Method and apparatus for operating a machine with a tool

Abstract

A method for operating a machine with a tool, wherein the following steps are performed: a) detecting at least one operating data point (S1-S4) about the machine and / or the tool, b) calculating at least one estimate (F1-F4) from the at least one operating data point (S1-S4) based on an ML model, c) determining an anomaly in the form of a time-discrete anomaly value (AS) based on the at least one estimate (F1-F4) by comparing with a predetermined comparison value for at least one estimate and identifying a conformity of the estimate comparison, d) storing the time-discrete anomaly value (AS) in a memory and aggregating a time course of the anomaly value (AS) into a smoothed anomaly value (SS), e) comparing the smoothed anomaly value (SS) with at least one predetermined comparison value for the anomaly and identifying a conformity of the anomaly comparison, f) outputting a control operation (O1-O4) to the machine based on the smoothed anomaly value (SS).

Description

Technical Field

[0001] The present invention relates to a method and apparatus for operating a machine having cutting tools. Background Technology

[0002] Tool break-down monitoring (TBM) is an important technology used in automated production lines and unmanned workshops for CNC machining. Tool breakage can lead to damage to the tool, machine, and / or the workpiece being machined, and can result in machine downtime.

[0003] Current solutions often take into account sensor data from machine tools, such as force and vibration sensors, and monitor machine operation in a threshold-based manner, that is, monitor the minimum and maximum values ​​of signals such as spindle power.

[0004] Here, because hard-coded alarm limits do not take into account specific operational behaviors, these solutions are often complex and may cause false alarms. Furthermore, even when a critical situation is reached, alarm output is not always triggered.

[0005] Unexpected machine stoppages may occur due to false alarms or no alarms, and damage to the machine, tools, or workpieces may occur.

[0006] In publication US 2020 / 0104224 A1, anomalies are detected based on recorded data in a manner in which the identified anomalies are trained by means of machine learning and used separately for comparison when other anomalies with increased noise components in the base sensor data are identified, thereby improving the anomaly identification rate during workpiece production and accordingly identifying the workpiece.

[0007] However, it was not discussed whether anomalies that increase over time or are caused by machine wear or production parameters should be reduced or avoided.

[0008] If excessively high feed rates are present in CNC milling cutters, then the tools used should be protected from damage or excessive wear, for example. By intervening appropriately in the production or manufacturing process, the availability of the tools or machines should be improved. Thus, the manufacturing process can be adapted to the increasingly aging or worn condition of the tools. Summary of the Invention

[0009] The objective of this invention is to provide a method and apparatus for automatically monitoring a machine with cutting tools, thereby enabling it to operate with improved usability.

[0010] In other words, it should enable the predictable and pre-defined operation of machines with cutting tools.

[0011] Preventative operation is understood as the ability to identify tool breakage and stop the current process, such as drilling or milling, in order to avoid damage to the workpiece.

[0012] At this point, the machine may stop, but the workpiece can be processed again using new cutting tools.

[0013] In predictive operation, efforts are made to forecast tool breakage in a timely manner in order to avoid damage to both the workpiece and the tool.

[0014] Machine downtime may occur, but workpiece processing can continue using the same or new cutting tools.

[0015] Under pre-defined operating conditions, tool breakage should be predicted in a timely manner and countermeasures should be taken to avoid tool breakage and machine downtime.

[0016] The task according to the invention is solved by a method of the type mentioned at the beginning, wherein the following steps are performed:

[0017] a) Detect at least one operational data point relating to the machine and / or the cutting tool.

[0018] b) Calculate at least one estimate from at least one running data point based on an ML model.

[0019] c) Based on the at least one estimate, anomalies in the form of time-discrete outliers are determined by comparing them with predetermined comparison values ​​for the at least one estimate and identifying the consistency of the estimate comparisons.

[0020] d) Store the discrete time outliers in memory and aggregate the temporal variation of the outliers into smooth outliers.

[0021] e) Compare the smoothed outliers with at least one predetermined comparison value for the outliers, and identify the consistency of the outlier comparisons.

[0022] f) Output control operations to the machine based on the smoothed outliers, wherein preferably the control operations act on the operation of the machine and / or the tool, the operation being detected by at least one running data point.

