Device for estimating tool life

Abstract

A tool life estimation device (100) for estimating a life of a tool used by a machine tool (1) for machining a workpiece, the device comprising: a state observation unit (112) configured to acquire machining information indicating a machining status from log data recorded during operation of the machine tool (1) and recorded in a state where the tool life still remains sufficiently, and configured to create input data based on the acquired machining information; a learning unit (111) configured to construct a learning model in which clusters of the machining information are created, the learning model being constructed by unsupervised learning using the input data created by the state observation unit (112); and a learning model storage unit (114) configured to store the learning model, characterized in that the machining information comprises two sections, a first section corresponding to a first time period of the log data before a predetermined time (t1), and a second section corresponding to a second time period of the log data between the predetermined time (t1) and the end of the tool life, and the tool life estimation device (100) is configured to exclude the second portion of the machining information from use in the learning process.

Description

Background of the inventionField of the invention

[0001] The present invention relates to a device for estimating a tool life and to a machine tool comprising such a device. Description of the related prior art

[0002] Typically, the cutting edge of a tool used in a machine tool wears over time when used in machining, resulting in an increase in cutting resistance. Likewise, as wear progresses, the machining accuracy of the tool deteriorates, making it difficult to maintain the specified machining accuracy required for a workpiece. Thus, the tool reaches the end of its service life. The tool that has reached the end of its service life must be replaced with another one, otherwise, it is impossible to continue machining.However, if the tool reaches the end of its life during automatic machine tool operation, the absence of an operator on-site may prevent the tool from being replaced immediately when the tool reaches its end of life, which is one of the factors hindering the efficiency of the machining cycle. Therefore, a technique for predicting tool life is important.

[0003] It is difficult to estimate the tool life of a machine tool, as it varies depending on the workpiece being machined and the machining conditions. Although it is possible to develop a method for estimating tool life based on machining time and the number of machining passes, the accuracy of the estimation using this method is not high, and in some cases, the operator must check the tool each time to determine the tool life.

[0004] As a conventional technique related to estimating tool life, the Taylor life equation is known (Japanese Patent Application Laid-Open No. JP-H11-170102 A). When estimating tool life using the Taylor life equation, it is possible to estimate the tool life under various machining conditions by defining a constant based on the machining conditions, such as the tool to be used in machining and the workpiece material, and applying the defined constant to the Taylor life equation. Furthermore, a technique for estimating a tool life based on machining time, the number of machining passes, and the like has also been proposed (Japanese Patent Application Laid-Open No. JP-H11-170102 A).JP 2002-224 925 A etc.).

[0005] However, when the tool life is to be predicted using the Taylor life equation, there is a disadvantage in that the constant of the equation must be calculated according to the machining conditions, which makes the determination of the constant complicated for a machine tool whose machining conditions change frequently, which in turn makes it difficult to apply the constant to the equation.

[0006] When estimating tool life based on machining time, number of machining passes, and the like, it is also necessary to record the machining passes and the number of machining passes for each tool. Furthermore, since the estimation method for estimating tool life relies on so-called rule-of-thumb calculations based on actual machining, a problem arises in that predicting tool life is difficult in situations where machining conditions change frequently.

[0007] DE 692 12 048 T2 discloses a system for predicting tool life. This system uses sensors to collect multiple machine data, evaluates them using a computer, and transforms them into feature values. This data is used for tool life prediction, for example, using an influence diagram and categorizing features based on decision boundaries for an impending "long life" or "short life" of the tool.

[0008] Furthermore, a system for optimizing tool maintenance, detecting tool wear, and the frequency of tool correction is known from EP 1 205 830 A1, DE 196 43 383 A1, and DE 10 2016 011 532 A1, respectively. JP 2015 - 203 646 A discloses a training (machine learning) method for lifetime prediction using only normal data (tool data). Summary of the invention

[0009] Therefore, an object of the present invention is to provide a tool life estimating apparatus capable of estimating the life of a tool used in a machine tool according to the change in machining conditions.

[0010] According to the present invention, in a manufacturing facility having a machine tool of the manufacturing industry, machining information indicating a machining status is collected from the machine tool, and a machine learning device is caused to learn a status in which tool life remains based on the machining information that has been collected.When the learning by the machine learning device is completed, the machine learning device is caused to estimate whether the machining status while machining is being performed by the machine tool corresponds to a status in which tool life remains, and the fact that an end of the service life of a tool is approaching is communicated to the machine tool that has been estimated to be in a machining status outside the status in which tool life remains.

