WO2021110301A1

WO2021110301A1 - Method and apparatus for determining a state of a machine

Info

Publication number: WO2021110301A1
Application number: PCT/EP2020/077993
Authority: WO
Inventors: Ruben KAPP; Joachim Soubari; Juergen Sojka; Mario Herrera; Martin Kessler
Original assignee: Robert Bosch Gmbh
Priority date: 2019-12-06
Filing date: 2020-10-06
Publication date: 2021-06-10
Also published as: DE102019219084A1

Abstract

The invention relates to a method (10) for determining a state (20) of a machine (11), characterized by the following features (17): - the machine (11) is continuously monitored (14) by sensors (12, 13), - measurement signals (15) from the sensors (12, 13) are supplied to a computing unit (16), - features (17) of the measurement signals (15) extracted by the computing unit (16) are evaluated on the basis of a model (19), and - the state (20) is derived from the features (17) by means of the model (19).

Description

description

title

Method and device for determining a state of a machine

The present invention relates to a method for determining a condition of a machine and detecting anomalies in its components. The present invention also relates to a corresponding device, a corresponding computer program and a corresponding storage medium.

State of the art

During the ferry operation of a motor vehicle according to the state of the art, all exhaust-influencing systems and other important control units, the data of which are accessible through their software, are monitored for the purpose of on-board diagnosis (OBD). Relevant methods include, for example, electrical diagnoses such as the detection of short circuits to ground or battery, cable breaks or implausible voltages between certain lines as well as plausibility checks, balancing and gradient monitoring of a wide variety of measurement signals supplied by sensors.

Errors detected in this way are displayed to the driver via a control lamp and permanently stored in the respective control unit so that they can later be queried by a specialist workshop via standardized interfaces.

DE102013200573A1, for example, relates to a method and an associated device for the active diagnosis of components of an exhaust gas cleaning system of an internal combustion engine, which at least one Has catalytic converter, with a first exhaust gas probe arranged in front of the catalytic converter and a second exhaust gas probe arranged behind the catalytic converter, wherein during preconditioning with an air-fuel mixture with a rich air-fuel ratio and in a subsequent measurement phase with a lean air Fuel ratio is operated and where an oxygen storage capacity of the catalyst is determined from an integrated oxygen input into the catalyst during the measurement phase until a lean air-fuel ratio is detected at the second exhaust gas probe, or where the internal combustion engine is used to determine the oxygen storage capacity the preconditioning is operated with a lean air-fuel ratio and operated in the subsequent measurement phase with a rich air-fuel ratio and the oxygen storage capacity from an integrated oxygen discharge from the catalytic converter during the Measurement phase until a rich air / fuel ratio is detected on the second exhaust gas probe.

Disclosure of the invention

The invention provides a method for determining a state of a machine, a corresponding device, a corresponding computer program and a corresponding storage medium according to the independent claims.

The approach according to the invention is based on the knowledge that diagnostic systems according to the prior art are typically not able to monitor components that are not relevant to exhaust emissions or that have corresponding sensors. The detection of anomalies in the components and a prediction of component failures is not readily possible, since generic diagnostic systems are generally only designed to detect faults that have already occurred. The prediction of errors, however, is not part of the functional scope of such diagnostic systems. In addition, measured variables, the course of which over time could indicate anomalies and future errors, are not recorded or are only recorded on the basis of accumulated values, so that no historical database is available for such a vehicle-internal error prediction. Against this background, a basic idea of the proposed solution is to use machine components that are conventionally not monitored or monitored to a limited extent - z. B. Motor or other drive components - to be monitored with the help of structure-borne sound or piezo sensors in order to determine and predict, among other things, mechanical failures due to cracks in the material or bearing wear on a pump, the complete failure of individual components or reaching other intervention thresholds.

