Method for providing a predicted binary process signal

Abstract

The invention relates to a method for providing a predicted binary process signal. In order to compensate for the signal delay for the binary process signal (S1), it is proposed to temporarily store other process signals (S2, S3, S4) of other sensors (2, 3, 4) for a preset time period (TB), and in the learning phase (20), at the switching time point (t S ), that is, at the time point at which the binary process signal shows a first edge change (F1) from logic zero to logic one or a second edge change (F2) from logic one to logic zero, the temporarily stored value of the process signal at the learning time point (t S -Tx) is fed as an input signal pattern of the excitation to a neural network (NN), wherein the learning time point is obtained by subtracting the prediction time period (Tx) from the switching time point.

Description

Field of the Invention

[0001] The present invention relates to a method for providing a predicted binary process signal of a first sensor for an automation control device that controls an industrial process, wherein an input component detects other process signals of other sensors in addition to the binary process signal of the first sensor to be predicted.

[0002] The present invention also relates to an input component that is designed to be connected to a first sensor and to detect the binary process signal of the first sensor, and that is also designed to detect other process signals. Background Art

[0003] A binary process signal is, for example, a feedback signal regarding the state of a control variable from a sensor of a control or regulation system, which is, for example, a switch position (on / off), a valve position (open / closed), or a motion state of a motor (rotating / stationary).

[0004] For example, the motion of a production line or a drive is controlled.

[0005] For example, the flow rate of a medium through a pipeline, the liquid level in a container, or the rotational speed or torque of an electric motor is regulated.

[0006] Such a control or regulation system is, for example, applied in a process control system. The process control system is used for the automation of processes in technical facilities. An automated process can, for example, be a process of a method or manufacturing technology or a process for generating or distributing electrical energy. The process control system is typically hierarchically structured in multiple layers (see, for example, EP 3 125 053 B1). At the lowest level, i.e., the so-called field level, the state of a technical process is detected by field devices configured as sensors (such as pressure transducers, temperature sensors, level sensors, flow sensors), or the process is deliberately influenced by field devices configured as actuators (such as position regulators for regulating valves).

[0007] Above this level is a control and / or regulation level with control and / or regulation devices, wherein a processing unit (central processor), which is typically part of a programmable controller, executes control and / or regulation functions close to the field as real-time as possible, and wherein the device receives the values of process variables as input parameters from the sensors and sends them as output parameters, such as commands, to the actuators.

[0008] At the process management level, which is still above this level, higher-level control and regulation takes place in a master computer, and an operator system consisting of one or more operating consoles enables the operation and observation of the process by the operators of the facility.

[0009] Data exchange between field devices (such as input components) and the processing unit of a programmable controller is usually carried out via a fieldbus (such as PROFIBUS DP or PROFINET). Since field devices usually do not have a corresponding fieldbus connection themselves, the field devices are connected to the digital fieldbus via decentralized peripheral stations. The peripheral station usually consists of an interface module (main component) for connecting to the digital fieldbus and a plurality of peripheral components for connecting field devices (mainly digital and analog input and output units). Each input or output unit can only have a separate so-called "channel" for connecting a single field device here. However, the unit can also have multiple channels for connecting multiple field devices (the number in the normal case is, for example, 8, 16 or 32).

[0010] The peripheral components are usually directly located on-site at the location of the field devices, while the programmable controller is located in a more central position of the facility. A highly future-oriented control and regulation design even provides cloud-based control or regulation devices.

[0011] In a typical sequence, the values of the input parameters provided by the input units of the peripheral components are periodically read in, processed by the processing unit in turn, and values for the output parameters are generated. These values for the output parameters are finally written into the output units.

[0012] What is disadvantageous for process signals is that the (apparent) "actual" value of the process signal provided by the input component at the time point of processing in the automation control device is already outdated and no longer current. This delay is regarded as the (additional) dead time during control or regulation. The reasons for this delay are, for example:

[0013] - The entire process of signal processing in the input unit, such as analog-to-digital conversion and filtering,

[0014] - Transmission via the sometimes slow fieldbus to the fieldbus controller,

[0015] - Transmission from the fieldbus controller to the automation control device or the processing unit, and

[0016] - The delay caused by first reading in the values of all or at least one group of input units before processing.