[0011] Furthermore, a tool life estimating apparatus according to the present invention that estimates a life of a tool used in machining a workpiece by a machine tool comprises a state observation unit configured to acquire machining information indicating a status of machining in a state where the life of the tool still remains sufficiently, the machining information being acquired from log data recorded while the machine tool is operated, and to create input data based on the acquired machining information;a learning unit configured to build a learning model in which clusters of the machining information are created through unsupervised learning using the input data created by the state observation unit; and a learning model storage unit configured to store the learning model.

[0012] Also, the tool life estimating apparatus according to the present invention, which estimates the life of a tool for use in machining a workpiece by a machine tool, comprises a learning model storage unit configured to store a learning model in which clusters of machining information are created by unsupervised learning based on the machining information indicating a status of machining in a state where the life of the tool still remains sufficiently, the machining information being acquired while the machine tool is operated;a condition observation unit configured to acquire machining information indicating the machining status from log data recorded while the machine tool is operating and to create input data based on the acquired machining information; and an estimation unit configured to estimate the tool life from the input data created by the condition observation unit.

[0013] Furthermore, the machine tool according to the present invention includes an alarm unit configured to issue an alarm based on the results of estimation of the life of the tool by the tool life estimation device.

[0014] According to the present invention, by estimating the tool life using the machine learning device, it is not necessary to calculate a constant according to the machining conditions such as the life equation, and the tool life does not need to be recorded for each machining condition, thus making it possible to estimate the tool life with high accuracy according to various situations. Short description of the drawings

[0015] The above-described and other objects and features of the present invention will become clear from the following description of embodiments with reference to the accompanying drawings. Fig. 1 is a schematic functional block diagram of an apparatus for estimating tool life at a time of learning according to an embodiment of the present invention; Fig. 2 is a diagram for describing machining information for use in a machine learning process according to an embodiment of the present invention; Fig. 3 is a diagram showing an example in the case where a multilayer neural network is used as a learning model; Fig. 4 is a diagram showing an example in the case where an autoencoder is used as a learning model; Fig. 5 is a diagram showing clusters of machining information in the case where tool life still remains; Fig. 6 is a schematic functional block diagram of the tool life estimating apparatus at the time of estimating the tool life according to an embodiment of the present invention; Fig. 7 is a diagram showing the relationship between machining information and the clusters in the case where the tool life is estimated to be still remaining; Fig. 8 is a diagram showing the relationship between the machining information and the clusters in the case where the tool life is estimated to be near its end; and Fig. 9 is a diagram for describing machining information for use in a machine learning process according to another embodiment of the present invention. Detailed description of the preferred embodiments

[0016] Embodiments of the present invention will be described below with reference to the drawings.

[0017] Fig. 1 shows a schematic functional block diagram of a tool life estimation apparatus at the time of learning according to an embodiment of the present invention. The tool life estimation apparatus 100 of this embodiment is configured to perform a machine learning process in a manufacturing facility having one or more machine tools, based on log data collected from at least one of the machine tools and stored in a log data storage unit 200.

[0018] The machining information acquired from a machine tool operating in a manufacturing facility is recorded as log data along with time in the log data storage unit 200. The machining information includes the types of tools used in machining, the workpiece material, the type of coolant, the tool feed rate, the spindle rotation speed, the cutting edge temperature, the cutting time aggregation / cutting distance aggregation for each tool, the cutting resistance (amplifier current value of an axis / spindle), and the like. The log data storage unit 200 can record pieces of machining information collected from a plurality of machine tools as the log data. Furthermore, the log data storage unit 200 can be configured as a typical database.

