Method for correcting a measured value of a modeled workpiece

Abstract

The invention relates to a computer-implemented method for correcting at least one measured parameter of a modeled workpiece, wherein the method comprises the following steps: - Capturing geometric parameters of the workpiece (S10); - Simulating at least one measured quantity (S12) and a first error for a first input vector using the geometric parameters; - Producing a test quantity of the workpiece (S14); - Measuring at least one measured variable (S16) with an actual error for each of the manufactured workpieces in the test batch; and - Determining a correction value for the measured quantity by comparing the at least one simulated measured quantity with the at least one measured measured quantity (S18).

Description

[0001] The invention relates to a method for correcting a measured parameter of a modeled workpiece and is therefore in the field of workpiece production, in particular simulation-based production of workpieces.

[0002] In the context of this invention, simulation-based production is a production process in which the workpieces to be manufactured are first modeled and their behavior is simulated. The simulation can, in particular, include the prediction of one or more measured parameters of the workpiece.

[0003] The invention can be used to characterize the product's behavior at a very early stage of development. If the product's function can be modeled, this model can be used to optimize the production process and eliminate development steps that can be replaced by the information provided by the model.

[0004] In the early stages of workpiece development, there is often limited data available from actual prototypes, as these prototypes are costly. However, this stage of development determines whether further investment should be made in setting up production for the workpiece. Currently, estimating individual parameters with sufficient accuracy requires a large dataset. The necessary amount of data typically only becomes available with the start of series production. By this time, however, the production investment has already been made – even if the workpiece does not behave as desired or intended.

[0005] The invention is therefore based on the objective of proposing a method that enables the estimation of a measured quantity without relying on a large amount of data relating to the workpiece.

[0006] The problem is solved by the subject matter of the independent claims. Disclosure of the invention

[0007] According to a first aspect of the invention, this problem is solved by a computer-implemented method for correcting at least one measured quantity of a modeled workpiece, wherein the method comprises the following steps: - Recording geometric parameters of the workpiece; - Simulating at least one measured quantity and a first error for a first input vector using the geometric parameters; - Producing a test batch of the workpiece; - Measuring at least one parameter with an actual error for each of the manufactured workpieces in the test batch; and - Determining a correction value for the measured quantity by comparing at least one simulated measured quantity with at least one measured quantity.

[0008] Capturing the geometric parameters of the workpiece can be achieved, for example, through electronic transfer of product parameters or through user input. In particular, the geometric parameters can be received by a system specializing in workpiece modeling. Such a system could be, but is not limited to, a CAD system.

[0009] When simulating at least one measured quantity, a value is determined that represents a function of the workpiece. For this purpose, several input parameters, collectively referred to as an input vector, are fed into a model that underlies the function of the workpiece. The input vector can, in particular, have two or more dimensions.

[0010] The measured quantity can preferably be determined using a physical model of the function.

[0011] The idea is to use simulation programs to obtain functional data of the workpieces, simulating the manufacturing distribution of the various components of the workpiece to be characterized. The result is a measured quantity with a specific distribution.

[0012] The measured value is subject to two uncertainties: the model error and the manufacturing variation. When measuring a single workpiece, it is generally impossible to predict the magnitude of these errors for that specific part. However, if several parts are measured in a statistically relevant or representative sample, it can be assumed that the manufacturing variation will change only slightly or not at all. In this case, the difference between the measurement results and the simulation results is due to the model error, from which a correction value can be determined.

[0013] The model error is an error that arises from the physical model underlying the simulation. No physical model can perfectly represent reality. Every model therefore represents an approximation of the real-world conditions concerning the workpiece. Furthermore, the closer physical models are to reality and the more boundary terms or correction factors the simulation incorporates, the more complex and therefore more resource-intensive they are to implement in a simulation. In fluid dynamics, for example, such a boundary term could be the frictional loss, which occurs when the flowing fluid rubs against the wall of a channel, thus reducing the measured quantity, 'flow'.

[0014] Simulations are typically simplified to save resources, particularly electronic resources such as storage capacity or computing power. To compensate for this simplification, the simulated measurement can be corrected or introduced with an error. However, for the reasons mentioned above, the magnitude of this error is difficult to estimate. Therefore, in accordance with this invention, a correction of the simulated measurement is performed.

[0015] The correction value can, for example, be a value that is subtracted from or added to the simulated value. In some embodiments, however, the correction value can also describe a factor by which the simulated measurement is multiplied to correct it.

