating that the process was accurately estimated with a high probability. Therefore, it can be said that the method can be used to detect anomalies in grinding processes, for example. Furthermore, it is expected that the accuracy rate will further improve as machine learning is improved in the future.

[0082] In the above experiment, the grinding process was divided into seven steps, from step 1 to step 7, but there are cases where detailed division into steps is not necessary. For example, the grinding process may be divided into three main steps, step 1, step 4, and step 6.

[0083] Fig. 9 is a chart showing the experimental results when grinding is roughly divided into three processes. As in Fig. 8, Fig. 9 shows the probability of the estimated process output by PC3 for the vibration data of each process input to PC3 (estimation model).

[0084] For example, when vibration data of the first process was input to PC3, it was estimated with a probability of 90.45% that the input vibration data was from the first process.

[0085] The average accuracy rate of the overall output (estimated process) for the input vibration data was over 91%, indicating that the process could be estimated with a high degree of probability. Therefore, it can be said that the method can be used to detect anomalies in grinding processes, for example. Furthermore, it is expected that the accuracy rate will further improve as machine learning is improved in the future.

[0086] [Variation 1] In the embodiment, the PLC 30 acquires the estimated process from the PC 3 and determines whether the estimated process and the instructed process match. However, the PC 3 may determine whether the estimated process and the instructed process match. In the first modification, an example in which the PC 3 determines whether the estimated process and the instructed process match will be described.

[0087] 10 is a flowchart showing the procedure of the process executed by the PLC 30 and the PC 3 according to Modification 1. The process shown in this flowchart starts when the grinding process starts.

[0088] When the grinding process is started, the PLC 30 transmits a start signal indicating that the grinding process has started to the PC 3 (S31).

[0089] When the PC 3 receives the start signal from the PLC 30, the PC 3 starts acquiring vibration data from the sensor 14 (S41). The PC 3 accumulates the vibration data acquired from the sensor 14.

[0090] Next, the PLC 30 transmits a signal indicating the currently instructed process to the PC 3 (S32). That is, the PLC 30 notifies the PC 3 of the process that is currently being instructed to be executed by each device (the workpiece supporting device 10, the grindstone driving device 20) of the processing device 2.

[0091] When the signal indicating the instructed process is received, the PC 3 estimates the process currently being performed by the processing device 2 (S42). Specifically, the PC 3 extracts vibration data for a predetermined period from the accumulated vibration data, counting from the time when the signal indicating the instructed process was received, inputs the extracted vibration data to an estimation model, and estimates which grinding process the extracted vibration data indicates (obtains an estimated process). The PC may also perform an FFT on the extracted vibration data, and input the FFT-applied data to the estimation model to obtain the estimated process.

[0092] When the estimated process step is obtained, the PC 3 compares the estimated process step with the instructed process step (S43), and then transmits the comparison result in S43 to the PLC 30 (S44).

[0093] When the comparison result is received, the PLC 30 judges whether the comparison result indicates a match between the estimated process and the specified process, or whether the comparison result indicates a mismatch between the estimated process and the specified process (S33). If the comparison result indicates a mismatch (NO in S33), the PLC 30 judges that some abnormality has occurred in the grinding process, and outputs an abnormality (S34). On the other hand, if the comparison result indicates a match (YES in S33), the PLC 30 skips S34 and proceeds to S35.

[0094] In S35, the PLC 30 judges whether or not all the grinding processes have been completed. If all the grinding processes have not been completed (NO in S35), the PLC 30 returns the process to S32. On the other hand, if all the grinding processes have been completed (YES in S35), the PLC 30 transmits an end signal indicating that all the grinding processes have been completed to the PC 3 (S36).

[0095] When the PC 3 receives the end signal, it stops obtaining the vibration data from the sensor 14 and ends the process (S45).

[0096] As described above, it is also possible to determine whether the estimated process and the instructed process match with each other by using the PC 3. Even in this case, the same effects as those of the embodiment can be achieved.

[0097] [Variation 2] In the embodiment and the first modified example, when the estimated process and the instructed process match, it is determined that the grinding process is performed normally. However, as explained in Fig. 6 of the embodiment, the PC3 calculates a score for each process for the input vibration data, and sets the process with the highest score as the estimated process.

