DE102019200482A1 - Method and device for the automated machining of a workpiece with a machine tool - Google Patents

Abstract

The invention relates to a method for machining one or more workpieces (2) with an automated machine tool (1), a machining head (6) and a workpiece (2) being moved relative to one another along a machining trajectory, the machining trajectory being determined by machining parameters, with the following steps, which are each carried out for machining one or more workpieces (2): - machining (S1) the workpiece (2) according to predetermined machining parameters along the machining trajectory; - measuring (S2) one or more machining variables, in particular one or more courses of the processing sizes during the processing of the workpiece (2) along the processing trajectory; - optimization (S4) of the processing parameters using an optimization method based on a cost function and at least one boundary condition, a modeled processing size and one of these modeled processing The uncertainty variable assigned to the measurement variable is determined using a machining model which is provided as a regression model, depending on the machining parameters considered in the optimization, the boundary condition depending on a range of values dependent on a predetermined probability around the modeled machining variable, the modeled machining variable, the uncertainty variable and the specified probability indicates the range of values around a modeled machining size.

Description

Technical field

The invention relates to methods for the automated control of a machine tool for machining a workpiece. In particular, the present method relates to the optimization of a machining trajectory according to which the workpiece is machined by the machine tool.

Technical background

When machining workpieces using automated machine tools, various factors can influence the machining time, the quality of the machining and the quality of the machined workpiece. As a rule, a machining trajectory is specified for the machining, with which a machining head of the machine tool moves relative to the workpiece to be machined. The machining trajectory is traversed according to several machining parameters which are decisive for the machining time, the quality of the machining and the quality of the machined workpiece.

The processing methods can include machining techniques such as drilling, milling, sawing, welding and the like, cutting techniques such as e.g. Laser cutting, heat treatments, e.g. Soldering, sintering, etc., as well as connection techniques such as Bonding, gluing, etc., and the like, and generally include other methods in which a tool moves relative to a workpiece, thereby material, i. H. executes material and / or shape-changing measures.

Disclosure of the invention

According to the invention, a method for machining a workpiece with a machine tool according to claim 1 and a corresponding device and a machine tool according to the independent claims are provided.

Further configurations are specified in the dependent claims.

• - Bearbeiten des Werkstücks gemäß vorgegebener Bearbeitungsparameter entlang der Bearbeitungstrajektorie;
• - Messen einer oder mehrerer Bearbeitungsgrößen, insbesondere einen oder mehrere Verläufe der Bearbeitungsgrößen während des Bearbeitens des Werkstücks entlang der Bearbeitungstrajektorie;
• - Optimieren der Bearbeitungsparameter mithilfe eines Optimierungsverfahrens basierend auf einer Kostenfunktion und mindestens einer Randbedingung,

wobei eine modellierte Bearbeitungsgröße und eine dieser modellierten Bearbeitungsgröße zugeordnete Unsicherheitsgröße mithilfe eines Bearbeitungsmodells, das als ein Regressionsmodel vorgesehen ist, abhängig von den bei der Optimierung betrachteten Bearbeitungsparametern bestimmt werden, wobei die Randbedingung von einem von einer vorgegebenen Wahrscheinlichkeit abhängigen Wertebereich für die modellierte Bearbeitungsgröße abhängt, wobei die modellierte Bearbeitungsgröße, die Unsicherheitsgröße und die vorgegebene Wahrscheinlichkeit den Wertebereich um eine modellierte Bearbeitungsgröße angeben.According to a first aspect, a method for machining a workpiece with an automated machine tool is provided, wherein a machining head and a workpiece are moved relative to one another along a machining trajectory, the machining trajectory being determined by machining parameters, with the following steps:

• - Machining the workpiece according to predetermined machining parameters along the machining trajectory;
• - Measuring one or more machining variables, in particular one or more courses of the machining variables during machining of the workpiece along the machining trajectory;
• Optimization of the machining parameters using an optimization method based on a cost function and at least one boundary condition,

wherein a modeled machining variable and an uncertainty variable associated with this modeled machining variable are determined with the aid of a machining model which is provided as a regression model, depending on the machining parameters considered in the optimization, the boundary condition depending on a range of values for the modeled machining variable which is dependent on a predetermined probability, where the modeled processing size, the uncertainty size and the specified probability indicate the range of values around a modeled processing size.

