System and method for computer-assisted machining

Abstract

The disclosed subject matter relates to a computer-assisted machining method comprising: A) providing a machine tool (MT) with a batch of material (M) with uncertain machinability, B) specifying a set of machining (c1) conditions comprising a machining speed, C) inserting a tool (TL) into the machine tool (MT) wherein the inserted tool (TL) has a predetermined type and a predetermined wear, D) machining the batch of material (M) with the machine tool (MT), the wear of the inserted tool (TL) is monitored during machining, and a first tool life (Tc1) of the tool (TL) is determined, E) steps B, C and D are repeated to determine a second tool life (Tc2), while a different machining speed is set in step B and a tool of the same type is inserted in step C, F) on the basis of the machining speed, the different machining speed, the first tool life and the second tool life (Tc1, Tc2), the tool life of the tool (TL) is determined. Determining coefficients of a model associated with the batch of material (M), G) determining the processability of the batch of material based on the model.

Description

Technical Field

[0001] The subject matter disclosed herein relates to systems and methods for computer-aided machining. Background Art

[0002] In machining, especially in subtractive machining (eg cutting) of metallic materials, significant deviations in machinability between material batches of the same specification may be observed.

[0003] Methods known from the prior art are used to solve this problem, for example by integrating a classification routine in the process, which classifies the material during the process (see German patent application DE 10 2020 201 077 A1). In DE 10 2020 201 077 A1, a process monitoring system uses a lookup table and a database to find a match for the process parameter vectors collected during the process. In this way, the processability of the material is obtained from the database. However, if the material batch is new, i.e. no match or only a poor match is found in the database, only little process optimization is possible, since no ground truth data about this material is available.

[0004] Machinability can be defined as the property of a workpiece or material that can be machined under given conditions (DIN 6583, 1981, S.1). For example, it can be quantified by a machinability index, where steel can be used as a reference.

[0005] In other words, machinability is the ease with which a metal can be cut (machined), allowing material to be removed with a satisfactory finish. Materials with good machinability (free-machining materials) require little power to cut, can be cut quickly, easily give a good finish, and do not cause significant wear on the tool.

[0006] Until now, in such cases of material batches with unknown behavior, material characterization experiments were carried out before the actual machining operation in order to determine the machinability of the respective material batch. This process is quite resource-intensive.

[0007] Therefore, there is a need for a procedure that allows avoiding the material characterization process before the actual processing process, thereby eliminating the previous testing effort. Summary of the invention

[0008] This need is met by a computer-aided machining method that includes:

[0009] A) providing a machine tool with a material batch with uncertain machinability, e.g. a metal or metal alloy batch,

[0010] B) specifying a set of processing conditions including a processing speed (e.g., fully automatically),

[0011] C) inserting a tool into the machine tool (e.g. fully automatically), wherein the inserted tool is of a predetermined type and has a predetermined wear,

[0012] D) machining a material batch with a machine tool, monitoring the wear of an inserted tool during machining, and determining a first tool life of the tool, e.g. in minutes,

[0013] E) repeating steps B, C and D to determine a second tool life, while setting a different machining speed in step B and inserting the same type of tool in step C,

[0014] F) determining (e.g., calculating) coefficients of a model associated with or describing the batch of material based on the machining speed, the different machining speeds, the first and second tool lives,

[0015] G) determining the processability of the material batch based on the model.

[0016] The factors may be referred to as batch-specific factors because they are associated with a specific batch of material.

[0017] In other words, the model determined in step F is a machinability model for the material batch, which initially has an uncertain machinability.

[0018] Those skilled in the art will recognize that the term "processing conditions" may refer to any processing parameters (processing speed, axis feed, etc.) and also to any other environmental factors that may affect the processing process, such as temperature, humidity, vibration, etc. of the processing area.

[0019] In an embodiment, at the same time, in step E, a different processing speed is selected, and the remaining processing conditions, especially the processing (eg, cutting) parameters, may remain unchanged.

