WO2020106201A1

WO2020106201A1 - Method, Computer Program, Control Unit for Detecting Faults in a Driver-Assistance System and Vehicle

Info

Publication number: WO2020106201A1
Application number: PCT/SE2019/051136
Authority: WO
Inventors: Paola MAGGINO
Original assignee: Scania Cv Ab
Priority date: 2018-11-23
Filing date: 2019-11-12
Publication date: 2020-05-28
Also published as: SE1851450A1

Abstract

Faults are detected in a driver-assistance system of a vehicle by obtaining sensor data (DS) in a processor (112). Sensor data (DS) are divided into data segments (d,), wherein each segment (di) is associated with a particular feature (f1, f2,..., fn). The processing involves predicting data segment content (d',) in an error-free operation of the driver-assistance system (120). Each of the data segments (di) is processed in a respective one of an artificial neural network models (ANN1, ANN2, ANNn) depending on the feature with which the respective data segment (d,) is associated to obtain a respective predicted data segment content (f1'(di), f2'(dj), fn'(dk)). For each feature (f1 , f2,..., fn), the predicted data segment content (f1(di), f2'(dj), fn'(dk)) is compared with corresponding data segment content (f1(dj), f2(dj), fn(dk)) divided from the obtained sensor data (DS) to derive a respective difference measure (e1, e2,..., en).

Description

Method, Computer Program, Control Unit for Detecting Faults in a Driver-Assistance System and Vehicle

TECHNICAL FIELD

The invention relates generally to the enhancement of advanced driver-assistance systems. In particular, the present invention concerns a method for detecting faults in a driver-assistance system of a vehicle, a control unit implementing this method and a vehicle containing the control unit. The invention also relates to a computer program product and a non-volatile data carrier.

BACKGROUND

Automatic detection and handling data concerning obstacles in a driver-assistance system is a highly complex task. Moreover, it is a fundamental that a driver-assistance system is robust and reliable. Consequently, any faults in the driver-assistance sys tem, as such, must be detected promptly and accurately. The prior art contains examples of such solutions of which some are listed below.

US 2008/0161989 describes a system on a moving object for monitoring components or subsystems, which includes sensors for obtaining a value of a measurable characteristic of the com ponent or subsystem and generating a signal indicative or repre sentative of the value, and a processor operatively connected to the sensors for receiving the signal from each sensor and ana lyzing the value of the measurable characteristic to determine that the component or subsystem has a fault condition, e.g., an actual or potential fault or failure. A communications unit is coupled to the processor and transmits a diagnostic or prognos tic message relating to the determination of the fault condition of the component or system to a remote site, upon direction or command by the processor. The processor may be part of a dia gnostics module and configured to recognize a predetermined fault condition, using for example pattern recognition technolo gies.

EP 2 790 165 discloses a method, an apparatus and a computer program product for quality determination in data acquisition. A signal processing apparatus and method for processing at least one detection signal are described. The apparatus contains a re ceiver configured to receive the at least one detection signal from at least one sensor, a model generator configured to gene rate a model of signal sequences based on previously received detection signals while taking into consideration a topology of the sensors and/or environmental condition in the vicinity of the at least one sensor, comparing means configured to compare the at least one detection signal with the model of signal se quences generated by the model generator, and a prediction means configured to predict quality of the at least one detection signal based on a comparison of the at least one detection sig nal with the model generated by the model generator.

US 2016/0350194 shows a health management solution, and more particularly a method and system for artificial intelligence based diagnostic and prognostic health management of host systems. In an embodiment, the system includes a memory to store instructions, and a neural network controller coupled to the memory. The neural network controller is configured by the inst ructions to monitor a plurality of unique patterns generated in real-time. The plurality of system parameters is indicative of a system-level performance of the host system. The neural net work controller is configured by the instructions to compare the plurality of unique patterns with a plurality of predetermined pat terns corresponding to the plurality of system parameters to de- tect potential anomalies in the host system and one or more subsystems of the plurality of subsystems, where the one or mo re subsystems are responsible for contributing to the potential anomalies in the host system.