The machining of a workpiece in an automated machining process is usually carried out by moving a machining head of a tool relative to the workpiece, so that the material or shape of the workpiece is changed. The movement of the machining head takes place along a machining trajectory and can be characterized by machining parameters. The machining trajectory can be defined as a trajectory parameterized by the machining parameters or can be divided into trajectory sections which are e.g. linear section profiles are defined by respective separate processing parameters. Depending on the type of machining method, the machining parameters can be the feed rate, a feed force of the machining tool on the workpiece, a tool speed (in the case of rotating machining heads, such as for example when drilling), a machining temperature of the machining head and the like.

The trajectory sections of the machining trajectory are defined by section times or section sections. The section times or section sections can have the same or different durations or lengths. The processing method carries out processing in each of the trajectory sections in accordance with the processing parameters specified for this. Parameter transitions can be smoothed by interpolation according to their physical meaning. For example, a feed rate or a tool temperature cannot be changed suddenly.

Depending on the choice of the processing parameters of the processing in the individual processing trajectories, the speed, duration and / or the quality of the processing is better or worse. It can therefore be provided that the cost function considered for the optimization depends on the total time for the processing and / or one or more quality parameters that indicate the quality of the processing and that depend on the modeled processing size, the one or more quality parameters in particular is determined by a course of the modeled processing variable, and in particular can be specified as an average, maximum value, minimum value of the course of the modeled processing variable for the entire processing trajectory or separately for one, some or all of the trajectory sections.

Such a machining process is optimized in accordance with the cost function and taking the boundary condition into account. The boundary condition can be considered separately during the optimization process or as part of the cost function. The optimization can be carried out automatically by an optimization process in which the quality of the processing with regard to a quality parameter with the optimized processing parameters is ensured by one or more corresponding boundary conditions.

The boundary conditions are defined with regard to the processing sizes. Since the processing sizes cannot be measured in the optimization computing process, they are provided as a modeled processing size according to the processing model. Since the machining model is often faulty, regression models are used as machining models for the above method, which in addition to the model value also provide an uncertainty variable that indicates the reliability of the model value. By defining the boundary condition with a range of values in which the true value of the machining size under consideration lies with a probability predetermined by the boundary condition, it is possible to avoid the quality of the machining for too long by avoiding machining variables that are outside the range of values. Although it is possible that the true value of the processing variable is temporarily outside the value range defined by the boundary condition, this is only the case in comparatively short periods of time due to the given probability.

In accordance with the above method, the machining target, which is predetermined by minimizing the total costs, is now optimized with probabilistic compliance with the boundary conditions. For this purpose, a gradient of the processing model of the target size is calculated and followed so far that the boundary conditions are met at least with a certain predetermined probability, i.e. that the limit value for the processing size in question lies with a predetermined probability within a range of values for the processing size in question.

For example, a certain feed rate of the tool requires a certain feed force (machining size) that acts on the workpiece. When machining a workpiece, this feed force can be recorded against the feed speed (machining parameters) and a corresponding machining model can be taught. As a boundary condition during optimization, the feed force can be limited for reasons of tool protection or wear reduction. Training the machining model results in a regression between the feed force and the feed speed with an uncertainty margin. The boundary condition that applies to the optimization can now be specified so that with a certain probability a specified limit value for the feed force (machining quantity) is not exceeded. This method enables an improved trajectory determination for machining trajectories of machine tools.

Such a processing method can be optimized automatically by an optimization process, in which a cost function is also defined in accordance with the quality parameters and the processing parameters for each trajectory section are optimized based on the cost function.

The cost function specifies a cost value depending on the quality parameters, each of which represents an evaluation of the machining process or a part thereof. For example, the quality parameters can include the time required for machining a workpiece, the occurrence of vibrations during machining of the workpiece, the material wear of the machining head and the machining precision with respect to the machined workpiece, a temperature of the workpiece, a temperature of the machining head, and / or a feed force, that affects the machining head.

