DE102022209239A1 - Method for diagnosing a device based on artificial intelligence - Google Patents

Abstract

The invention relates to a method for diagnosing a device, the method having the following steps: - Receiving data that represents a state of the device, - Assigning the device to one of the following three classes on the basis of the received data using a classification algorithm:* Class "1": Device is OK,* Class "2": Device has an error,* Class "3": It cannot be decided whether the device has an error, - whereby in the case in which the Class “3” device is assigned, further data is subsequently received that represents whether the device is OK or whether the device has an error, and the classification algorithm is then trained with training data that includes the data and the other data becomes.

Description

State of the art

According to the prior art, devices, such as motor vehicle components, which may be defective and potentially represent a warranty claim, are dismantled in workshops and sent to the manufacturer. Subsequent investigations by the manufacturer will determine whether the part is actually defective or not. The procedure then proceeds accordingly, for example a decision is made as to whether the case is covered by a guarantee.

Disclosure of Invention

The present invention is based on the inventors' observation that the procedure described at the outset is complex. As a rule, the manufacturer receives a large number of devices that are to be examined within a specified period of time. In addition, as a rule, many examination results are of the same type, so that despite a great deal of effort, only a few new findings can be gained that go beyond the error of the individual device.

The basic idea of the present invention is therefore to diagnose the device using a method that is based in particular on data that is already available, instead of physically shipping the device in question. This significantly simplifies the diagnosis of the devices to be examined.

• - Empfangen von Daten, die einen Zustand der Vorrichtung repräsentieren,
• - Zuordnen der Vorrichtung auf Basis der empfangenen Daten durch einen Klassifikationsalgorithmus zu einer der drei folgenden Klassen:
  • * Klasse „1“: Vorrichtung ist in Ordnung,
  • * Klasse „2“: Vorrichtung weist einen Fehler auf,
  • * Klasse „3“: Es kann nicht entschieden werden, ob die Vorrichtung einen Fehler aufweist.

The method according to the invention has the following steps:

• - receiving data representing a state of the device,
• - Allocation of the device based on the received data by a classification algorithm to one of the following three classes:
  • * Class "1": Device is OK,
  • * Class “2”: device has an error,
  • * Class "3": It cannot be decided whether the device has a fault.

It is provided according to the invention that in the case in which the device is assigned to class "3", further data are subsequently received which represent whether the device is in order or whether the device has an error, with the classification algorithm subsequently using Training data, which include the data and the further data, is trained.

Optionally, class "2" can have multiple subclasses associated with different types of errors.

Instead of or in a further development of the assignment to class "2", assignment to one of the subclasses of class "2" can then take place.

Furthermore, it can then be provided that the further data also represent the type of error in the event of an error.

• - Empfangen von Daten, die einen Zustand der Vorrichtung repräsentieren,
• - Zuordnen der Vorrichtung auf Basis der empfangenen Daten durch einen Klassifikationsalgorithmus zu einer der folgenden Klassen:
  • * Klasse „1“: Vorrichtung ist in Ordnung,
  • * Unterklassen der Klasse „2“: Vorrichtung weist einen der betreffenden Unterklasse zugeordneten Fehler auf,
  • * Klasse „3“: Es kann nicht entschieden werden, ob die Vorrichtung einen Fehler aufweist,

wobei in dem Fall, in dem die Vorrichtung der Klasse „3“ zugeordnet wird, nachfolgend weitere Daten empfangen werden, die repräsentieren, ob die Vorrichtung in Ordnung ist oder welchen Fehler die Vorrichtung aufweist, und wobei nachfolgend der Klassifikationsalgorithmus mit Trainingsdaten, die die Daten und die weiteren Daten umfassen, trainiert wird.A further development therefore relates to a method for diagnosing a device, with the further developed method having the following steps:

• - receiving data representing a state of the device,
• - Allocation of the device based on the received data by a classification algorithm to one of the following classes:
  • * Class "1": Device is OK,
  • * Subclasses of class “2”: device has an error associated with the relevant subclass,
  • * Class "3": It cannot be decided whether the device has a fault,

where in the case in which the device is assigned to class "3", further data is subsequently received that represents whether the device is OK or what error the device has, and where the classification algorithm is then used with training data that contains the data and the further data comprise, is trained.