[0020] In an embodiment, the tools inserted in step C and step E are identical, i.e. two copies of the same type of tool. In an embodiment, the tools used are both new, i.e. unused. In an embodiment, both tools have the same predetermined wear before being inserted into the machine tool.

[0021] Those skilled in the art will recognize that the term "monitoring wear during machining" has the following meaning in the context of the present disclosure. Machining (e.g., cutting) may include one or more interruptions of the process, wherein the machining is stopped and the wear of the cutting tool is analyzed, for example, by computer vision methods. If the wear status of the cutting tool is good, the machining is continued.

[0022] In an embodiment, the model is a Taylor model. The batch-specific coefficients are coefficients of the Taylor model.

[0023] In an embodiment, the machining is cutting and the cutting conditions further include feed rate and / or cutting depth, and the method further includes:

[0024] - determining at least one further tool life of the cutting tool, and

[0025] - Determining coefficients of an extended model associated with / describing the material batch.

[0026] In an embodiment, the extended model is an extended Taylor model.

[0027] In an embodiment, the method further comprises adding the machinability and coefficients of the model to a knowledge database.

[0028] In an embodiment, the tool is a drill, a milling tool, a cutting tool.

[0029] In an embodiment, the determination of the first tool life and / or the second tool life is performed without removing the tool from the machine tool.

[0030] In an embodiment, step D comprises the following sub-steps:

[0031] D1) interrupting the processing at least once, preferably at regular intervals,

[0032] D2) acquiring at least one image of the tool during interruption of machining, in particular without removing the tool from its position, and

[0033] D3) Based on at least one image of the tool, determining the wear of the tool.

[0034] In particular, when determining the first and / or second tool life without removing the tool from the machine tool, the time for determining machinability can be reduced. Interruptions can be performed every 20-30 seconds and the time to remove the tool from the machine tool and insert it again to continue machining can be considerable.

[0035] In an embodiment, step D3 (determining the wear of the tool based on at least one image of the tool) comprises the following sub-steps:

[0036] D31) segmenting at least one image of the tool using the trained image segmentation model,

[0037] D32) calculate the flank wear width of the tool, and

[0038] D33) Determine whether the side wear width meets a predefined side wear width threshold associated with wear of the tool.

[0039] The image segmentation model can be trained on a training dataset comprising images of tools, each image being associated with one or more tool features and / or defects. Such features and / or defects can be, for example, flank wear (level of flank wear), chipping, fracture, etc.

[0040] In an embodiment, training may be performed by providing an image of the tool (particularly, an image showing the tool in landscape orientation), annotating the image according to features and / or defects of the tool and generating a mask for the image, and training an (untrained) model on the image of the tool and the generated mask.

[0041] The image segmentation model may be based on a pattern recognition algorithm, such that the trained image segmentation model may be based on a trained pattern recognition algorithm.

[0042] It will be appreciated that the image segmentation model may also be based on an untrained algorithm or a trained neural network that is trained to determine the wear of the tool, in particular by determining the wear of its flanks.

[0043] In an embodiment, the method further comprises: determining processing parameters based on the processability of the material batch and processing the material batch.

[0044] The above needs are also met by providing a computer program comprising instructions which, when executed by a computing device associated with a machine tool, cause the computing device and the machine tool to implement the above method.

[0045] The above needs are also met by providing a computing device comprising such a computer program.

[0046] In an embodiment, the computing device comprises an edge device, particularly an edge device that is (communicatively and / or operationally) connected to a digital control device (eg, CNC).

[0047] The above needs are also met by providing a system for machining, the system comprising a machine tool and a computing device associated with (eg, operatively connected to) the machine tool. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The above and other objects and advantages of the present invention will be apparent upon consideration of the following detailed description of certain aspects which only indicate several possible ways in which it may be practiced. The description is made in conjunction with the accompanying drawings, in which like reference numerals refer to like parts throughout, and in which:

[0049] Figure 1 A system for processing batches of material is shown,

[0050] Figure 2 A flowchart showing the processing method, and

[0051] Figure 3 It is the machining process that determines the tool life.