[0019] In a machine tool as an object from which log data is to be collected, the workpiece is machined while controlling the respective drive units provided in the machine tool, and the states of the drive units and a detection value by a sensor are obtained from signals obtained from the respective units. Log data related to the machining process of the machine tool is created and stored in a non-volatile memory unit of the machine tool, a storage device as an external device, or the like. The log data is created in such a way that the temporal transitions of the operating statuses of the individual drive units and values of temperature and the like detected by sensors can be perceived.Likewise, the log data includes various pieces of information (such as a tool change / replacement operation) input by an administrator operating the machine tool or by maintenance personnel responsible for a required response to the occurrence of an abnormality in the machine tool via a machine operation panel. In this way, the log data stored in the non-volatile memory unit or the like of the machine tool is collected and sent to the log data storage unit 200 via a network or the like, or via an external storage device or the like carried by an operator such as maintenance personnel of the machine tool. The collection may be performed sequentially each time the log data is created, or may be performed periodically at an appropriate interval.

[0020] Next, an outline of the learning process performed by the tool life estimation device 100 will be described. After that, the individual features of the tool life estimation device 100 will be described.

[0021] Fig. 2 is a diagram for describing machining information for use in machine learning according to an embodiment of the present invention. The tool life estimating device 100 according to this embodiment collects the machining information at the time point when the tool life still remains sufficiently from the log data stored in the log data storage unit 200 and performs unsupervised learning based on the machining information. The unsupervised learning by the tool life estimating device 100 according to this embodiment is performed using pieces of machining information obtained by dividing the machining information recorded in the log data storage unit 200 every predetermined unit time.The unsupervised learning process by the tool life estimation device 100 according to this embodiment is performed for the purpose of generating clusters of the machining information at the time point when the tool life still remains sufficiently. For this purpose, as shown in FIG. Fig. 2, the tool life estimating apparatus 100 excludes the portion of the machining information corresponding to a span of time between (i) the time at which the operator determines that the tool reached the end of the tool life and replaced it with another tool and (i) a previous time preceding the time of replacement by a predetermined time t1 (for example, 1 hour) (i.e., the machining information immediately before the end of the tool life) in the log data stored in the log data storage unit 200, and extracts the remaining portions of the machining information for use in the learning process.The method for excluding the portion of the machining information immediately before the end of the tool life of the tool life estimating apparatus 100 of this embodiment may be configured to exclude a portion of the machining information in which an abnormal value occurs (since a portion of the machining information may have an abnormal value immediately before the end of the tool life, for example, in a case where a predetermined value is particularly prominent in comparison with the chronologically preceding and subsequent portions of the machining information).

[0022] According to the "unsupervised learning process," simply by providing a large amount of input data to the learning device, it is possible to learn the specific distribution exhibited by the input data and to perform compression, sorting, shaping, and the like on the input data without the need to provide corresponding teaching output data. The tool life estimation device 100 according to this embodiment may use, for example, a principal component analysis (PCA), a support vector machine (SVM), a neural network, or the like as the unsupervised learning algorithm. As also described in Fig. 3, a method of so-called deep learning can be used using a plurality of intermediate layers of a neural network, in which case a well-known autoencoder, as in Fig. 4 shown, so that the rest is learned.

[0023] Fig. Figure 5 shows a diagram that again shows, as an example, the clusters of machining information in the case where tool life still remains. Fig. 5. For the sake of simplicity, the example includes three pieces of information—feed rate, spindle rotation speed, and cutting resistance—as machining information. However, in reality, machining information is expressed as information that encompasses even more dimensions.

[0024] Next, the individual features of the tool life estimation device 100 will be described. The tool life estimation device 100 includes a learning unit 111, a state observation unit 112, and a learning model storage unit 114.

[0025] The learning unit 111 is a functional unit that performs the unsupervised learning process based on the input data obtained by the state observation unit 112, constructs a learning model, and stores the learning model in the learning model storage unit 114. The learning model constructed by the learning unit 111 is configured as a model for sorting (clustering) to distinguish the machining information at the time when the tool life still remains sufficiently from the remaining portions of the machining information, as shown in Fig. 5. As described above, the algorithm of the learning model constructed by the learning unit 111 may be any as long as it can sort and distinguish the machining information at the time when the tool life still remains sufficiently from its remaining portions.

[0026] The state observation unit 112 creates the input data from the log data stored in the log data storage unit 200 and outputs the created input data to the learning unit 111. In the tool life estimation apparatus 100 according to this embodiment, the input data are pieces of machining information for each unit time acquired from the machine tool.For the machining information expressed as a numerical value, the state observation unit 112 uses the numerical value as the input data of the learning unit 111 on an unprocessed basis, and for the machining information indicated by information other than numerical values, such as a character string, it is to store in a storage unit not shown a conversion table for converting the individual character strings into numerical values and convert the information other than numerical values into numerical values using the conversion table to in turn include it in the input data.

