JP2008087094A - Tool attaching abnormality detecting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To detect a problem of an attaching state of a tool, or an attachment of a wrong type tool while the tool is attached to a driving device. <P>SOLUTION: In a machine tool wherein a tool Y is detachably attached to a driving device X, a competitive learning type neural network 1 classifies the characteristic amount of output of a vibration sensor 2 closely attached to the driving device X, thereby detecting the problem of the attaching state of the tool Y to the driving device X. For the detection of the attaching state, the output of the vibration sensor 2 while the tool Y is attached to the driving device X and made to idle is used. If the characteristic amount of the output of the vibration sensor 2 is given to the competitive learning type neural network 1 after learning, it can be classified whether the attaching state of the tool Y is normal or abnormal. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a tool mounting abnormality detection that detects an abnormality such as a failure in the mounting state of a tool or a wrong type of tool in a machine tool in which a tool can be attached to and detached from a driving device having a driving source such as a motor. It relates to the device.

  Conventionally, in an automatic tool changer for a numerically controlled machine tool, a device for monitoring an abnormality in a change operation of a tool changer arm has been proposed (for example, see Patent Document 1). This device detects the torque waveform pattern of the drive motor at each position during the exchange operation of the tool exchange arm, and compares it with the torque waveform pattern of the drive motor during normal operation of the tool exchange arm. It is to detect. To compare torque waveform patterns, the error amount of the torque value at the same position during the tool change arm change operation is obtained, and whether there is an abnormality in the tool change operation is determined by whether this error amount is within the allowable range. Yes.

The device described in Patent Document 1 solves the technical problem of preventing damage to machine tools and workpieces by monitoring the presence or absence of abnormal tool change operations caused by manufacturing errors of the tool change arm. Yes. In Patent Document 1, a torque waveform pattern corresponding to a tool to be replaced is stored for a plurality of tools stored in a tool magazine in consideration of the weight difference of the tool, and is adjusted to the tool to be replaced at the time of tool replacement. The use of a torque waveform pattern is also described.
Japanese Patent Laid-Open No. 11-333657

  By the way, the technique described in Patent Document 1 detects the presence or absence of an abnormality in the automatic tool changer. If the automatic tool changer is operating normally, the tool is normally attached to the drive unit. It is based on the premise that However, even when an automatic tool changer is used, the work for storing the tool in the tool magazine is performed manually, and therefore the tool location and orientation may be wrong when storing the tool in the tool magazine. In such a case, even if the automatic tool changer is operating normally, the tool is erroneously attached to the drive unit, so that there is a problem that normal machining is not performed. Therefore, it is desired to detect an abnormality such as a failure in the mounting state of the tool or a wrong type of tool in a state where the tool is mounted on the driving device.

  The present invention has been made in view of the above-mentioned reasons, and the purpose of the present invention is to detect abnormalities in the installation state of the tool or the wrong type of tool when the tool is attached to the drive device. It is to provide a detection device.

  The invention of claim 1 comprises a vibration sensor for detecting vibration generated from the drive device in a machine tool in which the tool is detachably attached to the drive device, and a plurality of parameters from a target signal which is an output of the vibration sensor. A feature extraction unit that extracts a feature amount; a neural network that classifies the category of the feature amount extracted by the feature extraction unit; and a determination unit that determines a state of attachment of the tool to the drive device using a classification result by the neural network. The neural network uses the feature quantity of the target signal when the tool is idled with the tool attached to the drive device as learning data, and learning is performed with the tool attachment state as a category, and the determination unit drives the tool. The feature value of the target signal obtained when idling while attached to the device is inspected in the neural network after learning. Attachment state of the tool by the classification result of the neural network when given by the and judging whether normal or abnormal.

  According to a second aspect of the present invention, in the first aspect of the invention, the neural network uses, as learning data, only the feature amount when the state of attachment of the tool with respect to the drive device is normal, and the determination unit uses the learned neural network. It is characterized in that when the neuron that fires when the inspection data is given to the network does not belong to the normal category, it is determined that the attachment state of the tool to the drive device is abnormal.

  According to a third aspect of the present invention, in the second aspect of the present invention, the machine tool is configured to automatically replace a plurality of tools stored in a tool magazine in a specified order, and the neural network is classified by tool. The learning unit is configured to change the neurons belonging to the normal category in synchronization with an instruction to change the tool in the machine tool.

