Method for determining a quality feature during machining
Patent Information
- Application Number
- PCT/DE2026/100274
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-03-27
- Filing Date
- 2026-03-03
- Publication Date
- 2026-10-01
Smart Images
Figure DE2026100274_01102026_PF_FP_ABST
Abstract
Description
[0001] METHOD FOR DETERMINING A QUALITY CHARACTERISTIC
[0002] DESCRIPTION OF EXCITING MACHINING
[0003] Technical field
[0004] The present invention relates to a method for determining a quality characteristic of a workpiece that is machined by machining.
[0005] State of the art
[0006] Machining can involve processes such as milling, but more generally also turning or drilling. As explained in detail below, a preferred application lies in the manufacture of components for turbomachinery, particularly for aircraft engines. Their components can be exposed to high temperatures and significant mechanical loads during operation. This can apply, for example, to components located within the gas duct, but also to components outside of it, such as rotor discs. Regardless of the specific application, the high demands placed on the components may necessitate the use of special materials and / or strict criteria for dimensional accuracy. This can result in specific requirements for the machining of such components. This is intended to illustrate an advantageous application environment, but not to limit the scope of the topic at this stage.
[0007] Description of the invention
[0008] The present invention addresses the technical problem of providing an advantageous method for determining a quality characteristic of a workpiece undergoing machining. This is solved by the features according to claim 1. In addition to machining the workpiece with a machining tool, the method determines the toolpath of the machining tool and at least one dynamic process parameter, e.g., the spindle torque (see below for further details). Based on the toolpath, together with the shape or geometry of the machining tool ("tool geometry"), a virtual workpiece contour is determined. Subsequently, a deviation of this virtual workpiece contour from a target state is determined. This comparison is performed at at least some points along the virtual workpiece contour, i.e., at specific points and not necessarily across the entire contour.As explained in detail below, these evaluation or measuring points can be defined in such a way that geometric features are measured on the actual workpiece after machining and / or have been measured on identical workpieces.
[0009] Using a machine learning system (MLS), a quality characteristic is predicted based on the deviation of the virtual workpiece contour from the target state, specifically a geometric deviation of the machined workpiece from the target state. This quality characteristic, or geometric deviation, is the output value of the trained MLS. In addition to the deviation of the virtual workpiece contour, at least one dynamic process parameter of the machining operation is used as input. This could be, for example, the torque of the machining tool, such as a spindle torque measured during machining (see below for details).
[0010] By considering both the deviation of the virtual workpiece contour and at least one process parameter via the MLS, the influence of both the machine and process factors can be taken into account. Even though the geometry of the finished workpiece, i.e., its deviation from a target state, shows a strong correlation with the dynamic variables, i.e., at least one dynamic process parameter, this relationship can be difficult to model analytically in detail. By training the MLS on the geometric deviations measured in machined workpieces, the inventors have obtained an MLS with sufficiently accurate prediction quality, i.e., with an acceptable deviation between the measured and predicted geometric deviation.
[0011] The total deviation results from the sum of the machine's positioning inaccuracy and the influence of the dynamic process parameter (e.g., the effects of process forces). Process forces can be measured, for example, by torque and are considered "dynamic" because they can fluctuate during machining. Generally, "dynamic" means, for example, that the relevant process parameter results directly and / or immediately from the ongoing interaction between the tool and the workpiece. The dynamic process parameter can be subject to fluctuations during machining; in particular, it can be a process force or a quantity directly, immediately, and / or correlated with it.
[0012] The positioning accuracy, which can optionally be recorded additionally, does not result directly from the interaction between tool and workpiece, for example, and is not a dynamic process parameter.
[0013] In general, geometry- and process parameter-based prediction of geometric deviations can be advantageous, for example, with regard to throughput in manufacturing, by reducing the number of actual measurements required on the workpiece. In other words, measurements and acceptance tests after machining can be at least partially virtualized or specifically supported (unnecessary measurements can be reduced, and critical measurements can be initiated selectively). Furthermore, workpiece information generated or collected early on, e.g., during manufacturing, can increase process transparency and reveal potential problems. These could include, for example, positioning problems on the axis(es) of the machining center, which may arise due to wear.By considering geometric and dynamic process parameters, the positioning behavior of the machine tool is captured ("how is it set on the axis(s)?") and, at the same time, the forces acting or having acted on the machining tool at a given position are depicted. Preferred embodiments can be found throughout the disclosure and especially in the dependent claims, whereby the description of the features does not always differentiate in detail between process, application, or device aspects; in any case, the disclosure is implicitly to be read with regard to all claim categories.