[0017] This problem particularly occurs in the automation control devices of facilities that are widely distributed in space in methods or manufacturing technologies, especially in facilities that can extend over several square kilometers, such as in facilities in the chemical industry, oil and gas industry, metal industry, mines, power plants, transportation infrastructure (airports, tunnels), etc.

[0018] To compensate to some extent for the poor control and regulation quality due to outdated values, alternatively, the processing cycle can be shortened. However, more frequent circulation will particularly impose a load on the processing unit. In addition, a larger transmission bandwidth is also required.

[0019] Another method is so-called clock synchronization, in which detection, processing, and output are strictly clocked. The delay between detection and processing is not eliminated, but rather determined. Here, the disadvantages are complex project planning and poor changeability during continuous operation. In addition, components that are usually specially provided for this purpose and are therefore expensive are generally required. Summary of the Invention

[0020] Therefore, the object of the present invention is to reduce the influence of the above-mentioned delay.

[0021] For a method for providing a predicted binary process signal of a first sensor for an automation control device for controlling an industrial process, wherein the input component detects other process signals of other sensors in addition to detecting the binary process signal to be detected of the first sensor, the object is achieved as follows, namely, for a preset time period, the signal change process is temporarily stored from the process signal respectively, and the signal change process has corresponding time allocation values of the signal change process, and wherein, in order to compensate for the delay between the actual occurrence of the binary process signal at the first sensor and the subsequent processing in the automation control device, a learning phase is executed, wherein, in the learning phase, at the switching time point, that is, at the time point when the binary process signal shows a first edge change from logic zero to logic one or a second edge change from logic one to logic zero, the temporarily stored value of the signal change process at the learning time point is fed as an input signal pattern of the excitation to the neural network, wherein the learning time point is obtained by subtracting the prediction time period from the switching time point, wherein the corresponding edge change is assigned to the input signal pattern, and in the running phase, other signal change processes are monitored, so that the current value from the other signal change processes is fed as an input signal pattern to be evaluated to the neural network, and when the learned input signal pattern is consistent with the fed input signal pattern, the relevant edge change is provided as a predicted value for the binary process signal to the automation control device.

[0022] For this purpose, the predicted value is preferably located in the future within the prediction time period, wherein the prediction time period is adjusted according to the delay between the detection of the variable and the generation from the input variable to the output variable.

[0023] Since the accuracy of the prediction decreases as the time period increases, the time period is preferably equal to or less than the sum of these delays. Advantageously, an optimization is adjusted between the processing of the value as current as possible (the time period is as large as possible) and the accuracy of the prediction (the time period is as small as possible).

[0024] Advantageously, methods for pattern recognition from the field of artificial intelligence are used. According to the invention, such pattern recognition is used to predict binary signals. Sensors (binary and analog) in an automation facility are typically connected to an automation control device via a fieldbus. Peripheral stations may also be present in the automation facility, which consist of a so-called main component and one or more input components. The input components detect sensor signals and ultimately forward the signals to the automation control device via the fieldbus. The advantage of this method is that a dynamic model of the facility does not have to be established for predicting binary signals. The algorithms used do not require knowledge of any physical conditions. The method can therefore be applied generally.

[0025] An advantage also lies in that monitoring during the operating phase is carried out on the side of the input component or on the side of the peripheral component assigned to the input component, where the predicted binary process signal is forwarded from there to the automation control device via fieldbus communication.

[0026] In another design of the method, it is proposed that the actual process signal occurring after the prediction at the switching time point during the operating phase is used to continuously improve the prediction of the learning process running continuously in the background.

[0027] Advantageously, the neural network is designed as a self-organizing map. An artificial neural network is referred to as a self-organizing map, a cocoon map, or a cocoon network. For unsupervised learning methods, the artificial neural network is a powerful tool for data mining.

[0028] The object mentioned at the beginning is also achieved by an input component that is designed to connect to a first sensor and to detect the binary process signal of the first sensor. The input component is also designed to detect other process signals. The input component includes: a memory designed to temporarily store the signal change process from the process signal for a preset time period respectively; a learner having a neural network and a trigger, the learner being designed to, at the switching time point, that is, at the time point when the binary process signal shows a first edge change from logic zero to logic one or a second edge change from logic one to logic zero, feed the stored value of the signal change process at the learning time point as an input signal pattern of the excitation to the neural network, where the learning time point is obtained by subtracting the prediction time period from the switching time point, the learner is also designed to assign the corresponding edge change to the input signal pattern; a monitor designed to monitor other signal change processes, so as to feed the current value from other signal change processes as an input signal pattern to be evaluated to the neural network, and in the case where the learned input signal pattern is consistent with the fed input signal pattern, provide the relevant edge change as a predicted value for the binary process signal to the automation control device.

