JP7748871B2 - Machining abnormality detection method and machining abnormality detection device - Google Patents

Description

This disclosure relates to a method and apparatus for detecting machining abnormalities from machining data in machine tools.

For example, there is a technology that uses machining data related to machine control acquired from a numerical control device of a machine tool to identify machining abnormalities. Machining data refers to information such as motor rotation command values and position command values, and torque data of the motor that controls the machine, which change as machining progresses. Machining abnormalities can be detected by reading an algorithm derived from a prior analysis of the machining data or a machine learning, probabilistic, or statistical model that has learned normal machining data, predicting the machining data, and determining the prediction error, which is the difference between the predicted value and the actual value, as the degree of abnormality. The degree of abnormality may be a vector for the number of pieces of machining data used, or may be a single norm.
5 shows the configuration of a conventional machining abnormality detection device, which includes a numerical control device (hereinafter abbreviated as "NC") 1a, a machining abnormality degree calculation unit 1b, a machining abnormality determination unit 1c, and a threshold value storage unit 1d.
In order to determine machining abnormalities, the machining abnormality degree calculation unit 1b inputs the machining data acquired from the NC 1a into an algorithm derived by analyzing the machining data or a machine learning, probabilistic, or statistical model that has learned normal machining data, and calculates the degree of machining abnormality.
The machining abnormality determination unit 1c compares the degree of machining abnormality acquired from the machining abnormality degree calculation unit 1b with a preset threshold value, and determines that a machining abnormality has occurred when the degree of machining abnormality exceeds the threshold value.
The threshold value storage unit 1d stores a threshold value.

6 shows the flow of the method for detecting a machining abnormality in the machining abnormality detection device. First, in S1, the machining abnormality degree calculation unit 1b calculates the degree of machining abnormality from machining data created by the NC 1a as machining progresses.
Next, in S2, the machining abnormality determination unit 1c compares the degree of machining abnormality calculated by the machining abnormality degree calculation unit 1b with the threshold value acquired from the threshold value storage unit 1d, and if the threshold value is larger than the degree of machining abnormality (No in S2), it continues machining, but if it is smaller (Yes in S2), it assumes that a machining abnormality has occurred and automatically interrupts or stops machining, or replaces the tool, etc.
When using a machine learning, probabilistic, or statistical model that has learned normal machining data in this way for machining abnormality detection, in order to improve the accuracy of detecting machining abnormalities, a plurality of models that have learned machining data under specific machining conditions may be prepared, and a model may be selected depending on the current machining, as shown in Patent Document 1. The machining conditions are determined by the type of tool, the type of workpiece, etc.

Japanese Patent Application Laid-Open No. 2019-67136

When detecting machining abnormalities using algorithms derived from prior analysis of machining data or machine learning, probabilistic, and statistical models that have learned from normal machining data, the degree of abnormality when machining the same part next time will change depending on the accuracy of the previous machining stitches.
If the tool used in the previous machining operation has wear that does not require abnormality detection, the wear will cause a negligible change in the accuracy of the machining stitches. However, depending on the sensitivity of the algorithm or model, an abnormality may be detected unnecessarily in the next machining operation. This can lead to unintended machine shutdowns or the replacement of tools that are still usable, resulting in a decrease in the machine's operating rate. The technology disclosed in Patent Document 1 only performs abnormality detection that takes into account the current machining operation, and is therefore unable to detect abnormalities that take into account the conditions before and after machining.
Conversely, if the tool used in the previous machining operation has wear that requires abnormality detection, the necessary abnormality may not be detected due to the sensitivity of the algorithm or model, and an abnormality may be detected in the next machining operation. In this case, even though the tool used in the previous machining operation is the one to be excluded as a cause of the machining abnormality, the tool used in the next machining operation may be examined, which may lead to an incorrect response. If the response is incorrect, for example, a machining abnormality may occur again even after the tool used in the next machining operation is replaced. In other words, a tool that has not experienced wear may be replaced as an abnormality, while a tool that has experienced wear may be used without being replaced. This may cause poor machining accuracy again, accelerating the wear of the next tool and resulting in the use of more tools. Furthermore, if a machining abnormality occurs again, time is spent identifying the cause and taking measures, resulting in a deterioration in machine availability and machining efficiency. The technology disclosed in Patent Document 1 cannot identify the underlying machining process that caused the detection of the machining abnormality or the tool used in that process.

