DE102019220453A1 - Method and device for creating a safety function for measuring a manufacturing system - Google Patents

Abstract

According to the invention, a computer-implemented method for operating a manufacturing process, in particular a plastic molding process, is provided, with the following steps: Determination of process parameters for operating a manufacturing system using a Bayesian optimization method, the Bayesian optimization method being trainable with process parameters of the manufacturing process Quality function, which is successively specified, the process parameters being assigned to a quality measure, the method iteratively applying a predefined acquisition function to improve the quality function; performing the manufacturing process as a function of the trained quality function.

Description

Technical area

The invention relates to residues and methods for measuring technical systems. The invention also relates to carrying out a measurement for a technical system, taking into account a safety function.

Technical background

Machine learning methods can be used to model complex technical processes, such as manufacturing processes for products and processing processes for materials, in order to predict process behavior. In order to create a corresponding process model, training data must be determined, which consists of input variables and corresponding output variables, so that the process model learns to map the input variables to the output variables.

Active learning is a strategy for efficiently determining this training data by iteratively updating the system model with new training data and then selecting a training data point, in particular with the help of an acquisition function, as the measurement point that contains the greatest information content for creating the process model.

The training data is usually obtained by measuring on a test stand or the like. Depending on the complexity of the technical process, care must be taken during the measurement that the measurement points are not located outside a so-called safety area, outside of which the technical process cannot be carried out safely. In particular, the technical process cannot be carried out reliably outside the safety area or the test stand or the product to be manufactured can be damaged or destroyed by the operating point specified by the measurement point.

However, the security area is not known in advance. The manual creation of security areas is not possible due to the large number of process parameters. As a rule, therefore, only a very small area has so far been specified as safe by expert knowledge. In addition, certain safety areas are laborious to create manually and are nevertheless imprecise, since the limits of the technical process are often not exhausted in order not to endanger the process and / or the test bench.

However, this leads to an inadequate measurement result, so that when the process model is trained, it is only created for a limited range of input variables. If a process is to be optimized based on the process model, it may not be possible to find all possible optimals. Inaccurately determined safety areas can also lead to the discarding of permissible, measurable measurement points if these lie outside the specified safety area. This means that the process model can only be created imprecisely.

Disclosure of the invention

According to the invention, a method for measuring a manufacturing process, in particular a plastic casting process, according to claim 1 and a corresponding device and a manufacturing system, in particular a plastic casting system, in accordance with the independent claims are provided.

Further refinements are given in the dependent claims.

• - Bestimmen von Prozessparametern zum Betreiben eines Fertigungssystems mithilfe eines Bayes'schen Optimierungsverfahrens, wobei das Bayes'sche Optimierungsverfahren auf einer mit Prozessparametern des Fertigungsprozesses trainierbaren Qualitätsfunktion basiert, die sukzessive präzisiert wird, wobei die Prozessparameter einem Qualitätsmaß zugeordnet werden, wobei das Verfahren iterativ zum Verbessern der Qualitätsfunktion eine vorgegebene Aquisitionsfunktion anwendet;
• - Durchführen des Fertigungsprozesses abhängig von der trainierten Qualitätsfunktion.

According to a first aspect, a computer-implemented method for operating a manufacturing process, in particular a plastic molding process, is provided with the following steps:

• Determination of process parameters for operating a manufacturing system using a Bayesian optimization method, the Bayesian optimization method being based on a quality function that can be trained with process parameters of the manufacturing process and that is successively specified, the process parameters being assigned to a quality measure, the method iteratively for improvement applies a predetermined acquisition function to the quality function;
• - Execution of the manufacturing process depending on the trained quality function.

To optimize a manufacturing process, such as a plastic casting process, a number of process parameters must be set on a manufacturing system, such as a plastic casting system, depending on the starting materials or starting products, such as the plastic material used. These process parameters include, for example, a material pressure, a material temperature, a flow rate at one or more positions in the production system, which determine an operating state.

In addition, operating states of the production system are known which are not permitted due to a risk to the production system and should be avoided at all costs. These operating states arise in certain areas of the input variable space for the certain process parameters.

To optimize the manufacturing process, it is provided that the next measurement point to be measured is determined using a data-based quality function. The quality function is trained in order to map the process parameters of the production system to a product quality, in particular to a quality measure that specifies this. This quality function is used to optimize the process parameters for the manufacturing process. For this, however, it is necessary that the quality function is measured over the largest possible input variable range so that the optimization of the quality function can take into account all optima.