[0023] In statistics, an anomaly (also called an "abnormal measure") is an observation or event that is so different from other events that it raises suspicion that it is produced by another mean.

[0024] Anomalies in large data records may follow very complex patterns that are difficult to identify in most cases.

[0025] This invention enables the predictable and pre-defined operation of machines with cutting tools under conditions of high availability.

[0026] Unlike existing technologies, this method can not only reliably identify unwanted anomalies, but also introduce corresponding countermeasures in a timely manner, thereby improving the availability of tools or machines so that anomalies do not continue to occur as the manufacturing process is adapted accordingly.

[0027] For example, the manufacturing process can be modified accordingly after an anomaly is identified, or alternative machining of the workpiece can be initiated, such as by using another tool, changing process parameters such as rotation speed or feed rate of the CNC milling machine, or also increasing coolant flow.

[0028] In other words, the identified anomalies can be used to provide, for example, an indication of the degree of wear on a machine or cutting tool.

[0029] Therefore, the interest lies not in anomalies in the workpiece itself to improve existing technology, but in the machines or tools used to manufacture the workpiece. Anomalies are used as a corresponding measure of this.

[0030] This is achieved by identifying anomalies based on the estimated values ​​in step c). In other words, the estimated values ​​themselves are used as a measure to identify anomalies, rather than directly evaluating the current data point. For example, this could lead to the timely identification of wear on machines or cutting tools.

[0031] In one improved embodiment of the invention, at least two operational data points are detected and further processed.

[0032] This allows for more precise monitoring of machine operation and detection of parameter combinations.

[0033] In one improved embodiment of the invention, at least two estimates are calculated and further processed.

[0034] This allows for the identification of anomalies in a redundant manner.

[0035] In one improved embodiment of the invention, at least one operating data point is specified as rotational speed or displacement speed, torque, current consumption, or temperature.

[0036] This allows for precise and remarkably simple monitoring of machine operation.

[0037] In one improved embodiment of the invention, at least one estimate is calculated using an LSTM network.

[0038] This allows for the simple identification of anomalies.

[0039] In one improved embodiment of the invention, time-discrete outliers are identified using the "isolated forest" algorithm.

[0040] This allows for the identification of anomalies in a simple and particularly effective way.

[0041] Isolation forest is an unsupervised learning algorithm used to identify anomalies. It is based on the principle of isolating anomalies, rather than the most common technique used to analyze normal points.

[0042] The most common technique used to identify anomalies is based on creating a profile of what is considered "normal": anomalies are reported as instances in data records that do not conform to the normal profile.

[0043] Isolation forests employ a different approach: instead of attempting to model for regular events, they explicitly isolate outliers in the dataset. The main advantage of this approach lies in the possibility of using scanning techniques to a extent that is not permitted by distribution-based methods, thereby creating very fast algorithms with minimal storage requirements.

[0044] In one improved embodiment of the invention, outliers are smoothed by constructing a time average.

[0045] This allows for accurate and simple identification of anomalies.

[0046] In one improved embodiment of the invention, the comparison value in step e) is determined by a machine learning-based method, such as a decision tree, Bayesian network, neural network, multi-class support vector machine, or k-nearest neighbor classification.

[0047] In an improved embodiment of the invention, at least one number of running features is considered during the comparison in step f).

[0048] The task according to the invention is also solved by means of a device of the type mentioned at the beginning, said device comprising...

[0049] A sensor module having at least one sensor device for performing step a),

[0050] The estimation module used to perform step b).

[0051] The anomaly detection module used to perform step c).

[0052] The storage module used for storage in step d).

[0053] The decision module used to perform step e), and

[0054] The output module used to execute step f),

[0055] The device described herein is configured to perform the method according to the invention.

[0056] In one improved embodiment of the invention, an auxiliary data module is further included, which is applied during the execution of step e).

[0057] It is clear that the modules are directly or indirectly connected to each other for communication.

[0058] In one improved embodiment of the invention, the estimation module, anomaly detection module, storage module, and decision module are placed on the edge of the device, which is further included and connected to it.

[0059] Because local operation is possible, this allows for monitoring of the operation with exceptionally high availability.