The present disclosure therefore aims to provide a machining abnormality detection method and device that can detect machining abnormalities while taking into account abnormalities in the tool before and after use in machining.

In order to achieve the above object, a first configuration of the present disclosure is a machining abnormality detection method for determining a machining abnormality from machining information when machining is performed using a plurality of tools based on a numerical control command, the method comprising:
a previous machining information acquisition step of acquiring previous machining information, which is information related to the tool or machining information which is data associated with the tool, and which is information related to machining abnormality determination of a previously used tool;
a current machining information acquisition step of acquiring current machining information, which is information regarding machining abnormality determination of a currently used tool that is different from the previously used tool, among the machining information;
a current processing information correcting step of correcting the current processing information using the pre-processing information;
an abnormality determination step of determining whether or not there is a machining abnormality based on the current machining information corrected in the current machining information correction step;
If a machining abnormality is determined in the abnormality determination step, a tool identification step is executed to identify the tool causing the machining abnormality based on the previous machining information and the current machining information .
Another aspect of the first configuration is characterized in that, in the above configuration, in the tool identification step, likelihoods of tool abnormality determination are calculated based on the previous machining information and the current machining information, and the calculated likelihoods are compared to identify the tool that is causing the machining abnormality.
Another aspect of the first configuration is characterized in that, in the above configuration, a tool graph update step is further executed in which the previous machining information and the current machining information are held as a tool graph having a graph structure, and the tool graph is updated in accordance with the progress of machining.
Another aspect of the first configuration is characterized in that, in the above configuration, the tool graph updating step automatically updates the tool graph by using a machine learning algorithm using a directed graph.

In order to achieve the above object, a second configuration of the present disclosure is a machining abnormality detection device that determines a machining abnormality from machining information when machining is performed using a plurality of tools based on a numerical control command,
a previous processing information acquiring means for acquiring previous processing information, which is information relating to the tool or processing information which is data associated with the tool, and which is information relating to processing abnormality determination of a previously used tool;
a current machining information acquiring means for acquiring current machining information, which is information relating to machining abnormality determination of a currently used tool different from the previously used tool, from among the machining information;
a current processing information correcting means for correcting the current processing information using the previous processing information;
an abnormality determination means for determining whether or not there is a machining abnormality based on the current machining information corrected by the current machining information correction means;
The present invention is characterized by comprising a tool identification means for identifying the tool causing the machining abnormality based on the previous machining information and the current machining information when the abnormality determination means determines that a machining abnormality has occurred.
Another aspect of the second configuration is characterized in that, in the above configuration, the tool identification means calculates likelihoods of tool abnormality determination based on the previous machining information and the current machining information, and compares the calculated likelihoods to identify the tool that is causing the machining abnormality.
Another aspect of the second configuration is characterized in that, in the above configuration, the configuration further comprises a tool graph update means for holding the previous machining information and the current machining information as a tool graph having a graph structure and updating the tool graph in accordance with the progress of machining.
Another aspect of the second configuration is that, in the above configuration, the tool graph update means automatically updates the tool graph by using a machine learning algorithm that uses a directed graph.

According to the present disclosure, by creating and managing data linking before and after a machining process, it becomes possible to perform anomaly detection that takes into account abnormalities in the tools used before and after machining. Therefore, even if there is a change in the machining marks that does not require detection due to minor wear or the like of the previous tool, machining abnormalities for the next tool can be robustly determined.
Furthermore , when a machining abnormality is determined, the tool causing the machining abnormality is identified based on the previous machining information and the current machining information, so that it is possible to estimate the underlying machining process that caused the detection of the machining abnormality and the tool used during that process, thereby improving the continuity of machining.
According to another aspect of the present disclosure, in addition to the above effects, the tool is identified by comparing the likelihood of tool abnormality determination, making it possible to trace back past machining processes and estimate the cause of the machining abnormality.
According to another aspect of the present disclosure, in addition to the above effects, the tool graphs of the previous machining information and the current machining information are updated according to the progress of machining, so that in estimating the cause of a machining abnormality, the estimation accuracy can be optimized according to the progress of machining, and when a machining abnormality occurs, the tool that contributed to the detection of the abnormality can be estimated.

