EP4186156A1

EP4186156A1 - Monitoring an event in a power converter

Info

Publication number: EP4186156A1
Application number: EP21786792.8A
Authority: EP
Inventors: Fabian Hecht; Cornel-Marian Rosca
Original assignee: Siemens AG
Current assignee: Innomotics GmbH
Priority date: 2020-09-30
Filing date: 2021-09-28
Publication date: 2023-05-31
Also published as: CN116249946A; EP3979479A1; WO2022069430A1

Abstract

The invention relates to a method for monitoring an event in a power converter. Record data is used after starting the power converter, wherein the record data does not indicate an error for a period of time after the power converter is started, and the period of time can be determined. Messages are assigned to error sources, and a degree of probability is calculated for an assigning process.

Description

description

Event monitoring at a power converter

The invention relates to the detection of critical events in a power converter.

A power converter makes it possible to operate an electrical machine, such as a motor, with a variable speed. When the electrical machine is used as a generator, a power converter can also be used to convert the electrical current. In this way, for example, a grid feed-in can be implemented. For example, for a motor application, the mains power with constant frequency and voltage is converted into power with variable frequency and voltage.

The power converter is, for example, a converter, a rectifier or an inverter. The power converter can be water-cooled and/or air-cooled. The power converter is used, for example, in applications with high demands on reliability and quality. Examples of applications for power converters are:

• Industrial pumps and fans

• Oil and gas pumps and compressors, e .g . B. electric submersible pumps and high speed compressors

• Boiler fans (induced draft and forced draft) for power generation

• Clean water and waste water pumps

• Applications with multiple motors and synchronous transfer

(e.g. pipelines in the oil and gas industry).

These application examples often relate to use of the power converter in which high power levels in particular are required. In particular, these are outputs in the one-digit, two-digit or three-digit megawatt range. Converters which relate to the medium voltage are then preferably used for this purpose. These are medium-voltage to designate converters. A voltage greater than or equal to 1000 V can be regarded as medium voltage. Voltages of 4000 V or 6000 V can also be referred to as medium voltage. The power converter is, for example, a variable-frequency drive (VFD). The Sinamics Perfect Harmony GH180 is an example of a VFD.

If the power converter has problems, the problem is detected and documented in a log file or saved . The log file represents a log, with the detected and documented problems being log entries. These log entries can be warnings to alert a user to potentially critical events. These log entries can also be errors to alert a user to an error or to document this error. Critical events are also errors, for example. Error messages or Warning messages generated. Errors can lead to failure of the power converter or have led . However, since a power converter failure usually results not only in a fault but in a cascade of fault and warning log events, the technical cause of the power converter failure is obscured and can only be deduced by power converter experts who understand the history of the Analyze log events.

One object of the invention is to improve event monitoring of the power converter.

A solution to the problem results from a method according to claim 1 or. according to a method according to claim 6 or. in an event monitoring according to claim 10 . Configurations result, for example, according to claims 2 to 5 , 7 to 9 and 11 .

In a method for event monitoring in a power converter, log data of the power converter are used, with the log data being data does not show any error for a period of time after the start, the period of time being determinable, the power converter having at least two types of errors or has warnings, a first type of error or Warning type, which depends on the type of converter, so are determined in particular by this and a second type of error or. Warning type, where the second error type has errors or . the second type of warning has warnings that can be defined by a user, that is, are determined by the user, with an evaluation of a combination of errors or Warnings of the respective first type and the respective second type are used and/or an evaluation of a combination of errors or Warnings of the respective second type are used. The power converter can thus be designed, for example, by a user in such a way that the user defines individual messages for the power converter. An example of a message is an error or a warning, i.e. an error message or a warning message. Through the individual messages, an individual event monitoring can be created, which is dependent on the messages created individually by the user. This improves event monitoring and can make it more accurate. The messages defined by the user are stored, for example, in an SOP (System Operating Program). defined there. Such user-defined messages can be marked as such when the message is displayed. User-defined messages are based, for example, on signals from I/O interfaces that arise in a system in which the power converter is integrated. Signals on which the user-defined messages are based can be digital signals or analog signals, for example. User-defined messages can be generated, for example, based on a single signal and/or based on a combination of signals. Such signals can, for example, concern an emergency stop, opening a door, blowing a fuse, undervoltage, overvoltage, fault current, overcurrent, fan failure, failure of a power module of the power converter, the bypass of a power module of the power converter, an insulation error, an insulation warning, a communication error in particular of a power module of the power converter, an error or a warning for cooling the power converter, a pre-charging of the power converter, etc. . It can be seen that individually created messages are important for a power converter that is integrated into individual use (industrial plant). These messages created individually by a user can be used advantageously for event monitoring of the power converter in its individual environment, ie the industrial installation. This improves the quality of monitoring. The user of the power converter is a person who operates the power converter. This operation can be carried out, for example, by an operator of the power converter or by a commissioner or the like.

