JP7154100B2 - Machining abnormality detection device for machine tools - Google Patents

Description

TECHNICAL FIELD The present invention relates to a device for detecting a processing abnormality in a machine tool that is equipped with a cutting tool to machine a work material.

When a cutting tool is used to machine a work material, tool damage such as breakage of the cutting tool may occur due to overload, biting of chips, progress of wear of the cutting edge, and the like. In this case, the desired machining cannot be achieved and re-machining will take time, and the product will be defective, leading to losses. In addition, there is a risk of collision between the machine and the work material due to uncut parts.
In order to avoid the above danger, a technique has been proposed to detect a machining abnormality and stop the machine.
Patent Document 1 discloses a technique of recording a motor load waveform during normal machining as a reference waveform, and determining that machining is abnormal when the load during machining deviates from the reference waveform by a certain amount or more. According to such a technique, since the waveform during normal machining in the target machining can be used as a reference, abnormal machining can be detected under arbitrary machining conditions. However, since the divergence from the reference waveform is used as an index, there is a problem that the detection sensitivity cannot be increased when there is variation in the waveform during normal machining.

On the other hand, in Patent Document 2, for the purpose of diagnosing plant abnormalities, a signal pattern during abnormal operation is learned by a neural network in advance, and abnormality is detected by comparing the learning result with the signal waveform at the time of diagnosis. A method is proposed. According to such a method, it is possible to learn the "features" that appear in the signal waveform when it is normal or abnormal, and it is possible to detect an abnormality with higher accuracy than when the signal waveform itself is monitored.
However, machining with a machine tool is affected by a large number of parameters as machining conditions. In other words, in addition to the type and size of cutting tools such as drills and end mills, the shape and material of the work material, the rotational speed of the cutting tool or work material, the axial feed speed of the machine tool, the presence or absence of cutting fluid to be used, etc. Machining is affected by the type, relative motion trajectory of the tool and work material, as well as the type and installation environment of the machine to be used. was difficult.

Recent advances in machine learning technology have made it possible to construct models that can be applied to different conditions to some extent in general by training deep neural networks with high expressive power using data collected under multiple conditions. I understand. On the other hand, it is impossible to construct a universal model applicable to all conditions.
In order to address such a problem, there has been proposed a technique of improving diagnostic performance by adding newly collected data and updating a diagnostic model, as in Patent Document 3. Using such techniques is expected to improve the performance of the model.

JP 2014-172102 A JP-A-7-234988 JP 2006-163517 A

However, in order to improve the generalization ability, the number of input data required for learning and the computing power of the computer will increase. In addition, it is difficult to assume that all machining conditions have common "features", and there is a limit to improvement through model generalization.

SUMMARY OF THE INVENTION Accordingly, it is an object of the present invention to provide a machining abnormality detection apparatus capable of detecting machining anomalies with high precision in machining of a machine tool with a large number of combinations of operating conditions.

In order to achieve the above object, the invention according to claim 1 is a machining abnormality detection device for detecting the occurrence of an abnormality in the machining in a machine tool for machining a work material using a cutting tool. ,
Anomaly detection model holding means for holding a plurality of anomaly detection models for diagnosing the presence or absence of the anomaly;
model construction condition holding means for holding, in association with the abnormality detection model, processing conditions used when diagnostic information used to construct the abnormality detection model is collected for each of the abnormality detection models held in the abnormality detection model holding means; When,
Each of the abnormality detection models is associated with the abnormality detection model as to whether it is a generalized model that can be used without being limited to specific machining conditions or a specialized model that can be used only with specific machining conditions. a model property holding means for holding the
Abnormality detection for selecting the specialized model when the specialized model matching the machining conditions of the machining to be diagnosed is held by the abnormality detection model holding means, and otherwise selecting the generalized model. and model selection means.
The invention according to claim 2 is the configuration according to claim 1, wherein the machining conditions are the type of the cutting tool, the shape of the cutting tool, the spindle rotation speed of the machine tool, and the feed speed of the drive shaft of the machine tool. , the presence and type of cutting fluid used in the machining, the shape of the work material, the material of the work material, the trajectory of relative motion between the cutting tool and the work material, and the machine tool to be used. characterized by comprising
The invention according to claim 3 is the configuration according to claim 1 or 2, wherein the abnormality detection model selection means is an NC program used in the machine tool and/or the cutting tool used in the machining held by the machine tool. The machining conditions corresponding to the machining to be diagnosed are specified based on the information about the tool.
According to a fourth aspect of the invention, in any one of the first to third aspects, the abnormality detection model is constructed by machine learning.
According to a fifth aspect of the present invention, in the configuration of any one of the first to fourth aspects, diagnostic information holding means for holding diagnostic information used when diagnosing whether or not there is an abnormality, and when an erroneous diagnosis occurs, it is possible to input an erroneous diagnosis. diagnosis result input means, and when the generalized model is selected by the abnormality detection model selection means, a new specialized model corresponding to the machining conditions at the time of machining is newly created using the diagnostic information. an anomaly detection model updating means for updating the selected specialized model using the diagnostic information when the specialized model is selected by the anomaly detection model selecting means. and