The above method uses a trainable machining model about the physical interrelationships when machining a workpiece in order to improve the optimization of the machining trajectory. This is made possible by choosing the model as a machining model that on the one hand provides a regression curve of the model values and corresponding uncertainties for the individual model values of the regression curve. The trainable model thus enables the safe optimization of the cost functions. When using z. B. a gradient descent method, the model can not only calculate a gradient for the optimization, but also the distance of its safety limit, in which the modeled machining parameters are specified, can be determined.

Basically, the optimization process is carried out by optimizing the machining parameters in accordance with an optimization goal, which can be specified, for example, as minimizing a machining time or generally as minimizing costs according to a cost function, and boundary conditions, such as B. the quality and machining, and wear of the tool can take into account. The processing parameters are optimized by minimizing costs, while the boundary conditions are observed.

It can be provided that the machining model is selected as a trainable machining model, a learning process being carried out on the basis of the one or more machining variables measured with respect to machining with the predetermined machining parameters. It can thereby be provided that the trainable machining model is updated after each or after several machining runs, the machining result being taken into account in the learning process depending on the machining parameters used.

The trainable machining model can be any regression model that can also provide an uncertainty in the model value with respect to each of the machining parameters. In addition, regression models are advantageous, which also specify a gradient with respect to each of the processing parameters. In particular, a Gaussian process model, a Bayesian linear regression model, a regularized kernel ridge regression model, and / or a Bayesian neural network and the like can be used.

Figure list

• 1 eine schematische Darstellung einer automatisierten Werkzeugmaschine, die zur Durchführung eines Verfahrens zum Bearbeiten von Werkstücken ausgebildet ist,
• 2 eine schematische Darstellung eines Flussdiagramms zur Veranschaulichung des Verfahrens zur Bearbeitung von Werkstücken durch die Werkzeugmaschine der 1; und
• 3 ein Diagramm zur Veranschaulichung der Berücksichtigung eines Unsicherheitswert für eine modellierte Bearbeitungsgröße.

Embodiments are explained below with reference to the accompanying drawings. Show it:

• 1 1 shows a schematic representation of an automated machine tool that is designed to carry out a method for machining workpieces,
• 2nd is a schematic representation of a flowchart to illustrate the method for machining workpieces by the machine tool of the 1 ; and
• 3rd a diagram illustrating the consideration of an uncertainty value for a modeled processing variable.

Description of embodiments

1 shows a machine tool 1 with a machining head 6 that over a wave 7 can be driven in rotation. The machining head 6 can in particular be designed as a cutting tool through which a workpiece 2nd can be edited. The machining head 6 is powered by a drive unit 3rd operated by suitable movement of the machining head 6 of the tool the workpiece 2nd processed material or shape changing. For example, the machining head 6 a cutting tool such as a drill, a milling head, a laser, a heat treatment tool such as a soldering pin, a sintering tool, a welding head (resistance welding), etc., an ultrasonic head (ultrasonic bonding), an adhesive nozzle or the like.

The drive unit 3rd is still able to propel the machining head 6 to effect the workpiece 2nd to work along a route. The machining head 6 can continue with a sensor system 9 connected to a force between the machining head 6 and the workpiece 2nd acts, a temperature of the machining head 6 and / or the workpiece 2nd or to measure further resulting output variables. Furthermore, the workpiece 2nd with a sensor system 8th be provided, which can measure a vibration, a temperature and / or another property of the workpiece during processing.

Using a control unit 10th that the drive unit 3rd so that the workpiece 2nd The sensor values of the sensor systems can be processed in accordance with a processing trajectory 8th , 9 can be evaluated as described below.