  According to a fourth aspect of the present invention, in any one of the first to third aspects, the neural network is a competitive learning type neural network.

  According to the configuration of the first aspect of the present invention, the vibration of the driving device when the tool is idled with the tool normally attached to the driving device is detected by the vibration sensor, and the feature amount of the output of the vibration sensor at this time On the other hand, when network learning is performed, and abnormalities related to tool attachment are determined, the feature value of the output of the vibration sensor when the tool is attached to the drive unit and idled is classified by the neural network. It is possible to determine whether or not there is an abnormality in the state before mounting the workpiece and machining the workpiece. In other words, the wrong type of tool attached to the drive unit, the failure of the tool mounting state such as the displacement of the tool mounting position on the drive unit, or the wrong mounting direction of the tool with respect to the drive unit, or the damage of the tool It is possible to detect an abnormality of a driving device or the like as an abnormality.

  According to the configuration of the invention of claim 2, since only the feature amount when the tool mounting state is normal is used as the learning data and it is determined as abnormal except for normal, the type of abnormality cannot be classified. Learning data can be easily collected, and the period from introduction of the device to actual use can be shortened. In addition, by determining that an abnormality other than normal is abnormal regardless of the type of abnormality, even if the type of abnormality is unknown, it is determined on the safe side, and the possibility of damage to the machine tool or workpiece can be reduced.

  According to the configuration of the invention of claim 3, even when the types of tools are different, it is possible to respond by changing the neurons belonging to the normal category, so that a machining operation is performed by automatically exchanging a large number of tools as in a machining center. Even in this case, it is possible to monitor the abnormality related to the attachment of the tool for each tool.

  According to the configuration of the invention of claim 4, since a competitive learning type neural network is used, the configuration is simple, and learning data for each category is collected, and learning is simply performed by simply providing learning data for each category. Can do.

  In the embodiment described below, as a machine tool, a tool to be used is automatically selected from a plurality of tools held in a tool magazine, such as a machining center and a turning center, according to a program, and the selected tool is used as a drive device. Although what is automatically mounted is assumed, even if it is a single-function machine tool, the technical idea of this invention is applicable if a tool is detachably mounted with respect to a drive device. The drive device may be any device that uses a motor as a drive source, and an appropriate transmission mechanism such as a gear box or a belt may be provided between the drive source and the tool. In the following, a spindle including a housing is assumed as a driving device.

  As shown in FIG. 1, the tool attachment abnormality detection apparatus described in the present embodiment uses an unsupervised competitive learning type neural network (hereinafter simply referred to as “neural network” unless otherwise required) 1. . As a neural network, it is possible to use a back-propagation type with supervision, but a competitive learning type neural network is simpler in configuration than a back-propagation type neural network, and learning data for each category is used. This method is suitable for this kind of application in that it only needs to be used for learning, and learning can be strengthened by additional learning even after learning once.

  As shown in FIG. 2, the neural network 1 includes two layers of an input layer 11 and an output layer 12, and each neuron N <b> 2 of the output layer 12 is coupled to all the neurons N <b> 1 of the input layer 11. Have. Although the neural network 1 is assumed to be realized by executing an appropriate application program on a sequential processing type computer, a dedicated neurocomputer can also be used.

  The operation of the neural network 1 includes a learning mode and an inspection mode. After learning using appropriate learning data in the learning mode, a feature amount (inspection) including a plurality of parameters generated from an actual target signal in the inspection mode. Data) category.

  The degree of connection (weighting coefficient) between the neuron N1 in the input layer 11 and the neuron N2 in the output layer 12 is variable. In the learning mode, the learning data is input to the neural network 1 to learn the neural network 1, and the input layer The weighting coefficients of the 11 neurons N1 and the neurons N2 of the output layer 12 are determined. In other words, each neuron N2 in the output layer 12 is associated with a weight vector whose element is a weight coefficient between each neuron N1 in the input layer 11. Therefore, the weight vector has the same number of elements as the neuron N1 of the input layer 11, and the number of parameters of the feature quantity input to the input layer 11 matches the number of elements of the weight vector.