[0052] Reference signs used in the drawings and claims are for illustrative purposes and should not be construed as limiting the corresponding claim features. DETAILED DESCRIPTION

[0053] Figure 1 A system is shown on which the method disclosed in the present invention can be performed. The system comprises a machine tool MT for machining, for example subtractive machining, such as milling, drilling, cutting, etc. In particular, the machine tool MT is a CNC machine tool.

[0054] In the machine tool MT there is a (piece of) material batch M to be machined by a tool TL inserted in the machine tool MT. The tool may be, for example, a cutting tool.

[0055] The machine tool MT usually comprises one or more spindles (not shown here for simplicity) which are driven by a feed drive (eg a motor) during the machining process.

[0056] The system further comprises an edge device EDG communicatively (in particular also operatively) coupled to the machine tool MT.

[0057] The edge device EDG is suitable and configured to receive processing data from the machine tool and is also configured to control the machine tool in particular. In the latter case, the edge device EDG may have the functionality of an NC device for the machine tool. In addition, it (edge ​​device EDG) can also be designed as an NC device including the machine tool MT.

[0058] The data received by the edge device EDG may include signals for each spindle and feed drive, such as position, current, acceleration, torque, etc. These signals may be NC signals, ie, from an NC device of the machine tool MT (not shown here for simplicity).

[0059] The system may further comprise a (digital) camera CAM for computer vision purposes. The camera CAM may be carried by the machine tool MT, for example, for capturing live pictures and / or videos during eg machining or during pauses in machining.

[0060] In particular, the camera CAM may be configured to take images of the tool TL to analyze its wear. In order to perform the tool wear analysis, the edge device EDG may include an executable routine, software TWA (e.g., in the form of an app), which can be run on the edge device and includes instructions that implement an algorithm for analyzing the wear of the tool TL when executed by the edge device EDG. The algorithm may be based on, for example, a function or model previously trained on a predetermined training set of image data associated with the wear of the tool. In an embodiment, the function or model may be best matched with an image from a database DB, which is associated with the system, in particular with the edge device EDG, to determine the wear of the tool TL. To this end, the images in the database DB may be marked accordingly. Such images may also be used to train the previously mentioned function or model, i.e., the predetermined training set may include images of / from the database DB or consist of images of / from the database DB. The term "predetermined" may refer to the label or label value of the image selected for training. For example, only images whose labels have side wear values ​​above (or below) a certain determined threshold may be selected to form a training data set.

[0061] Generally, the edge devices EDG considered in the context of the present disclosure should be considered as industrial computers. This means that they include all common functionalities of such industrial computers, interfaces such as communication interfaces, software, etc.

[0062] about Figure 2 The machining method disclosed in the present invention begins with providing a batch of material with uncertain machinability to a machine tool (step A). ​​The material can be a metal or a metal alloy.

[0063] This can be done fully automatically. For example, the edge device can communicate with a robot R and be configured to command the robot R to grab a piece of material M and place it into a machine tool MT.

[0064] Those skilled in the art will appreciate that a bidirectional arrow indicates bidirectional communication.

[0065] The edge device EDG may further comprise a software module or program NIM configured to determine whether the material batch M has a novel behavior, ie its processability is unknown, implying that optimal processing parameters for this material batch M are yet to be determined.

[0066] The novelty identification module NIM may be based on a trained model - a novelty detection model. In particular, the novelty detection model may be trained on a set of historical data associated with respective machining states, such that the novelty detection model calculates a score - a novelty score - and generates high scores for data points (i.e. data points (current, voltage, axis and / or tool center point position, torque) at a specific time) that match the distribution of the historical data and low scores for data points that do not match the distribution. In other words, the novelty of the samples M of the material batch may be represented by the novelty score. Novelty scores above a certain predetermined threshold but within a decision boundary are then considered known, while samples outside the decision boundary are considered novel. In this way, the novelty of the material batch M may be determined fully automatically.

[0067] A machining state may be defined as a number (all) of machining parameters that may be adjusted fully automatically or by an operator (eg axis feeds, etc.).

[0068] Otherwise, the novelty of the material batch M may be determined by the operator of the machine tool MT.The material batch may simply be marked as having uncertain machinability.

