JP2018025936A

JP2018025936A - Machine tool

Info

Publication number: JP2018025936A
Application number: JP2016156772A
Authority: JP
Inventors: 加藤　隆幸; Takayuki Kato; 隆幸加藤
Original assignee: Okuma Corp; Okuma Machinery Works Ltd
Current assignee: Okuma Corp
Priority date: 2016-08-09
Filing date: 2016-08-09
Publication date: 2018-02-15

Abstract

PROBLEM TO BE SOLVED: To solve the problem in which: even with damage to a tool at a level in which a user determines that the damage can be sharpened, when an abnormality score of actual data exceeds an abnormal score of normal data, a prior art machine tool performs abnormal processing, which leads to an increase in cost, and to provide a machine tool that can reduce the cost.SOLUTION: A user sets damage levels at which abnormal processing of displaying an alarm, retreating a tool, and stopping a machine are respectively executed, and a machine tool executed the abnormal processing according to the set values. The tool can thus be used as long as possible according to the user and situation, and cost can be reduced.SELECTED DRAWING: Figure 1

Description

本発明は、回転工具の異常を検知する工作機械に関する。 The present invention relates to a machine tool that detects an abnormality of a rotary tool.

図２は、従来技術の一形態を表すブロック図である。
正常データ前処理部１１は、正常加工時の主軸負荷時系列データＮＤを入力し、一定区間（たとえば工具１回転毎のデータ）を１まとまりとするデータＮＲに変換する。正常データ学習部１２は、一定区間のデータＮＲを入力し、それが多層パーセプトロンの出力と一致するようにオートエンコーダあるいは変分オートエンコーダで学習し、正常データモデルＮＭを構築する。実データ前処理部３１は、実加工時の主軸負荷時系列データＲＤを入力し、一定区間のデータを１まとまりとするデータＲＲに変換する。実データ異常スコア計算部３２は、実データ前処理部３１で変換した、一定区間のデータＲＲを、正常データ学習部１２で構築した正常データモデルに入力し、出力と入力との差分の最大値あるいは平均値を異常スコアＲＳとして出力する。ここで、異常スコアを計算する毎に異常スコアの大きさに応じてオンライン学習を行うこともある。異常処理実行部３６は、異常スコアＲＳが正常時の異常スコアを越えたら、アラームを表示する、工具を退避する、あるいは機械を停止するなどの処理を実行する。 FIG. 2 is a block diagram illustrating an embodiment of the prior art.
The normal data preprocessing unit 11 inputs the spindle load time series data ND at the time of normal machining, and converts the data into a data NR that makes a certain section (for example, data for each rotation of the tool) as one unit. The normal data learning unit 12 inputs the data NR of a certain interval, learns with an auto encoder or a variational auto encoder so that it matches the output of the multilayer perceptron, and constructs a normal data model NM. The actual data preprocessing unit 31 receives the spindle load time-series data RD at the time of actual machining, and converts the data in a certain section into data RR that is a unit. The actual data abnormality score calculation unit 32 inputs the data RR in a certain section converted by the actual data preprocessing unit 31 to the normal data model constructed by the normal data learning unit 12, and the maximum value of the difference between the output and the input Alternatively, the average value is output as an abnormal score RS. Here, every time the abnormal score is calculated, online learning may be performed according to the magnitude of the abnormal score. When the abnormality score RS exceeds the normal abnormality score, the abnormality processing execution unit 36 performs processing such as displaying an alarm, retracting the tool, or stopping the machine.

特開平５−１６９３５５号公報JP-A-5-169355

Jinwon An, Sungzoon Cho. “Variational Autoencoder based Anomaly Detection using Reconstruction Probability”. SNU Data Mining Center. 2015. P 3-9Jinwon An, Sungzoon Cho. “Variational Autoencoder based Anomaly Detection using Reconstruction Probability”. SNU Data Mining Center. 2015. P 3-9

しかしながら、上記従来の形態では、ユーザが未だ削れると見なすレベルの工具損傷であっても、実データの異常スコアが正常データの異常スコアを超えると、異常処理を行ってしまう。したがって、必要以上に早い時期に新しい工具に交換しなければならなくなる等、コストの点で問題があった。 However, in the above-described conventional form, even if the tool damage is at a level that the user still considers to be scraped, if the abnormal score of the actual data exceeds the abnormal score of the normal data, an abnormal process is performed. Therefore, there was a problem in terms of cost, such as having to replace with a new tool at an earlier time than necessary.