By using a trainable, data-based security model, it can be determined for each measuring point to be measured, which corresponds to a set of process parameters, whether or not this leads to an impermissible operating state of the manufacturing system. Correspondingly, a measurement point can be discarded if it does not guarantee sufficient process reliability.

Since the permissible and non-permissible operating areas are not defined right from the start of the measurement process, a data-based safety function is trained that indicates for each measurement point whether this leads to permissible or impermissible operation of the manufacturing process. The safety function is updated with each survey with a survey point by determining a safety measure for each survey point. A measurement point is then only selected according to the current safety function, so that an impermissible operating state of the manufacturing process is avoided.

The selection of a measurement point to be measured next can be derived accordingly from the quality function, in particular with the aid of an acquisition function. Overall, the determination of new measurement points depending on a successively more precise safety model enables measurement of the quality function for the manufacturing process, such as the plastic casting system, without the manufacturing process being operated during the manufacturing process for the new measurement at operating points that are not permitted.

Furthermore, the method can iteratively apply the acquisition function and additionally a constraint to improve the quality function, the constraint being determined by the safety function, which maps a process parameter set to a safety measure, the constraint being determined as a function of the safety measure and a safety criterion.

In particular, the safety function can correspond to a separately trainable safety function, which in particular is also designed as a Gaussian process model.

It can be provided that the process parameters for the manufacturing process include a heating power and a pump power for a plastic casting process and one or more material parameters of the casting material.

Furthermore, the quality measure can depend on operating parameters of the production process and in particular on one or more component properties and can in particular be determined by an evaluation function.

• - Bestimmen von Prozessparametern zum Betreiben eines Fertigungssystems mithilfe eines Bayes'schen Optimierungsverfahrens, wobei das Bayes'sche Optimierungsverfahren auf einer mit Prozessparametern des Fertigungsprozesses trainierbaren Qualitätsfunktion basiert, die sukzessive präzisiert wird, wobei die Prozessparameter einem Qualitätsmaß zugeordnet werden, wobei das Verfahren iterativ zum Verbessern der Qualitätsfunktion eine vorgegebene Aquisitionsfunktion anwendet;
• - Durchführen des Fertigungsprozesses abhängig von der trainierten Qualitätsfunktion.

According to a further aspect, a device for operating a manufacturing process is provided, the device being designed to:

• Determination of process parameters for operating a manufacturing system using a Bayesian optimization method, the Bayesian optimization method being based on a quality function that can be trained with process parameters of the manufacturing process and that is successively specified, the process parameters being assigned to a quality measure, the method iteratively for improvement applies a predetermined acquisition function to the quality function;
• - Execution of the manufacturing process depending on the trained quality function.

According to a further aspect, a manufacturing system is provided with a manufacturing device that can be controlled by process parameters, and with the above device, the process parameters being determined by the device.

Figure list

• 1 eine schematische Darstellung eines Kunststoffgießsystems zur Herstellung von Kunststoffgussbauteilen;
• 2 ein Flussdiagramm zur Veranschaulichung eines Verfahrens zum Vermessen des Kunststoffgießsystems; und
• 3 eine Darstellung von Vermessungspunkten und einen Sicherheitsbereich.

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

• 1 a schematic representation of a plastic casting system for the production of plastic cast components;
• 2 a flow chart to illustrate a method for measuring the plastic molding system; and
• 3 a representation of survey points and a safety area.

Description of embodiments

1 shows a schematic representation of a manufacturing system 1 using the example of a plastic casting system with a plastic casting device 2 , a storage container for plastic material 3 and a casting apparatus that supports a mold to produce a cast component. During the casting process, the plastic material is liquefied and fed to the mold in order to form the cast component. The casting process is carried out using a control device 10 controlled by process parameters selected to form an injection molded component according to predetermined specifications.

The process parameters correspond to process settings that can be permanently specified, such as a heating power, a pump pressure or a pump power or the like, and / or which can be specified as a time curve, such as a heating power curve, etc. Furthermore, the process parameters can also include the material properties of starting materials based on material properties, such as B. viscosity, flow rate, molecular weight and the like specify.