[0060] Such an improved device can therefore be a distributed system having components, such as sensor devices, located at the machine, and other modules (such as auxiliary data modules) included in and executed there by edge computing devices. The various components and modules are interconnected and communicate with each other.

[0061] With such a distributed system, the modules mentioned can be used together to run on multiple machines, which makes the system simpler, cheaper, and more reliable.

[0062] Furthermore, there is no need to store runtime data in the cloud, thereby improving data security and confidentiality. Attached Figure Description

[0063] The invention will now be described in more detail with reference to embodiments shown in the accompanying drawings. In the drawings:

[0064] Figure 1 Symbolic illustrations showing embodiments of the method according to the present invention,

[0065] Figure 2 Showing according to Figure 1 Example of a flowchart for the decision-making module.

[0066] Figure 3 A symbolic diagram of the control circuit is shown.

[0067] Figure 4-5 An example showing a time-based graph illustrating the change in current in a machine.

[0068] Figure 6-7 An example of a diagram showing current consumption based on borehole depth.

[0069] Figure 8 An example of feed rate is shown based on the borehole depth. Detailed Implementation

[0070] Figure 1 An embodiment of the invention is shown in the form of a device according to the invention, which is configured to perform the method according to the invention.

[0071] The method comprises the following steps:

[0072] a) Detect at least one operating data point S1-S4 related to the machine and / or cutting tools.

[0073] b) Calculate at least one estimate F1-F4 from at least one running data point S1-S4 based on the ML model.

[0074] c) Based on at least one estimate F1-F4, identify anomalies in the form of time-discrete outliers AS by comparing them with predetermined comparison values ​​for at least one estimate and identifying the consistency of the estimate comparisons.

[0075] d) Store the discrete outliers AS in memory and aggregate the temporal variations of the outliers AS into smoothed outliers SS.

[0076] e) Compare the smoothed outlier SS with at least one predetermined comparison value for the outlier and identify the consistency of the outlier comparison.

[0077] f) Output control operations O1-O4 to the machine based on smoothed outliers SS, wherein preferably the control operations act on the operation of the machine and / or the tool, the operation being detected by at least one operation data point S1-S4.

[0078] Therefore, a control loop can optionally be created to monitor continuous operation and to ensure the output quality of the workpiece despite time-varying machine or tool conditions, by monitoring and accordingly adapting the manufacturing parameters of the machine or tool.

[0079] Clearly, by means of the described method, the material tolerances of the workpiece blank, such as alloy ratios, can also be monitored accordingly and included in the manufacturing process adapted accordingly.

[0080] It is particularly beneficial if at least two running data points S1-S4 are detected and further processed, as the probability of anomaly detection is increased. Thus, for example, multiple data points can be aggregated into a common data source, and these multiple data points can constitute new data points.

[0081] Because control operations can optionally act on the operation of the machine and / or the cutting tools, which is detected through at least one operational data point, future anomalies can be reduced or avoided, for example, by timely replacement of worn cutting tools. Figure 2 Other implementation schemes for this are listed in the description.

[0082] Subsequently, it is also beneficial if at least two estimates F1-F4 are calculated and further processed.

[0083] The operating data points S1-S4 can be, for example, rotational speed or displacement speed, torque, current consumption, or temperature.

[0084] Therefore, it is possible to perform combined evaluations of multiple manufacturing parameters, such as the synergistic effect of workpiece temperature and tool temperature at a manufacturing time point.

[0085] For example, the estimated values ​​F1-F4 can be calculated using an LSTM network.

[0086] For example, the "isolation forest" algorithm can be used to calculate time-discrete outliers (AS).

[0087] For example, smoothed outliers (SS) can be determined by constructing a time average.

[0088] The comparison values ​​in step e) can be determined, for example, by machine learning-based methods such as decision trees, Bayesian networks, neural networks, multi-class support vector machines (SVMs), or k-nearest neighbor (kNN) classification.

[0089] In addition, at least one running feature number K1, K2 may be considered during the comparison in step f).

[0090] Equipment for operating machines with cutting tools includes

[0091] • A sensor module SM having at least one sensor device SM1-SM4 for performing step a),

[0092] • The estimation module FCM used to perform step b),

[0093] • The anomaly detection module ADM is used to perform step c).

