DE102022003261A1 - Circuit arrangement and method for detecting the health status of electrical consumers - Google Patents

Abstract

The invention relates to a circuit arrangement (1) for a motor vehicle, comprising at least one consumer group (VG1 to VGn) and an artificial intelligence (KI, KI_T), which is configured to determine a normal operating point for each consumer group (VG1 to VGn) in a training mode Consumers (R_V1 to R_Vn) to learn by operating the consumers (R_V1 to R_Vn) in a standard operating mode, the artificial intelligence (Kl, KI_T) trained in this way being further configured to operate in an application mode in a consumer group (VG1 to VGn) to distinguish between normal operation and error cases at the consumer level, whereby each consumer group (VG1 to VGn) has a microcontroller (µC) which is used to determine a total current (IG1 to IGn) through a shunt (R1) in a supply line (L) to the consumers (R_V1 to R_Vn) is configured, whereby the artificial intelligence (KI, KI_T) can be supplied with the determined total currents (IG1 to IGn) of all consumer groups (VG1 to VGn) as well as vehicle bus signals (BS) as input signals, with the total currents in the standard operating mode (IR1bis IRn) together with a label “error-free” and the vehicle bus signals (BS) as training data can be supplied to the artificial intelligence (KI) as input variables, with the trained artificial intelligence (KI_T) receiving unlabeled total currents (IG1bis IGn) and bus signals in application mode (BS) can be supplied as input variables.

Description

The invention relates to a circuit arrangement according to the preamble of claim 1 and a method for detecting a health status of electrical consumers according to the preamble of claim 7.

Single semiconductor fuses with built-in logic (e.g. SPI interface) and shunt are cost-intensive and do not offer the possibility of early detection of errors and anomalies. Such semiconductor fuses also have increased internal consumption and require more data lines and/or connection points (e.g. SPI).

• einen Speicher, um Anweisungen und mehrere vorgegebene Muster zu speichern; und eine Neuronales Netz-Steuereinheit, die mit dem Speicher gekoppelt ist, wobei die Neuronales Netz-Steuereinheit durch die Anweisungen konfiguriert ist zum Überwachen mehrerer eindeutiger Muster, die in Echtzeit erzeugt werden und einem Satz Systemparameter des Host-Systems entsprechen, wobei die Systemparameter mindestens eines oder mehrere der Folgenden umfassen:
  • Eingangsparameter, Steuerparameter, Rückkopplungsparameter und Ausgangsparameter, wobei die eindeutigen Muster die Leistung des Host-Systems auf Systemebene in Echtzeit anzeigen und jedes eindeutige Muster für ein entsprechendes Untersystem des Host-Systems eindeutig ist, wobei die eindeutigen Muster eine Identifizierung von Fehlern ermöglichen, die mit einem oder mehreren Untersystemen verbunden sind, ohne von physikalischen Sensoren zum Detektieren eines Fehlers in dem Untersystem abzuhängen, und der Satz Systemparameter lediglich eine Untermenge von mehreren Systemparametern des Host-Systems enthält; Vorkonfigurieren der vorgegebenen Muster, wobei zum Vorkonfigurieren der vorgegebenen Muster die Neuronales Netz-Steuereinheit durch die Anweisungen konfiguriert ist zum Erfassen von Trainingsdaten, die die Leistung des Host-Systems auf Systemebene in einem normalen Betriebszustand der Untersysteme und in mehreren anomalen Betriebszuständen des oder der Untersysteme aus den Untersystemen umfassen; und
  • Extrahieren mehrerer Merkmalsvektoren aus den Trainingsdaten, wobei die Merkmalsvektoren die vorgegebenen Muster aufweisen, die eine oder mehrere mögliche Anomalien in dem Host-System anzeigen;
  • Vergleichen der eindeutigen Muster mit den vorgegebenen Mustern, die dem Satz Systemparameter entsprechen; und
  • Detektieren einer oder mehrerer möglicher Anomalien in dem Host-System und mindestens eines fehlerhaften Untersystems aus den Untersystemen auf der Grundlage des Vergleichs, wobei das mindestens eine fehlerhafte Untersystem für einen Beitrag zu der oder den möglichen Anomalien in dem Host-System verantwortlich ist.