[0027] By means of the above-described structure, the tool life estimating apparatus 100 is able to perform the learning operation with respect to the machining information (machining state) at the time point at which the tool life still remains sufficiently at the time of operation of the machine tool, and thereby construct the learning model.

[0028] Next, the tool life estimating apparatus 100 for estimating the life of the tool using the learning model that has been constructed will be described.

[0029] Fig. 6 shows a schematic functional block diagram at the time of estimating the tool life by the tool life estimating device 100 according to an embodiment of the present invention. The tool life estimating device 100 according to this embodiment estimates the life of the tool used in the machine tool 1 based on the machining information, which includes the information regarding the operating statuses of the individual drive units provided in the machine tool 1 and the information of the sensor (or sensors) of the units acquired via the input / output unit 17 included in the machine tool 1 and constituting the surrounding environment. The input / output unit 17 outputs information acquired internally or externally of the machine tool 1 to the processors located internally and externally of the machine tool 1.Although only the input / output unit 17 and the alarm unit 23 are in . Fig. 6 as the functional blocks provided in the machine tool 1, the machine tool 1 in practice comprises the respective components of a typical machine tool, such as a processor, such as a CPU, and a drive unit.

[0030] The tool life estimating device 100 includes a state observation unit 112, a learning model storage unit 114, and an estimation unit 115.

[0031] During the operation of the machine tool 1, the state observation unit 112 acquires the machining information used as the input data at the time of learning as described above via the input / output unit 17 and creates the input data based on the acquired information, and outputs it to the estimation unit 115.

[0032] Using the learning model stored in the learning model storage unit 114, the estimation unit 115 determines to which of the clusters of machining information the input data (machining information) input from the state observation unit 112 belongs, and thereby estimates the tool life. As shown in Fig. 7, if the input data belongs to the cluster of machining information at the time when the tool life still remains sufficiently, then the estimation unit 115 estimates that the life of the tool used in the currently operating machine tool 1 still remains sufficiently. As further shown in Fig. 8, if the input data input from the state observation unit 112 does not belong to the cluster of machining information at the time point where the tool life still remains sufficiently, then the estimation unit 115 estimates that the tool used in the currently operating machine tool 1 is about to reach the end of its life.

[0033] In this way, the estimation unit 115 outputs to the input / output unit 17 of the machine tool 1 the result of the tool life estimation using the input data created based on the machining information obtained from the machine tool 1. The input / output unit 17 instructs the alarm unit 23 to emit an alarm sound if it is determined that the result of the tool life estimation input from the estimation unit 115 indicates that the tool is about to reach the end of its life.

[0034] When the alarm unit 23 is instructed by the input / output unit 17 to issue an alarm, the alarm unit 23 notifies the operator that the tool is near the end of its service life using a lamp arranged on a machine control panel, an indication by means of a display device, a sound, or the like. Regarding the alarm provided by the notification by means of the alarm unit 23, as in Fig. 2, in a case where the machining information in the interval back to a predetermined time t1 before the end of the tool life is used in the learning process of the machining information at the time point at which the tool life still remains sufficiently, the specific time as the alarm on the display device may be indicated, for example, as “the end of the tool life will be reached in t1 hours.”

[0035] As described above, the tool life estimation device 100 can estimate the tool life during the operation of the machine tool 1 using the learning model obtained as a result of learning based on the machining information in at least one or more machine tools 1. Furthermore, when the tool life estimation device 100 estimates that the end of the tool life has been reached and an alarm is issued, the operator of the machine tool 1 is enabled to systematically stop the operation of the machine tool in response to the estimation result and replace the tool with another one.

[0036] While the embodiments of the present invention have been described above, the present invention is not limited to the examples of the above-described embodiments. The present invention can be implemented in various modes with appropriate modifications.