In one embodiment of the method, log data after a start of the converter, in particular a successful start of the converter, is used to generate event monitoring in a power converter, with the log data for a period of time after the start showing no error, the time period being determinable. The power converter is a medium-voltage power converter, for example. The log data relate, for example, to status messages, warning messages and/or error messages. Such reports relate to the following elements, for example: a controller for the converter, a controller for the converter, a temperature sensor, an air volume sensor, an ammeter, a voltmeter, a power semiconductor, etc. A successful start of the power converter is, in particular, a start in which no error messages and/or warning messages are generated.

In one embodiment of the method, the log data are marked, the marking, in particular temporal information, such as a time stamp or concerns a sequence of logged messages, with a message tion is in particular an error and / or a warning, in particular log data after the start of the converter or be used after a reset (reset) of the power converter.

In one embodiment of the method, a machine learning model, ie an artificial intelligence, is used. The artificial intelligence determines the most probable technical root causes of a failure or malfunction. of an error in an automated manner. Using artificial intelligence, a log event history can be analyzed automatically. The log event history corresponds to the log data.

In one embodiment of the method, an algorithm for machine learning is used, which can categorize the technical root cause, ie the source, of an error recognized as such. Different steps can be carried out for this purpose. In one step, failures of a historical stack of logbook data, which can also be referred to as log data, can be identified by an expert. In a further step, a machine learning algorithm can be trained to categorize error events into predefined cause categories. There is therefore a categorization into sources for errors. The artificial intelligence can be trained in such a way that, based on a cascade of messages (one or a large number of: status messages, warning messages, error messages), it concludes on a source of error, i.e. H . this indicates . In one embodiment, several error sources can also be specified, with a probability for the correctness of this information being calculated or is issued .

In one refinement of the method, the log data is thus identified, with the identification relating to a chronological sequence of logged messages. The log data shows these same messages. The chronological sequence results from a cascading of messages. Of the- such messages may be interdependent or independent of one another. In particular, the messages that are considered in their sequence are of different types. So there are messages which are determined by the type of converter and messages which are determined by the user of the converter.

One task of artificial intelligence is to recognize interdependent messages and to assign them to a causal source.

In one embodiment of the method, historical data of one or more power converters, in particular of the same type, are searched for errors in a database in an automated manner in one step. For example, the errors are listed with the following information about the error, i.e. distinguished from one another:

Errors that occurred up to 1 hour after the first error in the event of a failure. Designed for best results, this time interval is a typical time period when errors cause side effects or require human intervention to resolve the issue; These errors affect e.g. B. : an input-related error (e.g. a problem in the power supply to the drive) , a cooling system-related error (e.g. an air or water cooling failure) and/or a power cell communication error (e.g. a internal drive electronics) ; The information gathered contributes to the accuracy of the presented approach;

Warnings that occurred before the first failure, specifically up to three hours before the first failure; this time interval ensures that the most critical information (messages) causally related to the fault is collected and selected after expert advice and engineering of the results; However, different time intervals could be selected for different types of failures, since they have a different time up to failure is expected; For refrigeration system related failures, even a week's time for root cause analysis could be important (eg, a leak in the coolant may result in alarms days or weeks before actual failure of the refrigeration system can be diagnosed); The time between a first and last error; This time is important because it can separate failures from a combination of failures that occur almost simultaneously and are therefore likely to be causally related, from a series of failures caused by human intervention.