EP 3 098 681 B1 describes a health management system for diagnosing and prognosticating a host system, where the host system has several subsystems and the health management system includes the following:

• a memory to store instructions and a plurality of predetermined patterns; and a neural network controller coupled to the memory, the neural network controller configured by the instructions to monitor a plurality of unique patterns generated in real time and corresponding to a set of system parameters of the host system, the system parameters being at least include one or more of the following:
  • Input parameters, control parameters, feedback parameters and output parameters, wherein the unique patterns indicate real-time system-level performance of the host system and each unique pattern is unique to a corresponding subsystem of the host system, the unique patterns enabling identification of errors associated with one or more subsystems without depending on physical sensors to detect a fault in the subsystem, and the set of system parameters includes only a subset of a plurality of system parameters of the host system; Preconfiguring the predetermined patterns, wherein to preconfigure the predetermined patterns, the neural network controller is configured by the instructions to acquire training data that demonstrates system-level performance of the host system in a normal operating state of the subsystems and in multiple abnormal operating states of the subsystem(s). from the subsystems include; and
  • Extracting a plurality of feature vectors from the training data, the feature vectors having the predetermined patterns indicating one or more possible anomalies in the host system;
  • comparing the unique patterns with the predetermined patterns corresponding to the set of system parameters; and
  • Detecting one or more possible anomalies in the host system and at least one faulty subsystem among the subsystems based on the comparison, the at least one faulty subsystem being responsible for contributing to the possible anomaly(s) in the host system.

The invention is based on the object of specifying a novel circuit arrangement and a novel method for detecting the health status of electrical consumers.

The object is achieved according to the invention by a circuit arrangement with the features of claim 1 and by a method for detecting a health status of electrical consumers with the features of claim 7.

Advantageous embodiments of the invention are the subject of the subclaims.

A circuit arrangement according to the invention for a motor vehicle comprises at least one consumer group and an artificial intelligence, which is configured to learn a normal operating point of the consumers in a training mode for each consumer group by operating the consumers in a standard operating mode, the artificial intelligence trained in this way Intelligence is further configured to distinguish between normal operation and error cases at the consumer level in an application mode in a consumer group. According to the invention, each consumer group has a plurality of consumers which are connected to an operating voltage via a common supply line and via a shunt, each consumer group also having a microcontroller which is used to measure a voltage drop across the shunt and to calculate a total current corresponding to the voltage drop the shunt is configured, whereby the artificial intelligence can be supplied with the determined total currents of all consumer groups as well as vehicle bus signals as input signals, with the total current of the consumer groups together with a label “error-free” and the vehicle bus signals as training data for the artificial intelligence in the standard operating mode Input variables can be supplied, with the trained artificial intelligence receiving unlabeled total consumer flows in application mode groups and bus signals can be supplied as input variables.

In one embodiment, the consumers are connected to the supply line for the operating voltage via a respective semiconductor fuse.

In one embodiment, the microcontroller is further configured to control the semiconductor fuse via a respective control output.

In one embodiment, the semiconductor fuse is designed as a field effect transistor, in particular MOSFET.

In one embodiment, the artificial intelligence is designed as a neural network.

The circuit arrangement can be part of a motor vehicle.

According to one aspect of the present invention, a method for detecting a health status of a plurality of electrical consumers is proposed, comprising a circuit arrangement comprising at least one consumer group and an artificial intelligence, wherein in a training mode the artificial intelligence learns a normal operating point of the consumers for each consumer group by operating the consumers in a standard operating mode so that the trained artificial intelligence can distinguish between normal operation and fault cases at the consumer level in an application mode in a consumer group. According to the invention, each consumer group has a plurality of consumers that are connected to an operating voltage via a common supply line and a shunt, each consumer group also having a microcontroller that measures a voltage drop across the shunt and calculates a corresponding total current through the shunt, where the determined total currents of all consumer groups as well as vehicle bus signals are fed to the artificial intelligence as input signals, whereby in the standard operating mode the total current of the consumer groups together with a label “error-free” and the vehicle bus signals are fed to the artificial intelligence as training data as input variables, whereby the In this way, unlabeled total streams of consumer groups and bus signals are fed to trained artificial intelligence as input variables in application mode.

In one embodiment, the artificial intelligence is also provided with total flows of the consumer groups together with the label “faulty” and the vehicle bus signals as training data in training mode, which represent error cases for individual or all consumers.

In an alternative embodiment, a binary classification method is used in the training mode, which only requires one class labeled as “error-free”.

In one embodiment, conclusions about anomalies and signs of aging of the consumer are drawn from the determined state of health.

The present invention enables early fault and anomaly detection of electrical components in a motor vehicle, so that preventative measures can be taken against failures and errors. In this way, full protection of the electrical consumers using semiconductor fuses, each of which has an integrated current measurement via a shunt, can be dispensed with, which saves costs since the total number of shunts to be used is reduced.

The solution according to the invention reduces component failures by deriving preventative measures (analogous to predictive maintenance). It is possible to provide the acquired and evaluated data on a vehicle interface and send it to a backend. In this way, field data and information about the fleet in the field can be obtained. Thanks to the solution according to the invention, fewer components are required than with “intelligent” semiconductor fuses. This results in cost savings through fewer electronic components per semiconductor fuse and through fewer data lines and/or connection points.

