Module of a programmable logic controller

Abstract

The invention relates to a module of a programmable logic controller, comprising: - a controller structure in which an artificial intelligence (3) is implemented, and - a controller (6) for a controlled system (8) which is controlled or can be controlled by means of at least one manipulated variable (9) for the controller (6) and outputs at least one controlled variable (13), at least one reference variable (4) or the difference between the at least one reference variable (4) and an input variable or a modified input variable being suppliable to the controller (6) and / or to the artificial intelligence (3). An analog front end (2) is additionally provided for detecting analog signals from at least one sensor (10), which detects the controlled variable (13), and for processing said signals into the at least one input variable or the modified input variable and providing same to the artificial intelligence (3) and / or to the controller (6), the analog front end (2) being provided as part of the module or as a separate unit. Fig. 1

Description

[0001] Description

[0002] Module of a programmable logic controller

[0003] The invention relates to a module of a programmable logic controller with the features of claim 1.

[0004] In decentralized peripheral systems, such as programmable logic controllers (PLCs), process control tasks are performed by multiple modules. For example, an analog temperature controller might have an analog input module, an analog output module, and optionally several auxiliary modules (digital inputs and outputs). The control variables are transmitted from these modules via fieldbus to a CPU, processed there by a control algorithm in a user program, and the results are then transferred back to the output module. Currently, there is no controller module that performs this entire task "onboard," thereby reducing communication overhead and simultaneously improving control dynamics.

[0005] In arXiv:2201 .06961 , AI for Closed-Loop Control Systems, Schöning, Riechmann, Pfisterer, 2022, regulations based on artificial intelligence are described.

[0006] In the article entitled "AI for Closed-Loop Control Systems: New Opportunities for Modeling, Designing, and Tuning Control Systems" by Schöning, Riechmann and Pfisterer.

[0007] PROCEEDINGS OF THE 31ST ACM SIGSOFT INTERNATIONAL SYMPOSIUM ON SOFTWARE TESTING AND ANALYSIS, ACM, NEW YORK, NY, USA, February 18, 2022 (2022-02-18), pages 318-323, XP058825059, DOI: 10.1145 / 3529836.3529952 ISBN: 978-1-4503-9379-9, proposes the use of artificial intelligence (AI) in closed-loop control systems (CLCS).

[0008] The application note by Michael Seidl: "Analog Front-End Design With Texas Instruments' Tooling Landscape", June 15, 2022 (2022-06-15), pages 1-20, XP093238120, found on the internet: URL: https: / / web.archive.org / web / 20220615132515 / https: / / www.ti.com / 1it / an / sbaa534 / sbaa534.pdt?ts=1655299515583, describes in detail an analog front end for precise ADCs.

[0009] EP 2 184 654 A1 discloses a control system for industrial processes that combines a conventional controller with an expert system, wherein a selector selects either the manipulated variables calculated by the controller or the output values ​​determined by the expert system for process control based on the operating conditions of the process.

[0010] The invention is based on the objective of providing a novel control system in the form of a module of a programmable logic controller.

[0011] The problem is solved according to the invention by a module of a programmable logic controller with the features of claim 1.

[0012] Advantageous embodiments of the invention are the subject of the dependent claims.

[0013] According to the invention, a

[0014] A module of a programmable logic controller is proposed, comprising:

[0015] - a controller structure in which artificial intelligence is implemented,

[0016] - a controller for a controlled system which can be controlled or is controlled by means of at least one manipulated variable of the controller and outputs at least one controlled variable, wherein at least one reference variable or a difference between the at least one reference variable and an input variable or a modified input variable can be supplied to the controller and / or the artificial intelligence, wherein furthermore an analog front end is arranged for acquiring analog signals from at least one sensor that detects the controlled variable and for processing these signals to the at least one input variable or modified input variable and making them available to the artificial intelligence and / or the controller, wherein the analog front end is designed as part of the module or as a separate unit, wherein the artificial intelligence is configured for fine-tuning the at least one manipulated variable of the controller.