In one refinement of the method, reports are assigned sources of errors. In this way it can be achieved that consequential errors can be distinguished from a causal error.

In one embodiment of the method, information about the chronological sequence of messages and about a categorized causation is given to an expert for training the artificial intelligence, who determines the root cause of each failure and, for example, notes it in a list as at least one of the following root cause categories, which in particular includes all cover critical components of the power converter:

Power Gell (power cell / power semiconductor module) related (e.g. failure due to a Power Gell problem),

Power Gell communication related (e.g. failure due to power cell communication problems), system related, cooling related (e.g. failure due to a cooling system problem), input related (e.g. failure due to input problems from an operator on site to drive the power converter), power related (e.g failure due to problems in the application driven by the drive) , Precharge related (e.g. failure due to precharge problems when starting the converter) , Loss of drive related (e.g. failure due to excessive drive losses) , Manual stop (e.g. cascades of faults initiated by manually turning off the drive) .

In one embodiment of the method, a probability is calculated for an assignment. In this way, you can approach the elimination of an error in a more targeted manner without forgetting possible but improbable failures.

In one refinement of the method, a list of the errors, ie the log data, with expert labels is used as input for an algorithm for machine learning in the sense of a monitored learning approach in one step. The machine learning algorithm is optimized to be able to categorize previously unknown failures, returning the categories with a categorization probability that is e.g. the root cause, i.e. the source of the failure: e.g. cooling related failure (90 %) , system related error (10%) .

In one embodiment of the method, artificial intelligence is trained and event monitoring is thus generated.

In a method for event monitoring in a power converter, messages are recorded, with the messages having a time stamp and an identification, with an event being recognized by the sequence of the messages. A specific causal source of error can be inferred from a specific cascading. Where causally related errors can be identified.

In one embodiment, a data collection can be carried out using a digital platform, in particular in a cloud, to optimize drive systems, motors, converters. be laid . The data relate in particular to log files of corresponding machines, such as a power converter or a motor or a transformer. In particular, this data can be used to categorize the root cause of a selected error using a machine learning algorithm. For example, log data obtained from a power converter can contain the following information: the time stamp of the log event, the severity of the log event: info, error or warning, a text of the log event, the trigger of the log event.

So e.g. B. a failure of a drive is indicated by the occurrence of one or more errors after a certain error-free time after a successful restart. The error-free time is defined as at least seven days. This is a characteristic time that e.g. B. was set during development to ensure the drive was in a regular operating mode. In this case, the drive has at least one power converter.

In one refinement of the method, the identification has a report type, a text and/or a source of the report. In this way, a uniqueness can be ensured.

In one embodiment of the method, the event is an initial error, with artificial intelligence being used to identify the event, with the artificial intelligence being a cloud application. In this way, the same quality of artificial intelligence can always be used, regardless of location.

In one embodiment of the method, event monitoring of the type described is used. The event monitoring Calculation of a power converter can therefore have artificial intelligence and/or access it.

Event monitoring of a power converter has a communication device and data processing for carrying out one of the methods described.

It is now possible to extract the cause of a power converter failure from the log events without in-depth technical knowledge. Until now, it was only possible for experts to analyze the log files of the power converter and derive the cause of a failure. As a result, z. B. The following is possible: a user can quickly analyze the failure of his power converter and get his application up and running again more quickly; the user can check the failures of his drive, which has at least one power converter, in the past in order to optimize his system for fewer downtimes;

Converters can be checked more easily and service measures can be suggested more easily; a faster restart of a power converter is possible because the time for root cause analysis by experts is drastically reduced.