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

• 1 eine schematische Ansicht einer elektrischen Schaltung einer Verbrauchergruppe,
• 2 eine schematische Ansicht einer Schaltungsanordnung, umfassend mehrere Verbrauchergruppen gemäß 1 sowie eine künstliche Intelligenz in einem Trainingsfall, und
• 3 eine schematische Ansicht einer Schaltungsanordnung, umfassend mehrere Verbrauchergruppen gemäß 1 sowie eine künstliche Intelligenz in einem Anwendungsfall.

Show:

• 1 a schematic view of an electrical circuit of a consumer group,
• 2 a schematic view of a circuit arrangement, comprising several consumer groups according to 1 as well as artificial intelligence in a training case, and
• 3 a schematic view of a circuit arrangement, comprising several consumer groups according to 1 as well as artificial intelligence in one use case.

Corresponding parts are provided with the same reference numbers in all figures.

1 is a schematic view of an electrical circuit of a consumer group VG1, in particular for a motor vehicle. The consumer group VG1 has a plurality of consumers R_V1 to R_Vn, which are connected on the one hand to a ground potential GND and on the other hand via a respective semiconductor fuse M1 to Mn to a supply line L for an operating voltage V+. The supply line L is connected to the operating voltage V+ via a shunt R1. The semiconductor fuses M1 to Mn are designed, for example, as field effect transistors, in particular MOSFETs. Furthermore, a microcontroller µC is provided, which is configured to control the semiconductor fuse M1 to Mn via a respective control output C1 to Cn and to measure a voltage drop across the shunt R1 using two measuring inputs CS-, CS+.

The microcontroller µC measures the voltage applied to the shunt R1 and thereby calculates a total current I _G1 of the consumer group VG1 formed from the consumers R_V1 to R_Vn connected in parallel. In addition, the microcontroller µC can control individual semiconductor fuses M1 to Mn in the event of an error and thus specifically switch individual consumers R_V1 to R_Vn on and off. The microcontroller µC takes on a measuring, control and safety function and protects the cables and components.

2 is a schematic view of a circuit arrangement 1, comprising several consumer groups VG1 to VGn, which are as in 1 can be trained.

The circuit arrangement 1 further comprises an artificial intelligence Kl, in particular a neural network. For the training of the artificial intelligence Kl, selected vehicle bus signals BS and the total currents I _R1 to I _Rn of the consumer groups VG1 to VGn determined by the microcontrollers µC of the consumer groups VG1 to VGn are made available to it. The vehicle bus signals BS provide information about the overall condition of the vehicle. As a result, a normal operating point of the consumers R_V1 to R_Vn of the consumer group VG1 to VGn can be learned for each consumer group VG1 to VGn by operating the consumers R_V1 to R_Vn in a standard operating mode and the total current I _R1 to I _Rn of the consumer groups VG1 to VGn together with a label “error-free” and the vehicle bus signals BS are made available as training data to the artificial intelligence Kl.

In addition, in the training case, the artificial intelligence AI can be provided with total currents I _R1 to I _Rn of the consumer groups VG1 to VGn together with the label “faulty” and the vehicle bus signals BS as training data, which represent error cases of individual or all consumers R_V1 to R_Vn.

Alternatively, binary classification methods, which only use one class (e.g. the error-free one), can also be used (one-class learning).

Through machine learning, in particular monitored learning, the artificial intelligence AI is trained in such a way that in a consumer group VG1 to VGn a distinction can be made between normal operation and error cases at consumer level.

The result of training the artificial intelligence AI for each consumer R_V1 to R_Vn is a trained artificial intelligence KI_T with a health model. Each consumer can have their own health model or a common health model for several or all consumers.

3 is a schematic view of a circuit arrangement 1 in an application, comprising several consumer groups VG1 to VGn, which are as in 1 can be trained.

The circuit arrangement 1 further includes the trained artificial intelligence KI_T.

The trained artificial intelligence KI_T is provided with the vehicle bus signals BS and the total currents I _G1 to I _Gn of the consumer groups VG1 to VGn determined by the microcontrollers µC of the consumer groups VG1 to VGn.

In the application case, the trained artificial intelligence KI_T is provided with unlabeled total currents I _G1 to I _Gn of the consumer groups VG1 to VGn as well as bus signals BS, in particular selected bus signals BS. The trained artificial intelligence KI_T can use this information to draw conclusions about the health status GS of individual or all consumers R_V1 to R_Vn of individual or all consumer groups VG1 to VGn.

This allows conclusions to be drawn about anomalies and signs of aging.