Thus, different solutions are known for supervising and diagno- sing complex technical arrangements, such as driver-assistance systems, inter alia by using artificial intelligence based diagnos tic mechanisms. However, there is room for improving these su pervision systems, for example in terms of reliability and speedi- ness.

SUMMARY

One object of the present invention is therefore to offer an en hanced solution for identifying faults in a driver-assistance sys tem. According to one aspect of the invention, this object is achieved by a method performed in a control unit to detect faults in a dri ver-assistance system of a vehicle. The method involves obtai ning sensor data in at least one processor. The sensor data des cribe spatio-temporal relationships between the vehicle and obs- tacles located in a sector relative to the vehicle, for example in a cone in front of the vehicle. Further, the method involves divi ding the sensor data into data segments in the at least one pro cessor. Each of the data segments is associated with a particu lar feature from a set of features containing at least two featu- res. Here, each feature has a probability distribution for detec ting obstacles within the sector. A respective artificial neural network model is configured to process the data segments. The processing involves predicting data segment content in an error- free operation of the driver-assistance system. Specifically, each of the data segments is processed in a respective one of the artificial neural network models depending on the feature with which the respective data segment is associated. As a re sult, respective predicted data segment content is obtained. For each of the features, the predicted data segment content is com- pared with corresponding data segment content divided off from the obtained sensor data to derive a respective difference mea sure. Additionally, the method involves generating a fault diag nosis report based on the difference measures. Provided , of course, that the artificial neural network models ha ve been trained properly, this method is advantageous because it enables discovery of rare and critical error scenarios very promptly, i.e. with exceptionally short latency from occurrence of an error.

According to one embodiment of this aspect of the invention, the features reflect a respective longitudinal relative position, a res pective latitudinal relative position, a respective longitudinal re lative velocity and/or a respective latitudinal relative velocity of the obstacles, if any, that are located in the sector. This means that the vehicle surroundings are modeled in a relevant manner. Preferably, the sector contains at least two zones, and the features further reflect in which of the at least two zones each of the obstacles is located. According to another embodiment of this aspect of the invention , the method involves generating the artificial neural network mo dels by, for each of the at least two zones, training a basic neu ral-network model with training data in the form of segmented sensor data representing error-free operation of the driver-assis- tance system. The training involves evaluating a capability for the basic neural-network model to predict a data segment con tent based on an input data segment. One or more parameters in the basic neural-network model are then adjusted until the capability for the basic neural-network model to predict the data segment content lies within an accuracy threshold. Thereby, the artificial neural network models can be made highly reliable and accurate predictors.

According to yet another embodiment of this aspect of the in vention , the method further involves producing the segmented sensor data by: obtaining raw sensor data from the driver-assis tance system, which raw sensor data reflect a sequence of events following in succession after one another in time; and preprocessing the raw sensor data by interpolating data samples in the raw sensor data so as to fill out any temporal gaps in the raw sensor data with one or more synthetic data samples whose content is based on a respective content of temporally neigh boring data samples in the raw sensor data. Thereby, the trai ning avoids being based on non-representative input, and can thus be made very efficient.

According to still another embodiment of this aspect of the in vention, the sensor data contains a radar signal (radar = radio detection and ranging), a lidar signal (lidar = light detection and ranging), a sonar signal (sonar = sound navigation and ranging) and/or a video signal. Namely, this enables a relevant basis for registering and keeping track of any obstacles in vicinity of the vehicle.

According to a further aspect of the invention the object is ac hieved by a computer program containing instructions which, when executed on at least one processor, cause the at least one processor to carry out the above-described method.