[0094] • The storage module MEM that performs the storage in step d),

[0095] • The decision module DM for performing step e), and the output modules OM1-OM4 for performing step f).

[0096] • Auxiliary data module AUXM, which is used when performing steps to consider the running feature numbers K1, K2.

[0097] For better overview, the machine is not shown in this figure. It is clear that sensor devices are correspondingly connected to the machine to detect the machine's sensor data.

[0098] Output modules OM1-OM4 indicate the machine's operating status, provide information about necessary maintenance or repair work, and can also be composed of a common output module OM.

[0099] Figure 2 An example flowchart from the decision module DM in the above figure is shown.

[0100] The storage module provides smoothed outlier values ​​SS to the decision module DM.

[0101] Check whether the smoothed outlier SS is within the range of values, i.e., greater than the boundary value of 0.8, which corresponds to a first predetermined comparison value for the outlier.

[0102] Therefore, consistency in identifying abnormal comparisons is crucial.

[0103] If this does not apply, then there is no exception, and the machine can continue to run.

[0104] However, if the boundary value is exceeded, an attempt is then made to more accurately determine the cause of the anomaly.

[0105] This is done by comparing it with runtime characteristic parameters.

[0106] Check whether the cutting force CF (English: "cutting force") is within a pre-defined threshold TH. CF the following.

[0107] If so, further check whether the machine's maximum set runtime (OPH) has been reached by executing operations against a pre-defined threshold (TH). OPH The comparison is as follows. Considering the remaining runtime (RUL) and the runtime feature number K1 provided by the auxiliary data module AUXM, the runtime number OPH is further examined to see if it is still within the remaining runtime (RUL).

[0108] The feature numbers K1 and K2 can also be another machine, namely the so-called "key performance indicator" (KPI).

[0109] If this applies, the output module OM1 can output O1 "Inspection tool pick-up (Werkzeug-Aufnahme)".

[0110] Otherwise, the output module OM2 can output O2 "displacement tool".

[0111] If the cutting force CF is within a pre-defined threshold TH CF The above further verifies whether the cutting speed CS is within a pre-defined threshold TH. CS the following.

[0112] If this applies, O3 "Check Lubrication" can be output via output module OM3.

[0113] Otherwise, you can output O4 "Check CNC program / model" through the output module OM4.

[0114] The cutting force CF, cutting speed CS, and maximum set running hours OPH can be provided, for example, through the storage module MEM, or directly from the corresponding sensor devices SM1-SM4, which are connected to the machine respectively.

[0115] Figure 3 An example of a control circuit for a machine having a device according to the invention is shown, namely, outputting control operations O1-O4 to the machine based on smoothed outliers SS, wherein the control operations optionally act on the operation of the machine and / or the cutting tool, the operation being detected by at least one operation data point S1-S4.

[0116] The control system CON detects relevant variables through the sensor module SM to monitor various operating characteristics.

[0117] The control system CON provides the data to the edge platform EPF, where the running data points S1-S4 are evaluated by means of one or more applications APP.

[0118] In other words, in this example, the estimation module FCM, the anomaly detection module ADM, the storage module MEM, and the decision module DM can optionally be placed on the edge platform EPF.

[0119] The application (APP) includes machine learning-based algorithms and controls the machine via an output module (OM) using a corresponding actuator or display.

[0120] This creates a device with a control circuit, which allows a machine with cutting tools to be operated in an optimal manner.

[0121] The components shown illustrate an example of a distributed networking system deployed on an edge EPF and close to the machine. Other distributions of the components may also be advantageous upon request.

[0122] Figures 4 to 8 Examples of the nominal operating parameters (Betriebsnenngrößen) that can constitute operating data points S1-S4 are shown.

[0123] Figure 4 This shows the machine's current consumption and, in this example, the time-varying current consumption of the CNC machine's spindle.

[0124] The measured current is plotted on curve M1.

[0125] The predicted smoothed outliers can be seen in curve P1, which can form an example of the smoothed outlier SS according to the above figure.

[0126] Figure 5 This shows the time-varying process of another current consumption of the machine.

[0127] The measured current is plotted on curve M2.