  On the other hand, in the inspection mode, when the inspection data whose category is to be determined is given to the input layer 11 of the neural network 1, among the neurons N2 of the output layer 12, the neuron N2 having the smallest Euclidean distance between the weight vector and the inspection data. set a fire. If a category is associated with the neuron N2 of the output layer 12 in the learning mode, the category of the inspection data can be known from the category of the position of the fired neuron N2.

  The neuron N2 of the output layer 12 is associated one-to-one with each region of the two-dimensional clustering map 4 having, for example, 6 × 6 regions. Therefore, if the learning data category is associated with each area of the clustering map 4 in the learning mode, the category corresponding to the neuron N2 fired by the inspection data can be known from the clustering map 4. The clustering map 4 functions as an output unit that outputs the classification result obtained by the neural network 1.

  When associating a category with each region of the clustering map 4 (substantially each neuron N2 of the output layer 12), the learned neural network 1 is operated in the opposite direction from the output layer 12 toward the input layer 11, and output. The data given to the input layer 11 is estimated for each neuron N2 of the layer 12, and the category of learning data having the closest Euclidean distance to the estimated data is used as the category of the neuron N2 in the output layer 12. In other words, as the category of each neuron N2 in the output layer 12, the category of learning data having the minimum Euclidean distance from the weight vector of each neuron N2 is used. Thereby, the category of learning data is reflected in the category of each neuron N2 of the output layer 12.

  By giving a large number (for example, 150) of learning data for each category, a category having a high similarity is arranged at a close position on the clustering map 4. That is, among the neurons N2 in the output layer 12, the neurons N2 fired corresponding to the learning data belonging to the same category gather at close positions on the clustering map 4 to form a cluster composed of the set of neurons N2.

  Note that the clustering map 4 having the original meaning is formed after the learning and the clusters are associated with each other. However, in the present embodiment, even before the learning, the clustering map 4 is called and is not particularly distinguished. The learning data given to the neural network 1 in the learning mode is stored in the learning data storage unit 5 and is read from the learning data storage unit 5 and given to the neural network 1 as necessary.

  By the way, since the information to be detected by the neural network 1 is whether or not the tool Y is normally attached to the drive device X, it is sufficient that the category can be classified as normal or abnormal. When the type of abnormality is to be detected, the type of abnormality is also a category. In addition, since a plurality of tools Y are replaced and mounted on the driving device X, it is necessary to distinguish between normal and abnormal for each tool Y. Therefore, if an attempt is made to identify the type of abnormality, many types of categories are required, and it takes time and effort to collect learning data. Therefore, in this embodiment, only the normal state category is set for each tool Y, and normal When deviating from the state, it is determined as abnormal. However, when learning data can be collected for various abnormal states, the type of abnormality may be set as a category.

  In order to determine the mounting state of the tool Y with respect to the driving device X, the output of the vibration sensor 2 attached so as to be in close contact with the driving device X is used as a target signal to be inspected. As the vibration sensor 2, an acceleration pickup is used, and the vibration sensor 2 is attached in close contact with the outer surface of the housing 13 of the driving device X as shown in FIG. In attaching the vibration sensor 2 to the drive device X, it is required that the operation of the drive device X is not hindered and the attachment state is not changed by the operation of the drive device X. Therefore, the vibration sensor 2 is fixed to the housing 13 of the driving device X by adhesion or welding. Further, in order to enable the vibration sensor 2 to be attached and detached, a configuration in which a holder that removably couples the vibration sensor 2 is fixed to the housing 13 may be employed.

  The electrical signal output from the vibration sensor 2 is given to the feature extraction unit 3 as a target signal, and the feature extraction unit 3 extracts a feature quantity having a plurality of parameters for the target signal. The feature amount can be appropriately extracted according to the attribute of interest in the target signal, but in this embodiment, attention is paid to the frequency component (power for each frequency band) of the target signal. Since the drive device X is a spindle, the output of the vibration sensor 2 has periodicity, but the extracted feature value changes depending on which part on the time axis the feature value is extracted from the vibration sensor 2 output. Therefore, before the feature extraction unit 3 extracts the feature quantity, a pre-process for aligning the parts from which the feature quantity is extracted from the output of the vibration sensor 2 is necessary.