[0017] In one embodiment, the analog front end is configured for digitizing and / or processing the sensor signals into a voltage value, a current value, an FFT spectrum and / or an impedance spectrum.

[0018] In one embodiment, the artificial intelligence is configured as a neural network or as an interpretable and explainable algorithm, in particular as a support vector machine or a learning vector quantizer. In one embodiment, the artificial intelligence is arranged and configured to take over the function of the controller and to provide the at least one manipulated variable in its place, wherein a switch is arranged which is configured to switch between the at least one manipulated variable of the controller and the at least one manipulated variable of the artificial intelligence in order to supply the latter to the controlled system.

[0019] In one embodiment, the switch is configured to switch depending on at least one threshold value. The switch can be implemented by software.

[0020] In one embodiment, the artificial intelligence is configured to provide the at least one manipulated variable or at least one correction value for the at least one manipulated variable, in particular for the selection of classes in a classification problem or for the selection of numerical attribute values ​​in a regression problem, in response to the at least one input variable, to recorded process data stored in the control system, and to disturbances that influence the controlled system.

[0021] In one embodiment, the artificial intelligence is configured to label certain temporally successive measured values ​​during a learning phase and assign them to the classes.

[0022] In one embodiment, the control behavior of the controller can be learned based on experience.

[0023] Exemplary embodiments of the invention are explained in more detail below with reference to drawings.

[0024] It shows:

[0025] Fig. 1 shows a schematic view of an embodiment of a micro-Kl module, and

[0026] Fig. 2 shows a schematic view of another embodiment of a micro-Kl module.

[0027] Corresponding parts are marked with the same reference symbols in all figures.

[0028] Figure 1 is a schematic view of a micro-AI module 1. A micro-AI module 1, as defined in the present invention, is a module of a programmable logic controller (PLC) that includes an analog front end 2 for acquiring analog signals from at least one sensor 10 and optionally for processing these signals, for example, into FFT spectra and / or impedance spectra, as well as a controller structure in which an artificial intelligence 3, for example, a neural network, is implemented. Furthermore, the micro-AI module 1 behaves like a normal electronic module of a modularly designed programmable logic controller. The designation micro-AI module 1 results from the minimalist approach of an AI (artificial intelligence 3) on a decentralized peripheral, particularly minimalist with regard to size, power dissipation, and manufacturing costs, while simultaneously offering high performance for specific, frequently occurring applications.

[0029] The Micro-Kl module 1 also has a controller 6 for a controlled system 8, which can be controlled by means of at least one manipulated variable 9 of the controller 6 and outputs at least one controlled variable 13, which can be detected by the at least one sensor 10.

[0030] The Micro-Kl module 1 is configured to acquire at least one analog input variable (measured value) from the at least one sensor 10, which is optionally processed in the analog frontend 2 into a modified input variable, for example as a voltage value, current value, or in the form of an FFT spectrum and / or an impedance spectrum, particularly in digital form, and to provide the input variable or the modified input variable to the artificial intelligence 3 integrated in the Micro-Kl module 1. Furthermore, disturbance variables 5, which influence the controlled system 8, are also fed to the artificial intelligence 3. In Figure 1, the artificial intelligence 3 (for example, a support vector machine, a learning vector quantizer, or similar) is designed to take over the function of the controller 6 and provide the at least one manipulated variable 9 in its place, whereby the control behavior of the controller 6 can be learned based on experience.

[0031] Furthermore, at least one reference variable 4 or a difference between the at least one reference variable 4 and the input variable or the modified input variable can be supplied to the controller 6 and / or the artificial intelligence 3. At least one static parameter 12 can also be supplied to the controller 6. Recorded process data 7 stored in the Micro-Kl module 1 can also be supplied to the artificial intelligence 3.

[0032] Furthermore, a switch 11 is provided, which is configured to switch between the at least one manipulated variable 9 of the controller 6 and the at least one manipulated variable 9 of the artificial intelligence 3, in order to supply these to the controlled system 8. To ensure functional safety, the switchability via the switch 11 creates a fallback path, whereby the artificial intelligence 3 can be deactivated, for example, depending on at least one threshold value.

