METHOD AND DEVICE FOR CONTROLLING A PROCESS WITHIN A SYSTEM, IN PARTICULAR A GRINDING PROCESS IN A GRINDING DEVICE

Abstract

A method for controlling a process within a system, in particular a grinding process in a grinding device, comprising the following steps: .detecting 1 of state variables (s.sub.t) of the system; creating 2 of at least two process models (PM), each describing the effects of actuating actions (a.sub.t) on the state variables (s.sub.t) of the system, wherein the structure of the at least two process models (PM) differs from each other; controlling 3 of the process within the system by executing actuating actions (a.sub.t) under consideration of predefined control objectives and the process model, which currently provides the best prediction for the process running in the system. A device for carrying out the method according to the invention.

Claims

1. A method for controlling a process within a system, in particular a grinding process in a grinding device, comprising: detecting state variables (s.sub.t) of the system; creating at least two process models (PM) which each describe effects of actuating actions (a.sub.t) on the state variables (s.sub.t) of the system, a structure of the at least two process models (PM) differing from one another; controlling of the process within the system by executing actuating actions (a.sub.t) under consideration of predefined control objectives and a process model of the at least two process models which currently provides a best prediction for the process within in the system.

2. The method according to claim 1, wherein the state variables (s.sub.t) are detected by sensors of the grinding device.

3. The method according to claim 1, wherein the at least two process models (PM) are continuously adapted by the effects of actuating actions (a.sub.t) on the state variables (s.sub.t) of the grinding device.

4. The method according to claim 1, wherein the at least two process models (PM) are created by a test run of the grinding device with at least an exemplary execution of actuating actions (a.sub.t).

5. The method according to claim 1, wherein the at least two process models (PM) are each created by a computer-based neural network and are continuously adapted.

6. The method according to claim 1, wherein the at least two process models (PM) each take into account several assumed future effects of actuating actions (a.sub.t+1) on the state variables (s.sub.t+1) of the grinding device.

7. The method according to claim 1, further comprising the steps of: creating a situation assessment (SB) by quality functions (u.sub.t), which assesses the state variables (s.sub.t) with respect to the control objectives; and deciding whether a control is required at the current time to adapt the state variables (s.sub.t) to the control objectives.

8. The method according to claim 7, wherein the effects of different actuating actions (a.sub.t) on the state variables (s.sub.t) are calculated several time steps in advance and an overall quality (Q) is evaluated.

9. The method according to claim 1, wherein the method is performed in real time to control the process within the system.

10. The method according to claim 1, wherein at least one of said at least two process models (PM) is based on the method of numerical fluid mechanics.

11. The method according to, wherein similar state variables (s.sub.t) of the system are grouped together and wherein at least one process model (PM) is produced for each group.

12. A device for implementing the method according to claim 1, comprising a grinding device to be controlled with sensors for detecting condition variables (s.sub.t), actuating devices for carrying out actuating actions (a.sub.t) and a control device connected to the grinding device.

13. The method according to claim 1, wherein the state variables (s.sub.t) are detected by manual and/or automatic sample evaluations.

14. The method according to claim 1, wherein the at least two process models (PM) are created by expert knowledge.

15. The method according to claim 7, wherein the quality functions (u.sub.t) are established by statistical method.

16. The method according to claim 15, wherein the statistical method includes artificial intelligence and/or fuzzy logic.

17. The method according to claim 1, wherein the at least two process models (PM) are based on or trained on evolutionary strategies, genetic algorithms, genetic programming or evolutionary programming.

18. The method according to claim 17, wherein the at least two process models (PM) are based on or trained on the evolutionary strategies, the genetic algorithms, the genetic programming or the evolutionary programming within a framework of an autonomous selection and/or optimisation process.

19. The method according to claim 11, wherein a group and a new process model are created if state variables (s.sub.t) of the system are detected which are not yet classified in any group.

20. The method according to claim 12, wherein the control device comprises a computing device.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In the following, the method in accordance with the invention is explained in more detail on the basis of the exemplary embodiment shown in FIG. 1.

[0035] FIG. 1 illustrates the sequence of an inventive method for controlling a process within a system, in particular a grinding process in a grinding device.

DETAILED DESCRIPTION

[0036] FIG. 1 shows the sequence of an inventive process for controlling a process within a system, in particular a grinding process in a grinding device.

[0037] In a first step, the state variables (s.sub.t) of the system are recorded 1. The state variables (s.sub.t) of the system are recorded 1, for example, by means of sensors of the grinding device or manual and/or automatic sample evaluations. It is preferable to record 1 the state variables (s.sub.t) by means of sensors, as this can take place continuously. The grinding process within a grinding device can, for example, be monitored by means of a microphone and/or automatic measurement of the starting product. In the event that state variables cannot be recorded 1 directly with sensors, samples can be taken manually and/or automatically and evaluated, and the result of the evaluation can be made available to the method according to the invention.

[0038] In the next step of the method according to the invention, at least two process models (PM) are created, each of which describes the effects of actuating actions (a.sub.t) on the state variables (s.sub.t) of the system. According to the invention, the structure of the at least two process models (PM) differs from each other.