[0029] Advantageously, the input component has a fieldbus interface, wherein there is a transmitter designed to send the predicted value with a higher priority than other messages.

[0030] In a further refinement of the input component, the learner is also designed for this purpose to work in the background of the monitor and use the actual process signal that occurs after the prediction at the switching time point for continuously improving the prediction of the learning process running continuously in the background. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The drawings illustrate embodiments of the invention and show here:

[0032] Figure 1 A diagram showing the evaluation of the process signal over time;

[0033] Figure 2 Process signals superimposed on top of each other to determine the change in the input signal pattern over time; and

[0034] Figure 3 An input component for detecting a process signal and predicting a binary process signal. DETAILED DESCRIPTION

[0035] According to Figure 1 , the automation control device 10 is connected to the peripheral component 12 via the fieldbus F. The peripheral component 12 has an input component 11, which is divided into a first input component 11a, a second input component 11b, and a third input component 11c. A first sensor 1 providing a first process signal S1 is connected to the first input component 11a. In addition, a second sensor 2 providing a second process signal S2 is connected to the first input component 11a. A third sensor 3 providing a third process signal S3 is connected to the second input component 11b. A fourth sensor 4 providing a fourth process signal S4 is connected to the third input component 11c. A prediction is to be made for the first sensor 1 or its first process signal S1. For this purpose, for a preset time period t B , the process signals S1, S2, S3, S4 respectively provided with signal change processes x 1(t) , x 2(t) , x 3(t) , x 4(t) and their corresponding time allocation values x1, x2, x3, x4 are temporarily stored in the peripheral component 12 having a memory 19.

[0036] To compensate for the delay between the actual occurrence of the binary process signal S1 at the first sensor 1 and the subsequent processing in the automation control device 10, a learning phase 20 is performed. In the learning phase 20, at the switching time point t SThat is, at the time point when the binary process signal S1 shows the first edge change F1 from logic zero to logic one or the second edge change F2 from logic one to logic zero, the signal change process x 1(t) , x 2(t) , x 3(t) , x 4(t) at the learning time point t s -Tx of the temporary value x 2( t s- Tx ) , x 3( t s- Tx ) , x 4( t s- Tx ) is conveyed as the input signal patterns M1,..., M5 for excitation to the neural network NN (see also Figure 3 ). The learning time point is obtained by subtracting the prediction time period Tx from the switching time point t s , where the corresponding edge changes F1, F2 are also assigned to the input signal patterns M1,..., M5.

[0037] Subsequently, a retrospective observation 33 of the past is performed for the learning phase for the signal change processes S2, S3, S4. In the case of the first signal change process S1 to be predicted, a prospective observation 32 is performed for the prediction.

[0038] The process signals S1, S2, S3, S4 are displayed superimposed on each other in Figure 2 . If at the switching time point t S the first edge change F1 occurs in the first process signal S1 of the first signal change process x 1(t) , then for the other process signals S2, S3, S4, the values from the past t S -Tx are conveyed to the neural network NN for the first input signal pattern M1 of the excitation to be input. In the case of other edge changes, the extraction of the historical values for the other input signal patterns M2, M3, M4, M5 is observed in a similar manner.

[0039] Figure 3 The input component 11 is shown, which is designed to connect to the first sensor 1 and detect the binary process signal S1 of the first sensor 1. The input component is also designed to detect the other process signals S2, S3, S4. The input component 11 has a memory 19, and the memory is designed to temporarily store the signal change processes x 1(t) , x 2(t) , x 3(t) , x 4(t) from the process signals S1, S2, S3, S4 respectively for a preset time period TB.

[0040] To provide the input signal patterns M1, …, M5 for the neural network NN during the learning phase 20, the input component 11 has a learner 21. The learner 21 also has a trigger 22 which, at the switching time point t S , i.e., at the time point when the binary process signal S1 shows a first edge change F1 from logic zero to logic one or a second edge change F2 from logic one to logic zero, transfers the signal change process x 1(t) , x 2(t) , x 3(t) at the learning time point t s -Tx of the temporary values x 2( t s- Tx ) , x 3( t s- Tx ) , x 4( t s- Tx ) as the input signal patterns M1, …, M5 serving as stimuli to the neural network NN. The learning time point is obtained by subtracting the prediction period Tx from the switching time point.