[0016] As a result, the proposed method determines a measured quantity with an error, whereby the measured quantity is corrected by the correction value, thus making a more accurate prediction for the workpieces with respect to the measured quantity.

[0017] The invention solves the problem by using a method that requires only the limited data from a test batch and does not depend on serial production of the workpieces. Workpiece production can then commence or not, depending on the correction value and whether the determined difference between the simulated and the actual measured value is acceptable. The difference between the measured values ​​can be acceptable, in particular, if the corrected measured value meets the requirements placed on the workpiece.

[0018] In one embodiment, measuring the at least one measured quantity with the actual error includes forming an average of three or more measurement results for each workpiece in the test set and / or forming an average of all measurement results for all workpieces.

[0019] In one embodiment, the workpiece is an injector or a similar device for supplying a fluid.

[0020] Within the scope of this invention, a fluid can be, for example, a gas, a liquid or a flowable solid, such as fine sand or glass granules.

[0021] Various requirements can be placed on an injector regarding its characterizing parameters. For example, these requirements can include the overall size of the injector, its maximum injection capacity, and / or a maximum injection quantity. In particular, the injector can be electronically controlled, so that its function, the injection of a fluid, can be switched on and off.

[0022] Especially in the field of fluid dynamics, measured quantities relating to the flow of fluids in pipes and / or through openings are difficult to predict. Fluid dynamic boundary effects, such as turbulence at edges, friction effects of the fluid with the channel wall, and / or effects due to high viscosity, are difficult to model and often require an unjustifiable amount of effort to incorporate into a simulation.

[0023] If the method is designed for injectors and thus for these fluid dynamic boundary effects, an improved prediction regarding the measured variables and their errors is advantageously achieved.

[0024] In another embodiment, the at least one measured quantity comprises a value with the unit volume.

[0025] If the workpiece is controlled to allow a certain flow of fluid for a specific time, then a measurement parameter for the workpiece can be a volume.

[0026] The input vector can, for example, have two dimensions: a pressure on the fluid and a time the injector is active. The pressure can be caused, for example, by the gravitational force on the fluid or by an artificially generated pressure. The geometric parameters, including the corresponding tolerances or geometric errors, are also known. From the input vector and the geometry of the injector, the volume injected by the injector for the time and pressure specified in the input vector can be determined.

[0027] The input vector can, for example, without being limited to, comprise pairs of two values, one each for pressure and time. For example, without being limited, the following can be used as input vectors for the measured quantity "volume" of an injector: (2000 bar, 500 µs), (1200 bar, 500 µs) or (2000 bar, 3000 µs).

[0028] As previously described, fluid dynamic boundary effects are difficult to predict. This is especially true when very high accuracy in determining the quantity of a fluid is required; in such cases, knowing the injected volume as precisely as possible is crucial. This can play a significant role, for example, in fuel injection into an engine or in the dosing of medications. Particularly in the pharmaceutical field, an excessively high or low dose of an active ingredient dissolved in a fluid can be a matter of life and death.

[0029] The method according to this embodiment allows the flow of a workpiece, in particular an injector, to be precisely determined.

[0030] In one embodiment, the determination of the correction value is carried out using a system trained with machine learning.

[0031] If the function of the workpiece is based on simple technical relationships, the error of the measured quantity, and in particular the model error, may be known by its nature. For example, the error may correspond to a known distribution function, especially a Gaussian distribution. For more complex technical relationships, such as in fluid dynamics or thermodynamics, the boundary effects mentioned above can occur, which, individually or in combination, no longer follow a Gaussian distribution.

[0032] Machine learning systems can react to such situations because they are trained to recognize patterns or systematic structures in the input data. Therefore, using a machine learning-trained system advantageously allows even more complex distributions to be considered when determining the correction value, distributions that cannot be described mathematically, or at least not easily.

[0033] In one embodiment, the machine learning-trained system comprises an artificial neural network.

[0034] Artificial neural networks, due to their structure consisting of neurons arranged in layers, are particularly well-suited for pattern recognition. The model error that the proposed method aims to correct is based on physical laws. These laws can generally be described mathematically, albeit not easily, resulting in a pattern. This pattern can be recognized very effectively by an artificial neural network.

[0035] In a further embodiment, the method also includes the step of determining a correction value of the at least one measured quantity for a second input vector.

[0036] If the artificial neural network that determines the correction value is trained on the pattern behind the boundary terms of the measured quantity, it can be used to predict the measured quantity for further input vectors. Each input vector comprises its own set of parameters that are specified when using the workpiece.