According to the invention of claims 1 to 3, it has both a generalized model that can be used universally for a plurality of machining conditions and a specialized model that is limited to specific machining conditions but has high detection accuracy. Which model should be used during machining can be appropriately selected based on the associated machining conditions during learning. If a high-precision anomaly detection model corresponding to the machining conditions has already been constructed, the specialized model is appropriate, and if the machining conditions are unknown, a generalized model that has inferior performance but can be used for general purposes is appropriate. can be selected to Therefore, in the machining of a machine tool that involves a huge number of combinations of operating conditions, the abnormality detection capability can be easily maximized and the machining abnormality can be detected with high accuracy.
According to the invention of claim 4 or 5, in addition to the above effects, when the anomaly detection model being selected cannot be correctly identified, without excessively requesting the number of data and computational capacity required for learning, Anomaly detection performance can be improved.

1 is a block configuration diagram of a machine tool; FIG. 6 is a flowchart of processing abnormality detection processing; 8 is a flowchart of anomaly detection model selection processing;

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block configuration diagram showing an example of the processing abnormality detection device of the present invention. A cutting tool 2 is attached to the spindle of the machine tool 1 so that a work material 3 on a table can be machined.
A machining abnormality detection device 10 is installed side by side with this machine tool 1 . However, the machining abnormality detection device 10 may be included in a control device (not shown) of the machine tool 1 .
The processing abnormality detection device 10 is provided with diagnostic information acquiring means 11 for acquiring control information of the machine tool 1 and measurement signals of various sensors (not shown) as diagnostic information. The diagnostic information also includes attribute information (tool type, tool shape, material, etc.) of the cutting tool 2 held by the control device of the machine tool 1 and attribute information (material, shape, etc.) of the work material 3 .
When starting machining, the machine tool 1 reads an NC program describing the details of machining into the controller, interprets it, and determines the operation of the machine. The NC program stores information related to machining, such as the rotational speed of the cutting tool 2, the feed shaft speed of the machine tool 1, and whether or not cutting fluid is used. In general, the control device also stores the attribute information of the cutting tool and the work material mounted on the machine tool.

The diagnostic information acquisition means 11 acquires the above information from the control device of the machine tool 1 and sends it to the abnormality detection model selection means 12 . Based on the sent information, the abnormality detection model selection means 12 selects the type, shape, and dimensions of the cutting tool 2, the shape and material of the work piece 3, the rotation speed of the cutting tool 2, and the feed of the drive shaft of the machine tool 1. Specify processing conditions such as speed and whether or not cutting fluid is used. Depending on the information held by the machine tool 1, the type of cutting fluid, attribute information of the machine tool 1 (lathe, machining center, vertical, horizontal, etc.), installation environment information, and the like may be added to the machining conditions.
The anomaly detection model holding means 13 holds a plurality of anomaly detection models for diagnosing the presence or absence of anomaly, which are pre-constructed by machine learning. It is specified as either a generalized model that can be used for diagnosing the presence or absence of an abnormality without being limited to machining conditions, or a specialized model that can be used for diagnosing the presence or absence of an abnormality by limiting to specific machining conditions.
The model construction condition holding means 15 associates the processing conditions used when diagnostic information used for constructing the abnormality detection model with respect to each abnormality detection model held by the abnormality detection model holding means 13 is associated with the abnormality detection model. holding.
Further, the processing abnormality detection device 10 includes diagnosis result input means 18 for allowing the operator to input the occurrence of misdiagnosis, abnormality detection model updating means 17 for updating the abnormality detection model based on the input, and A diagnostic information storage means 19 is provided for storing the diagnostic information obtained.