The plastic molding system is provided with a sensor system in order to detect operating parameters, such as a pressure, a temperature, a flow rate and the like, and in particular their temporal progressions at different positions of the plastic molding system. These operating parameters represent a result of the implementation of the manufacturing process and can only be influenced indirectly by specifying the process parameters.

To set the manufacturing process, the process parameters should be optimized. For this purpose, a Bayesian optimization method is used that optimizes the process parameters based on a quality function. In doing so, one or more safety criteria should be taken into account, which can be defined in particular with regard to the operating parameters or their temporal progressions.

The control unit is used to determine optimized process parameters 10 implements a function which, in conjunction with the flowchart of the 2 is described in more detail. The method can be implemented in a data processing device as a software algorithm and / or hardware algorithm.

In step S1 becomes an initially permissible and safe test process parameter set of the manufacturing system 1 given. The test process parameter set contains process parameters for controlling the manufacturing process.

Furthermore, at least one safety criterion is specified, which is defined, for example, with regard to one or more operating parameters. The safety criterion can, for example, specify limit values for operating parameters and / or process parameters. If these limit values are exceeded, the manufacturing process is considered unsafe or impermissible because damage or destruction of the component to be manufactured or the manufacturing system is to be expected.

In step S2 a component to be manufactured is manufactured with the process parameters which are determined by the test process parameter set and which specify a measurement point. The operating parameters, such as material pressure, flow rate, material temperature for an injection molding process, are measured or otherwise determined, which characterize the manufacturing process. Furthermore, component properties are determined that determine the quality of the component, such as breaking strength, dimensional accuracy and the like.

A Bayesian optimization process is now started.

Bayesian optimization methods for the determination of optimized process parameters apply test process parameters iteratively and optimize them in an efficient way. A quality function is modeled using a Gaussian process regression in order to model the performance of the manufacturing process as a function of its process parameters. The Gaussian process model is created based on the results of a measurement on one or more operating parameters determined as a function of the respective test process parameters or is updated cyclically in each case.

The quality function indicates a quality measure depending on company sizes and, if necessary, on process parameters of the manufacturing process as well as on the component quality. For this purpose, an evaluation function can be defined that connects the individual operating parameters and the component quality with one another in order to provide a quality measure that is suitable as a quantitative evaluation measure. Furthermore, for example, the influences of the individual Operating parameters and the component properties are weighted with appropriate weightings in order to determine their respective influence on the quality measure.

Basically, the problem is to find process parameters that lead to an optimized production of the component to be manufactured in a manufacturing process, i.e. e.g. to an injection molding process with which a component can be manufactured cost-effectively that meets the specified quality requirements. For this purpose, a quality function (cost function) that is dependent on the process parameters is evaluated.

In general, Bayesian optimization is used when uncertainty (e.g. due to measurement noise) is to be taken into account. This is also suitable if an unknown function f a so-called "black box" function, such as the quality function in the present case, is to be optimized. This unknown function f can only be evaluated and observed (possibly affected by noise) for a value x. The observed value y results as y = f (x) + e, where e denotes the noise.

In addition, it is assumed that every evaluation or measurement of the unknown function f is “expensive”, ie it causes costs, in the sense that the implementation of a test method for measuring the unknown function causes a certain, in particular high effort, as is the case with a Evaluation of the manufacturing process or the estimation of the component quality is the case with an injection molding system. Due to the "expensive" measurement of the unknown function, it is desirable that only comparatively few measurements have to be carried out for the optimization or that the costs for the measurements (especially determined by the time and material expenditure) are as low as possible.

Under certain assumptions, such as the continuity of the unknown function, the quality function can be approximated with a data-based function model, such as a Gaussian process regression. A Gaussian process is a universal function approximator that is used as a surrogate function for the unknown quality function. In general, Gaussian processes are understood to mean temporal, spatial or any other functions whose function values can only be modeled with probabilities due to incomplete information. With the help of functions of the expected values, variances and covariances, a Gaussian process describes the function values as a continuum of correlated random variables in the form of a high-dimensional normal distribution. A random sample of these results in a random function with certain desired properties.

Usually numerical optimization algorithms are based on a large number of very cheap evaluations / measurements of the optimization functions. However, if the function evaluations are complex, such as B. performing an evaluation of the process quality of the manufacturing process and / or the component quality of the component to be manufactured, the conventional optimization algorithms can no longer be used. Instead, Gaussian process models are used, which include a modeling of the quality function in the form of a surrogate function. The quality function describes the behavior of the system and indicates a quality value depending on the parameters with which the manufacturing process is operated.