It can be provided that output data is subsequently output, which represents the class and possibly the subclass to which the device was assigned by the classification algorithm.

The data can be, for example, the following: error memory entries, for example in an error memory of an engine control unit of a motor vehicle, information memory entries, information about a vehicle condition, e.g. a voltage of an electric battery of the vehicle, information about the presence or absence of certain components in a Vehicle, for example special equipment, a specification of a software version and/or a hardware version of the device in question.

Additionally or alternatively, the data can also represent information about whether a symptom or several symptoms with regard to the device apply in each case or whether they do not in each case. The symptom/symptoms can be objectively observable malfunctions of the device, for example in the case of the device as an internal combustion engine, knocking noises or rough engine operation of the internal combustion engine. On the one hand, such symptoms can be observed humanely, for example, and the relevant data can be generated manually. On the other hand, the observation of the symptoms and/or the generation of the relevant data can also be fully or partially automated.

In particular, it can be data that can be read out of a vehicle using a diagnostic tester or is read out as part of the method.

In particular, the data can be read from a data memory of the device or from a target system comprising the device.

The additional data can in particular represent the result of an examination of the device, for example an examination of the device in a laboratory belonging to the manufacturer. The further data can then be received on the basis of manual input.

The classification algorithm can be a machine learning algorithm, for example based on the use of a neural network.

The method according to the invention is not limited to devices of one type; it can also relate to devices that belong to an ensemble of a large number of types. An individual classification algorithm is preferably used or further trained for each variety. When processing an individual device, the classification algorithm provided for the type of device is then used (mixture of experts).

• - Vorzuordnung der Vorrichtung auf Basis der empfangenen Daten zu der Klasse „1“ oder zu der Klasse „2“ bzw. zu einer Unterklasse der Klasse „2“,
• - Bestimmung eines Konfidenzmaßes der erfolgten Vorzuordnung,
• - Zuordnung der Vorrichtung gemäß der Vorzuordnung, falls das Konfidenzmaß einen vorgegebenen Schwellwert überschreitet,
• - Zuordnung der Vorrichtung zu der Klasse „3“ falls das Konfidenzmaß den vorgegebenen Schwellwert unterschreitet.

For example, it can be provided that the classification algorithm has the following steps:

• - pre-assignment of the device based on the received data to class "1" or to class "2" or to a subclass of class "2",
• - Determination of a confidence level of the pre-allocation that has taken place,
• - Assignment of the device according to the pre-assignment if the confidence level exceeds a predetermined threshold value,
• - Assignment of the device to class "3" if the confidence level falls below the specified threshold value.

This development has the advantage that the degree of security with which a device is assigned to classes "1" and "2", i.e. with which it is decided whether the device is rated as error-free or defective and with which, if necessary, a Error type is assigned, can be selected on the specification of the threshold.

The method according to the invention can be implemented on a computer. Accordingly, the invention also includes a computer program which includes instructions which, when executed by a computer, cause the computer to carry out the method according to the invention, and a computer-readable data carrier on which this computer program is stored. Finally, the method according to the invention also includes a computer which includes such a computer-readable data carrier and an evaluation and control unit which includes such a computer-readable data carrier and also means for carrying out the method steps according to the invention.

In the drawing show the 1 and 2 an embodiment of the method according to the invention based on diagrams.

1 shows schematically an exemplary embodiment of the method according to the invention using a flowchart.

Based on the fact that a device, for example a component of a motor vehicle such as a display or an auxiliary heater, is considered potentially faulty in a workshop (method step V1), the device-related data from a fault memory of the device or a target system comprising the device, for example of the motor vehicle, read out and received by a computer (method step V2), which is located, for example, in the sphere of the manufacturer and/or is attributable to a cloud.