FIG. 1 is a block diagram of a machining abnormality detection device according to a first embodiment. 1 is a flowchart of a processing abnormality detection method according to the first embodiment. FIG. 10 is a block diagram of a processing abnormality detection device of a second embodiment. 10 is a flowchart of a processing abnormality detection method according to a second embodiment. FIG. 1 is a block diagram of a conventional machining abnormality detection device. 1 is a flowchart of a conventional method for detecting abnormalities in machining.

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.
[Form 1]
Fig. 1 shows an example of the configuration of a machining abnormality detection device 10. This machining abnormality detection device 10 differs from the conventional machining abnormality detection device shown in Fig. 5 in that it includes a current tool abnormality degree data storage unit 2a, a previous tool abnormality degree data storage unit 2b, an abnormal tool estimation unit 2c, and a correction amount acquisition unit 2d. The machining abnormality determination unit 1c is an example of the abnormality determination means of the present disclosure.
The current tool abnormality degree data storage unit 2a stores the number or ID of the tool currently being used for machining, the latest abnormality degree which is the machining abnormality degree calculated the last time machining was performed with that tool, and the initial abnormality degree which is the machining abnormality degree calculated the first time machining was performed with that tool. The current tool abnormality degree data storage unit 2a, together with the machining abnormality degree calculation unit 1b, is an example of a current machining information acquisition means of the present disclosure.
The previous tool abnormality degree data storage unit 2b stores abnormality degree data (including the latest abnormality degree and the initial abnormality degree) of the tool used in the immediately preceding machining. The previous tool abnormality degree data storage unit 2b, together with the machining abnormality degree calculation unit 1b, is an example of a previous machining information acquisition means of the present disclosure.
The abnormal tool estimating unit 2c estimates which tool has an abnormality when the machining abnormality determining unit 1c determines that a machining abnormality has occurred. The abnormal tool estimating unit 2c is an example of a tool identifying means of the present disclosure.
The correction amount acquiring unit 2d calculates and acquires a correction amount by which the degree of machining abnormality of the tool used immediately before acts on the machining abnormality degree calculated for the current machining, from the ratio between the initial abnormality degree stored in the previous tool abnormality degree data storing unit 2b and the latest abnormality degree. The correction amount acquiring unit 2d is an example of a current machining information correcting means of the present disclosure.

FIG. 2 shows the flow of a machining abnormality detection method in the machining abnormality detection device 10 of FIG.
First, in S11, the machining abnormality calculation unit 1b calculates the machining abnormality degree from machining data created by the NC 1a as machining progresses.
Meanwhile, in S12, the NC 1a determines whether or not a tool has just been replaced, and if so, in S13, copies the current tool abnormality data in the current tool abnormality data storage unit 2a to the previous tool abnormality data in the previous tool abnormality data storage unit 2b (previous processing information acquisition step).
Next, in S14, the number or ID of the tool currently being used for machining, obtained from NC 1a, is stored in the current tool abnormality data storage unit 2a, and in S15, the machining abnormality degree calculated in S11 is stored in the current tool abnormality degree data storage unit 2a as the initial abnormality degree immediately after tool replacement (current machining information acquisition step).

On the other hand, if it is determined in S12 that the tool has not been replaced immediately, in S16 the machining abnormality degree calculated in S11 is stored in the current tool abnormality degree data storage unit 2a as the latest abnormality degree obtained during the last machining operation using the current tool (current machining information acquisition step).
Next, in S17, the correction amount acquisition unit 2d calculates a correction amount for offsetting the degree of abnormality of the tool used immediately before from the ratio between the initial abnormality degree stored in the previous tool abnormality degree data storage unit 2b and the latest abnormality degree, and the calculated correction amount is applied to the latest abnormality degree acquired in S16 to determine the corrected machining abnormality degree (current machining information correction step).
Next, in S18, the corrected machining abnormality degree is compared with the threshold value acquired from the threshold value storage unit 1d (abnormality determination step), and if the threshold value is greater than the corrected machining abnormality degree (No in S18), it is determined that there is no machining abnormality and machining is continued.On the other hand, if the determination in S18 shows that the threshold value is smaller than the corrected machining abnormality degree (Yes in S18), it is determined that there is a machining abnormality, and in S19, the rate at which the correction amount contributed to the detection of the machining abnormality is obtained to identify the tool causing the abnormality (tool identification step), and the interruption or stop of machining, or replacement of the identified tool, etc. is automatically performed.