For this purpose, the sensor signals detected in this way can be monitored in a computing unit such as a central vehicle control unit (VCU) or another electronic control unit (ECU) in the vehicle and prepared for further processing by using a suitable algorithm is calculated. The signals can be used in the processing unit or electronic control unit in the form of transient signals, diagnostic features or statistical features or their frequencies or as a spectogram or spectral bands or prepared for further analysis. The data and features from the signal can be evaluated with the help of machine learning - for example by means of artificial neural networks - or on the basis of data and physical models with the aim of determining the current status of the monitored component, anomalies in the monitored components to detect their remaining service life or to predict the expected point in time at which their functional impairment becomes noticeable to the driver.

For intensive calculations, the signals from the sensors can be sent to the cloud in the form of suitable data via a telematics unit in the vehicle or a mobile phone, and features can be extracted in the cloud.

In this way, unpredictable failures as well as unplanned workshop visits or even car breakdowns can be avoided. This in turn increases the comfort of the vehicle driver and any passengers. The targeted detection of the anomaly or defects made possible according to the invention Component and type of error make it easier to plan any visits to the workshop and, if necessary, save time when troubleshooting in the workshop or at the original equipment manufacturer (OEM).

One advantage of the monitoring of the condition of engine components and systems according to the invention lies in the possibility of recognizing or even predicting defects and signs of wear as well as their localization (pinpointing). Relevant information can be provided over the air using web services and other cloud solutions. An OBD interface is not absolutely necessary for this.

The measures listed in the dependent claims allow advantageous developments and improvements of the basic idea specified in the independent claim.

Brief description of the drawings

Exemplary embodiments of the invention are shown in the drawings and explained in more detail in the description below. It shows:

Figure 1 shows the flow chart of a method according to a first embodiment.

Figure 2 schematically shows a control device according to a second embodiment. Embodiments of the invention

FIG. 1 illustrates the basic sequence of a method according to the invention (10). In order to measure the condition (20) of components of a machine (11), structure-borne sound sensors (12) such as vibration velocity sensors, vibration displacement sensors or vibration acceleration sensors are used. Piezoceramics are for this purpose particularly suitable. In particular, the on-board sensors already provided in conventional vehicles come into consideration, in the case of gasoline engines, for example, the knock sensor (24) built into the engine block or crankcase, or in the case of CRI2 or CRI3 diesel engines, the needle closing sensors built into the injectors , NCS; 25). One or more structure-borne sound sensors (12) or one or more piezo sensors (13) can be used or signals coming from the ECU or other components can be used to increase the quality of the prediction (21) and the quality of the analysis.

The sensors (12) measure the structure-borne noise, i.e. mechanical oscillation, and convert this into electrical voltage. The resulting voltage signal (15) can be electrically amplified or preprocessed via analog filters, depending on the application. The signal (15) is converted into a digital data stream for further digital processing.

The digital data stream is processed in a computing unit (16) in the vehicle. Various methods and algorithms are used for processing and analysis. The signal (15) represented by the data can, for example, be analyzed at the time and frequency level for the presence of certain frequencies and ranges / sections or for certain threshold values to be exceeded. The data and / or features (17) extracted in this way can ultimately serve as input for physical models or statistical models (19) formed by machine learning (18).

In view of the limited computing power of the computing units (16) on board conventional vehicles, more computationally intensive analyzes can be outsourced. For this purpose, the extracted data and / or features (17) are preferably packet-oriented via a transmission unit (22) in the form of a communication interface {connectivity control unit, CCU) or telematics unit (telematics unit, TCU) to an external computer, for example in the cloud ( 23) sent. The feature data (17) are then stored on the server and can be combined with other fleet data and prepared for further processing. The data and / or features (17) are stored in the cloud (23) for this purpose, for example through machine learning (18) based methods are evaluated with regard to possible faults in the machine (11) in order to make a relevant prediction (21). The described merging of the feature data (17) obtained in different vehicles of a fleet improves the models (19) used in the context of the method (10), but requires external computers, e.g. B. in the cloud (23).