However, the motor vehicle can also communicate with a cloud, for example, which can be assigned to the sphere of a manufacturer, while the vehicle is “in the field”, for example at any location, in particular outside of a workshop. This communication can include, for example, the transmission of data relating to the device, for example diagnostic and/or vehicle data, in particular the data already mentioned above, which represent a state of the device. Such communication can, for example, take place continuously and/or at certain intervals, i.e. in particular “live”, so to speak.

It is also possible, in particular later, for example in a workshop, to operate a computer, for example a tablet computer or a smartphone, which is set up, for example using a corresponding computer program, in particular an app, the vehicle and the device mentioned above to receive relevant data, for example from the cloud, which is attributable to the sphere of a manufacturer, for example.

In order to make this possible, the recording, for example by means of digital photography and image recognition, of an identification number of the vehicle and/or a part number of the device and their transmission to the cloud can be provided.

• - Zuordnen der Vorrichtung auf Basis der empfangenen Daten durch einen Klassifikationsalgorithmus zu einer der drei folgenden Klassen:
  • * Klasse „1“: Vorrichtung ist in Ordnung,
  • * Klasse „2“: Vorrichtung weist einen Fehler auf,
  • * Klasse „3“: Es kann nicht entschieden werden, ob die Vorrichtung einen Fehler aufweist,
  unmittelbar von diesem Computer durchgeführt werden.

Insofar as the data mentioned above, which represent a state of the device, were then received by the computer, for example the tablet computer or the smartphone, in particular in the workshop, the method steps can:

• - Allocation of the device based on the received data by a classification algorithm to one of the following three classes:
  • * Class "1": Device is OK,
  • * Class “2”: device has an error,
  • * Class "3": It cannot be decided whether the device has a fault,
  be performed directly from this computer.

Provision can be made for the computer to access the current date, which is made available, for example, by a clock that is included in the computer's hardware or is available in the form of software.

It can be envisaged that the method step envisaged below in the case where the device is assigned to class "3" and which consists in receiving further data representing whether the device is in order or whether the device has an error and that the classification algorithm is trained with training data comprising the data and the further data is executed on a computer other than the last-mentioned computer. For example, the other computer may be located at a location that is not in a workshop but is in a sphere of the manufacturer of the device.

Alternatively, however, this method step can in principle also be carried out on the first-mentioned computer, which is, for example, a tablet computer or a smartphone.

In the example, it is provided that, based on this data, a classification algorithm makes a pre-allocation (method step V3) that class "1" device, which corresponds to a fault-free part, or class "2", which corresponds to a faulty part, and possibly assigned to a subclass of class "2" corresponding to one of the error types assigned to it. In addition, in this method step, the algorithm calculates a confidence measure φ, which is higher the more unequivocally or reliably the pre-allocation is evaluated by the classification algorithm.

In the example, it is subsequently provided that the confidence measure φ is compared with a predefined threshold value (method step V4).

The device is then assigned according to the pre-assignment if the confidence measure φ exceeds a predefined threshold value. Since the pre-assignment is considered reliable based on the high confidence measure φ, the pre-assignment can be retained (method step V4a).

Alternatively, or if the confidence measure ϕ falls below the specified threshold value, provision is made for the device to be assigned to class “3”, which expresses the fact that it is not possible to decide with sufficient certainty whether the device has an error (method step V4b). So the pre-mapping is not retained. How safe is safe enough can be defined quantitatively by specifying the threshold value.

In the case in which the device is assigned to class "1", the method can conclude with the output of the information and/or the transmission of the information to the workshop that the device is considered error-free by the manufacturer (method step V5) .

In the case in which the device is assigned to class "2", the process can conclude with the output of the information and/or the transmission of the information to the workshop that the device is considered defective by the manufacturer (process step V5) .

In the case where the device is classified as class “3”, in this example provided that the part is physically brought into the sphere of the manufacturer and is examined there individually with the result of whether the part is actually free of defects or whether it is defective, and what type of error is present in this case (step V6). Correspondingly, further data are generated that represent this examination result. Provision is then made for the classification algorithm to be trained with training data that includes this data and this additional data (method step V7).