[0037] An injector can be classified, for example, by its volume, although volume alone is not very informative. A small injector, that is, an injector with a small injection orifice, can inject a larger volume than a large injector with a large injection orifice if the pressure behind the fluid is high enough and / or the injector is active for a sufficient duration.

[0038] Consequently, the so-called operating points, i.e., points in an n-dimensional parameter space, are generally always specified along with the measured quantity, where n corresponds to the dimension of the input vector. For example, for the measured quantity "volume" of an injector, the operating point is specified with a pressure (on the fluid) and a time during which the injector injects. It is assumed that the area of ​​the injector opening remains constant. In other embodiments of an injector, the opening can also be variable, for example, via an iris aperture, so that in this case the opening area or parameters from which the opening area can be determined would also have to be specified along with the input vector.

[0039] Since edge effects can have different impacts on the measured quantity at different operating points, i.e., with different input vectors, this embodiment advantageously enables predictions about the workpiece's measured quantity to be made by applying the artificial neural network to a second or, more generally, additional input vector, without requiring the quantity to be measured again for each workpiece in the test set. In other words, the measurement and testing effort for workpieces intended for operation at multiple operating points is reduced because fewer measurements are required.

[0040] In one embodiment, the test quantity of workpieces comprises a number of 25 or less, preferably 20 or less.

[0041] Advantageously, reducing the number of test pieces to 25 or fewer can allow for an early determination of the correction value and thus a decision regarding any further investments in mass production. This can significantly reduce the scope of development and its associated costs.

[0042] In one embodiment, the test quantity is limited to a number of workpieces that just allows for a statistically relevant statement about the measured quantity of the workpieces.

[0043] The number of workpieces in the test set can be determined, in particular, by the degrees of freedom of the input vector. Generally speaking, the higher the dimension of the input vector, the larger the number of workpieces in the test set should be.

[0044] In another aspect, the invention relates to a computer program with program code for executing a computer-implemented method as described above when the computer program is executed on a computer.

[0045] In another aspect, the invention relates to a computer-readable data carrier containing program code of a computer program for carrying out a computer-implemented method as described above when the computer program is executed on a computer.

[0046] In another aspect, the invention relates to a system for correcting at least one measured parameter of a modeled workpiece, wherein the system has means to carry out the method as described above.

[0047] The means for measuring at least one measured quantity can, for example, include a measuring or test bench.

[0048] In summary, the present invention provides a method and a device for correcting at least one measured parameter of a modeled workpiece.

[0049] The described configurations and training programs can be combined in any way desired.

[0050] Further possible embodiments, developments and implementations of the invention also include combinations of features of the invention described previously or subsequently with regard to the exemplary embodiments that are not explicitly mentioned. Brief description of the drawings

[0051] The accompanying drawings are intended to provide a further understanding of the embodiments of the invention. They illustrate embodiments and, in conjunction with the description, serve to explain the principles and concepts of the invention.

[0052] Other embodiments and many of the aforementioned advantages become apparent with reference to the drawings. The elements depicted in the drawings are not necessarily shown to scale.

[0053] They show: Fig. 1 schematically illustrates the process of a method for correcting at least one measured parameter of a modeled workpiece according to an embodiment; Fig. 2 a schematic section of an injector; Fig. 3. In perspective, an injector tip; and Fig. 4 schematically illustrates the process of a method for correcting at least one measured parameter of a modeled workpiece according to a further embodiment.

[0054] In the figures of the drawings, identical reference symbols denote identical or functionally equivalent elements, parts or components, unless otherwise stated.

[0055] Fig. Figure 1 schematically shows the sequence of a method for correcting at least one measured parameter of a modeled workpiece according to an embodiment of the invention.

[0056] The process starts with step S10, in which the geometric parameters of a modeled workpiece are acquired. This acquisition can be fully automated, with the geometric parameters then being received from another system. This other system can be an external system that communicates with the system executing the process, or it can be a subsystem of the system executing the process.

[0057] Alternatively, the geometric parameters can be entered manually by a user. Preferably, the method is executed as part of a system used to model the workpiece, so that the geometric parameters of the modeled workpiece can be loaded directly from memory.

[0058] In step S12, at least one measured quantity is simulated from the geometric parameters. In some embodiments, multiple measured quantities can be simulated. The simulation comprises a virtual and / or mathematical model of the workpiece's function. For example, if the workpiece is an injector, its function is to inject a fluid. The measured quantity for an injector could, for instance, be a flow rate with the unit volume per unit time, or a volume if an injection time is specified.