[0039] The at least two process models (PM) are preferably adapted continuously by the effects of actuating actions (a.sub.t) on the state variables (s.sub.t) of the grinding device. For this purpose, the effects of the previous actuating actions (a.sub.t−1) on the previous state variables (s.sub.t−1) of the system are taken into account by the method according to the invention when recording 1 the state variables (s.sub.t) of the system. The recorded changes of the state variables (s.sub.t) of the system can be used to continuously adapt the at least two process models (PM), for example by taking into account the change of the state variables (s.sub.t) from the previous state variables (s.sub.t−1) of the system and the effects of the actuating actions (a.sub.t−1) of the previous point in time.

[0040] The at least two process models (PM) are created, for example, by means of a test run of the system with at least exemplary execution of possible actuating actions (a.sub.t) and/or by expert knowledge 2. The two process models (PM) created can subsequently be continuously adapted by the effects of actuating actions (a.sub.t) on the state variables (s.sub.t) of the system, as explained in more detail above.

[0041] The at least two process models (PM) can be created by means of a computer-based neural network and, in particular, continuously adapted (trained). In particular, neural networks can be created and/or trained by means of known evolutionary strategies, genetic algorithms, genetic programming or evolutionary programming, preferably within the framework of an autonomous selection and/or optimization process.

[0042] In order to improve the accuracy of the method according to the invention, the at least two process models (PM) can each take into account several assumed future effects of actuating actions (a.sub.t+1) on the state variables (s.sub.t+1) of the grinding device.

[0043] Consequently, the at least two process models (PM) advantageously taking into account past effects of actuating actions (a.sub.t−1) on the state variables (s.sub.t−1) of the system and current effects of actuating actions (s.sub.t) on the state variables (s.sub.t) of the system and in predicting the assumed future effects of actuating actions (a.sub.t+1) on the state variables (s.sub.t+1) of the system.

[0044] According to a preferred variant of the method according to the invention, the at least two process models (PM) are created and/or optimized under consideration of different time scales. Thus, for example, one of at least two process models (PM) can be optimized for short-term effects of actuating actions (a.sub.t) on the state variables (s.sub.t) of the system, while the other of at least two process models (PM) can be optimized for long-term effects of actuating actions (a.sub.t) on the state variables (s.sub.t) of the system.

[0045] After a further advantageous variant of the method according to invention, similar state variables (s.sub.t) of the system are grouped together, whereby at least one process model (PM) is created for each group.

[0046] The classification of similar state variables (s.sub.t) of the system into groups is preferably self-learning and self-optimizing. For example, cost functions are taken into account that compare the costs of grouping state variables (s.sub.t) of the system into a group with the costs of creating a new group.

[0047] For example, a new group and a new process model (PM) is created 2 when system state variables (s.sub.t) are captured that have not yet been assigned to a group and/or that cannot be assigned to an existing group taking into account the above cost functions.

[0048] It is useful to assign a unique identification code to each group.

[0049] In the next step of the method according to the invention, the process within the system is controlled 3 by carrying out actuating actions (a.sub.t) under consideration of given control objectives and the process model (PM), which currently provides the best prediction for the process running in the system.

[0050] Preferably, the inventive method is executed in real time to control the process within the system. According to a particularly practical variant of the invention, at least two process models (PM) are based on the method of numerical fluid mechanics (CFD=Computational Fluid Dynamics).

[0051] In a useful way, the at least two process models (PM) are also based on or trained on evolutionary strategies, genetic algorithms, genetic programming or evolutionary programming, preferably within the framework of an autonomous selection and/or optimisation process.

[0052] According to a further variant, the method according to the invention comprises the following steps of the process, which are not described in more detail in FIG. 1: [0053] creating a situation assessment (SB) using quality functions (u.sub.t) that assesses the state variables (s.sub.t) against the control objectives; and [0054] deciding whether a control is required at the current time to adapt the state variables (s.sub.t) to the control objectives.

[0055] The situation evaluation (SB) takes into account, by means of the quality function (u.sub.t), the costs associated with the actuating actions (a.sub.t) and compares these with the current state variables (s.sub.t) and the specified control objectives. On the basis of such a situation evaluation (SB) it can be decided whether a control is necessary and/or useful at the current time to adapt the state variables (s.sub.t) to the control objectives or whether an adaptation of the state variables (s.sub.t) to the control objectives by means of actuating actions (.sub.at) can be dispensed with.

[0056] It is useful to calculate the effects of different actuating actions (a.sub.t) on the state variable (s.sub.t) several time steps in advance and to evaluate an overall quality (Q). On the basis of the overall quality (Q) it can be decided whether a control is required at the current time in order to adapt the state variable (s.sub.t) to the control objectives or whether an adaptation of the state variable (s.sub.t) to the control objectives can be dispensed with initially. Since the total quality (Q) considers several time steps in advance, short-term changes in the state variable (s.sub.t), for example, can be filtered out, since such short-term changes in the state variable (s.sub.t) only have a small influence on the total quality (Q).

[0057] The quality function (u.sub.t) is established, for example, by means of statistical methods, such as artificial intelligence (AI approaches in particular) or according to the rules of fuzzy logic.

REFERENCE CHARACTER LIST

[0058] 1 detecting state variables (s.sub.t) [0059] 2 creating the process models (PM) [0060] 3 controlling the process within the system