The processing is controlled by the control unit 10th corresponding to a machining trajectory that is determined by machining parameters. The processing trajectory can comprise several trajectory sections. The machining parameters specify machining specifications for machining the workpiece. The machining parameters can, for example, specify an infeed, a speed, a depth of cut, a feed speed and / or the like for each trajectory section of the machining trajectory. Are different parameter values of the processing parameters for different trajectory sections provided, the transition between the parameter values of the machining parameters can be smoothed in order to avoid sudden changes in the parameter values. In particular, linear interpolation can in particular be carried out during the transition of the processing between the trajectory sections between the parameter values. In this way, parameter jumps that are not technically feasible can be avoided. Alternatively, the processing trajectory can be defined without sections, the course of the processing trajectory being defined by courses of one or more processing parameters.

During processing according to the processing trajectory, the control unit can 10th the sensor values related to the machining head 6 (about the sensors 9 ) and with regard to the workpiece (via the sensors 8th ) capture, take into account and store temporarily.

Using the flowchart of 2nd The following is the procedure for operating the machine tool 1 described in more detail.

In step S1 becomes the workpiece 2nd processed in accordance with an initial machining trajectory or by an optimization, with corresponding machining parameters. For this purpose, for example, the feed rate, an infeed, a speed and / or a depth of cut can be specified in order to carry out an exemplary sawing process with a circular saw. For a soldering process, for example, the feed rate, a solder material supply and a soldering temperature can be specified.

Simultaneously with the machining, which is done by moving the machining trajectory through the machining head 6 done in step S2 Sensor values of the sensors 8th , 9 detected, which relate to the resulting machining variables, such as one against the feed direction on the workpiece 2nd acting feed force, a rotational torque for driving the circular saw with the predetermined speed, a tool temperature, a workpiece temperature during machining, a vibration of the workpiece during machining and the like. The courses of the respective processing variables are saved according to the processing trajectory.

Based on the processing parameters used and the measured processing sizes in step S3 trains a machining model that models the machining process. The processing model can be, for example, a Gaussian process, a Bayesian linear regression, regularized kernel ridge regression, Bayesian neural networks and the like. In general, the machining model is selected such that it can provide a regression of at least one of the machining variables based on the machining parameters. The machining model is also selected such that, in addition to the estimated model values of a machining variable, it can also output an uncertainty value for the model value of the machining variable in question. The uncertainty value defines, for example by specifying a factor of a standard deviation, an uncertainty range around the model value of the machining model, within which the true value of the machining variable in question lies with a predetermined uncertainty.

According to an approximation, the following are in step S4 optimized the processing parameters for the trajectory sections. For this purpose, in accordance with a target, which can correspond, for example, to minimizing a cost function, with the aid of a suitable optimization method, such as a gradient descent method or the like, the machining parameters used for traversing the machining trajectory are optimized.

The cost function can define costs depending on a processing time, one or more quality parameters of the processing and / or the like. For example, the quality parameters for one, several or all trajectory sections can be the respective time periods for the machining, a respective measure for the course of the occurrence of vibrations during the machining of the workpiece (measured as vibration by the sensor system) 8th ), for the material wear of the machining head (determined by the course of the feed force, which corresponds to the force acting on the machining head due to the feed and by the sensor system 9 is measured) and for the machining precision with respect to the machined workpiece, which indicates how the machined workpiece deviates from a target state of the workpiece. The course of machining variables, such as the course of the feed force, the course of the occurrence of vibrations, etc., can be specified as the mean, maximum value, minimum value of the relevant machining size for the entire machining trajectory or separately for one, some or all of the trajectory sections.

The new optimized machining parameters are now made available for subsequent machining of another workpiece.

In addition to the specified cost or target function, the optimization method can also take one or more boundary conditions into account, which can be specified with regard to one or more processing variables. A boundary condition can be a limit for a Specify the machining size that results from the machining parameters considered during the implementation of the optimization process. Optimization solutions in which the limit value is exceeded or undershot by the expected value of the machining size are accordingly not permitted. For this purpose, the modeled machining size is determined as a model value from the trainable, in particular self-learning, machining model in accordance with the machining parameters under consideration and is taken into account accordingly during the minimization of the target size / costs according to the cost function.