[0069] In the following, the specific example of cutting will be considered. However, in view of the above, it will be appreciated that this example may be extended to any type of machining mentioned in relation to the machine tool MT.

[0070] See also Figure 2 , the next step (step B) of the method is to specify a (first) set of machining (here cutting) conditions c1 including a machining (here cutting) speed. This can be done fully automatically based on historical data, which may include expert knowledge, or by an operator. For example, the operator may request to select a set of initial cutting conditions c1.

[0071] Given this set of initial cutting conditions c1, a new cutting tool TL is inserted (step C). The insertion itself can be carried out fully automatically by the machine tool MT, or the operator can start a procedure for inserting the tool TL. It should be understood that the cutting tool TL does not have to be new or unused. However, it is of a predetermined or known type and has a predetermined degree of wear.

[0072] In a further step (step D), the material batch M is processed using a cutting tool. During the entire processing process, monitoring of the tool and analysis of its wear are performed, the process is stopped once the cutting tool is worn and damaged, and the tool life Tcl of the tool is thus determined. That is, in order to perform tool wear analysis, an edge device EDG with app TWA can be used.

[0073] In an embodiment, the cutting process is interrupted to take an image of the tool using a camera CAM. A person skilled in the art will appreciate that the cutting process may be interrupted several times to analyze the wear of the tool TL. Preferably, the process is interrupted at regular intervals. The image acquisition can be performed without removing the tool TL from the machine tool MT. Based on the captured images of the tool, the wear of the tool may be determined by the tool wear analysis software TWA. As mentioned above, the tool wear analysis software TWA may include a trained function or perform a best match comparison to determine the wear of the tool TL.

[0074] In one embodiment, determining tool wear based on at least one image of the tool includes: segmenting at least one image of the tool through a trained image segmentation model, calculating the side wear width of the tool, and determining whether the side wear width meets a predefined side wear width threshold associated with the wear of the tool.

[0075] As explained above, the image segmentation model may be trained on a predetermined image dataset, wherein each image is associated with a feature and / or defect (e.g., flank wear, chipping, fracture, etc.) of the tool T. The image segmentation model may be based on a pattern recognition algorithm. Once the measured flank wear width meets a predetermined threshold value v b,th , the tool is considered to be worn and damaged, and the tool life Tc1 is determined.

[0076] During machining, a tuple K1 of ground truth data can be generated, which consists of the selected cutting conditions c1, the correspondingly achieved tool life Tcl and the feature vector Xc1 calculated from the observed process data. By doing so, initial information on the machinability of the novel material batch M of c1 is obtained. For example, the tuple K1 can be stored in a database DB to train / improve the model used in the novelty identification module NIM and / or in a classifier of material batches (such as in DE 10 2020 201 077 A1).

[0077] In an embodiment, the operator may be asked whether there is enough information to determine a model that can be used to determine the processability of the material batch. Those skilled in the art will appreciate that this step may also be performed in a fully automated manner. Figure 2 , "n" represents the number of runs performed during the machining process, such as the number of tool lives measured, and "N" represents the number of runs required to derive the model.

[0078] For example, a simple Taylor model can be used. In order to achieve process optimization based on the Taylor model, at least two support points are required, and therefore the tool life under different cutting conditions is required.

[0079] Therefore, in the next step (step E), a further new or unused cutting tool is inserted, wherein the further cutting tool has the same tool type as the previous cutting tool. A person skilled in the art will appreciate that the further tool need not be new but must have a predetermined wear.

[0080] In addition, by using the cutting speed v c2 The material batch is processed under the other cutting condition c2 and the wear of the other tools is monitored to determine the second tool life Tc2.

[0081] In the embodiment, the second cutting condition c2 is only when the cutting speed v c The above is different from the first cutting condition c1, while the other cutting conditions remain unchanged.

[0082] In the next step (step F), based on the first and second cutting speeds (v c1 and v c2 ) and the first and second tool lives (Tc1 and Tc2) determine the parameters of the Taylor model. The calculated parameters determine the Taylor model associated with the material batch M that has been processed as described above.