  Therefore, the feature extraction unit 3 uses a timing signal (trigger signal) synchronized with the operation of the drive device X, or uses a waveform characteristic of the target signal (for example, a start point and an end point of a group of target signals). As a result, the target portion output from the vibration sensor 2 is cut out (segmented) in the time axis direction. Here, at the time of segmentation, a signal for one section may not be extracted from the target signal in one cycle, but may be divided into a plurality of sections for each appropriate unit time. The feature extraction unit 3 extracts a set of feature amounts including a plurality of parameters for each section.

  In order to perform the above-described preprocessing, the feature extraction unit 3 includes a buffer that temporarily stores a target signal given from the vibration sensor 2. In the preprocessing of the feature extraction unit 3, noise is reduced by limiting the frequency band as necessary. Furthermore, it also has a function of converting the target signal output from the vibration sensor 2 into a digital signal.

  In order to extract frequency components in the feature extraction unit 3, an FFT (Fast Fourier Transform) technique or a filter bank made up of a large number of bandpass filters is used. Which frequency is used for the feature amount is appropriately selected according to the target drive device X and the abnormality to be extracted.

  The feature amount obtained from the feature extraction unit 3 is stored in the learning data storage unit 5 when learning data is collected before the learning mode, and the neural network 1 is extracted every time the feature amount is extracted in the inspection mode. The neural network 1 classifies the category of the inspection data by using the feature amount as the inspection data.

  Here, the data stored in the learning data storage unit 5 is referred to as a data set, and each data included in the data set corresponds to a state in which the tool Y is normally attached to the driving device X. That is, it is assumed that the learning data storage unit 5 stores a data set of normal category data. The number of data constituting the data set is arbitrary as long as it can be stored in the learning data storage unit 5, but it is desirable to use about 150 data for learning of the neural network 1 as described above.

  Since the learning data storage unit 5 stores only the data set associated with the normal category, when learning is performed with each data set stored in the learning data storage unit 5 using the neural network 1 as the learning mode. The neural network 1 performs learning only in the normal state. In other words, since the category of the clustering map 4 is only in the normal state, the operation of setting the category by operating in the reverse direction after learning as described above can be omitted.

  If learning is performed as described above, a weight vector having a weight coefficient as an element with each neuron N1 in the input layer 11 is set in each neuron N2 in the output layer 12. Accordingly, when learning data of one category is given using the neural network 1 as the inspection mode, the neuron N2 corresponding to the category is fired. However, since there is variation in learning data, the number of neurons N2 that fire for one category of learning data (data set) is not limited to one, and a cluster of a plurality of neurons N2 is formed.

  After the learning of the neural network 1 in the learning mode, when the inspection data is given by using the neural network 1 as the inspection mode, the inspection data indicates that the mounting of the tool Y on the driving device X is normal. In the neural network 1, the neuron N2 belonging to the normal category fires, and the clustering map 4 at the position indicates that it is normal. On the other hand, if the test data is not in the normal state, the neuron N2 at the normal position in the clustering map 4 does not fire, so it can be determined that the state is abnormal. That is, the clustering map 4 functions as a determination unit.

  When the determination result by the clustering map 4 indicates an abnormal state, it is desirable to present the user to the abnormal state using appropriate notification means. In order to present an abnormal condition, the lamp may be turned on or an alarm sound may be generated.

  The neural network 1, the feature extraction unit 3, the clustering map 4, and the learning data storage unit 5 are housed in one housing 14 together with the notification unit, and are arranged separately from the vibration sensor 2. The housing 14 and the vibration sensor 2 are connected via a sensor line 15. The sensor wire 15 is attached to the vibration sensor 2 and is detachably connected to the housing 14 by a connector or the like. Further, it is desirable to use a shielded wire for the sensor line 15 so that a noise component is not mixed into the target signal through the sensor line 15.

  In actual use, if the feature amount is extracted from the output of the vibration sensor 2 during the period when the tool Y is cutting the workpiece, the feature amount includes a component due to the difference in the workpiece, and it becomes difficult to classify the category. If the workpiece is machined while the tool Y is not normally attached to the drive device X, the workpiece is not only damaged, but also the drive device X and the tool Y are damaged, so that the tool Y is attached to the drive device X. Before driving the workpiece, the drive device X is idled, and the output of the vibration sensor 2 obtained at this time is used as the target signal. The same applies to the learning data as well as the inspection data. In the inspection mode, when it is detected that the mounting state of the tool Y is abnormal during idling, the drive device X is stopped and the abnormality is notified before the tool Y contacts the workpiece.