[0098] Referring again to FIG. 6, for example, it is assumed that the score b4 of the fourth process is the highest score among the calculated scores. In this case, the fourth process is output as the estimated process. However, it is also assumed that the score b4 is a relatively low score. In such a case, there is a possibility that the process estimation for the input vibration data is inappropriate. If anomaly detection is performed using an inappropriate estimated process, it may lead to erroneous detection of anomaly, and the accuracy of anomaly detection may decrease.

[0099] Therefore, in the second modification, in addition to comparing the estimated process with the command process, the score of the estimated process is compared with a preset threshold. Specifically, even if the estimated process and the command process match, the PLC 30 detects an abnormality if the score of the estimated process is equal to or less than the threshold. The threshold can be set appropriately depending on the device and / or environment to which the anomaly detection system 1 is applied.

[0100] Fig. 11 is a flowchart showing the procedure of the process executed by PLC 30 and PC 3 according to Modification 2. This flowchart is obtained by adding the process of S50 to the flowchart of Fig. 7 and changing the process of S25 to a process of S60. As the other processes are similar to those in the flowchart of Fig. 7, the same numbers are used and the description thereof will not be repeated.

[0101] When the PC3 obtains the estimated process in S23, the PC3 transmits the estimated process and its score to the PLC 30 (S60). To explain this in detail using the example of Fig. 6, if the score b4 of the fourth process is the highest score among the calculated scores, the estimated process (fourth process) and its score b4 are transmitted to the PLC 30.

[0102] The PLC 30 determines whether the estimated process and the instructed process match (S13), and if they match (YES in S13), determines whether the score of the estimated process is equal to or less than a threshold (S50).

[0103] If the score is equal to or less than the threshold (YES in S50), the PLC 30 advances the process to S14 and outputs an abnormality, whereas if the score is greater than the threshold (NO in S50), the PLC 30 skips S14 and advances the process to S15.

[0104] As described above, the anomaly detection system 1 according to the second modification compares the score of the estimated process with a preset threshold value in addition to comparing the estimated process with the instructed process. If the score of the estimated process is equal to or less than the threshold value, that is, if there is a possibility that the process estimation is inappropriate, an anomaly is output. This makes it possible to prevent erroneous detection of an anomaly due to an inappropriate estimated process. Therefore, it is possible to prevent a decrease in the accuracy of anomaly detection.

[0105] It should be noted that Modification 1 can be applied to Modification 2. That is, the comparison between the estimation process and the instruction process, and the determination of whether the score of the estimation process is equal to or less than the threshold value may be performed by the PC3.

[0106] The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present invention is defined by the claims, not the above description, and is intended to include all modifications within the meaning and scope of the claims. [Explanation of symbols]

[0107] 1 Anomaly detection system, 2 Processing device, 3 PC, 3a CPU, 3b memory, 4 Communication cable, 10 Work support device, 11 Stage, 12 Fixing device, 13 Rotating axis, 14 Sensor, 20 Grinding wheel drive device, 21 Grinding wheel, 22 Rotating axis, 23 Motor, 24 Mechanism, 30 PLC, 30a CPU, 30b Memory, 31 Acquisition unit, 33 Process control unit, 35 Comparison unit, 37 Output unit, 40 Display device, 301 Acquisition unit, 303 Processing unit, 305 Output unit.

Claims

1. A processing device that performs grinding or cutting on an object to be processed by sequentially executing a plurality of predetermined steps; An estimation device that inputs data on vibrations generated in the processing device during the processing into an estimation model trained by machine learning to estimate which process the processing device is executing, An anomaly detection system in which an anomaly is detected when an instructed process to be executed by the processing device differs from an estimated process estimated by the estimation device.

2. The estimation model is generated by a learning process using a learning dataset, The training data set includes a plurality of training data, The anomaly detection system according to claim 1 , wherein each of the plurality of learning data includes information indicating a process and data related to the vibration when the process is performed.

3. the estimation model outputs, for each process, a likelihood indicating which process among the plurality of processes the input relates to; 3. The anomaly detection system according to claim 1, wherein the estimation device determines a process having a highest probability among the output possibilities as the estimated process.

4. The anomaly detection system according to claim 3 , wherein even if the indication step and the estimation step coincide, an anomaly is detected if the likelihood of the estimation step is less than a threshold value.

5. 3. The anomaly detection system according to claim 1, wherein the data regarding the vibration is data obtained by accumulating vibrations occurring in the processing device over a predetermined period of time, or data obtained by performing an FFT (Fast Fourier Transform) on the accumulated data.

Images