For this purpose, after measuring the quality function at a measuring point, which is determined by a test process parameter set x and by observing / measuring the corresponding operating parameters and the component quality, i.e. the respective quality measure (depending on the test method, the maximum temperature cycles, the load-bearing capacity, in particular in terms of shape the maximum tensile force or the maximum shear force), a model of the quality function can be established using the Gaussian process. One property of the Gaussian process is that in areas around the measured test process parameter set x the model prediction is very good and the quality function is approximated well. This is reflected in a low level of uncertainty in the functional model. Far away from evaluation points, the model predictions about the quality function become poor and the uncertainty increases with increasing distance from the measurement point determined by the test process parameter set x.

step S3 provides for the training of the quality function designed as a Gaussian process model. The training takes place with the test process parameters x and an associated determined quality measure, in particular by updating the Gaussian process model and in particular the hyperparameters. The quality function provides the quality measure directly. Alternatively, the quality function can be trained to map one or more operating parameters, which are summarized in the quality measure with the aid of the evaluation function and made evaluable. This has the advantage that the quality function can be used to better understand how the operating parameters are influenced by the process parameters.

In step S4 a safety function S (x) is trained, which is designed as a further data-based function, in particular also from a Gaussian process model. The safety function is trained with the process parameters of the respective test process parameter set x and an associated safety measure fs = S (x). The safety measure fs can be derived from the measured operating parameters and the component properties in conjunction with one or more predetermined safety criteria with the aid of a predefined safety assessment function SB. For example, a safety measure fs greater than 0 can indicate a safe implementation and a safety measure fs less than 0 indicates an unsafe implementation of the manufacturing process. For example, the safety measure fs can depend on a distance between an operating variable and a predetermined limit value or specify a measure of how well one or more of the operating variables meet the required safety criterion. For example, a difference between a material temperature and a temperature limit can determine the degree of safety, the temperature limit specifying a limit that damages or destroys the production facility. Furthermore, the level of the safety measure can depend on the difference between the material temperature and the temperature limit.

A possible strategy to optimize the quality function is to evaluate the quality function in many different places (e.g. on a regular grid) and to assume the lowest observed function value as the result of the optimization. This procedure is inefficient and many measurement processes with the test procedures are necessary with a correspondingly high level of effort to find the optimum.

Instead of this approach, the Gaussian process of the quality function is used to select a new test process parameter set. This is done in step S5 In each recursion, a new test process parameter set for measuring the quality function is selected in such a way that, on the one hand, the Gaussian process model of the quality function is improved by reducing the uncertainty of the corresponding Gaussian process. For this purpose, the process parameters of the test process parameter set are usually selected in areas in which the quality function has not yet been evaluated (exploration); these can be areas with high predictive variance, for example. On the other hand, the new test process parameter set for measuring the quality function is selected so that the goal of optimizing the quality function, ie minimizing or maximizing it, is achieved as quickly as possible or with the smallest possible number of measurements with the process parameters. For this purpose, test process parameter sets are preferred which, based on the Gaussian process, promise low or high functional values (exploitation).

These two opposing criteria are weighed up by what is known as an acquisition function. In the Bayesian optimization method, the measurements for determining the quality function are optimized with the aid of the acquisition function in such a way that they do not necessarily have the lowest uncertainty overall, but rather the greatest possible information about the position of the optimum, i.e. H. have those process parameters at which the highest possible quality value can be achieved.

The acquisition function uses parameters of the quality function, which is described by a Gaussian process model, such as the Gaussian process mean µ (x) and the Gaussian process standard deviation σ (x). One example is the so-called Lower Confidence Bound (LCB) acquisition function or the Upper Confidence Bound (UCB) acquisition function (depending on whether a high quality measure is better or worse), which are described as follows: LCB (x) = µ (x) - kσ (x) or UCB (x) = µ (x) + kσ (x). In practice, the factor k is often set to a constant value, for example k = 2. This new criterion can be efficiently minimized or maximized using common gradient-based methods and the minimum of LCB (x) or the maximum from UCB (x) then forms the new test process parameters for the quality function. It should be noted here that an optimization domain must be defined for the optimization of the acquisition function, in which the next test process parameters are searched for. This domain is typically chosen based on experience and / or expert knowledge.