  As described above, by determining the attachment state of the tool Y to the driving device X using the output of the vibration sensor 2 at the time of idling, when the tool Y is tilted or the attachment direction of the tool Y is wrong ( It is possible to detect an abnormal state such as when the tool Y is rotated 180 degrees in the rotation direction of the tool Y, and to stop the driving device X before the tool Y contacts the workpiece.

  As shown in FIG. 4, when a plurality of types of tools Y are stored in the tool magazine 16, the tools Y mounted on the drive device 4 are sequentially replaced by the tool changer arm 17 in accordance with instructions of a preset program. Prepares a data set of learning data to be stored in the learning data storage unit 5 for each tool Y, and causes the neural network 1 to learn using the learning data for each tool Y. In the neural network 1 learned in this way, in the clustering map 4, there are regions in which categories of normal states are associated with a plurality of types of tools Y, so it is necessary to determine whether each tool Y is normal or abnormal. It is.

  Since which tool Y is attached to the drive device 4 is known by the program, an area selection unit 6 for selecting an area of the clustering map 4 according to an instruction of the program is provided in addition to the clustering map 4 as a determination unit. The region selection unit 6 selects which region of the clustering map 4 is used as a normal state for each tool Y, and the neuron N2 in the region selected in the clustering map 4 is ignited for the inspection data given to the neural network 1. If not, the notification means notifies the abnormality.

  As described above, in a machine tool that automatically selects a plurality of tools Y stored in the tool magazine 16 according to the instructions of the program and mounts them on the drive device X, the tool magazine 16 is manually selected for maintenance or the like. When storing the tool Y, there is a possibility that the storage position of the tool Y may be wrong. However, it is highly possible to detect the error of the tool Y even for this type of error. It is possible to prevent the drive device X from being damaged.

  In the above example, the output of the vibration sensor 2 is used as the target signal. However, when the drive source of the driving device X is a motor, the load current of the motor can be used as the target signal. When the motor is servo-controlled, the output of an encoder provided in the motor may be used as the target signal.

It is a block diagram which shows Embodiment 1 of this invention. It is a schematic block diagram of the neural network used for the same as the above. It is a schematic block diagram same as the above. It is a schematic block diagram same as the above.

Explanation of symbols

1 Neural network (competitive learning type neural network)
2 Vibration sensor 3 Feature extraction unit 4 Clustering map (determination unit)
DESCRIPTION OF SYMBOLS 5 Learning data memory | storage part 6 Area | region selection part 11 Input layer 12 Output layer 13 Housing 14 Case 15 Sensor line 16 Tool magazine 17 Tool exchange arm N1 neuron N2 neuron X Drive device

Claims (4)

  In a machine tool in which a tool is detachably mounted on a drive device, a vibration sensor that detects vibrations generated from the drive device, and feature extraction that extracts a feature amount including a plurality of parameters from a target signal that is an output of the vibration sensor A neural network that classifies the category of the feature amount extracted by the feature extraction unit, and a determination unit that determines the mounting state of the tool on the driving device using the classification result by the neural network. The learning is performed using the feature amount of the target signal when it is idled in a state where it is attached to the drive device as learning data, and the attachment state of the tool is categorized, and the determination unit is idle while the tool is attached to the drive device. When the feature value of the target signal obtained when Mounting abnormality detection apparatus for a tool, characterized in that the classification result of-menu neural network attachment state of the tool to determine whether the normal or abnormal.   The neural network uses, as learning data, only the feature amount when the state of attachment of the tool to the driving device is normal, and the determination unit is normal when the neuron fired when the inspection data is given to the learned neural network The tool attachment abnormality detection device according to claim 1, wherein when it does not belong to the category, it is determined that the attachment state of the tool to the drive device is abnormal.   The machine tool is configured to automatically replace a plurality of tools stored in a tool magazine in an instructed order, the neural network is learned using tool-specific learning data, and the determination unit includes: 3. The tool attachment abnormality detection device according to claim 2, wherein neurons belonging to a normal category are changed in synchronization with an instruction to change a tool in a machine tool.   The tool attachment abnormality detection device according to any one of claims 1 to 3, wherein the neural network is a competitive learning type neural network.

Images