A large number of relevant models (19) and algorithms can be considered. A separate model (19) is typically used for each state (20) to be predicted. In individual cases, a single model (19) may nonetheless be used to predict (21) multiple component failures.

This procedure is illustrated by a specific example: The NCS (25) of a CRS3 injector monitors (14) at a point of the combustion cycle - defined with respect to the top dead center of the cylinder in question - over a defined period of time, e.g. B. 400 ms. The features (17) of the scanned signal (15) extracted by the ECU (16) are transmitted to the VCU (22) and z. B. stored compressed by an encoder. In the ECU (16) or VCU (22), the signal (15) or its features (17) can already be checked for plausibility and validity, so that only permissible signals (15) are processed further. This monitoring (14), transmission, compression and storage takes place over the duration of the driving cycle at defined time intervals and only when the vehicle is in a defined state, e.g. B. only in overrun phases. So many features (17) are collected until a predetermined amount of data is reached. The features (17) are then transmitted to the cloud (23).

The feature data (17) are decompressed in the cloud (23) and merged with inventory data in a database. Prediction models (19) are then applied to the new database and new predictions (21) are made. For this purpose, the feature data (17) are propagated through a convolutional neural network for the purpose of anomaly detection. Frequency and amplitude, for example, serve as examined parameters. In order to improve the model (19), further measured variables are also transmitted from the vehicle. One should think about Cooling water temperature, rail pressure and ambient temperature. The CNN used as model (19) in the present application example classifies the underlying signal (15) as normal or abnormal on the basis of the feature data (17) present at its input. If an anomaly is detected in this way over a certain number of measurements, the remaining range can be calculated using a further algorithm. This information can be transmitted or otherwise made available to the driver or owner of the vehicle.

This method (10) can, for example, in software or hardware or in a mixed form of software and hardware, for example in one

Control unit (30) can be implemented, as the schematic illustration in FIG. 2 illustrates.

Claims

Expectations

1. Method (10) for determining a state (20) of a machine (11), characterized by the following features (17):

- the machine (11) is continuously monitored (14) by sensors (12, 13),

- Measurement signals (15) from the sensors (12, 13) are fed to a computing unit (16),

- Features (17) of the measurement signals (15) extracted by the computing unit (16) are evaluated and based on a model (19)

- The state (20) is derived from the data and / or features (17) by means of the model (19).

2. The method (10) according to claim 1, characterized by the following feature:

- Using the model (19), a prediction (21) is also made with regard to a possible failure of the machine (11).

3. The method (10) according to claim 1 or 2, characterized by at least one of the following features:

- The model (19) is a statistical model (19) or formed by machine learning (18)

- The model (19) is a physical partial model of the monitored (14) machine (11).

4. The method (10) according to any one of claims 1 to 3, characterized by the following features:

- The data / features (17) are uploaded to a cloud (23) by a transmission unit (22) and

- The evaluation takes place in the cloud (23).

5. The method (10) according to any one of claims 1 to 4, characterized by at least one of the following features:

the arithmetic unit (16) is a computer, microcontroller, microprocessor or the arithmetic unit (16) is a vehicle or something else

Control unit.

6. The method (10) according to any one of claims 1 to 5, characterized by at least one of the following features:

- the sensors (12, 13) comprise structure-borne sound sensors (12), - the sensors (12, 13) comprise piezo sensors (13) or

- The measurement signals (15) are supplemented by additional signals in the arithmetic unit (16).

7. The method (10) according to any one of claims 1 to 6, characterized by one of the following features:

- The sensors (12, 13) comprise a knock sensor (24) or

- The sensors (12, 13) comprise a needle closing sensor (25).

8. Computer program which is set up to carry out the method (10) according to one of claims 1 to 7.

9. Machine-readable storage medium on which the computer program according to claim 8 is stored.

10. Device (30) which is set up to carry out the method (10) according to one of claims 1 to 7.