According to another aspect of the invention, the object is achie ved by a non-volatile data carrier containing such a computer program. According to yet another aspect of the invention, the above ob ject is achieved by a control unit adapted to be included in a ve hicle for detecting faults in a driver-assistance system of the ve hicle. The control unit contains at least one processor configu red to obtain sensor data describing spatio-temporal relation- ships between the vehicle and obstacles located in a sector re lative to the vehicle. The at least one processor is further confi gured to divide the sensor data into data segments, and asso ciate each data segment with a particular feature from a set of features containing at least two features. Each feature has a probability distribution for detecting obstacles within the sector. A respective artificial neural network model implemented in the at least one processor is configured to process the data seg ments. This processing involves predicting data segment con- tent in an error-free operation of the driver-assistance system. Each of the data segments is processed in a respective one of the artificial neural network models depending on the feature with which the respective data segment is associated. As a re- suit, a respective predicted data segment content is obtained. For each features, the at least one processor is further configu red compare the predicted data segment content with correspon ding data segment content divided from the obtained sensor da ta to derive a respective difference measure. Based on the diffe- rence measures, in turn, the at least one processor is configured to generate a fault diagnosis report.

The advantages of this control unit, as well as the preferred em bodiments thereof, are apparent from the discussion above with reference to the method to detect faults in a driver-assistance system of a vehicle.

According to still another aspect of the invention, the object is achieved by a vehicle including the proposed control unit for de tecting faults in a driver-assistance system of the vehicle.

Further advantages, beneficial features and applications of the present invention will be apparent from the following description and the dependent claims.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention is now to be explained more closely by means of preferred embodiments, which are disclosed as examples, and with reference to the attached drawings.

Figure 1 schematically depicts a vehicle in which an embo diment of the invention is implemented;

Figure 2 shows a block diagram over a processor accor ding to one embodiment of the invention;

Figure 3 shows a block diagram over a control unit accor ding to one embodiment of the invention; Figure 4 illustrates how a monitored sector in vicinity of a vehicle can be divided into zones according to one embodiment of the invention;

Figure 5 illustrates, schematically, how an artificial neural network may be trained according to one embodi ment of the invention;

Figure 6 illustrates, by means of a flow diagram, the gene ral method according to the invention.

DETAILED DESCRIPTION

Figure 1 schematically depicts a vehicle V in which an embodi ment of the invention is implemented in an onboard control unit 1 10. Figure 2 shows a block diagram over a processor 1 12 ac cording to one embodiment of the invention. For reasons of cla rity, Figure 2 shows only one processor 1 12. However, accor- ding to the invention, the processing functionality may equally well be distributed between two or more processors. Figure 3 shows a block diagram over the control unit 1 10 according to one embodiment of the invention.

The control unit 1 10 may contain a processing unit 1 12 with pro- cessing means including at least one processor, such as one or more general-purpose processors. Further, the processing unit 1 12 is preferably communicatively connected to a data carrier 1 14 in the form computer-readable storage medium, such as a Random Access Memory (RAM), a Flash memory, or the like. The data carrier 1 14 contains computer-executable instructions, i.e. a computer program 1 15, for causing the processing unit 1 12 of the control unit 1 10 to perform in accordance with the embodiments of the invention as described herein, when the computer-executable instructions are executed on the at least one processor of the processing unit 1 12.

The control unit 1 10 is preferably implemented in one or more so-called ECUs (Electronic Control Units) in the vehicle V, which ECUs exchange data and instructions with other units, sensors and actuators in the vehicle V via a CAN (Controller Area Net work) bus, or analogous internal communications network.

The control unit 1 10 is adapted to detect faults in a driver-assis- tance system 120 of the vehicle V. To this aim, the control unit 1 10 contains at least one processor 1 12, which is configured to obtain sensor data DS describing spatio-temporal relationships between the vehicle V and any obstacles, e.g. another moving vehicle OB1 and a stationary barrel OB2, that are located in a sector S relative to the vehicle V. Thus, the obstacles OB1 and OB2 may be represented by other vehicles travelling in the sa me direction as the vehicle V or in any other direction, pedes trians and/or stationary objects on or beside the road.

