Method for monitoring by means of machine learning

Abstract

A method for monitoring an IO link system and/or at least one IO link device of the IO link system and/or a plant, a plant part and/or a process that works together with the IO link system is suggested. The current (I.sub.m), the voltage (U.sub.m) and/or the electrical power (P.sub.m) are here recorded (42) at at least one port of an IO link master of the IO link system. A monitoring of a condition (Z) and/or a detection of anomalies, errors, deviations and/or maintenance indicators and/or a prediction of a maintenance requirement, an error and/or an outage of the IO link system and/or of the at least one IO link device and/or of the plant, of the plant part and/or of the process in the IO link master occurs by means of a model (M) for the current, the voltage and/or the electrical power previously learned via machine learning.

Claims

1. Method for monitoring an IO link system and/or at least one IO link device (S1, S2, S3, S4, A1) of the IO link system and/or a plant, a plant part and/or a process that works together with the IO link system, the method comprising: recoding a the current (I.sub.m), the-voltage (U.sub.m) and/or the electrical power (P.sub.m) at at least one port (11) of an IO link master (1) of the IO link system, and monitoring of a condition (Z) and/or a detection of anomalies (30-33), errors (50-52), deviations and/or maintenance indicators and/or a prediction of a maintenance requirement, an error and/or an outage of the IO link system and/or of the at least one IO link device (S1, S2, S3, S4, A1) and/or of the plant, of the plant part and/or of the process occurs in the IO link master (1) by means of usage of a model (M) for the current (I.sub.e), the voltage (U.sub.e) and/or the electrical power (P.sub.e) previously learned via machine learning.

2. The method according to claim 1, wherein the model (M) is learned in the IO link master (1) via training data being recorded for the current (I.sub.e), the voltage (U.sub.e) and/or the electrical power (P.sub.e) and the corresponding conditions (Z), anomalies (30-33), errors (50-52), deviations and/or maintenance indicators, and the model (M) being calculated from these.

3. The method according to claim 2, wherein the model (M) learned in the link master (1) is transferred to at least one other IO link master.

4. The method according to claim 1, wherein the model (M) is pre-learned in an external system (PC) and transferred to the IO link master (1).

5. The method according to claim 1, wherein the model (M) is updated via the measured values (I.sub.m, U.sub.m, P.sub.m) while being used.

6. The method according to claim 1, wherein an evaluation and/or correction of the result of the usage of the model (M) and/or a confirmation or characterisation of the current condition (Z), anomaly (30-33), error (50-52), deviation and/or maintenance indicator can be undertaken by a user via an interface.

7. The method according to claim 1, wherein a temporal course of the current (I.sub.m), of the voltage (U.sub.m) and/or of the electrical power (P.sub.m) are recorded, and the temporal course and/or variables derived from the latter are used in the monitoring.

8. The method according to claim 1, wherein the current (I.sub.m), the voltage (U.sub.m) and/or the electrical power (P.sub.m) are measured at several ports (11) of the IO link master (1), and the values, their temporal course and/or variables derived from them are used in combination in the monitoring.

9. The method according to claim 1, wherein one of the following variants of machine learning is used: artificial neural networks decision-tree based methods; margin-based methods; cluster methods; ensemble methods; nearest neighbour methods; linear and/or non-linear regression methods.

10. The method according to claim 1, wherein a pattern recognition using the measured values for the current (I.sub.m), the voltage (U.sub.m) and/or the electrical power (P.sub.m) is undertaken on the basis of the model (M) in the event of a classification (44) of a condition (Z) and/or when anomalies (30-33), errors (50-52), deviations and/or maintenance indicators are detected, and/or when a maintenance requirement, an error and/or an outage is predicted.

11. The method according to claim 1, wherein a statistical test method using measured values for the current (I.sub.m), the voltage (U.sub.m) and/or the electrical power (P.sub.m) is undertaken when using the model (M) in the event of a classification (44) of a condition (Z) and/or when anomalies (30-33), errors (50-52), deviations and/or maintenance indicators are detected, and/or when a maintenance requirement, an error and/or an outage is predicted.