[0037] In the above embodiment, the modes of learning and use in a single tool life estimating apparatus 100 are shown, however, the learning model itself constructed by the learning unit 111 and stored in the learning model storage unit 114 is a set of pieces of data indicating the results of the learning operation, so that it is possible to configure the learning model to be shared with another tool life estimating apparatus 100, for example, via a storage device (not shown), a network, or the like.In such a configuration, in the learning process, in a state where a single learning model is shared among a plurality of tool life estimation devices 100, the respective tool life estimation devices 100 perform a learning process in parallel, making it possible to shorten the time required to complete the learning process. Furthermore, in the use of the learning model, it is also possible to estimate the tool life by the respective tool life estimation devices 100 using the shared learning model. The method for sharing the learning model is not limited to a specific method.For example, a learning model may be stored in a host computer of a factory and shared by the respective tool life estimating devices 100, or a learning model may be stored on a server installed by a manufacturer such that the learning model can be shared by the tool life estimating devices 100 of the clients.

[0038] In the above embodiment, the configurations of the tool life estimation device 100 at the time of learning and at the time of acquisition are described individually. However, the tool life estimation device 100 may simultaneously include the configuration at the time of learning and the configuration at the time of acquisition. In such a configuration, the tool life estimation device 100 may estimate the tool life and cause the controller of the learning unit 111 to perform further additional learning based on the information input to the machine tool 1 by an administrator or maintenance personnel.

[0039] Likewise, the above embodiment describes a case where the tool life estimation device 100 is configured as a separate unit independent of the machine tool 1. However, the tool life estimation device 100 may be configured as part of the control device of the machine tool 1.

[0040] Furthermore, in the above-described embodiment, the clusters of machining information are generated by machine learning at the time when the tool life is still sufficiently remaining. On the other hand, however, as shown in Fig.9, the machining information before the end of the tool life is reached is divided into, for example, parts such as a part of machining information corresponding to the period between a predetermined time t1 at which the end of the tool life is reached and a predetermined time t2 preceding the time t1; a part of information corresponding to the period between the predetermined time t2 and a predetermined time t3 preceding the time t2;and dividing a part of machining information before the predetermined time t3 preceding the time at which the end of the tool life is reached, so that the clusters of the respective time periods are created and it is determined to which cluster the machining information acquired at the time of machining by the machine tool 1 belongs, which makes it possible to make a more detailed estimate of how long it will take to reach the end of the tool life.;

[0041] Although the embodiments of the present invention have been described above, the present invention is not limited to the examples of the embodiments described above, and other modes can be implemented with modifications made thereto as needed.

Claims

[1] A tool life estimation device (100) for estimating a life of a tool used by a machine tool (1) for machining a workpiece, the device comprising: a state observation unit (112) configured to acquire machining information indicating a machining status from log data recorded during operation of the machine tool (1) and recorded in a state where the tool life still remains sufficiently, and configured to create input data based on the acquired machining information; a learning unit (111) configured to construct a learning model in which clusters of the machining information are created, the learning model being constructed by unsupervised learning using the input data created by the state observation unit (112); and a learning model storage unit (114) configured to store the learning model, characterized by , that the machining information comprises two sections, a first section corresponding to a first time period of the log data before a predetermined time (t1), and a second section corresponding to a second time period of the log data between the predetermined time (t1) and the end of the tool life, and the tool life estimation device (100) is configured to exclude the second portion of the machining information from use in the learning process. [2] A tool life estimation device (100) for estimating the life of the tool used by a machine tool (1) for machining a workpiece, the device comprising: a learning model storage unit (114) configured to store a learning model in which clusters of the machining information are created by unsupervised learning using machining information indicating a status of the recorded machining and recorded in a state in which the life of the tool still remains sufficiently during operation of the machine tool (1); a state observation unit (112) configured to obtain the machining information indicating the status of the machining from log data recorded while the machine tool (1) is operating, and configured to create input data based on the machining information obtained; and an estimation unit (115) configured to estimate the service life of the tool from the input data created by the condition observation unit (112), wherein the service life of the tool is estimated using the learning model, characterized by , that the machining information comprises two sections, a first section corresponding to a first time period of the log data before a predetermined time (t1), and a second section corresponding to a second time period of the log data between the predetermined time (t1) and the end of the tool life, and the tool life estimation device (100) is configured to exclude the second portion of the machining information from use in the learning process. [3] Machine tool (1), comprising: an alarm unit (23) configured to issue an alarm based on an estimation result of the life of the tool by the tool life estimation device (100) according to claim 2.

Images