More precisely, the at least one processor 1 12 is configured to divide the sensor data DS into data segments di, and associate each of these data segments d, with a particular feature f i , f2, ... , f_n from a set of features {f_m}. The set of features {f_m} contains at least two features each of which has a probability distribution for detecting obstacles within the sector S. A respective artificial neural network model ANN1 , ANN2, ... , ANNn is configured to process the data segments di. The processing involves predic ting data segment content d’, in an error-free operation of the driver-assistance system 120. The at least one processor 1 12 is configured to process each of the data segments d, in a respec- tive one of the artificial neural network models ANN1 , ANN2, ... , ANNn depending on the feature of said features with which the respective data segment d, is associated to obtain a respective predicted data segment content fi’(di), f2’(d_j) and f_n’(d_k) respectively. This means that the at least one processor 1 12 is configured to the content of data segments d, so that each segment is processed in an appropriate artificial neural network model.

For each of the features f i , f 2 , ... , fn, the at least one processor 1 12 is configured to compare the predicted data segment content fi’(di), f2’(d_j) and f_n’(d_k) with corresponding data segment content fi (di), f2(d_j) and f_n(d_k) respectively divided off from the obtained sensor data DS to derive a respective difference measure ei , ez, ... , e_n. Then, based on the difference measures ei , e2, e_n, the at least one processor 1 12 is configured to ge nerate a fault diagnosis report R, for instance in the form of one or more alarm signals or an indication of error-free operation de pending on the statuses of the difference measures ei , ei, ... ,

6 n Preferably, the features f i , f 2 , ... , f_n reflect: a respective longi tudinal relative position, a respective latitudinal relative position, a respective longitudinal relative velocity and/or a respective la titudinal relative velocity of the obstacles OB1 and OB2 that are located in the sector S to describe the spatio-temporal relation- ships between the vehicle V and the obstacles OB 1 and OB2. The sector S, in turn, may be embodied as a cone, for instance as illustrated in Figure 1 , in which one or more sensors (not shown) are configured to register the sensor data DS, via for ex ample radar signals, lidar signals, sonar signals and/or video signals.

According to one embodiment of the invention , the sector S con tains, i.e. is subdivided into, two or more zones Z01 , Z02, Z03, Z04, Z05, Z06, Z07, Z08, Z09, Z10, Z1 1 , Z12, Z13 and Z14, which zones represent different areas in front of the vehicle V as illustrated in Figure 4. In such a case, the features f 1 , f 2 , ... , fn further reflect in which of the at least two zones that each of the obstacles OB1 and OB2 respectively is located. In this embodi ment of the invention , the artificial neural network models ANN 1 , ANN2, ... , ANNn have preferably been generated as follows. Referring further to Figure 5, for each of the at least two zones

Z01 , Z02, Z03, Z04, Z05, Z06, Z07, Z08, Z09, Z10, Z1 1 , Z12, Z13 and Z14, a basic neural-network model ANN is trained with training data in the form of segmented sensor data that repre sent error-free operation of the driver-assistance system 120. In Figure 5, this training data is exemplified by segmented sensor data fi(di), fi(di+i) and fi (d ,+2) respectively. The training involves evaluating a capability for the basic neural-network model ANN to predict a data segment f’(di) based on an input data segment fi (d i) . For repeated inputs of training data, i.e. error-free seg mented sensor data fi (d ,) , fi (d ,+i ) and fi (d ,+2) , the output from the neural-network model ANN is compared to the input, and one or more parameters (p_m) in the basic neural-network model ANN are adjusted until the capability for the thus adjusted va- riant of the basic neural-network model ANN to predict the data segment content f’(di) lies within an accuracy threshold e_tn., i.e. an acceptable quality measure for the prediction.