The abnormality level held in the current tool abnormality level data and previous tool abnormality level data may be a scalar such as the maximum value of the machining abnormality level in the machining unit, or a machining abnormality level vector for the number of time-series samples. In the former case, the contribution rate calculated in S19 is, for example, a comparison of the ratio between the initial abnormality level and the latest abnormality level in the previous tool abnormality level data and the ratio between the initial abnormality level and the corrected machining abnormality level in the current tool abnormality level data. In the latter case, for example, a comparison of the cosine similarity between the initial abnormality level and the latest abnormality level in the previous tool abnormality level data and the cosine similarity between the initial abnormality level and the corrected machining abnormality level in the current tool abnormality level data. The tool with the largest contribution rate is the tool causing the abnormality.

In this way, the machining abnormality detection method (an example of the first configuration) and device (an example of the second configuration) of form 1 above acquire previous tool abnormality degree data (an example of previous machining information), which is information related to machining abnormality judgment of a previously used tool, and acquire current tool abnormality degree data (an example of current machining information), which is information related to machining abnormality judgment of a currently used tool, correct the current tool abnormality degree data using the previous tool abnormality degree data, and determine the presence or absence of a machining abnormality based on the corrected machining abnormality degree (an example of corrected current machining information).
According to this configuration, by creating and managing anomaly data that links before and after the machining process, it becomes possible to perform anomaly detection that takes into account abnormalities in the tools before and after the machining. Therefore, even if there is a change in the machining marks that does not require detection due to slight wear and tear on the previous tool, it is possible to robustly determine machining abnormalities for the next tool.
In particular, when a machining abnormality is determined, the tool causing the machining abnormality is identified based on the previous tool abnormality degree data and the current tool abnormality degree data, so that it is possible to estimate the underlying machining process that caused the detection of the machining abnormality and the tool used during that process, thereby improving the continuity of machining.

[Form 2]
Next, another configuration of the present disclosure using a directed graph will be described, where the same components as those in the first embodiment are denoted by the same reference numerals and redundant description will be omitted.
Fig. 3 shows an example of the configuration of the machining abnormality detection device 20. Unlike the machining abnormality detection device 10 shown in Fig. 1, the machining abnormality detection device 20 does not include the current tool abnormality degree data storage unit 2a and the previous tool abnormality degree data storage unit 2b, but instead includes a tool graph update unit 3a, a tool graph storage unit 3b, and a machining process chart storage unit 3c. These are examples of the tool graph holding means of the present disclosure.
The tool graph update unit 3a creates a tool graph in which data associated with tools used to machine the same machining portion is used as a node and edges connect the nodes in the order in which the tools are used. The tool graph update unit 3a also updates the tool graph using a machine learning algorithm.
The tool graph storage unit 3b stores the tool graphs created or changed by the tool graph update unit 3a.
The machining process chart storage unit 3c stores a machining process chart to be referenced for creating a tool graph.

A tool graph is a directed graph that represents the order of machining operations before machining from a machining program or machining process chart. Each node in this tool graph has a one-to-one number or ID corresponding to each individual tool, and also has a probability table showing the initial abnormality level of the tool, the abnormality probability of the tool, and other nodes connected to the edge of the node. For example, this probability table stores the conditional abnormality probability of tool B being abnormal after tool A is used when machining a part to be machined with tool B after tool A is used, for the number of combinations of each node. Note that the abnormality probability is the sum of all conditional abnormality probabilities, so the abnormality probability of each previous tool is required. However, since the tool that performs machining first in a machining process is not affected by the conditions of other tools, each abnormality probability can be calculated recursively from that tool.
The edge has a correction amount that increases or decreases the effect of a machining abnormality caused by that tool on the abnormality of the next tool used. By applying this correction amount to the machining abnormality degree of the next tool, an abnormality degree that is used to detect machining abnormalities taking into account before and after the process can be obtained. The correction amount depends on the abnormality probability of each tool.
In this form 2, when an actual machining abnormality occurs, the part is to probabilistically infer which tool abnormality is the root cause of the abnormality, so the tool graph is configured using a method that allows Bayesian estimation and similar estimations, such as a Bayesian network.