[0033] Figure 2 is a schematic view of another embodiment of a micro-Kl module 1, in which the artificial intelligence 3 can perform fine-tuning of the at least one manipulated variable 9 of the controller 6, for example by adding a respective correction value to the at least one manipulated variable 9. An analog front end 2 for acquiring analog signals from at least one sensor 10 and optionally for processing these signals, for example into FFT spectra and / or impedance spectra, as well as a controller structure in which an artificial intelligence 3, for example a neural network, is implemented, is arranged.

[0034] The Micro-Kl module 1 also has a controller 6 for a controlled system 8, which can be controlled by means of at least one manipulated variable 9 of the controller 6 and outputs at least one controlled variable 13, which can be detected by the at least one sensor 10.

[0035] The Micro-Kl module 1 is configured to acquire at least one analog input variable (measured value) from at least one sensor 10, which is optionally processed in the analog front end 2 into a modified input variable, for example as a voltage value, current value, or in the form of an FFT spectrum and / or an impedance spectrum, particularly in digital form, and to provide the input variable or the modified input variable and / or its difference to a reference variable 4 to the controller 6 and / or the artificial intelligence 3. Furthermore, disturbance variables 5, which influence the controlled system 8, are also fed to the artificial intelligence 3.

[0036] Furthermore, at least one static parameter 12 can be supplied to the controller 6. Additionally, the artificial intelligence 3 can be supplied with recorded process data 7 stored in the micro-Kl module 1.

[0037] The control system architectures in Figures 1 and 2 were derived from and modified from the paper "AI for Closed-Loop Control Systems" (source: arXiv:2201.06961, AI for Closed-Loop Control Systems, Schöning, Riechmann, Pfisterer, 2022). Since in these architectures the artificial intelligence 3 directly intervenes in the control loop 8 and may control actuators, in the present invention, unlike in the aforementioned paper, an algorithm (for example, a support vector machine, a learning vector quantizer, etc.) is implemented in the artificial intelligence 3. This algorithm is interpretable and explainable and does not represent a black box like a deep neural network.

[0038] The results determined by the artificial intelligence correspond to the reactions to the evaluated measured values, to previously recorded process data 7, and to the disturbance variables 5. The results are represented as numerical attribute values ​​because, in the case of a control loop where the artificial intelligence 3 influences the manipulated variable 9, this is referred to as a regression problem. However, if the artificial intelligence 3 is to influence the manipulated variable 9 with a predefined number of fixed attribute values ​​(classes) in the control loop, then this is referred to as a classification problem, where an unknown categorical attribute value is determined.

[0039] In a learning phase (regression problem) of artificial intelligence 3, certain temporally successive measurements (trends) are labeled, representing the numerical attribute values.

[0040] In a learning phase (classification problem) of artificial intelligence 3, certain temporally successive measured values ​​(trends) are labeled, which represent the categorical attribute values ​​(classes).

[0041] In Figure 1, either the artificial intelligence 3 or the controller 6 specifies the manipulated variables 9. In Figure 2, the artificial intelligence 3 merely corrects the manipulated variables 9 within predefined limits. The embodiment shown in Figure 2 thus fulfills the function of a behavior formerly known as fuzzy logic. However, the artificial intelligence 3 significantly expands the possibilities compared to fuzzy logic, enabling a generalization of the behavior. Through its learning capability, the control sequences are successively optimized and adapted to the controlled system 8. The output variables (manipulated variables 9) of the controller 6 are the setpoints assigned to the determined class, which can be transmitted via the programmable logic controller to actuators that may be part of the controlled system 8.