It is preferable if the segmented sensor data fi (d ,) , fi(di+i), fi(d_i+2) being used for the training have been preprocessed to enhance the data quality. For example, the enhancement may involve obtaining raw sensor data DS from the driver-assistance system 120, which raw sensor data DS reflect a sequence of events following in succession after one another in time. The raw sensor data DS is then preprocessed by interpolating data samples in the raw sensor data so as to fill out any temporal gaps in the raw sensor data with one or more synthetic data samples, where a content synthetic data samples is based on respective contents of temporally neighboring data samples in the raw sensor data. In other words, the raw sensor data DS is supplemented with suitable artificial data samples to avoid fee ding non-representative data into the basic neural-network model ANN, and its gradually adjusted variants, during the trai ning process.

In order to sum up, and with reference to the flow diagram in Figure 6, we will now describe the general method according to the invention for detecting faults in a driver-assistance system of a vehicle.

In a first step 610, sensor data are obtained, which describe spatio-temporal relationships between a vehicle and any obstac- les located in a sector relative to the vehicle.

A step 620, thereafter checks if a sufficient amount of sensor data have been received to form a data segment, and if so a step 630 follows. Otherwise, the procedure loops back and stays in step 620 until enough sensor data have been received to fill up a data segment. In step 630, the data segment formed in the previous step is associated with a particular feature from a set of features containing at least two features. Each of these featu res, in turn, has a probability distribution for detecting obstacles within the sector.

Subsequently, in a step 640, the data segment is processed in a particular artificial neural network model depending on the featu re of said features with which the data segment is associated to. The processing involves predicting data segment content in an error-free operation of the driver-assistance system. Thus, the processing results in a predicted data segment content being obtained.

A subsequent step 650, compares, for the features in question, the predicted data segment content with corresponding data segment content divided from the obtained sensor data to derive a difference measure between predicted data segment and the data segment content obtained from the sensor data.

Then, in a step 660, a fault diagnosis report is generated based on the difference measure. The fault diagnosis report may be generated gradually as new data segments are divided off from the obtained sensor data, and if the difference measure exceeds a threshold level, an alarm signal may be generated.

After step 660, the procedure loops back to step 610.

All of the process steps, as well as any sub-sequence of steps, described with reference to Figure 6 above may be controlled by means of at least one programmed processor. Moreover, although the embodiments of the invention described above with reference to the drawings comprise processor and processes performed in at least one processor, the invention thus also extends to com puter programs, particularly computer programs on or in a carrier, adapted for putting the invention into practice. The program may be in the form of source code, object code, a code intermediate source and object code such as in partially compiled form, or in any other form suitable for use in the implementation of the pro cess according to the invention. The program may either be a part of an operating system, or be a separate application. The carrier may be any entity or device capable of carrying the program. For example, the carrier may comprise a storage medium, such as a Flash memory, a ROM (Read Only Memory), for example a DVD (Digital Video/Versatile Disk), a CD (Compact Disc) or a semi conductor ROM, an EPROM (Erasable Programmable Read-Only Memory), an EEPROM (Electrically Erasable Programmable Read-Only Memory), or a magnetic recording medium, for ex ample a floppy disc or hard disc. Further, the carrier may be a transmissible carrier such as an electrical or optical signal which may be conveyed via electrical or optical cable or by radio or by other means. When the program is embodied in a signal which may be conveyed directly by a cable or other device or means, the carrier may be constituted by such cable or device or means. Alternatively, the carrier may be an integrated circuit in which the program is embedded, the integrated circuit being adapted for performing, or for use in the performance of, the relevant proces ses.

The term“comprises/comprising” when used in this specification is taken to specify the presence of stated features, integers, steps or components. However, the term does not preclude the presen- ce or addition of one or more additional features, integers, steps or components or groups thereof.

The invention is not restricted to the described embodiments in the figures, but may be varied freely within the scope of the claims.