12. The method according to claim 1, wherein additional IO link data (D) that is transferred from one or several of the IO link devices (S1, S2, S3, S4, A1) is used during learning (41) of the model (M), and/or in the usage of the learned model (M), and/or in a classification (44) of a condition (Z) and/or when anomalies (30-33), errors (50-52), deviations and/or maintenance indicators are detected, and or when a maintenance requirement, an error and/or an outage are predicted.

13. The method according to claim 1, wherein the monitoring is provided by the IO link master (1) having an electronic computing device that is equipped to carry out the monitoring.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Exemplary embodiments of the invention are depicted in the drawings and explained in more detail in the following description.

[0058] FIG. 1 shows a schematic depiction of an IO link system

[0059] FIGS. 2a and 2b respectively show a conventional IO link port according to the state of the art, wherein FIG. 2a shows a 4-pin port, and FIG. 2b shows a 5-pin port.

[0060] FIG. 3 is a diagram of a measured current strength over the time for a periodic process of a plant, in which occurring anomalies are marked.

[0061] FIG. 4 shows the diagram from FIG. 3, in which detected anomalies are marked via the inventive method.

[0062] FIGS. 5a and b show a flowchart of an embodiment of the method according to the invention.

[0063] FIG. 6 shows the association between IO link devices and errors, as well as data on the basis of which the errors can be detected.

EMBODIMENTS OF THE INVENTION

[0064] The embodiments for conditions, anomalies, errors, deviations and maintenance indicators are described separately from one another in the following. The described embodiments can transfer to the others and/or be combined with them, however.

[0065] FIG. 1 shows an IO link system having an IO link master 1 that is connected to a programmable control PLC and an external computer PC via a fieldbus FB. The computer PC serves both as external computing device and as interface to a user, via which the result of the monitoring can be output, and the user can give a reply. The IO link master 1 has a computing device (not depicted) that carries out the method according to the invention as described below. The IO link master 1 also has several link ports 11 (only one port is provided with a reference numeral for reasons of overview) that are connected to the IO link devices. Exemplary sensors S1, S2, S3 and S4 - S4 via an inductive coupler IC - and an actuator A1 are each connected to the IO link master 1 via one port 11 each.

[0066] FIG. 2a shows a conventional 4-pin IO link port according to the state of the art, which has four pins in total: pin 1, pin 2, pin 3, pin 4.

[0067] FIG. 2b shows a conventional 5-pin IO link port according to the state of the art that has five pins in total: pin 1, pin 2, pin 3, pin 4, pin 5. In the following, the pins are denoted according to the denotation in the connecter, as “pin 1” to “pin 4” or “pin 5”. The IO link port is specified according to the connection technology in IP65/67 in standard IEC 61131-9 in such a way that pin 1 and pin 3 are used to provide energy, and the data is transferred via pin 4. In the IO link port according to type B depicted in FIG. 2b, an additional energy provision is provided via pin 2 and pin 5. In a 4-pin port as shown in FIG. 2a, a current measurement takes place at pin 1 or pin 3, and a voltage measurement is carried out between pin 1 and pin 3. In a 5-pin port as depicted in FIG. 2b, a current measurement can additionally be carried out at pin 2 or pin 5, and a voltage measurement can additionally occur between pin 2 and pin 5.

[0068] In each of FIGS. 3 and 4, a diagram of a current strength I measured at port 11 of the IO link master 1 over the time t for a periodic process of a plant is depicted. In FIG. 3, anomalies 20 to 23 are marked. While the anomaly 20 manifests as a sharp peak (“outlier”), the anomalies 21, 22 and 23 represent changes in the course. A threshold value monitoring is conventionally carried out, in which the measured current strength I is compared with a threshold value that is here shown as SW for comparison. It can be seen that in this example only the anomaly 20, and thus the “outlier” can be detected in the threshold value monitoring.