[0014] According to a preferred embodiment, the tool geometry is determined by measuring the machining tool. This can be done, for example, with a scanner, either at a point or as a result of relative movement along a line. This creates a three-dimensional image of the machining tool's surface, or at least of a region of the surface. For this purpose, a laser scanner can be used, for example, with its laser beam scanning the surface of the machining tool. Regardless of the specific scanner type, the machining tool can be measured, for example, at least before the workpiece is machined. When determining the virtual component contour, the actual geometry of the machining tool is then taken into account; that is, any deviations of the machining tool from a target state, caused by wear or other factors, are considered.
[0015] According to a preferred embodiment, at least one process parameter is measured during the machining of the workpiece, preferably continuously or at least quasi-continuously. This allows, with respect to the toolpath, at least one additional process parameter to be available, for example, at those points on the virtual workpiece contour where its deviation is determined. The process parameter is sensor-acquired or read out during the machining process; see below for details.
[0016] According to a preferred embodiment, at least one process parameter includes the torque of the machining tool. This can also be evaluated indirectly, for example, via the motor current of the electric motor that rotates the machining tool during workpiece machining. Regardless of the specific measurement method, the spindle torque can be well-suited because it is easily measurable and also shows a strong correlation with the quality characteristic, in particular the geometric deviation. According to a preferred embodiment, the toolpath is read from the machine tool in which the workpiece is machined with the machining tool. The virtual workpiece contour is then determined based on the read toolpath.Preferably, the toolpath is read out during machining; that is, the positions set on the axis(es) of the machine tool for the machining tool are read out and saved during the machining of the workpiece (they can be used in situ or supplied for later evaluation). Preferably, the actual machining tool position is determined during machining, and the workpiece contour is calculated from this, together with the measured tool geometry (surface image of the machining tool).
[0017] According to a preferred embodiment, the tool orientation is also read from the machine tool. The tool orientation can be represented, for example, by one or more degrees of freedom, such as an angle of attack relative to the direction of travel and an angle of attack perpendicular to the direction of travel. Regardless of these details, by additionally considering the tool orientation, a contact point between the machining tool and the workpiece can be determined even in more complex processes or geometries, such as milling a three-dimensionally contoured surface. In such cases, the virtual workpiece contour can also be reliably determined.
[0018] According to a preferred embodiment, a parameter read from the machine tool is subjected to a data transformation, which can, for example, concern the toolpath and / or the tool orientation. Here, the data from the machine coordinate system, e.g., the positions of the tool axis(es) recorded over time, are transformed into a coordinate system of the workpiece. Thus, the toolpath and / or the tool orientation can be determined from the read-out axis positions. In other words, the respective contact point or, in the time integral, a contact area between the tool and the workpiece is calculated from the axis positions and, preferably, the measured tool geometry. According to a preferred embodiment, depending on the predicted geometric deviation, the machining of the workpiece is controlled and / or the machining process is interrupted. An interruption can, for example,This can be used for readjustment or part replacement, for example, if at least one process parameter indicates (excessive) wear. A regulatory intervention can, for example, involve adjusting the spindle speed.
[0019] According to a preferred embodiment, the machine learning system comprises a linear model, although in general a non-linear artificial neural network (ANN) can also be used.
[0020] As mentioned above, the total deviation is the sum of the machine's positioning inaccuracy and the dynamic process parameter (effects of process forces). In the case of a linear model, the deviation A can be linearly related to the input variables, e.g., torque D, machine positioning inaccuracy P, and optionally other input variables or process parameters:
[0021] A = xD + yP +z
[0022] The parameters x, y, z (constants) can be determined using the training process. These parameters are then fitted to the target variables using a regression procedure.
[0023] Other models or relationships, which can be determined in particular using artificial neural networks (ANN), are also possible.
[0024] D,P -> KNN -> A
[0025] The KNN can have, for example, 10-20 layers, e.g., exactly 12 layers in one implementation, including input and output layers. Each layer can have, for example, 5-10 nodes, e.g., exactly 7 nodes in one implementation. The training data can include the parameters D, P, and A, which are measured in production over a specific period, e.g., over one month in one implementation, for several components, e.g.,
[0026] 80 components were measured and / or determined. In one version, the data D(t, component), P(t, component), A(t, component) comprise between 40 and 200 pairs of values, e.g., 120.
[0027] The machine learning system can include one or more other models in addition to or as an alternative to the linear or KNN model (relationship / structure) mentioned above, e.g., one or more of the following models: random forest, decision tree, support vector machine, gradient boosting machines, ridge regression, lasso regression, polynomial regression.