The boundary condition is not only defined as a fixed limit for the processing size, but the uncertainty of the respective model value of the processing size specified by the corresponding uncertainty value can also be taken into account in a suitable manner during the optimization. In this case, the limit value (s) or the range of permissible values of the processing variable depends on the modeled value of the processing variable and its associated uncertainty value for a specific combination of parameter values.

A range of values around the model value of the machining size can be determined from the model value of the machining size for default values of a specific machining parameter or a specific machining parameter combination considered, given a probability specified by the boundary condition, which indicates that the actual value of the machining size during machining with the corresponding Machining parameter values with the specified probability will lie within the value range. This reflects the uncertainty of the modeling through the trainable machining model.

The processing model provides a point-wise posterior distribution for each individual point of the modeled processing trajectory, i.e. for a Gaussian process as the regression model, this corresponds to a Gaussian distribution.

wobei µ dem Modellwert der Bearbeitungsgröße entsprechend des Bearbeitungsmodells, p der vorgegebenen Wahrscheinlichkeit entspricht, dass der tatsächliche Wert der Bearbeitungsgröße innerhalb des durch z (vorgegeben als Quantil z der Standardabweichung σ) vorgegebenen Streuintervalls liegt. Aus der Auflösung der obigen Gleichung ergeben sich ein oberer und unterer Grenzwert x1, x2 für ein Streuintervall. Das Streuintervall gibt den Wertebereich der Bearbeitungsgröße an, innerhalb dem der tatsächliche Wert der Bearbeitungsgröße mit der vorgegebenen Wahrscheinlichkeit liegt.The range of values of the machining variable xE [x1, x2], within which the actual value of the machining variable lies with the specified probability, is determined from: $$p=\mathrm{2nd}\frac{1}{\sqrt{\mathrm{2nd}\pi }}{\displaystyle {\int }_{-\infty }^{\mathrm{Z.}}{e}^{-\frac{{\left(x-\mu \right)}^{\mathrm{2nd}}}{\mathrm{2nd}}}}-\mathrm{1,}$$

where µ corresponds to the model value of the machining variable corresponding to the machining model, p the predetermined probability that the actual value of the machining variable lies within the scatter interval predetermined by z (predetermined as the quantile z of the standard deviation σ). An upper and lower limit value x1, x2 for a scattering interval result from the resolution of the above equation. The scatter interval specifies the range of values of the processing size within which the actual value of the processing size lies with the specified probability.

As in 3rd As an example, the model curve specified by the machining model can be specified for a modeled machining variable x a course of the value range depending on an exemplary machining parameter z. By means of the model curve as the course of the modeled processing variable x over values of the exemplary processing parameter z.This indicates courses of the lower and upper range limits GU, GO of the range of values for the respective values of the modeled processing variable which, with the probability specified by the boundary condition, did not fall below or did not fall below should be exceeded. The lower / upper range limit GU, GO of the modeled machining size in question can then be used as a limit value for the boundary condition which specifies a certain measurement certainty. For example, the range boundaries result from the specification of the probability.

In this way, compliance with the boundary conditions for the currently considered values of the processing parameters is checked during the optimization. In order to obtain the respective range limit by specifying a probability, the equation for the cumulative density (distribution function) is used as the starting point, the desired probability is used and is resolved according to the standard deviation. This value (e.g. at 99% you get a width of z = 2.576 σ) is used with the posterior standard deviation of the Gaussian process. (So: upper limit = µ + z σ).

For example, in 3rd based on a modeled course of the resulting force (machining size x) on the tool at a certain feed speed, a range of values around the modeled resulting force is specified. The boundary condition is specified as a condition that the processing size is within a range of two standard deviations in this case (ie with a probability of approximately 95%).

If the machining parameter "feed speed" should be selected so that the force on the tool should not exceed 120 N with a probability of 95%, this results in a feed speed of 320 mm / min with a value of the upper limit of the value range of 120 N .

The above method can be carried out cyclically during the machining of workpieces by the machine tool. It can also be used in advance during the initial setting of machine tools to determine a set of initial machining parameters for machining workpieces. When used during an optimization process, the feed rate would then be limited to a value at which the force on the tool is not exceeded with a certain probability.