[0083] In other words, a processability model of the unknown material batch is established. The model can be stored in a database DB for further use.

[0084] In the next step (step G), the processability of the material batch M may be determined based on the Taylor model.

[0085] The above is the simplest case of the Taylor model which requires only two support points. In this case, only two runs are required during the machining process.

[0086] In an embodiment, additional cutting conditions (e.g., feed rate f and cutting depth a) may be used. p ) as an extended Taylor model.

[0087] Extended Taylor model with coefficients m, n, and q:

[0088] ln T=ln c t +m ln v c +n ln f+q ln a p

[0089] In order to determine the parameters of the extended Taylor model, more support points N are needed, ie more cutting runs are needed during the machining process. The number of support points depends on the number of model parameter coefficients.

[0090] Only when using the extended Taylor model can different feed rates f and / or cutting depths a be selectedp .

[0091] Once the minimum number N of support points K are generated, these support points can be used to calculate the corresponding model parameters of the Taylor model.

[0092] In an embodiment, tool life and cutting conditions v c , f and a p The dataset can be used to optimize a linear regression model using the least squares error to approximate the coefficient parameters c of the transformed extended Taylor model t , m, n and q.

[0093] The coefficient parameters can then be added to the knowledge base DB as a new instance for novelty detection and / or better classification models for further training (see Figure 2 ).

[0094] In an embodiment, when determining tool life, the set of initial cutting conditions (c) is kept constant for the remainder of the cutting to ensure a constant machining state.

[0095] refer to Figure 3 , the machining process for determining tool life is shown in more detail. Figure 3 It shows how to obtain a single support point K.

[0096] As discussed above, the cutting process starts at t0 and can be performed at regular intervals t1, t2, ..., t n When the cutting process is stopped, a tool wear analysis TWA is performed. The cutting tool is moved in front of the image acquisition unit CAM (see Figure 1 ), or move the image acquisition unit CAM to a position to acquire an image of the tool TL. Take a picture of the cutting tool, segment the picture using the image segmentation model, and use the picture to calculate the side wear width v b,1 . Wear the side b,1 With the predetermined threshold v b,th If the threshold value is reached, the tool is damaged. The above procedure is repeated each time the machining process is interrupted. Flank wear value v b,2 ,…v b,n is measured / determined and compared to the threshold value v b,th Compare until this value is reached. When the threshold v is reached b,th The total processing time t m The tool life T is limited.

[0097] In an embodiment, the set of all detected wear defects may be stored in a database DB and / or displayed to an operator for manual machinability assessment. Figure 3circle) and the total processing time t preferably calculated from the process data m Stored in, for example, a database DB.

[0098] At the end of the process, when the tool life T is determined, a set of ground truth information K consisting of tool lives up to T with the observed feature vector X under cutting conditions c is provided and preferably stored in a database DB.

[0099] In summary, the above-described method may be designed as a fully automated material characterization routine or an automated material characterization routine that guides an operator through an automated process for characterizing the processability of a new material batch.

[0100] The above-described method and system allow the step of material batch characterization to be integrated into the processing process, since not all material batch variations are known prior to processing. By integrating material batch characterization into the processing process, the testing effort can be reduced, thereby reducing costs. Furthermore, the technical system is thereby able to independently evaluate novel material batches without consulting a technical expert and requesting her or him to characterize a material batch with unknown processability.

[0101] After characterizing the novel material batch, optimized processing parameters may be suggested to an operator or adopted by the control of the machine tool.

[0102] Those skilled in the art will appreciate that there are several methods for describing the processability of a material.

[0103] In the present disclosure, a tool life based machinability assessment is used as ground truth machinability data. For this machinability determination method, the wear of the cutting tool is monitored with respect to the tool life. Since the rate of tool wear is generally dependent on the cutting conditions, at least the machining speed needs to be considered. While the tool usage time can be derived from the metadata of the machining operation, a visual tool condition monitoring system can also be used to obtain the tool condition / tool ​​wear condition.