[0028] The invention also relates to a method for manufacturing a component for a turbomachine, wherein the component is manufactured by machining a workpiece and a quality characteristic, in particular a geometric deviation, is predicted in the manner described herein. As explained at the outset, turbomachine components may be subject to special quality requirements, particularly aircraft engine components. The component may, for example, be a disk, in particular a turbine or compressor disk, or a blisk (bladed disk), in particular a turbine or compressor blisk. As explained in detail below, the method can be particularly advantageous for such critical and / or so-called Class 1 components.
[0029] A preferred embodiment relates to a method for monitoring the production of a component manufactured as described above. In other words, the manufacturing process may preferably include monitoring. If the quality characteristic reaches a predefined threshold, a countermeasure can be taken, such as recalibrating the production machine, performing maintenance, and / or inspecting and modifying production equipment. A maintenance measure could, for example, be replacing a worn axis of the production machine or replacing the machining or...
[0030] A cutting tool. A manufacturing aid can be, for example, a component holding device.
[0031] Alternatively or additionally, a countermeasure can also be taken at the software level, for example, by determining a different, e.g., more stable, toolpath (x(t),y(t),z(t),t) for an NC program and using it for machining subsequent components. Monitoring can be particularly advantageous in the production of critical components (see vome) where intervention in the manufacturing process is not desired. Immediately after the completion of a component, the quality characteristic, e.g., the deviation, of the manufactured component can be predicted or determined using the invention. Such predictions can be used within the framework of monitoring or, in particular, continuous process assessment or evaluation in manufacturing, e.g., for trend analysis or trend monitoring.When the trend of the quality characteristic reaches a critical threshold, this can be detected and countermeasures can be taken, see above.
[0032] The invention further relates to a method for manufacturing a turbomachine, in particular an aircraft engine, wherein at least one component of the turbomachine is manufactured by machining in a manner described herein.
[0033] The invention further relates to a method for training a machine learning system (MLS) to predict a quality characteristic of a workpiece that is machined using a machining tool. The training includes
[0034] a reading in of at least one process parameter that was recorded during the machining of a workpiece;
[0035] a reading of a virtual workpiece contour of the workpiece, which was determined based on a tool geometry and a toolpath;
[0036] Determining a deviation of the virtual workpiece contour from a target state at at least some points of the virtual workpiece contour;
[0037] a comparison of a geometric deviation measured on the machined workpiece relative to a target state with a geometric deviation predicted by the MLS based on the deviation of the virtual workpiece contour and at least one process parameter.
[0038] As described above, the MLS can be trained, in particular, to predict deviations in critical and / or so-called Class 1 components. For example, it can be trained to predict quality measures on blisks. The range within which the input variables of the training data should be varied can be determined using historical manufacturing data. For example, data is collected from a large number of components, e.g., from 100 components, and then applied to future components. The data can be collected or recorded during manufacturing and stored centrally, e.g., on a hard drive.
[0039] Furthermore, the invention relates to a system for machining a workpiece, which includes a machining center. The system is configured to determine a virtual workpiece contour and its deviation from a target state.
[0040] Preferably, the system also includes a machine learning system designed to predict geometric deviations.
[0041] The invention further relates to a computer program product comprising instructions which, when executed by a computer unit, cause it to predict a geometric deviation.
[0042] Brief description of the drawings
[0043] The invention will now be explained in more detail using an exemplary embodiment, without further distinction being made between the different claim categories.
[0044] In detail, it shows
[0045] Figure 1 shows a system for machining a workpiece and for determining a geometric deviation of the workpiece;
[0046] Figure 2 a,b a summary of some process steps in a flowchart;
[0047] Figure 2c shows a schematic representation of KNN;
[0048] Figure 3 shows a turbofan engine in a schematic longitudinal section to illustrate an application area. Preferred embodiment of the invention.
[0049] Figure 1 shows a system 10 comprising a machine tool 11 with a machining tool 12, in this example a milling tool 13. A workpiece 20 is machined with this tool by moving along a toolpath 15. The toolpath 15 is derived from an NC data set, based on which the machine tool 11 or its control unit 14 moves the machining tool 12.
[0050] In this process, during the machining of the workpiece, the coordinates a, b, c, which are assumed on the individual axes of the machine tool over time, are read from the machine tool 11. These machine-internal coordinates are transformed into a coordinate system of the workpiece, i.e., converted into positional coordinates x, y, z. This provides a virtual representation of the toolpath 15 traversed during the machining process. Additionally, a 3D scanner 30 measures the machining tool 12, thus creating a three-dimensional surface image of the machining tool 12.