[0059] The measured parameter simulated for a workpiece depends on the type of workpiece and its adjustable settings. These settings together define a multidimensional operating point, which, within the scope of this invention, is referred to as the input vector.

[0060] Returning to the example where the workpiece is an injector: Another adjustable parameter for operating the injector could be, for example, a pressure with which a fluid is forced through the injection orifice. Alternatively, a force acting on a piston, or a voltage applied to an electrically driven pump to force the fluid through the injection orifice, could be specified. The input vector always describes the possible settings for the intended operation of the workpiece.

[0061] When simulating at least one measured quantity, an attempt is made to predict the property of the workpiece characterized by the measured quantity(ies).

[0062] The input vector therefore specifies a location in the parameter space of the possible settings where the measured quantity is to be predicted.

[0063] An injector operated, for example, with variable pressure and for a variable time, i.e., with a two-dimensional input vector, can therefore be investigated for different pressures and injection times. The injected volume increases both with increasing pressure on the fluid and with increasing time the injector is active.

[0064] When simulating the measured quantity, an error is also determined, reflecting a certain inaccuracy in the measured value. This error arises, for example, from the tolerances of the machines manufacturing the workpiece and the geometry and / or components of the workpiece itself.

[0065] The workpiece's function, however, is only simulated, which can always be an approximation, but not an exact representation of reality. The simulation can be based, for example, on a mathematical model of the function or a physical simulation. In particular, the simulation can be based on physical, preferably thermodynamic and / or fluid dynamic, laws, so that the workpiece's function is virtually recreated within the simulation. This error is to be compensated for by a correction in the subsequent stages of the process.

[0066] In step S14, a test batch of the workpiece is produced. Production can be fully automated, either immediately following or in parallel with the simulation of at least one measured variable. Furthermore, production can preferably be carried out by 3D printing or sintering. In some embodiments, production can also comprise several sub-steps.

[0067] Sufficiently many workpieces are produced so that the evaluation of the measurement results from the next step yields just enough statistically significant findings. A trade-off must be made here. Producing fewer workpieces reduces production costs, measurement effort, but also the statistical significance of the measurement results. Conversely, producing many workpieces increases production costs, measurement effort, but also the statistical significance of the measurement results. The exact number of workpieces to be produced can therefore depend on the type of workpieces, the measured quantity, the manufacturing process, the production equipment, the measurement method, or other factors.

[0068] In step S16, the manufactured workpieces of the test batch are measured. At least one measurand is recorded using a measuring device or sensor corresponding to that measurand. Preferably, the measurement of the measurand follows directly after the production of the workpieces. For example, the workpieces can be clamped directly into a suitable measuring device or test bench by an industrial robot arm after production, so that the measurement can begin promptly after the workpieces are manufactured.

[0069] When measuring a quantity, each workpiece can be measured multiple times to increase the significance of the individual measurements. The combined measurement results of all manufactured workpieces reveal a measurement error. This measured error comprises the geometric error, caused by the geometry, components, and production equipment of the workpieces, and the model error, inherent in the model itself because it cannot adequately represent reality.

[0070] In step S18, the simulated and measured quantities, as well as their errors, are compared. This comparison can be performed, for example, by calculating a difference or a quotient.

[0071] In one embodiment, the measured quantities, their errors, and the model underlying the simulation of the measured quantity can all be processed by a system trained with machine learning to generate a new model for simulating the measured quantity. This new model compensates for the shift in the error of the measured quantity relative to the simulated quantity. For example, such a model can be based on the original model, with correction factors applied to the individual terms of the original model. This will be illustrated below using the following example: Fig. 2 and Fig. 3 illustrates this.

[0072] Fig. Figure 2 shows an injector 10 filled with a liquid 12. The liquid 12 is stored in a corresponding reservoir. A pressure P can be generated on the liquid 12 via a mechanism that is not relevant here and therefore not shown.

[0073] The injector 10 has a metering device 14, which can be, for example, a valve. The valve can, in particular, be electronically controlled.

[0074] The volume of liquid 12 exiting the injection port 16 can be adjusted by the pressure P and the time t that the dosing device 14 is open.

[0075] To characterize this injector 10, it is therefore of interest to know the volume for specific operating points, for example a pressure P of 2000 bar and an opening time t of 1000 µs. The input vector for the method proposed here would therefore consist of the parameters pressure and time, i.e., (2000 bar, 1000 µs).