[0104] The above embodiments of the present disclosure are presented for the purpose of illustration rather than limitation. In particular, the embodiments described with respect to the drawings are only a few examples of the embodiments described in the introduction. The technical features described with respect to the system can be applied to enhance the method disclosed in the present invention and vice versa.

Claims

1. A computer-aided machining method, include: A) Supplying a material batch (M) with uncertain machinability to a machine tool (MT), B) specifying a set of processing (c1) conditions including processing speed, C) inserting a tool (TL) into the machine tool (MT), wherein the inserted tool (TL) is of a predetermined type and has a predetermined wear, D) machining said material batch (M) with a machine tool (MT), monitoring the wear of said inserted tool (TL) during machining and determining a first tool life (Tcl) of said tool (TL), E) repeating steps B, C and D to determine a second tool life (Tc2), while setting a different machining speed in step B and inserting the same type of tool in step C, F) determining coefficients of a model associated with the material batch (M) based on the machining speed, the different machining speeds, the first tool life and the second tool life (Tc1, Tc2), G) determining the processability of the material batch (M) based on the model.

2. The method according to claim 1, in, The model is a Taylor model.

3. The method according to claim 1 or 2, in, The machining is cutting and the cutting conditions (c1, c2) also include feed rate (f) and / or cutting depth (a p ), and the method further comprises: - determining at least one further tool life of the cutting tool, and - determining coefficients of an extended model associated with (describing) the material batch, wherein the extended model is preferably an extended Taylor model.

4. The method according to any one of claims 1 to 3, in, When a different processing speed is selected in step E, the remaining processing conditions remain unchanged.

5. The method according to any one of claims 1 to 4, in, In step C and step E a duplicate of the same tool (TL) is used.

6. The method according to claim 5, in, The replica is either an unused replica or a replica having the same predetermined wear.

7. The method according to any one of claims 1 to 6, further comprising: include: The processability and the coefficients of the model are added to a knowledge database (DB).

8. The method according to any one of claims 1 to 7, in, The tool is a drill, a milling tool, a cutting tool.

9. The method according to any one of claims 1 to 8, in, The determination of the first tool life (Tc1) and / or the second tool life (Tc2) is performed without removing the tool from the machine tool.

10. The method according to any one of claims 1 to 9, in, Step D includes the following sub-steps: D1) interrupting the processing at least once, preferably at regular intervals (t 1 , t 2 ,…t n ) interrupt, D2) acquiring at least one image of the tool while the machining is interrupted, D3) Determining wear of the tool based on the at least one image of the tool.

11. The method according to claim 10, in, Step D3 includes the following sub-steps: D31) segmenting the at least one image of the tool by a trained image segmentation model, D32) calculates the side wear width (v b,1 , v b,2 ,…v b,n ),as well as D33) determining whether the flank wear width satisfies a predefined flank wear width threshold value (v) associated with wear of the tool (TL); b,th ).

12. The method according to claim 11, in, The image segmentation model has been trained on a training dataset comprising images of tools, with each image being associated with one or more tool features and / or defects.

13. The method according to claim 11 or 12, in, The training has been performed by providing an image of a tool, in particular an image showing the tool in a landscape orientation; annotating the image according to features and / or defects of the tool and generating a mask for the image; And train the model on the tool's images and generated masks.

14. The method according to any one of claims 11 to 13, in, The image segmentation model is based on a pattern recognition algorithm.

15. The method according to any one of claims 1 to 14, further comprising: include: - determining processing parameters based on said processability of said material batch (M) and processing said material batch (M).

16. A computer program comprising instructions which, when executed by a computing device (EDG) associated with a machine tool (MT), cause the computing device (EDG) and the machine tool (MT) to implement a method according to any one of claims 1 to 15.

17. A computing device comprising the computer program of claim 16.

18. The computing device of claim 17, in, The computing device includes an edge device (EDG), in particular, an edge device connected to a numerical control device.

19. A system for machining, comprising a machine tool (MT) and a computing device (EDG) according to claim 17 or 18 associated with the machine tool (MT).

Images