[0042] The learning capability of the controller 6 enables adaptive control at a new level. This represents a new level of quality compared to fuzzy controllers. In an exemplary embodiment for a classification problem, a heating controller is assumed. During the learning phase, the required room temperatures are labeled according to a perceived comfort level for typical weather conditions (measured values: outside temperature, wind, rain, air pressure, time of day, etc.). Many arbitrary temperature profiles can be recorded, with varying gradients of the measured values. Furthermore, it is possible to consider different types of people (sensitive to cold, prone to perspiration, active, rather passive, cozy in the evening, etc.) and define these as parameters. As a result, the heating system is significantly better adapted to individual needs.

[0043] In embodiments not shown, the analog front end 2 can be a separate unit that is not part of the micro-Kl module 1.

[0044] Further potential applications exist in the field of solar energy generation. Here, for example, individual characteristics of an installation site, including the efficiency curves of solar panels depending on the outside temperature, can be incorporated into the control behavior. The same applies analogously to the use of energy from battery banks, for example in electric vehicles.

[0045] The present invention enables the optimization of experience-based control processes based on a universal micro-intelligence module 1 of an automation device (for example, a programmable logic controller). By employing artificial intelligence 3, this behavior can be adapted very precisely to rapidly changing input conditions without requiring the controller 6 to be configured using purely analytical methods. Various conditions that are difficult to formally categorize can be taken into account.

[0046] Regardless of the grammatical gender of a particular term, persons with male, female or other gender identities are included.

Claims

Patent claims 1. Module of a programmable logic controller designed as a control system comprising: - a controller structure in which an artificial intelligence (3) is implemented, - a controller (6) for a controlled system (8) which can be controlled or is controlled by means of at least one manipulated variable (9) of the controller (6) and outputs at least one controlled variable (13), wherein at least one reference variable (4) or a difference between the at least one reference variable (4) and an input variable or a modified input variable can be supplied to the controller (6) and / or the artificial intelligence (3), wherein furthermore an analog front end (2) is arranged for acquiring analog signals from at least one sensor (10) that detects the controlled variable (13), and for processing these signals to the at least one input variable or modified input variable and making them available to the artificial intelligence (3) and / or the controller (6), wherein the analog front end (2) is designed as part of the module or as a separate unit, wherein the artificial intelligence (3) is configured for fine-tuning the at least one manipulated variable (9) of the controller (6).

2. Module of a programmable logic controller according to claim 1, wherein the analog front end (2) is configured for digitizing and / or processing the signals of the sensor (10) into a voltage value, a current value, an FFT spectrum and / or an impedance spectrum.

3. Module of a programmable logic controller according to claim 1 or 2, wherein the artificial intelligence (3) is implemented as a neural network or as an interpretable and explainable algorithm.

4. Module of a programmable logic controller according to claim 3, wherein the artificial intelligence (3) is configured as a support vector machine or a learning vector quantizer.

5. Module of a programmable logic controller according to one of the preceding claims, wherein the artificial intelligence (3) is arranged and configured to take over the function of the controller (6) in order to provide the at least one manipulated variable (9) in its place, wherein a switch (11) is arranged which is configured to switch between the at least one manipulated variable (9) of the controller (6) and the at least one manipulated variable (9) of the artificial intelligence (3) in order to supply the latter to the controlled system (8).

6. Module of a programmable logic controller according to claim 5, wherein the switch (11) is configured to switch depending on at least one threshold value.

7. Module of a programmable logic controller according to one of the preceding claims, wherein the artificial intelligence (3) is configured to provide the at least one manipulated variable (9) or at least one correction value for the at least one manipulated variable (9), in particular for the selection of classes in a classification problem or for the selection of numerical attribute values ​​in a regression problem, in response to the at least one input variable, to recorded process data (7) stored in the control system and to disturbance variables (5) which influence the controlled system (8).

8. Module of a programmable logic controller according to claim 7, wherein the artificial intelligence (3) is configured to label certain temporally successive measured values ​​during a learning phase and to assign these to the categorical attribute values ​​(classification problem) or the numerical attribute values ​​(regression problem).

9. Module of a programmable logic controller according to one of the preceding claims, wherein the control behavior of the controller (6) can be learned based on experience.

Images