Reference symbol list

1

Circuit arrangement

B.S

Bus signal

C1 to Cn

Control output

CS+

Measuring input

CS

Measuring input

GND

Ground potential

G.S

Health status

IG1 to IGn

Total current

IR1 to IRn

Total current

AI

artificial intelligence

KI_T

trained artificial intelligence

L

supply line

M1 to Mn

Semiconductor fuse

R_V1 to R_Vn

consumer

R1

shunt

VG1 to VGn

consumer group

V+

Operating voltage

µC

Microcontroller

QUOTES INCLUDED IN THE DESCRIPTION

This list of documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• EP 3098681 B1 [0003]

Claims (10)

Circuit arrangement (1) for a motor vehicle, comprising at least one consumer group (VG1 to VGn) and an artificial intelligence (KI, KI_T), which is configured to determine a normal operating point of the consumers (R_V1) for each consumer group (VG1 to VGn) in a training mode to R_Vn) by operating the consumers (R_V1 to R_Vn) in a standard operating mode, the artificial intelligence (AI, KI_T) trained in this way being further configured to operate in an application mode in a consumer group (VG1 to VGn). To distinguish between normal operation and fault cases at the consumer level, characterized in that each consumer group (VG1 to VGn) has a plurality of consumers (R_V1 to R_Vn) which have an operating voltage (V+) via a common supply line (L) and via a shunt (R1). _{) .} R1) is configured, whereby the artificial intelligence (KI, KI_T) can be supplied with the determined total currents (I _G , to I _Gn ) of all consumer groups (VG1 to VGn) as well as vehicle bus signals (BS) as input signals, in the standard operating mode Total current (I _R1 to I _Rn ) of the consumer groups (VG1 to VGn) together with a label “error-free” and the vehicle bus signals (BS) can be supplied as training data to the artificial intelligence (KI) as input variables, with the trained artificial intelligence (KI_T ) In the application mode, unlabeled total currents (I _G , to I _Gn ) of the consumer groups (VG1 to VGn) as well as bus signals (BS) can be supplied as input variables. Circuit arrangement (1) according to Claim 1 , characterized in that the consumers (R_V1 to R_Vn) are connected to the supply line (L) for the operating voltage (V+) via a respective semiconductor fuse (M1 to Mn). Circuit arrangement (1) according to Claim 2 , characterized in that the microcontroller (µC) is further configured to control the semiconductor fuse (M1 to Mn) via a respective control output (C1 to Cn). Circuit arrangement (1) according to Claim 1 , characterized in that the semiconductor fuses (M1 to Mn) are designed as field effect transistors, in particular MOSFET. Circuit arrangement (1) according to Claim 1 or 2 , characterized in that the artificial intelligence (AI, KI_T) is designed as a neural network. Motor vehicle, comprising a circuit arrangement (1) according to one of the preceding claims. Method for detecting a health status (GS) of a plurality of electrical consumers (R_V1 to R_Vn), comprising a circuit arrangement (1), comprising at least one consumer group (VG1 to VGn) and an artificial intelligence (Kl, KI_T), wherein in a training mode of The artificial intelligence (AI) learns a normal operating point for the consumers (R_V1 to R_Vn) for each consumer group (VG1 to VGn) by operating the consumers (R_V1 to R_Vn) in a standard operating mode, so that the trained artificial intelligence (KI_T ) in an application mode in a consumer group (VG1 to VGn) can distinguish between normal operation and error cases at consumer level, characterized in that each consumer group (VG1 to VGn) has a plurality of consumers (R_V1 to R_Vn) which have a common supply line (L ) and are connected to an operating voltage (V+) via a shunt (R1), each consumer group (VG1 to VGn) also having a microcontroller (µC) which measures a voltage drop across the shunt (R1) and a corresponding total current (I _G1 to I _Gn ) is calculated by the shunt (R1), whereby the determined total currents (I _G1 to I _Gn ) of all consumer groups (VG1 to VGn) as well as vehicle bus signals (BS) are fed to the artificial intelligence (Kl, KI_T) as input signals, whereby in the standard operating mode, the total current (I _R1 to I _Rn ) of the consumer groups (VG1 to VGn) together with a label “error-free” and the vehicle bus signals (BS) are supplied as training data to the artificial intelligence (AI) as input variables, where The artificial intelligence trained in this way (KI_T) is supplied with unlabeled total currents (I _G1 to I _Gn ) of the consumer groups (VG1 to VGn) and bus signals (BS) as input variables in the application mode. Procedure according to Claim 7 , characterized in that the artificial intelligence (AI) in training mode is also provided with total currents ( _IR1 to I _Rn ) of the consumer groups (VG1 to VGn) together with the label “faulty” and the vehicle bus signals (BS) as training data , which represent error cases of individual or all consumers (R_V1 to R_Vn). Procedure according to Claim 7 , characterized in that in training mode a binary classification method, which with Finally, a class labeled as “error-free” is used. Procedure according to one of the Claims 7 until 9 , characterized in that conclusions about anomalies and signs of aging of the consumers (R_V1 to R_Vn) are drawn from the determined state of health (GS).

Images