[0069] Detected anomalies 30 to 33 that have been detected via the method according to the invention are entered in FIG. 4. A comparison with FIG. 3 shows that the detected anomalies 30 to 33 contain both the anomaly 20 in the form of the “outlier” and the anomalies 21, 22 and 23 in the course. All anomalies 20 to 23 can thus be detected via the method according to the invention.

[0070] FIGS. 5a and 5b show a flowchart of the method according to the invention for an intelligent condition monitoring. The condition monitoring can be carried out for the entire IO link system, for the IO link devices S1, S2, S3, S4, A1 and/or for a plant, a plant part and/or a process that works together with the IO link system. In the following, the IO link devices S1, S2, S3, S4, A1 should be monitored.

[0071] FIG. 5a shows the learning and training phase for a model M. In this exemplary embodiment, the current I.sub.e, the voltage U.sub.e and/or the electrical power P.sub.e are each measured and recorded at the port 11 of the IO link master 1, in order to use these as training data for learning the model. The corresponding conditions Z.sub.e in which each of the IO link devices S1, S2, S3, S4, A1 are during the measurement are recorded. The training data I.sub.e, U.sub.e, P.sub.e is added to a pre-processing 40. In this exemplary embodiment, IO link data D that is sent over from the sensors S1, S2, S3, S4 to the IO link master 1 is included, and is also added to the pre-processing 40. Features can be extracted from the data or the data can be transformed, filtered, aggregated and/or otherwise processed in the pre-processing 40. The pre-processing 40 used is connected to the particular model M and implemented in the IO link master 1 together with said model. A learning 41 of the model M (also described as training) then occurs, in which the connection between the training data, and thus the current I.sub.e, the voltage U.sub.e, the electrical power P.sub.e and the IO link data D and the corresponding condition Z.sub.e is learned. The machine learning is based on pattern recognition, for example, in which patterns in the course of the electrical variables I.sub.e, U.sub.e, P.sub.e and the IO link data D are recognised and connected to the conditions Z.sub.e. This is provided by an artificial neural network, for example, but is not limited to this variant of machine learning. The electrical variables I.sub.e, U.sub.e, P.sub.e and the IO link data D measured for learning can also be maintained during the learning 41 in further exemplary embodiments.

[0072] In this exemplary embodiment, the pre-processing 40 and the learning 41 of the model M occurs directly in the computing device of the IO link master 1. The learned models M can be transferred from the IO link master 1 in which they are learned to other IO link masters, which are preferably used in IO link systems constructed in the same manner. In further exemplary embodiments, the learning 41 of the model M occurs on the external computer PC or via a cloud (not depicted), either of which is connected to the IO link master 1 via the fieldbus FB. The model M is finally transferred to the IO link master 1.

[0073] FIG. 5b shows the evaluation or operating phase. A measurement 42 of the electrical variables, and thus of the current I.sub.m, of the voltage U.sub.m and/or the electrical power P.sub.m at the ports 11 of the IO link master 1 occurs during operation. The current I.sub.m is provided in the IO link master 1 via the voltage drop at a small sense resistance and subsequent transformation of the voltage drop into a digital value. Electronic current measurement components that determine the current I.sub.m from a magnetic field measurement can alternatively be used. The voltage U.sub.m is determined directly via analogue/digital transformation. The electrical power P.sub.m is calculated as the product of the current I.sub.m and the voltage U.sub.m according to formula 1 (see above). The electrical variables I.sub.m, U.sub.m, P.sub.m measured during operation are added to a pre-processing 43. In this exemplary embodiment, the IO link data D that is sent over from the sensors S1, S2, S3 and S4 to the IO link master 1 is included, and also added to the pre-processing 43. The pre-processing 43 is carried out in the same manner as the pre-processing 40. A classification 44 of the current condition Z then occurs via usage of the model M and the electrical variables measured in operation, and thus the current I.sub.m, the voltage U.sub.m and the electrical power P.sub.m and the IO link data D via machine learning. Here, too, the machine learning is based on pattern recognition, for example, in which patterns are recognised in the course of the measured electrical variables I.sub.m, U.sub.m, P.sub.m and the IO link data D, and the condition Z is classified on the basis of the model M. This is provided by an artificial neural network, for example, but is not limited to this variant of machine learning. Alternatively, a statistical test method is carried out. The model M can be updated during its usage via the measured electrical variables I.sub.m, U.sub.m, P.sub.m and the IO link data D.