そこで、本発明は、上記問題に鑑みなされたものであって、コストの低減を図ることができる工作機械を提供しようとするものである。 Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to provide a machine tool capable of reducing the cost.

上記目的を達成するために、本発明のうち請求項１に記載の発明は、回転工具の異常を検知する工作機械であって、加工正常時に採取した時系列データを一定区間のデータを１まとまりとするデータに変換する正常データ前処理手段と、前加工正常時の一定区間毎のデータを１まとまりとするデータから教師無し学習により、実加工時のデータが正常時のデータとどれほど離れているかを計算できるモデルを構築する正常データ学習手段と、工具の損傷レベルを鑑定するために採取した鑑定データから一定区間毎のデータを１まとまりとするデータに変換する鑑定データ前処理手段と、前鑑定データの一定区間毎のデータを１まとまりとするデータを前正常データモデルに入力し、出力と入力の差分に基づき異常スコアとして出力する鑑定データ異常スコア計算手段と、１回の加工後に工具の損傷状態を有限個の損傷レベルに割り当てる損傷レベル鑑定手段と、前鑑定データを一定区間毎のデータを１まとまりとするデータに変換したデータを入力とし、前損傷レベルを教師入力として学習したモデルを構築する損傷レベル学習手段と、実加工時に採取した時系列データを一定区間毎のデータを１まとまりとするデータに分割する実データ前処理手段と、前実加工時の一定区間毎のデータを前正常データモデルに入力し、出力と入力の差分に基づき異常スコアとして出力する実データ異常スコア計算手段と、前損傷レベル学習モデルに前実データの異常スコアを入力して損傷レベルを出力する損傷レベル計算手段と、所望の異常処理を、それぞれ損傷レベルがいくつになったら実行するかを設定する異常処理切替実行手段と、前損傷レベルに応じて異常処理を実行する異常処理切替実行手段とを具備したことを特徴とする。
請求項２に記載の発明は、請求項１に記載の発明において、異常スコアとして、出力と入力の差分の最大値あるいは平均値を用いることを特徴とする。 In order to achieve the above object, the invention according to claim 1 of the present invention is a machine tool for detecting an abnormality of a rotary tool, and time series data collected during normal processing is grouped with data of a certain section. Normal data pre-processing means to convert to the data to be, and how far the data at the time of actual machining is different from the data at the time of normal by unsupervised learning from data that is a set of data for each fixed section at the time of normal pre-processing Normal data learning means for constructing a model capable of calculating the value, appraisal data preprocessing means for converting the data for each predetermined section into data that is collected as a unit from the appraisal data collected to appraise the damage level of the tool, and pre-appraisal Appraisal data is input to the pre-normal data model as a set of data for each fixed section of data, and output as an abnormal score based on the difference between the output and the input. Data obtained by converting the pre-assessment data into data that is a set of data for each predetermined section, and a damage level judgment means for assigning the damage state of the tool to a finite number of damage levels after one machining. Damage level learning means for constructing a model that has been learned using the previous damage level as a teacher input, and actual data preprocessing means for dividing the time-series data collected during actual machining into data that is a set of data for each predetermined section And the actual data abnormality score calculation means for inputting the data for each predetermined section at the time of the previous actual machining to the previous normal data model and outputting as an abnormal score based on the difference between the output and the input, and the previous actual data in the previous damage level learning model Damage level calculation means to input damage score and output damage level, and execute desired abnormality processing at each damage level And abnormality processing switching execution means for setting a Luke, characterized by comprising an abnormality processing switching execution means for executing the abnormality processing according to pre-injury levels.
The invention according to claim 2 is characterized in that, in the invention according to claim 1, the maximum value or the average value of the difference between the output and the input is used as the abnormality score.

本発明によれば、ユーザ・状況に応じて工具をできるだけ長く使用でき、コストを低減することができる。 According to the present invention, the tool can be used as long as possible according to the user and the situation, and the cost can be reduced.

本発明の一実施例を示すブロック図である。It is a block diagram which shows one Example of this invention. 従来技術の一形態を示すブロック図である。It is a block diagram which shows one form of a prior art.