[0128] The curve P2 shows smoothed outliers, which can be used to form examples of smoothed outliers SS according to the above figure.

[0129] Figure 6 and Figure 7 The time-varying process of the machine's current consumption I is shown based on the drilling depth DD.

[0130] Anomaly densities AD1-AD3 are plotted separately, and these anomaly densities can be used as criteria for detecting anomalies.

[0131] The anomaly densities AD1-AD3 can be derived from predicted estimates or outliers, but they can also be derived from smoothing outomas.

[0132] Figure 8 The feed rate variation is shown based on the borehole depth.

[0133] The anomalous densities AD1-AD3 are then plotted.

[0134] List of reference numerals in the attached diagram:

[0135] AD1-AD3 anomalous density

[0136] ADM Anomaly Detection Module

[0137] APP: Artificial intelligence applications on computing devices

[0138] AS (anomaly score) refers to discrete-time outliers.

[0139] AUXM auxiliary data module

[0140] CF cutting force

[0141] CON refers to the control system.

[0142] CS refers to cutting speed.

[0143] DD stands for drilling depth.

[0144] DM Decision Module

[0145] EPF computing device, "edgeplatform"

[0146] F1-F4 estimates

[0147] FCM estimation module, "forecasting module"

[0148] K1 and K2 feature numbers, "key performance indicator", KPI

[0149] The values ​​measured by Ml and M2

[0150] MEM storage module

[0151] Output values ​​of O1-O4

[0152] OM, OMI-OM4 output modules

[0153] The value of OPH runtime, or "operation hours".

[0154] Predicted values ​​of P1 and P2

[0155] The RUL value for “meaningful remaining time”, or “remaining useful life”.

[0156] S1-S4 Input Signals

[0157] SM sensor module

[0158] SM1-SM4 sensor devices

[0159] SS (Smoothed Outlier Score)

[0160] TH CF TH CS TH OPH threshold

[0161] V represents speed.

Claims

1. A method for operating a machine with cutting tools, characterized in that, Perform the following steps: a) Detect at least one operational data point relating to the machine and / or the cutting tool. b) Calculate at least one estimated value of a prediction from the at least one running data point based on a machine learning model. c) Based on the at least one estimate, anomalies in the form of time-discrete outliers are determined by comparing them with predetermined comparison values ​​for the at least one estimate and identifying the consistency of the estimate comparisons. d) Store the discrete time outliers in memory and aggregate the temporal variations of the discrete time outliers into smooth outliers. e) Compare the smoothed outliers with at least one predetermined comparison value for the outliers, and identify the consistency of the outlier comparisons. f) Output control operations to the machine based on the smoothed outliers, wherein the control operations act on the operation of the machine and / or the tool detected by the at least one operating data point, in a manner that monitors and accordingly adapts the manufacturing parameters of the machine or the tool.

2. The method of claim 1, wherein at least two running data points are detected and further processed.

3. The method of claim 1, wherein at least two estimates are calculated and further processed.

4. The method according to any one of claims 1-3, wherein the at least one operating data point is rotational speed or displacement speed, torque, current consumption or temperature.

5. The method according to any one of claims 1-3, wherein the at least one estimate is calculated by means of an LSTM network.

6. The method according to any one of claims 1-3, wherein the time discrete outliers are determined by means of the "isolated forest" algorithm.

7. The method according to any one of claims 1-3, wherein the smoothing of outliers is performed by constructing a time average.

8. The method according to any one of claims 1-3, wherein the comparison value in step e) is determined by a machine learning-based method, including decision tree, Bayesian network, neural network, multi-class support vector machine or k-nearest neighbor classification.

9. The method according to any one of claims 1-3, wherein at least one running feature number is considered during the comparison in step f).

10. An apparatus for optimizing the operation of a machine having cutting tools, wherein the apparatus is configured to perform the method according to any one of claims 1 to 9, characterized in that... The device has: A sensor module having at least one sensor device for performing step a), The estimation module used to perform step b), The anomaly detection module used to perform step c). The storage module used for storage in step d). The decision module for performing step e), and the output module for performing step f).

11. The device of claim 10, wherein the estimation module, the anomaly detection module, the storage module, and the decision module are placed on the edge included and connected by the device.

Images