In addition to the above acquisition functions, other acquisition functions are known, such as Expected Improvement (EI), Probability of Improvement (PI) or so-called entropy search methods that are based on information-theoretical considerations.

ermittelt, mit der Nebenbedingung, dass der Test-Prozessparametersatz bei einer Nutzung in dem Fertigungssystem zu einem Sicherheitsmaß fs führt, das eine sichere Durchführung des Fertigungsprozesses angibt, nämlich als modelliertes Sicherheitsmaß fs von größer als 0 für die obige beispielhafte Definition des Sicherheitsmaßes fs.The acquisition function is used in step S5 a new test process parameter set accordingly $$\begin{array}{c}x*=\underset{x}{\text{argmax}}\left(\mu \left(x\right)+\beta \sigma \left(x\right)\right)\\ s.t.\mathrm{S.}\left(x\right)=fs>0\end{array}$$

determined, with the secondary condition that the test process parameter set, when used in the manufacturing system, leads to a safety measure fs, which indicates a safe implementation of the production process, namely as a modeled safety measure fs greater than 0 for the above exemplary definition of the safety measure fs.

zu wählen. Dies entspricht ausschließlich einer Exploration ohne Exploitation.If the goal is not to find an optimal set of process parameters, but rather a description of the entire model that is as good as possible, e.g. to carry out further analyzes later (on weaknesses and strengths), it can make sense to focus more strongly instead of a Bayesian optimization method Active learning to focus and $x*=\underset{x}{\text{argmax}}\left(x/\sigma \left(x\right)\right)$

to choose. This corresponds exclusively to exploration without exploitation.

In step S6 a termination criterion is checked, for example the length of time that is to be expended for the optimization of the quality function, or the number of iterations or a suitable convergence criterion. If it results from this that the optimization should be ended (alternative. Yes), the procedure starts with step S7 continued, otherwise (alternative. No) the procedure continues with step S2 continued, with a measurement being carried out with the test process parameter set.

In step S7 the manufacturing process is operated with a process parameter set that results directly from the quality function. In particular, the process parameter set used can be determined by optimizing, in particular by maximizing the quality measure modeled by the quality function.

3 shows, for an intermediate state of the learning method for the quality function, schematically for two process parameters x1, x2 the safe operating area B, a trajectory T of measurement points and a process parameter area PB that could previously be mapped by the quality function.

Claims (9)

Computer-implemented method for operating a manufacturing process, in particular a plastic molding process, with the following steps: - Determination (S2, S4, S5, S6) of process parameters for operating a manufacturing system with the aid of a Bayesian optimization method, the Bayesian optimization method being based on a quality function that can be trained with process parameters of the manufacturing process and that is successively refined, the process parameters being a quality measure are assigned, the method iteratively applying a predetermined acquisition function to improve the quality function; - Carrying out (S7) the manufacturing process depending on the trained quality function. Procedure according to Claim 1 , the method iteratively applying the acquisition function to improve the quality function and additionally taking into account a secondary condition, the secondary condition being determined by a safety function (S) which maps a process parameter set to a safety measure (fs), the secondary condition being dependent on the safety measure (fs ) and a specified safety criterion. Procedure according to Claim 2 , wherein the safety function corresponds to a separately trainable safety function (S), which in particular is also designed as a Gaussian process model. Method according to one of the Claims 1 to 3 , wherein the process parameters for the manufacturing process include a heating power and a pump power for a plastic molding process and one or more material parameters of the molding material of the plastic molding process. Method according to one of the Claims 1 to 4th , wherein the quality measure depends on operating parameters of the manufacturing process and in particular on one or more component properties and is determined in particular by an evaluation function. Device for operating a manufacturing process, the device being designed to: - Determination (S2, S4, S5, S6) of process parameters for operating a manufacturing system with the aid of a Bayesian optimization method, the Bayesian optimization method being based on a quality function that can be trained with process parameters of the manufacturing process and that is successively refined, the process parameters being a quality measure are assigned, the method iteratively applying a predetermined acquisition function to improve the quality function; - Carrying out (S7) the manufacturing process depending on the trained quality function. Manufacturing system (1) with a manufacturing device that can be controlled by process parameters, and with a device according to Claim 6 . Computer program with program code means which is set up to implement a method according to one of the Claims 1 to 5 execute when the computer program is executed on a computing unit. Machine-readable storage medium with a computer program stored thereon Claim 8 .

Images