図１は、本発明の一実施例のブロック図である。
正常データ前処理部１１、正常データ異常スコア学習部１２、実データ前処理部３１、実データ異常スコア計算部３２は、従来技術と同様である。ただし、正常データと実データとして主軸負荷を例に挙げているが、主軸速度、振動加速度などその他のデータを用いてもよい。 FIG. 1 is a block diagram of an embodiment of the present invention.
The normal data preprocessing unit 11, the normal data abnormality score learning unit 12, the actual data preprocessing unit 31, and the actual data abnormality score calculation unit 32 are the same as those in the related art. However, although the spindle load is taken as an example as normal data and actual data, other data such as spindle speed and vibration acceleration may be used.

鑑定データ前処理部２１は、工具の損傷レベルを鑑定するために採取したデータＴＤを入力し、工具１回転毎のデータを１まとまりとするデータＴＲに変換する。
鑑定データ異常スコア計算部２２は、鑑定データ前処理部２１で変換した工具１回転毎のデータＴＲを、正常データ異常スコア学習部１２で構築した正常データモデルＮＭに入力し、出力と入力の差分の最大値あるいは平均値を異常スコアＴＳとして出力する。ここで、異常スコアを計算する毎に異常スコアの大きさに応じてオンライン学習を行っても良い。 The appraisal data pre-processing unit 21 inputs data TD collected in order to appraise the damage level of the tool, and converts the data for each rotation of the tool into data TR.
The appraisal data abnormality score calculation unit 22 inputs the data TR for each rotation of the tool converted by the appraisal data preprocessing unit 21 to the normal data model NM constructed by the normal data abnormality score learning unit 12, and the difference between the output and the input Is output as an abnormal score TS. Here, every time the abnormal score is calculated, online learning may be performed according to the magnitude of the abnormal score.

損傷レベル鑑定部２３は、１回の加工終了後に人が工具を詳しく観察する、画像データあるいは形状測定器などで定量化して、例えば１０段階に分割された工具損傷レベルのいずれに相当するか判断し、その損傷レベルＴＬを割り当てる。ここで、工具に複数の刃が付いている場合、損傷レベルＴＬとして、最も損傷した刃のレベルとしても、全ての刃の損傷レベルの合計としてもよい。 The damage level appraisal unit 23 determines whether it corresponds to, for example, a tool damage level divided into 10 stages by quantifying with an image data or a shape measuring instrument, etc., in which a person observes the tool in detail after one processing is completed. And assigning the damage level TL. Here, when the tool has a plurality of blades, the damage level TL may be the most damaged blade level or the total damage level of all blades.

損傷レベル学習部２４は、異常スコアＴＳを入力データ、損傷レベルＴＬを教師データとしてニューラルネットワークで学習し、損傷レベルモデルＬＭを構築する。
損傷レベル計算部３３は、実データ異常スコア計算部３２から入力した異常スコアＲＳを、損傷レベル学習部２４で構築した損傷レベルモデルＬＭに入力し、損傷レベルＲＬを出力する。 The damage level learning unit 24 learns with a neural network using the abnormal score TS as input data and the damage level TL as teacher data, and constructs a damage level model LM.
The damage level calculation unit 33 inputs the abnormality score RS input from the actual data abnormality score calculation unit 32 to the damage level model LM constructed by the damage level learning unit 24, and outputs the damage level RL.

異常処理設定部３４は、アラームを表示する、工具を退避する、機械を停止するなどの異常処理を、それぞれ損傷レベルがいくつになったら実行するかの情報ＡＳをユーザが設定する。
異常処理切替実行部３５は、損傷レベル計算部３３から入力した損傷レベルＲＬと異常処理設定部３４で設定された情報ＡＳに基づいて設定された異常処理を実行する。 In the abnormal process setting unit 34, the user sets information AS indicating how many abnormal levels, such as displaying an alarm, retracting a tool, and stopping the machine, are executed.
The abnormality process switching execution unit 35 executes the abnormality process set based on the damage level RL input from the damage level calculation unit 33 and the information AS set by the abnormality process setting unit 34.