[0051] Based on the tool geometry 32 determined in this way, a virtual workpiece contour 40 can be defined together with the toolpath 15 (see dotted line). By comparing the virtual workpiece contour 40 with a target state 41 (dashed line), a deviation 45 can be determined. The deviation 45 represents, for example, the wear-related influence of the machine tool 11 and the machining tool 12.
[0052] In addition, at least one process parameter 25 is read out during the machining of the workpiece 20, in the present example a torque 26 of the machining tool 12. This represents a dynamic process influence.
[0053] Using a machine learning system 50, a quality characteristic 60 can then be determined based on at least one process parameter 25, which is recorded during machining, and the virtual workpiece contour 40 or deviation 45. This characteristic is a geometric deviation 61 of the machined workpiece 20 from a target state. During the training of the MLS 50, the deviation 65 is measured on correspondingly machined components 20, i.e., at least at some points 65.1-65.3.
[0054] The input data consists of the geometric deviation from the target model (distance) and the torque at the point of deviation. Using these two parameters, a linear model is created that describes the geometric measurement data (deviations from the target) measured on a coordinate measuring machine as the target variable. Data from a production history, comprising 50-100 components, is used to train this model (parameter adjustment).
[0055] Figure 2a summarizes some steps in determining a quality characteristic of a machined workpiece. During machining 100 of the workpiece, the toolpath along which the machining tool moves is determined 101. Furthermore, the tool geometry is measured 102, whereby the virtual workpiece contour is determined based on the tool geometry and the toolpath 103. Based on the virtual workpiece contour, the deviation from a target state is determined 104.
[0056] During machining, at least one process parameter is determined 105, preferably the spindle torque. Based on this at least one process parameter and the deviation of the virtual component contour, the geometric deviation of the workpiece from the target state is then predicted 106.
[0057] Figure 2b summarizes the process steps for training the MLS. At least one process parameter, preferably the spindle torque, is read in. Furthermore, the virtual workpiece contour is read in 111 and its deviation from the target state is determined 112. The geometric deviation is measured at at least some points on the machined workpiece 115, and this measurement is compared with the geometric deviation predicted based on the process parameter and the deviation 120. Figure 2c illustrates a simplified representation of an Artificial Neural Network (ANN) with an input layer (left in the figure) and an output layer (right in the figure). Several hidden layers are arranged between them, but only two hidden layers are shown here. An actual ANN can, for example, have ten hidden layers, i.e., a total of twelve layers including the input and output layers.
[0058] Figure 3 shows an axial turbomachine 70, specifically an aircraft engine 71. Functionally, this engine is divided into a compressor 72, a combustion chamber 73, and a turbine 74. Both the compressor 72 and the turbine 74 are composed of several stages, each with a stator and a rotor (not referenced individually). In the compressor 72, intake air is compressed. This compressed air is then combusted in the combustion chamber 73 with added fuel, e.g., kerosene. The resulting hot gas is expanded in the turbine 74, and the kinetic energy gained is used to drive the compressor 72 and to generate thrust. Due to the high mechanical and / or thermal stresses, special quality requirements may apply to various components 75. Only a few rotor disks are referenced here as examples.
[0059] System 10 Machining machine 11 Machining tool 12 Milling tool 13 Control unit 14 Tool path 15 Workpiece 20 Process parameters (dynamic) 25 Torque 26 3D scanner 30 Determined tool geometry 32 Virtual workpiece contour 40 Target state 41 Deviation 45 Machine learning system (MLS) 50 Quality characteristic 60 Geometric deviation 61 Some locations (measurement) 65.1-65.3. Machining (workpiece) 100 Determining (toolpath) 101 Measuring (tool geometry) 102 Determining (virtual workpiece contour) 103 Determining (deviation from target state) 104 Determining (dynamic process parameter) 105 Predicting (geometric deviation) 106 Reading (dynamic process parameter) 110 Reading (virtual workpiece contour) 111 Determining (deviation) 112 Measuring (geometric deviation) 115 Adjusting 120 Coordinates (machine internal, axes) a, b, c Position coordinates x, y, z
Claims
REQUIREMENTS 1. Method for determining a quality characteristic (60) of a workpiece (20) that is machined, comprising the steps: Machining (100) of the workpiece (20) with a machining tool (12); Determining (101) a toolpath (15) of the machining tool (12) and at least one dynamic process parameter (25) during machining; Determining (103) a virtual workpiece contour (40) of the workpiece (20) based on a tool geometry (32) of the machining tool (12) and the detected toolpath (15); Determine (104) a deviation (45) of the virtual workpiece contour (40) from a target state (41) at at least some points of the virtual workpiece contour (40); Predictions (106) of a geometric deviation (65) of the machined workpiece (20) from a target state using a machine learning system (50), based on o the deviation (45) of the virtual workpiece contour (40); o of at least one dynamic process parameter (25).