Processing abnormality detection processing in the processing abnormality detection apparatus 10 configured as described above will be described based on the flowchart of FIG.
First, when machining is started in S1, diagnostic information is acquired from the control device of the machine tool 1 by the diagnostic information acquiring means 11 in S2. Next, in S3, the abnormality detection model selection means 12 specifies machining conditions based on the diagnostic information.
Next, in S4, an appropriate abnormality detection model is selected from the abnormality detection model holding means 13 based on the machining conditions specified in S3.

Selection of this abnormality detection model is performed based on the flowchart shown in FIG.
First, in S41, for each of the anomaly detection models held in the anomaly detection model holding means 13, the corresponding model characteristics (generalized model or specialized model) are obtained from the model characteristic holding means 14, and the model construction conditions (model The processing conditions used for collecting the teacher data used for construction) are obtained from the model construction condition holding means 15 .
Next, in S42, it is determined whether or not there is a specialized model that matches the machining conditions specified in S3. If there is a matching specialized model, that specialized model is adopted as an anomaly detection model in S43. On the other hand, if there is no matching specialized model, the generalized model is adopted as the anomaly detection model in S44.

Here, the generalized model refers to an abnormality detection model that can be used for general purposes without depending on specific machining conditions. In general, building a general-purpose model requires a large amount of machining data and a high-performance computer, so machine tool manufacturers provide it in advance. The intended range of "general purpose" does not necessarily include all machining conditions, such as "abnormality detection model for drills", "abnormality detection model for vertical machining centers", "abnormality detection model for aluminum material", etc. A model that includes a plurality of machining conditions may be treated as a "generalized model".
On the other hand, a specialized model refers to a model specialized for machining conditions that are often used by each machine tool user. For example, it is a model that can be used under limited conditions, such as using a specific machining program to machine a specific work material with a tool of a specific type and shape.

In S5, the machining abnormality diagnosis means 16 inputs the diagnosis information of the machine tool 1 obtained from the diagnosis information acquisition means 11 to the abnormality detection model selected by the abnormality detection model selection means 12, thereby determining the presence or absence of machining abnormality. Diagnose. As a result of the diagnosis, if it is determined in S6 that no machining abnormality has occurred, it is checked in S7 whether or not machining has been completed. That is, diagnosis is always performed during machining.
On the other hand, if a machining abnormality is detected in the determination in S6, in S8, the control device of the machine tool 1 is notified of the abnormality, and suitable abnormality processing such as issuing an alarm or stopping machining is performed. In addition, the diagnostic information used for diagnosis is stored in the diagnostic information holding means 19 in order to prepare for use in updating an abnormality detection model, which will be described later.

As described above, in the processing abnormality detection apparatus 10 of the above embodiment, the abnormality detection model holding means 13 for holding a plurality of abnormality detection models for diagnosing the presence or absence of an abnormality, and the individual abnormality detection models held by the abnormality detection model holding means 13 With respect to the detection model, a model construction condition holding means 15 for holding processing conditions when diagnostic information used for building an abnormality detection model is collected in association with the abnormality detection model, and an abnormality detection model holding means 13 for holding individual Regarding the anomaly detection model of , whether it is a generalized model that can be used without limiting to specific machining conditions or a specialized model that can be used by limiting to specific machining conditions is stored in association with the anomaly detection model. If a specialized model matching the machining conditions of the machining to be diagnosed is held by the model characteristic holding means 14 and the abnormality detection model holding means 13, the specialized model is held, and if not, the generalized model. It is characterized by comprising an anomaly detection model selection means 12 that selects.
As a result, it has both a generalized model that can be used for multiple machining conditions and a specialized model that is limited to specific machining conditions but has high detection accuracy. which model should be used during processing can be appropriately selected based on If a high-precision anomaly detection model corresponding to the machining conditions has already been constructed, the specialized model is appropriate, and if the machining conditions are unknown, a generalized model that has inferior performance but can be used for general purposes is appropriate. can be selected to Therefore, in the machining of a machine tool that involves a huge number of combinations of operating conditions, the abnormality detection capability can be easily maximized and the machining abnormality can be detected with high accuracy.