Claims

1 . A method performed in a control unit (1 10) to detect faults in a driver-assistance system (120) of a vehicle (V), the method comprising:

obtaining, in at least one processor (1 12), sensor data

(DS) describing spatio-temporal relationships between the ve hicle (V) and obstacles (OB1 , OB2) located in a sector (S) rela tive to the vehicle (V),

characterized by, in the at least one processor (1 12):

dividing the sensor data (DS) into data segments (d,);

associating each of said data segments (d,) with a particu lar feature (f ₁ , f ₂ , - - - , fn) from a set of features ({f }) containing at least two features each of which has a probability distribution for detecting obstacles within the sector (S), a respective artificial neural network model (ANN 1 , ANN2, ANNn) being configured to process the data segments (d ), the processing involving predic ting data segment content (d’,) in an error-free operation of the driver-assistance system (120);

processing each of the data segments (d,) in a respective one of the artificial neural network models (ANN 1 , ANN2, ANNn) depending on the feature of said features with which the respec tive data segment (d,) is associated to obtain a respective pre dicted data segment content (fi’(di), f₂’(d_j), f_n’(d_k));

comparing , for each of said features (f i , f ₂ , .. - , fn), the pre- dieted data segment content (fi’(di), f₂’(d_j) , f_n’(d_k)) with corres ponding data segment content (f i (d ,) , f₂(d_j) , f_n(d_k)) divided from the obtained sensor data (DS) to derive a respective difference measure (ei , e₂, ... , e_n); and

generating a fault diagnosis report (R) based on said dif- ference measures (ei , e₂, ... , e_n).

2. The method according to claim 1 , wherein said features (fi , f ₂ , ... , f_n) reflect at least one of: a respective longitudinal relative position , a respective latitudinal relative position, a respective longitudinal relative velocity and a respective latitudinal relative velocity of the obstacles (OB1 , OB2) located in the sector (S).

3. The method according to claim 2, wherein the sector (S) comprises at least two zones (Z01 , Z02, Z03, Z04, Z05, Z06, Z07, Z08, Z09, Z10, Z1 1 , Z12, Z13, Z14), and said features (fi , f ₂ , ... , fn) further reflect in which of said at least two zones each of the obstacles (OB1 , OB2) is located.

4. The method according to claim 3, comprising:

generating the artificial neural network models (ANN1 ,

ANN2, ANNn) by, for each of the at least two zones, training a basic neural-network model (ANN) with training data in the form of segmented sensor data (f(d,) , f(d,_+i ), f(di₊2), ... ) representing error-free operation of the driver-assistance system (120), the training involving evaluating a capability for the basic neural- network model (ANN) to predict a data segment content (f’(d ,)) based on an input data segment (f(d ,)) , and adjusting one or mo- re parameters (p_m) in the basic neural-network model (ANN) un til the capability for the basic neural-network model (ANN) to predict the data segment content (f’(d ,)) lies within an accuracy threshold (etn).

5. The method according to claim 4, further comprising

producing the segmented sensor data (fi(d,), fi (d,_+i ) , fi(di+₂), ... ) by:

obtaining raw sensor data (DS) from the driver-assis tance system (120), which raw sensor data (DS) reflect a sequence of events following in succession after one an- other in time; and

preprocessing the raw sensor data (DS) by interpola ting data samples in the raw sensor data so as to fill out any temporal gaps in the raw sensor data with one or more synthetic data samples whose content is based on a res- pective content of temporally neighboring data samples in the raw sensor data.

6. The method according to any one of the preceding claims, wherein the sensor data (DS) comprises at least one of: a radar signal, a lidar signal, a sonar signal and a video signal.

7. A computer program product (1 15) loadable into a non-vo latile data carrier (1 14) communicatively connected to at least one processing unit (1 12), the computer program product (1 15) comprising software for executing the method according any of the claims 1 to 6 when the computer program product (1 15) is run on the at least one processing unit (1 12).