[0076] Fig. Figure 3 shows the injection opening 16 in perspective with greater detail. The injection opening can, for example, be formed from a wall 18 that forms a hollow cylinder (tube). The injection opening 16 therefore has an inner diameter with a specific base area. Since the in Fig. Since the injection opening shown in section 3 has a circular base, the area of ​​the exit opening can be defined as the circular area A with A = π * R 2 will be calculated.

[0077] Depending on the material properties of the wall 18 and the viscosity of the liquid 12, the liquid 12 will flow out of the injection opening 16 at a specific flow rate v, as long as the metering device 14 allows flow. Using the flow rate v, the base area A, and the opening time t, the volume V of the injected liquid 12 can be determined: V = t * A * v.

[0078] This formula, however, assumes that the flow velocity v of the liquid 12 is constant, or at least constant on average, across the entire area A. At the edge of the wall 18, turbulence can occur during the injection of the liquid 12, which affects the flow velocity v. Depending on the combination of the wall 18 material and the type of injected liquid 12, other frictional forces can arise in the region of the wall 18, causing the liquid 12 to flow more slowly at the edge than in the center of the area A.

[0079] During the manufacturing of injector 10, the wall thickness, the radius R of the surface, and / or the control of the dosing device may be subject to errors. These errors can result in errors for the area A, the time t, and thus, overall, for the volume V. These and similar errors together constitute the so-called geometric or first error, which is taken into account when simulating the measured quantity.

[0080] The model error can be corrected, according to the concept of the invention, either by correcting the value of the measured quantity for the corresponding input vector or by adjusting the geometric parameters of the workpieces to be manufactured. If the latter is not possible due to specifications, particularly those arising from the use of the injector, and if the measured quantity does not achieve the desired value or precision, the workpiece can be discarded, thus saving the investment in mass production. The decision can depend on whether it is more important to know the measured quantity precisely for specific operating points or whether the measured quantity of the workpiece must achieve a specific value with a specific precision at specific operating points.

[0081] Fig. Figure 4 shows an embodiment of the method, which was already described in Fig.As outlined in section 1, step S18 is followed by a further step S20. In step S20, at least one measured value for a second input vector is determined. This is particularly relevant for workpieces that can be used under different conditions.

[0082] To illustrate this additional step S20, an injector is again considered which uses as its input vector a pressure P on the fluid to be injected and a time t for which the injector should be active.

[0083] If the injector has been analyzed for a first input vector, for example (2000 bar, 1000 µs), and a correction value for this input vector has been determined using an artificial neural network, then the artificial neural network has already successfully identified the model error underlying the simulation. With a corresponding correction by the artificial neural network, the inaccuracy caused by an inaccurate representation of reality in the simulation can therefore be compensated for. In other words, the artificial neural network can make predictions for further input vectors and predict the injected volume of an injector for a second input vector, for example (1800 bar, 1200 µs).

[0084] This additional process step simplifies the measurement setup for measuring the measurand, as the measurand only needs to be measured for one input vector. Determining the measurands for other input vectors is no longer necessary, thus reducing and simplifying development work.

Claims

[1] Computer-implemented method for correcting at least one measurement value of a modeled workpiece, the method comprising the following steps: - Recording of geometric parameters of the workpiece (S10); - Simulating the at least one measured variable (S12) and a first error for a first input vector using the geometric parameters; - producing a test quantity of the workpiece (S14); - measuring the at least one measured quantity (S16) with an actual error for each of the manufactured workpieces of the test set; and - Determining a correction value for the measured variable by comparing the at least one simulated measured variable with the at least one measured measured variable (S18). [2] The computer-implemented method of claim 1, wherein the workpiece is an injector (10). [3] Computer-implemented method according to one of claims 1 or 2, wherein the at least one measured variable comprises a value with the unit volume. [4] Computer-implemented method according to one of the preceding claims, wherein the determination of the correction value is carried out using a system trained with machine learning. [5] The computer-implemented method of claim 4, wherein the system comprises an artificial neural network. [6] A computer-implemented method according to claim 5, the method further comprising the step of: - Determining a correction value of the at least one measured variable for a second input vector (S20). [7] Computer-implemented method according to one of the preceding claims, wherein the test set comprises a number of workpieces of 25 or less, preferably of 20 or less pieces. [8] A computer program comprising program code for carrying out a computer-implemented method according to any one of claims 1 to 7 when the computer program is executed on a computer. [9] A computer-readable data carrier comprising program code of a computer program for carrying out a computer-implemented method according to any one of claims 1 to 7 when the computer program is executed on a computer. [10] System for correcting at least one measurement value of a modeled workpiece, the system comprising means for carrying out the method according to any one of claims 1 to 7.

Images