In addition, in the processing abnormality detection apparatus 10 of the above-described embodiment, misdiagnosis (oversight or overdetection of abnormality) may occur when the performance of the abnormality detection model is insufficient. In that case, it is possible to continuously improve the anomaly detection performance by updating the model.
As a means of detecting misdiagnosis, in addition to the operator's impression (e.g., an "abnormality" alarm was issued even though no abnormality was found in the tool, or the tool was broken without an "abnormality" alarm being issued), The quality determination result of a device for determining whether the workpiece is machined inside or outside the machine tool 1, such as a dimension measuring device (not shown), can be used. As the diagnostic result input means 18, a button that can be pressed by the operator or means for exchanging signals with a dimension measuring device or the like can be provided.

When the diagnosis result input means 18 inputs the occurrence of misdiagnosis, the abnormality detection model updating means 17 updates the abnormality detection model. At this time, if the abnormality detection model selected by the processing abnormality diagnosis means 16 at the time of misdiagnosis is a generalized model, re-learning of the generalized model requires cost for both input data and calculation load. To construct. On the other hand, if the specialized model is selected, the specialized model is re-learned using diagnostic information used as input at the time of diagnosis as additional teacher data. Since the specialized model is limited to the target machining conditions, the scale of the model is generally not so large, and the model can be sufficiently implemented with a general personal computer or the like.

In this way, the abnormality detection model selection means 12 is provided with the diagnosis information holding means 19 for holding the diagnosis information used when diagnosing whether or not there is an abnormality, and the diagnosis result input means 18 capable of inputting the misdiagnosis when an erroneous diagnosis occurs. If the generalized model was selected in , a new specialized model corresponding to the machining conditions during machining was constructed using the diagnostic information, and the specialized model was selected by the abnormality detection model selection means 12. In this case, by further providing an anomaly detection model updating means 17 for updating the selected specialized model using the diagnostic information, if the anomaly detection model being selected cannot be correctly identified, learning can be performed. Anomaly detection performance can be improved without requiring an excessive amount of data and computational power.

Reference Signs List 1 Machine tool 2 Cutting tool 3 Work material 10 Machining abnormality detection device 11 Diagnosis information acquisition means 12 Abnormality detection model selection means 13 Abnormality detection model Holding means 14 Model characteristics holding means 15 Model construction condition holding means 16 Machining abnormality diagnosis means 17 Abnormality detection model update means 18 Diagnosis result input means 19 Diagnosis information holding means.

Claims (5)

In a machine tool that processes a work material using a cutting tool, a machining abnormality detection device for detecting the occurrence of an abnormality in the machining,
Anomaly detection model holding means for holding a plurality of anomaly detection models for diagnosing the presence or absence of the anomaly;
model construction condition holding means for holding, in association with the abnormality detection model, processing conditions used when diagnostic information used to construct the abnormality detection model is collected for each of the abnormality detection models held in the abnormality detection model holding means; When,
Each of the abnormality detection models is associated with the abnormality detection model as to whether it is a generalized model that can be used without being limited to specific machining conditions or a specialized model that can be used only with specific machining conditions. a model property holding means for holding the
Abnormality detection for selecting the specialized model when the specialized model matching the machining conditions of the machining to be diagnosed is held by the abnormality detection model holding means, and otherwise selecting the generalized model. A machining abnormality detection device for a machine tool, comprising: model selection means; The machining conditions include the type of the cutting tool, the shape of the cutting tool, the spindle rotation speed of the machine tool, the feed rate of the drive shaft of the machine tool, the presence and type of cutting fluid used in the machining, and the workpiece. 2. The machine tool according to claim 1, characterized by including at least one of the shape of the material, the material of the work material, the locus of relative motion between the cutting tool and the work material, and the machine tool to be used. Machining abnormality detection device. The abnormality detection model selection means selects the machining condition corresponding to the machining to be diagnosed based on the NC program used by the machine tool and/or the information on the cutting tool used for the machining held by the machine tool. 3. The abnormal machining detection device for a machine tool according to claim 1 or 2, wherein: 4. A machining abnormality detection device for a machine tool according to claim 1, wherein said abnormality detection model is constructed by machine learning. diagnostic information holding means for holding diagnostic information used when diagnosing the presence or absence of an abnormality;
a diagnostic result input means capable of inputting a misdiagnosis when a misdiagnosis occurs,
When the generalized model is selected by the abnormality detection model selection means, the diagnosis information is used to construct a new specialized model corresponding to the machining conditions during machining, and the abnormality detection model selection means 5. An abnormality detection model updating means for updating the selected specialized model using the diagnostic information when the specialized model is selected in the above. 1. A processing abnormality detection device for a machine tool according to claim 1.

Images