FIG. 4 shows the flow of a machining abnormality detection method in the machining abnormality detection device 20 of FIG.
First, in S21, the machining abnormality calculation unit 1b calculates the machining abnormality degree from the machining data created by the NC 1a as machining progresses. Meanwhile, in S22, the NC 1a determines whether machining has started, a tool has been added, or a tool has been removed. If machining has started or something similar has occurred (Yes in S22), in S23, the tool graph update unit 3a reads the process sheet from the machining process sheet storage unit 3c and creates or updates the tool graph (pre-machining information acquisition step and tool graph update step). If a tool has been replaced or an initial abnormality degree, abnormality probability, and probability table for the replaced tool are initialized, and the probability table for the node with an edge directed from that tool is also initialized. Furthermore, if a tool has been added or removed, the node or edge for that tool is added or removed, and entries in the probability table for the remaining nodes are also added or deleted. If machining has not started or something similar has occurred, the process proceeds to S24.
Next, in S24, it is determined whether the tool currently being used for machining is a tool being used for the first time. If it is a tool being used for the first time (Yes in S24), in S25, the machining abnormality degree calculated in S21 immediately after the tool change is set as an initial abnormality degree in a node of the tool graph (current machining information acquisition step). If it is not a tool being used for the first time, the process proceeds to S26.
In S26, the machining abnormality degree determined in S21 is input to a node of the tool graph to update the abnormality probability of the current tool, and in S27, the conditional abnormality probabilities of the current tool and the immediately preceding tool are updated (tool graph update step).

Next, in S28, the correction amount obtained by the correction amount acquisition unit 2d from the abnormality probability of the tool used immediately before is applied to the machining abnormality degree obtained in S21 to obtain the corrected machining abnormality degree (current machining information correction step).
Next, in S29, the corrected machining abnormality degree is compared with the threshold value acquired from the threshold value storage unit 1d (abnormality determination step), and if the threshold value is greater than the corrected machining abnormality degree (No in S29), machining is continued.
On the other hand, if the threshold is smaller than the corrected machining abnormality degree (Yes in S29), it is deemed to be a machining abnormality, and in S30, the node storing the data of the tool that detected the abnormality is referenced from the number or ID of that tool, and the likelihood that the previous tool is the cause when an abnormality is detected in a subsequent tool is calculated from a probability table containing conditional abnormality probabilities and the abnormality probability for each tool. The tool with the highest probability is estimated to be the tool that is the root cause of the machining abnormality (tool identification step). If the abnormality probability is highest, the tool that detected the machining abnormality is estimated to be the root cause. Then, automatic measures such as interrupting or stopping machining, or replacing the tool estimated to be the root cause are performed.

In this way, the machining abnormality detection method and device of the above-mentioned form 2 acquires data from the previous tool node (an example of previous machining information), which is information related to the machining abnormality judgment of the previously used tool, and acquires data from the current tool node (an example of current machining information), which is information related to the machining abnormality judgment of the currently used tool, corrects the data from the current tool node using the data from the previous tool node, and judges whether or not there is a machining abnormality based on the corrected machining abnormality level (an example of corrected current machining information).
According to this configuration, by creating and managing anomaly data that links before and after the machining process, it becomes possible to perform anomaly detection that takes into account abnormalities in the tools before and after the machining. Therefore, even if there is a change in the machining marks that does not require detection due to slight wear and tear on the previous tool, it is possible to robustly determine machining abnormalities for the next tool.

In particular, when a machining abnormality is determined, the tool causing the machining abnormality is identified based on the conditional abnormality probability held in the node of the current tool, so that the underlying machining process that caused the detection of the machining abnormality and the tool used in that process can be estimated, thereby improving the continuity of machining. At this time, the likelihood that the previous tool was the cause when the abnormality was detected with the current tool is calculated, and the calculated likelihoods are compared to identify the tool causing the machining abnormality, so that it is possible to trace back past machining processes and estimate the cause of the machining abnormality.
Furthermore, the abnormality level data for each tool is held as a tool graph with a graph structure, and the tool graph is updated as the machining progresses. This allows the accuracy of estimation of the cause of a machining abnormality to be optimized as the machining progresses, and when a machining abnormality occurs, it is possible to estimate the tool that contributed to the detection of the abnormality.
Therefore, by operating and maintaining the tool graph, it is possible to detect machining abnormalities taking into account the machining process even if the machining target, the tool use order, or the workpiece material is changed.