[0074] Exemplary connections between the condition Z of the IO link device S1, S2, S3, S4, A1 or between a condition of a plant to be monitored and the current or power consumption are described in the following:

[0075] In inductive couplers IC, the efficiency of the energy transfer depends on the width of the air gap, the lateral offset and the angular offset of the two couplers and the temperature. Changes in the position of the coupler or the temperature can be recognised via intelligent condition monitoring that includes the current or power consumption of the inductive coupler IC in the monitoring.

[0076] In the case of optical distance sensors having adaptive transmission power of the transmission light source (e.g., LED or laser diode), a dirtying of the optical components or a change of the target can be recognised by means of the current or power consumption dependent on the transmission power.

[0077] In the case of IO link devices, the current or power consumption of an IO link device can change due to warming or aging (e.g., drying out of electrolyte condensers) of the electronic components. This typically very slow change can also be recognised by means of the intelligent condition monitoring. A prediction of a maintenance requirement can also be made, and an anticipatory maintenance can be planned.

[0078] The power consumption of LEDs is fundamentally dependent on the condition of the IO link device (e.g., the display of an alarm condition via a blinking LED). Each LED generates a current consumption of several milliamps only in the lit condition. Defective LEDs of an IO link device can thus be recognised due to deviations between the sensor signal and the current signal.

[0079] In sensors for measuring path and spacing (inductive, magnetostrictive, opto-electronic), the current consumption is often connected to the path or spacing in a non-linear manner if the current for energising the sensor element is automatically controlled, in order to compensate for the decay of the measured section, for example. This non-linear dependency can be adopted in the model.

[0080] In the case of actuating drives, an insufficient lubrication or corrosion can damage the smoothness of the actuator, which can lead to an increased current or power consumption of the actuating drive. The blockage of a drive is also visible in the course of the current.

[0081] In hydraulic or pneumatic plants, a change of the viscosity of the liquid for a magnetic valve can lead, among other things, to a change of the course of the current or the power when switching the valve. If the fluid pressure changes, then this leads to a change of the holding current or of the holding power. Changes of the fluid can thus be recognised without additional sensors.

[0082] These examples given above should only be understood as a portion of the various options for intelligent condition monitoring on an IO link master via the usage of the current, of the voltage and/or of the electrical power of the IO link port, and the invention is not limited to these. Far more complex tasks in the area of condition monitoring can also be solved by the method according to the invention, which have a less obvious connection between data and condition.

[0083] In FIG. 6, the IO link devices S1, S2, S3, A1, S4 & IC from FIG. 1 are listed, and are assigned in exemplary form to different errors (error cases) 50 to 52 by means of the different data types - the IO link data D or the current I, voltage U and/or electrical power P measured at the port 11. The error 50 can be directly detected from the IO link data D. The detection can also occur, however, via the current I, the voltage U and/or the electrical power P by means of the method according to the invention. A combination of the IO link data D and the data of the current I, of the voltage and/or of the electrical power leads to an improvement of the detection. The errors 51 and 52 cannot be detected via the IO link data. They can only be detected by means of the method according to the invention, via the current I, the voltage U and/or the electrical power P. The current I, the voltage U and/or the electrical power P are measured at several ports 11 of the IO link master 1 in order to detect the error 51, such that the data of the current I, the voltage U and/or the electrical current P of one or several of the sensors S1, S2, S3, of the actuator A1 and of the sensor S4 flows in together with the inductive coupler IC when the error 51 is detected. The error 52, however, is detected only from the data of the current I, of the voltage U and/or of the electrical power P of the sensor S4 together with the inductive coupler IC.