8. A non-volatile data carrier (1 14) containing the computer program product (1 15) of the claim 7.

9. A control unit (1 10) adapted to be comprised in a vehicle

(V) for detecting faults in a driver-assistance system (120) of the vehicle (V), the control unit (1 10) comprising at least one pro cessor (1 12) configured to obtain sensor data (DS) describing spatio-temporal relationships between the vehicle (V) and obs- tacles (OB1 , OB2) located in a sector (S) relative to the vehicle (V), characterized in that the at least one processor (1 12) is further configured to:

divide the sensor data (DS) into data segments (d,);

associate each of said data segments (d,) with a particular feature (f i , f 2 , ... , f_n) from a set of features ({fm}) containing at least two features each of which has a probability distribution for detecting obstacles within the sector (S), a respective artificial neural network model (ANN 1 , ANN2, ANNn) being configured to process the data segments (di), the processing involving predic- ting data segment content (d’i) in an error-free operation of the driver-assistance system (120);

process each of the data segments (d,) in a respective one of the artificial neural network models (ANN1 , ANN2, ANNn) de pending on the feature of said features with which the respective data segment (d,) is associated to obtain a respective predicted data segment content (fi’(di), f2’(dj), f_n’(d_k));

compare, for each of said features (f 1 , f 2 , ... , fn) , the predic ted data segment content (fi’(di), f2’(dj), f_n’(d_k)) with correspon- ding data segment content (fi (di), f2(d_j), f_n(d_k)) divided from the obtained sensor data (DS) to derive a respective difference measure (ei , ez, ... , e_n); and

generate a fault diagnosis report (R) based on said dif- ference measures (ei , ei , e_n).

10. The control unit (1 10) according to claim 9, wherein said features (fi , f 2 , ... , f_n) reflect at least one of: a respective longi tudinal relative position, a respective latitudinal relative position, a respective longitudinal relative velocity and a respective latitu- dinal relative velocity of the obstacles (OB 1 , OB2) located in the sector (S).

1 1 . The control unit (1 10) according to claim 10, wherein the sector (S) comprises at least two zones (Z01 , Z02, Z03, Z04, Z05, Z06, Z07, Z08, Z09, Z10, Z1 1 , Z12, Z13, Z14), and said features (fi , f .. fn) further reflect in which of said at least two zones each of the obstacles (OB1 , OB2) is located.

12. The control unit (1 10) according to claim 1 1 , wherein the artificial neural network models (ANN 1 , ANN2, ANNn) have been generated by, for each of the at least two zones, training a basic neural-network model (ANN) with training data in the form of segmented sensor data (f 1 (d ,) , fi (d,_+i ) , fi (di₊2), ... ) representing error-free operation of the driver-assistance system (120), the training involving evaluating a capability for the basic neural- network model (ANN) to predict a data segment (f’(d ,)) based on an input data segment (fi (d,)), and adjusting one or more para meters (p_m) in the basic neural-network model (ANN) until the capability for the basic neural-network model (ANN) to predict the data segment content lies (f’(di)) within an accuracy thres hold (eth).

13. The control unit (1 10) according to claim 12, wherein the segmented sensor data (fi (d,), fi (d,_+i ) , fi(di₊2), ... ) have been pro duced by: obtaining raw sensor data (DS) from the driver-assis tance system (120), which raw sensor data (DS) reflect a sequence of events following in succession after one an other in time; and

preprocessing the raw sensor data (DS) by interpola ting data samples in the raw sensor data so as to fill out any temporal gaps in the raw sensor data with one or more synthetic data samples whose content is based on a res pective content of temporally neighboring data samples in the raw sensor data.

14. The control unit (1 10) according to any one of the claims 9 to 13, wherein the sensor data (DS) comprises at least one of: a radar signal, a lidar signal, a sonar signal and a video signal.

15. A vehicle (V) comprising the control unit (1 10) according to any one of claims 9 to 14 for detecting faults in a driver-assis tance system (120) of the vehicle (V).