Claims (12)

Method for machining one or more workpieces (2) with an automated machine tool (1), a machining head (6) and a workpiece (2) being moved relative to one another along a machining trajectory, the machining trajectory being determined by machining parameters, with the following steps, which are each carried out for machining one or more workpieces (2): - Machining (S1) the workpiece (2) according to predetermined machining parameters along the machining trajectory; - Measuring (S2) one or more machining variables, in particular one or more courses of the machining variables during machining of the workpiece (2) along the machining trajectory; - Optimizing (S4) the machining parameters using an optimization method based on a cost function and at least one boundary condition, wherein a modeled machining variable and an uncertainty variable associated with this modeled machining variable are determined using a machining model that is provided as a regression model, depending on the machining parameters considered during the optimization become, the boundary condition depends on a range of values dependent on a predetermined probability around the modeled machining size, where the modeled processing size, the uncertainty size and the specified probability indicate the range of values around a modeled processing size. Procedure according to Claim 1 , the boundary condition being fulfilled if a predetermined limit lies within the value range for the machining size. Procedure according to Claim 1 or 2nd , wherein the machining trajectory is divided into trajectory sections in which the workpiece (2) is machined based on one or more machining parameters. Procedure according to Claim 3 , wherein the cost function considered for the optimization depends on one or more quality parameters which depend on the modeled machining size, the one or more quality parameters being determined in particular by a course of the modeled machining size or depending on it, and in particular as an average value, maximum value, Minimum value of the course of the modeled machining size for the entire machining trajectory or separately for one, some or all trajectory sections can be specified. Procedure according to one of the Claims 1 to 4th , wherein the machining sizes include a measure of one or more of the following sizes: - the occurrence of vibrations during the machining of the workpiece (2), - the material wear of the machining head (6); - a temperature of the workpiece (2); - A temperature of the processing head (6); - A feed force that acts on the machining head (6) and the workpiece (2); and - an indication of a machining precision with respect to the machined workpiece (2). Procedure according to one of the Claims 1 to 5 , wherein the machining parameters for a machine tool (1) comprise one or more of the following parameters: - a feed speed of the machining head (6), - a feed force, - a temperature of the machining head (6), - a rotational speed of the machining head (6); and - a performance of the machining head (6). Procedure according to one of the Claims 1 to 6 , wherein the machining model is selected as a machinable machining model, a learning process being carried out based on the one or more machining variables measured with respect to machining the workpiece (2) with the predetermined machining parameters. Procedure according to one of the Claims 1 to 7 , wherein the processing model comprises a Gaussian process model, a Bayesian linear regression model, a regularized kernel ridge regression model, and / or a Bayesian neural network. Device, in particular a control unit, for machining workpieces with a Automated machine tool (1), wherein a machining head (6) and a workpiece (2) are moved relative to one another along a machining trajectory, the machining trajectory being determined by machining parameters, the device being designed to process one or the other a plurality of workpieces (2): - to process the workpiece (2) along the machining trajectory in accordance with predetermined machining parameters; - to measure one or more machining variables, in particular one or more profiles of the machining variables during machining of the workpiece (2) along the machining trajectory; to optimize the machining parameters using an optimization method based on a cost function and at least one boundary condition, a modeled machining variable and an uncertainty variable associated with this modeled machining variable being determined using a machining model which is provided as a regression model, depending on the machining parameters considered in the optimization, wherein the boundary condition depends on a value range dependent on a predefined probability around the modeled machining size, the modeled machining size, the uncertainty size and the predefined probability indicating the value range around a modeled machining size. Machine tool (1) with a drive unit (3) for causing a workpiece (2) and a machining head (6) to move relative to one another and a device according to Claim 9 for controlling the drive unit (3) according to machining parameters along a machining trajectory. Computer program which, when executed on a data processing device, carries out all the steps of a method according to one of the Claims 1 to 8th carries out. Machine-readable storage medium on which a computer program Claim 11 is saved.

Images