2 shows the embodiment in a larger time frame. At the beginning of this time frame, at time T ₀ , a certain type of device is operated in the field for the first time. At this point in time, the classification algorithm in the example has not yet been trained. Although he pre-assigns a device to class "1" or "2" (corresponding to devices that are presumably error-free or are presumably defective) on the basis of the data received, the associated confidence measure is the value that corresponds to the greatest possible uncertainty, e.g value zero. Accordingly, at this point, all devices are classified as Class “3”, corresponding to total ignorance of whether the parts are defective or not.

These devices are examined in the manufacturer's laboratory and it is determined whether and, if so, to what extent errors are present, and further data representing this knowledge are obtained.

The data and the further data now represent labeled training data, since the data represent a state of the device and the further data represent the information as to whether and, if so, to what extent errors are present in a device.

The classification algorithm is then trained with this training data at time T ₁ . For this purpose, known algorithms of supervised learning can be used.

Again subsequently, at time T ₂ , the classification algorithm is already able to correctly classify a certain proportion of the devices to be diagnosed as faulty or fault-free, corresponding to classes “1” and “2”. Only in the remaining part of the devices to be diagnosed is the confidence measure ϕ so low or the uncertainty so high that a more detailed examination of the device has to be carried out as described above, corresponding to class "3".

This examination takes place at time T ₃ and at this time the classification algorithm can also be trained further with the training data newly obtained in this way, with the result that at the subsequent time T ₄ the proportion of devices that are definitely correct as faulty or be classified without error is increased again.

The procedure can be continued in this way. This results in a self-learning system that makes it possible to identify errors in devices that have already occurred in the past purely on the basis of data and only to carry out a laboratory test on devices that may have new types of error patterns based on the transmitted data.

Another embodiment may be based on the recognition of patterns in the data that represent a state of the device.

For example, a device, e.g. a display, is removed from a target system, e.g is. The result of this process, also referred to as a diagnosis, can be linked to the data that represent the status of the device (e.g. error code(s), vehicle model, mileage, etc.), which is "over- the-Air”. From the data linked in this way, data patterns can be determined using machine learning, which make it possible to classify the device as defective or non-defective.

For example, such linked data is determined with a diagnostic tester or sent wirelessly ("over the air") to a cloud via a connectivity unit. The linked data is, for example, compared to the data patterns determined using machine learning, either before creating a guided error search or in a guided error search as the first test step.

In this respect, three cases can occur, which are identified below with A, B and C and are explained in more detail.

A: A known data pattern was detected in the associated data that the device classifies as defective. The device can now be removed without guided troubleshooting and without further test steps.

B: A known data pattern was detected in the associated data that the device classifies as non-defective. No test steps need to be performed for this device. The device can be considered "okay". Any test steps included in guided troubleshooting can be skipped.

C: No known data pattern was detected in the linked data. Therefore, the device can be classified neither as defective nor as non-defective at this point. Instead, this information is obtained in other ways, such as through conventional, e.g., manual and guided, troubleshooting. The result is then in turn ready to successively enrich the treasure trove of known data patterns using suitable algorithms.

As mentioned above, the associated data may also expressly include data representing information about whether one or more symptoms apply to the device or not. Such symptoms of the devices can be observed, for example, in the premises of a workshop and the associated data can be generated at such locations.

Yet another embodiment may be based on training classifiers and using the trained classifiers.

For example, a device, e.g. a display, is removed from a target system, e.g is. The result of this process, also referred to as a diagnosis, can be linked to the data that represent the status of the device (e.g. error code(s), vehicle model, mileage, etc.), which is "over- the-Air”. A classifier or several classifiers can be trained on the basis of this data linked in this way.

For example, such linked data is determined with a diagnostic tester or sent wirelessly ("over the air") to a cloud via a connectivity unit. The classifier or classifiers now use this linked data in order to classify the device as defective or non-defective, for example. The classifier can be implemented on the diagnostic tester or a cloud or another platform. For example, the classifier can be implemented as part of a guided fault finding as a first test step or before the creation of the guided fault finding. Provision can be made for the classifier or classifiers to simultaneously determine a confidence level and to be assigned to one of the two classes mentioned only if the confidence level exceeds a confidence threshold, ie incorrect classification is ruled out with a certain degree of certainty.