１１・・正常時の主軸負荷時系列データの前処理部、１２・・正常時の主軸負荷時系列データの学習部、２１・・鑑定時の主軸負荷時系列データの前処理部、２２・・鑑定時の主軸負荷時系列データの異常スコア計算部、２３・・工具損傷レベル鑑定部、２４・・工具損傷レベル学習部、３１・・実加工時の主軸負荷時系列データの前処理部、３２・・実加工時の主軸負荷時系列データの異常スコア計算部、３３・・工具損傷レベル計算部、３４・・異常処理設定部、３５・・異常処理切替実行部、３６・・異常処理実行部、ＮＤ・・正常加工時の主軸負荷時系列データ、ＮＲ・・正常加工時の工具１回転毎のデータを１まとまりとするデータ、ＮＭ・・正常データモデル、ＴＤ・・工具の損傷レベルを鑑定するために採取したデータ、ＴＲ・・ＴＤを工具１回転毎のデータを１まとまりとするデータ、ＴＬ・・鑑定時の工具損傷レベル、ＬＭ・・損傷レベルモデル、ＲＤ・・実加工時の主軸負荷時系列データ、ＲＲ・・実加工時の工具１回転毎のデータを１まとまりとするデータ、ＲＳ・・実加工時の正常データモデルの出力と入力との差分の最大値あるいは平均値、ＲＬ・・実加工時の工具損傷レベル、ＡＳ・・各異常処理をそれぞれ損傷レベルがいくつになったら実行するかの情報。 11. Pre-processing unit for spindle load time-series data at normal time, 12.-learning unit for spindle load time-series data at normal time, 21.-Pre-processing unit for spindle load time-series data at appraisal, 22.- Anomaly score calculation unit of spindle load time series data at the time of appraisal, 23... Tool damage level appraisal unit, 24 .. Tool damage level learning unit, 31... Preprocessing unit of spindle load time series data at actual machining, 32 ..Score load time series data anomaly score calculation unit during actual machining, 33..Tool damage level calculation unit, 34..Abnormal process setting unit, 35..Abnormal process switching execution unit, 36..Abnormal process execution unit , ND ... Spindle load time series data during normal machining, NR ... Data for each rotation of tool during normal machining, NM ... Normal data model, TD ... Tool damage level Data collected for the purpose, TR・ Data for TD as a group of data per tool rotation, TL ・・ Tool damage level at appraisal, LM ・・ Damage level model, RD ・・ Main spindle load time series data during actual machining, RR ・・ Real Data that collects data for each rotation of the tool at the time of machining, RS ··· Maximum or average value of difference between normal data model output and input at actual machining, RL · · Tool damage level at actual machining , AS ... Information about how much damage level each abnormal process will be executed.

Claims

A machine tool for detecting abnormalities in rotating tools,
Normal data preprocessing means for converting time-series data collected during normal processing into data that is a set of data in a certain section;
Normal data learning means for constructing a model that can calculate how far the data at the actual processing is from the data at the normal processing by unsupervised learning from the data that is a set of data for each predetermined section at the time of normal preprocessing,
Appraisal data preprocessing means for converting the data for each predetermined section from the appraisal data collected in order to appraise the damage level of the tool,
Appraisal data abnormality score calculation means for inputting data that is a set of data for each predetermined section of the pre-examination data to the pre-normal data model and outputting as an abnormal score based on the difference between the output and the input,
A damage level appraisal means for assigning the damage state of the tool to a finite number of damage levels after one machining;
Damage level learning means for constructing a model in which the pre-assessment data is converted into data that is a set of data for each predetermined section as input and the previous damage level is learned as a teacher input;
Real data pre-processing means for dividing time series data collected at the time of actual processing into data that is a set of data for each predetermined section;
Actual data abnormality score calculation means for inputting data for each predetermined section at the time of previous actual machining to the previous normal data model, and outputting as an abnormal score based on the difference between the output and the input,
A damage level calculation means for inputting the abnormal score of the previous actual data to the previous damage level learning model and outputting the damage level;
An abnormal process switching execution means for setting the number of damage levels at which each of the desired abnormal processes is executed;
What is claimed is: 1. A machine tool comprising abnormality process switching execution means for executing an abnormality process according to a previous damage level.

The machine tool according to claim 1, wherein a maximum value or an average value of a difference between output and input is used as the abnormality score.