2. Method according to claim 1, wherein the tool geometry (32) is determined by measuring (102) the machining tool (12) and the virtual workpiece contour (40) is determined based on the measured tool geometry (32).
3. Method according to claim 1 or 2, wherein the at least one dynamic process parameter (25) is measured during the machining (100) of the workpiece (20).
4. Method according to one of the preceding claims, wherein the at least one dynamic process parameter (25) comprises a torque of the machining tool (12) and / or wherein the prediction (106) of a geometric deviation (65) of the machined workpiece (20) from a target state is carried out using a machine learning system (50), additionally based on a positioning accuracy.
5. Method according to one of the preceding claims, wherein the toolpath (15) is read out from a machine tool (11) in which the workpiece (20) is machined with the machining tool (12) and the virtual workpiece contour (40) is determined on the basis of the read-out toolpath (15).
6. Method according to one of the preceding claims, in which a tool orientation is read out from a machine tool (11) in which the workpiece (20) is machined with the machining tool (12) and the virtual workpiece contour (40) is determined on the basis of the read-out tool orientation.
7. Method according to claim 5 or 6, wherein a quantity read from the processing machine (11) is subjected to a data transformation, in particular from a coordinate system of the processing machine (11) into a coordinate system of the workpiece (20).
8. A method according to any of the preceding claims, wherein the machine learning system (50) comprises at least one of the following models: linear model, nonlinear model, artificial neural network (ANN), random forest, decision tree, support vector machine, gradient boosting machines, ridge regression, lasso regression, polynomial regression.
9. Method for manufacturing a component (75) for a turbomachine (70), wherein the component (75) is produced by machining (100) a workpiece (20) with a machining tool (12); a quality feature (160) of the workpiece (20) is predicted according to one of the preceding claims.
10. Method according to claim 9 wherein the component (75) is preferably a disk, in particular a turbine or compressor disk, or a blisk, in particular a turbine or compressor blisk.
11. Method for monitoring the production of a component (75) which is produced in a method according to claim 9 or 10, wherein, in particular when the quality characteristic reaches a predefined threshold, a countermeasure is taken, in particular one or more of the following measures are carried out: recalibration of the production machine, maintenance measures, inspection and modification of production aids.
12. Method for training a machine learning system (50) to determine a quality characteristic (60) of a workpiece (20) that is machined with a machining tool (12), comprising the steps: Reading (110) at least one dynamic process parameter (25) that was recorded during machining (100) of the workpiece (20); Reading (111) a virtual workpiece contour (40) of the workpiece (20), which was determined on the basis of a tool geometry (32) of the machining tool (12) and a toolpath (15); Determine (112) a deviation (61) of the virtual workpiece contour (40) from a target state (41) at at least some points of the virtual workpiece contour (40); Matching o a geometric deviation from a target state measured at at least some points of the machined workpiece (20); with a geometric deviation (65), which the machine learning system (50), based on o the deviation (45) of the virtual workpiece contour (40); o of at least one dynamic process parameter (25); predicts.
13. System (10) for machining a workpiece (20) and determining a quality characteristic (60) of the workpiece (20), with a machine tool (11) for machining the workpiece (20); wherein system (10) is configured to, to determine a toolpath (15) of a machining tool (12) and at least one dynamic process parameter (25) during machining; to determine a virtual workpiece contour (40) of the workpiece (20) based on a tool geometry (32) of the machining tool (12) and the detected toolpath (15); to determine a deviation (45) of the virtual workpiece contour (40) from a target state (41) at at least some points of the virtual workpiece contour (40).
14. System (10) according to claim 13, comprising a machine learning system (50); wherein the machine learning system (50) is configured to predict a geometric deviation (65) of the machined workpiece (20) from a target state, based on o the deviation (45) of the virtual workpiece contour (40); o of at least one dynamic process parameter (25).
15. Computer program product comprising instructions which, when executed by a computing unit of a system (10) according to claim 13, cause the computing unit to predict a geometric deviation (65) of the machined workpiece (20) from a target state, based on o the deviation (45) of the virtual workpiece contour (40); o of at least one dynamic process parameter (25).