In this respect, three cases can occur, which are again labeled A, B and C below and are explained in more detail.

A: Based on the linked data, the classifier classifies the device as defective. The device can now be removed without guided troubleshooting and without further test steps.

B: Based on the linked data, the classifier classifies the device as non-defective. No test steps need to be performed for this device. The device can be considered "okay". Any test steps included in guided troubleshooting can be skipped.

C: On the basis of the linked data, the classifier is unable to classify either "defective" or "non-defective", for example because a confidence measure does not exceed a threshold value in either case. Instead, this information is obtained in other ways, such as through conventional, e.g., manual and guided, troubleshooting. The result is then in turn available for further training of the classifier.

Again, the associated data may also expressly include data representing information about whether one or more symptoms apply to the device or not. Such symptoms of the devices can be observed, for example, in the premises of a workshop, and the associated data can be generated at such locations.

Claims (13)

Method for diagnosing a device, the method having the following steps: - receiving data representing a state of the device, - assigning the device to one of the following three classes based on the received data using a classification algorithm: * Class "1" : Device is OK, * Class "2": Device has a defect, * Class "3": It cannot be decided whether the device has a defect, - where in the case where the device is assigned to Class "3", below further data are received which represent whether the device is in order or whether the device has an error, and the classification algorithm is subsequently trained with training data which comprise the data and the further data. procedure after claim 1 , where the class "2" has several subclasses associated with different types of errors, and the method comprises the following steps: - Assigning the device to one of the following classes based on the received data by the classification algorithm * Class "1": device is OK, * one of the subclasses of class “2”: device has an error of the type of error assigned to the subclass, * class “3”: it cannot be decided whether the device has an error or which error it has, - where, in the event that the device is assigned to class "3", further data is subsequently received which represents whether the device is in order or whether the device has an error, the further data in in the latter case also represent the type of error, and the classification algorithm is then trained with the training data, which includes the data and the further data. Procedure according to one of Claims 1 or 2 , wherein output data is output which represents the class and possibly the subclass to which the device was assigned by the classification algorithm. A method according to any one of the preceding claims, wherein the data further represents: - error memory entries, - information store entries, - vehicle states, e.g. battery voltage, - special equipment, - Software versions and/or - Hardware revisions. Method according to one of the preceding claims, wherein the data are read from a data memory of the device or a target system comprising the device. A method according to any one of the preceding claims, wherein the further data represents the result of an examination of the device. A method according to any one of the preceding claims, wherein the further data is entered manually. procedure after claim 5 and 7 , the reading out and the inputting taking place at different locations from one another, for example in a workshop on the one hand and in a laboratory on the other hand. Method according to one of the preceding claims, wherein the classification algorithm has the following steps: - pre-assignment of the device based on the received data to class "1" or to class "2" or to a subclass of class "2", - Determination of a confidence measure (ϕ) of the pre-allocation that has taken place, - Allocation of the device according to the pre-allocation if the confidence measure (ϕ) exceeds a predetermined threshold value, - Allocation of the device to class "3" if the confidence measure (ϕ) falls below the specified threshold value. Computer program that is set up to carry out each step of the method according to one of the preceding claims or a combination of several, in particular two, computer programs that are set up together to carry out each step of the method according to one of the preceding claims. Electronic storage medium on which a computer program according to the preceding claim is stored, or a total of several electronic storage media on which all of the computer programs according to the preceding claim are stored. Evaluation and control unit, in particular computer, which comprises an electronic storage medium according to the preceding claim, or a combination of several evaluation and control units, in particular computers, which together comprise the entirety of several electronic storage media according to the preceding claim. Evaluation and control unit according to the preceding claim or a total of several evaluation and control units according to the preceding claim, characterized in that the evaluation and control unit or at least at least one of the evaluation and control units includes a clock.

Images