[0041] If the learning phase 20 ends, a monitoring phase 30 is introduced. For this purpose, the input component 11 has a monitor 31 which is designed to monitor further signal change processes x 2(t) , x 3(t) , x 4(t) , so as to transfer the current values x2, x3, x4 from the signal change processes x 2(t) , x 3(t) , x 4(t) as the input signal patterns M1, …, M5 to be evaluated to the neural network NN, and, if the learned input signal patterns M1, …, M5 are consistent with the transferred input signal patterns, provide, via the transmitter 14 and the fieldbus interface 13, the relevant edge changes F1, F2 as the predicted values PW for the binary process signal S1 for the automation control device 10.

Claims

1. A method for providing a predicted binary process signal of a first sensor (1) for an automation control device (10) controlling an industrial process, wherein, The input component (11) detects other process signals of other sensors in addition to the binary process signal to be predicted of the first sensor (1). For a preset time period (TB), the signal change process is temporarily stored from the process signals respectively, and the signal change process has corresponding time assignment values of the signal change process, and in order to compensate for the delay between the actual occurrence of the binary process signal at the first sensor (1) and the subsequent processing in the automation control device (10), a learning phase (20) is executed, where In the learning phase (20), at the switching time point (t S ), that is, at the time point at which the binary process signal exhibits a first edge change (F1) from logic zero to logic one or a second edge change (F2) from logic one to logic zero, the stored value of the signal change process at the learning time point (t S -Tx) is fed as an input signal pattern of the excitation to the neural network (NN), wherein the learning time point (t S ) minus the prediction time period (Tx) gives the learning time point (t S -Tx), wherein the respective edge change is assigned to the input signal pattern, and in the operating phase, other signal change processes are monitored, so that the current values from the other signal change processes are fed as input signal patterns to be evaluated to the neural network (NN), and in the case where the learned input signal pattern and the fed input signal pattern match, the relevant edge change is provided as a predicted value for the binary process signal to the automation control device (10), and the neural network (NN) is designed as a self-organizing map.

2. The method according to claim 1, wherein The monitoring in the operating phase is performed on the side of the input component (11) or on the side of the peripheral component (12) assigned to the input component (11), and the predicted binary process signal is forwarded via fieldbus communication from the side of the input component (11) or the side of the peripheral component (12) assigned to the input component (11) to the automation control device (10).

3. The method according to claim 1 or 2, wherein During the operation phase, the real process signal that occurs after the prediction at the switching time point (t S ) is used to continuously improve the prediction of the learning process that runs continuously in the background.

4. An input component (11) is designed to be connected to a first sensor (1) and to detect the binary process signal of the first sensor (1), and the input component (11) is also designed to detect other process signals. The input component (11) includes: a memory (19), which is designed to temporarily store the signal change process from the process signals respectively for a preset time period (TB); Learner (21), the learner (21) having a neural network (NN) and a trigger (22), the learner (21) being designed to, at a switching time point (t S ), that is, at the time point at which the binary process signal exhibits a first edge change (F1) from logic zero to logic one or a second edge change (F2) from logic one to logic zero, feed the stored value of the signal change process at a learning time point (t S -Tx) as an input signal pattern of the excitation to the neural network (NN), wherein the learning time point (t S ) minus a prediction time period (Tx) yields the learning time point (t S -Tx), the learner (21) being further designed to assign the corresponding edge change to the input signal pattern; a monitor (31), which is designed to monitor other signal change processes, so that the current values from the other signal change processes are fed as input signal patterns to be evaluated to the neural network (NN), and in the case where the learned input signal pattern and the fed input signal pattern match, the relevant edge change is provided as a predicted value for the binary process signal to the automation control device (10), and the neural network (NN) is designed as a self-organizing map.

5. The input component (11) according to claim 4, the input component (11) having a fieldbus interface (13), wherein, There is a transmitter (14) designed to send the predicted value with a higher priority than other messages.

6. The input component (11) according to claim 4 or 5, wherein, The learner (21) is also designed to operate in the background of the monitor (31) and use the true process signal that occurs after the prediction at the switching time point (t S ) to continuously improve the prediction of the learning process running continuously in the background.

Images