In contrast to manual analysis, the automated solution presented here is particularly capable of processing any number of errors at the same time.

A cloud analysis approach is also possible, since the log file data from the power converter is stored in the cloud. As a result z. B. provided the root cause analysis on scalable cloud instances in a suitable Python environment. These technical features contribute to the great advantage of the presented approach as they allow process automation. In one embodiment of the invention, the machine learning approach (labelling) derived from expert knowledge is used as a key algorithm for the artificial intelligence algorithm and is implemented as an automated module for cause analysis. In particular, error extraction is based on historical data, as is expert tagging.

The features of each claimed or The items described can be easily combined with one another. In the following, the invention is illustrated and explained in more detail by way of example with reference to figures. The features shown in the figures can be expertly combined to form new embodiments without departing from the invention. Show it

1 shows a first example and,

2 shows a second example.

The representation according to FIG. 1 shows two faults F1 and F2 occurring simultaneously (F1: Cooling Sys Mv Trip and F2: Cooling Sys Vfd Trip). These errors are recorded as soon as they occur within one hour of T2 and are recognized as errors. Up to three hours T3 before the first fault, the Al Coolnt Tank Level < 30 alarm was triggered twice. Algorithm A comes to the conclusion that this is a failure Q of the cooling system. In the period T3 z. B. a first error Fl can also occur, which has nothing to do with the failure Q.

The illustration according to FIG. 2 shows that three different faults F4, F5 and F6 occurred within two seconds T4: F4 input protection fault, F5 emergency stop remote and F6 emergency stop. No alarms are found three hours T3 before the first fault F4. By just looking at the "Input Protection Fault" an expert may conclude that the drive has been stopped by a serious drive-related problem. However, "Emergencia Stop Remote" and "Notstopp" lead to the conclusion of the algorithm that it is a manual stop of the drive. This is useful as it avoids searching for a problem in the event log and allows the user to directly restart the drive.

For example, as a result, if the machine learning algorithm knows the errors and warning that occurred during an outage, it can determine the cause of a previously unseen outage with >95% accuracy. The machine learning algorithm used is based, for example, on a C-support vector classification. The root cause analysis can be triggered selectively for a single drive failure (specified as a critical log event) so that it is accessible to service and the operator when needed.

Claims

patent claims

1 . Method for event monitoring in a power converter, using log data from the power converter, with the log data not showing any errors for a period of time after the start before event monitoring begins, the time period being determinable, the power converter having at least two types of errors or has warnings, a first type of error or Warning type, which depends on the type of converter and a second error type or. Warning type, where the second error type has errors or . the second warning type has warnings, which can be defined by a user, with an evaluation of a combination of errors or Warnings of the respective first type and the respective second type are used and/or an evaluation of a combination of errors or Warnings of the respective second type are used.

2 . The method of claim 1, wherein the log data are marked, wherein the marking is a temporal information or. relates to a sequence of logged messages, one message being an error and/or a warning, with log data being used in particular after the power converter has been started.

3 . Method according to claim 1 or 2, wherein reports are and/or are sources of errors.

4 . Method according to claim 3, wherein a probability is calculated for an assignment.

5. The method according to any one of claims 1 to 4, wherein an artificial intelligence is trained and in this way, in particular, an event monitoring is generated.

6. A method for event monitoring in a power converter, with messages being recorded, the messages a Have a time stamp and an identification, with an event being recognized by the sequence of messages of different types.

7. The method according to claim 6, wherein the identification has a report type, a text and/or a source of the report.

8. The method according to claim 6 or 7, wherein the event is an initial error, wherein an artificial intelligence is used to recognize the event, wherein the artificial intelligence is a cloud application.

9. The method according to any one of claims 6 to 8, wherein an event monitor generated according to claim 1 to 4 is used.

10. Event monitoring of a power converter, which has an artificial intelligence.

11. Event monitoring according to claim 10, wherein a method according to any one of claims 1 to 9 is used.