In the above-mentioned form 2, if an abnormality is detected in a certain tool, it is assumed that the cause is wear on the tool immediately before that. However, if the tool wear is caused by an abnormality in the machining process even earlier than that of the tool in question, the cause will be traced back to the analysis of the tool used in that previous process to further investigate the cause. Therefore, a tool graph that includes tools two or more steps before the currently used tool is also meaningful. However, if estimation is made using only the tool currently being used for machining and the tool that machined the same machining area immediately before that, anomaly detection can be performed in the same way by having multiple graphs with two nodes and one edge for each machining area.

1a: Numerical control device, 1b: Machining abnormality calculation unit, 1c: Machining abnormality judgment unit, 1d: Threshold value storage unit, 2a: Current tool abnormality data storage unit, 2b: Previous tool abnormality data storage unit, 2c: Abnormal tool estimation unit, 2d: Correction amount acquisition unit, 3a: Tool graph update unit, 3b: Tool graph storage unit, 3c: Machining process chart storage unit, 10, 20: Machining abnormality detection device.

Claims (8)

A machining abnormality detection method for determining a machining abnormality from machining information when machining is performed using a plurality of tools based on a numerical control command, comprising:
a previous machining information acquisition step of acquiring previous machining information, which is information related to the tool or machining information which is data associated with the tool, and which is information related to machining abnormality determination of a previously used tool;
a current machining information acquisition step of acquiring current machining information, which is information regarding machining abnormality determination of a currently used tool that is different from the previously used tool, among the machining information;
a current processing information correcting step of correcting the current processing information using the pre-processing information;
an abnormality determination step of determining whether or not there is a machining abnormality based on the current machining information corrected in the current machining information correction step;
a tool specifying step of specifying a tool causing the machining abnormality based on the previous machining information and the current machining information when the abnormality determination step determines that a machining abnormality has occurred. 2. The machining abnormality detection method according to claim 1, wherein in the tool identifying step, likelihoods of tool abnormality determination are calculated based on the previous machining information and the current machining information, and the calculated likelihoods are compared to identify the tool causing the machining abnormality. The machining abnormality detection method according to claim 1 or 2, further comprising a tool graph update step of holding the previous machining information and the current machining information as a tool graph having a graph structure and updating the tool graph according to the progress of machining. 4. The machining abnormality detection method according to claim 3 , wherein the tool graph updating step automatically updates the tool graph by using a machine learning algorithm that uses a directed graph. A machining abnormality detection device that determines machining abnormalities from machining information when machining is performed using a plurality of tools based on a numerical control command,
a pre -processing information acquiring means for acquiring pre-processing information, which is information relating to the tool or processing information which is data associated with the tool, and which is information relating to processing abnormality judgment of a previously used tool;
a current machining information acquiring means for acquiring current machining information, which is information relating to machining abnormality determination of a currently used tool different from the previously used tool, from among the machining information;
a current processing information correcting means for correcting the current processing information using the previous processing information;
an abnormality determination means for determining whether or not there is a machining abnormality based on the current machining information corrected by the current machining information correction means;
and a tool identifying means for identifying a tool causing the machining abnormality based on the previous machining information and the current machining information when the abnormality determining means determines that a machining abnormality has occurred. 6. The machining abnormality detection device according to claim 5, wherein the tool identification means calculates likelihoods of tool abnormality determination based on the previous machining information and the current machining information, and identifies the tool causing the machining abnormality by comparing the calculated likelihoods. 7. The machining abnormality detection device according to claim 5, further comprising a tool graph update means for holding the previous machining information and the current machining information as a tool graph having a graph structure and updating the tool graph in accordance with the progress of machining. 8. The machining abnormality detection device according to claim 7 , wherein the tool graph update means automatically updates the tool graph by using a machine learning algorithm that uses a directed graph.