US-12618907-B2 - Method and device for detecting a critical anomaly in a device battery based on machine learning methods

US12618907B2US 12618907 B2US12618907 B2US 12618907B2US-12618907-B2

Abstract

A method for monitoring a device battery for the presence of an anomaly in the battery operation. In some examples, the method includes providing temporal operational variable profiles of operational variables for the device battery; determining an input dataset of operational characteristics for a historical evaluation period as a function of the temporal operational variable profiles; and using an anomaly detection model with an encoder/decoder model to determine a reconstruction error for the input dataset. The encoder/decoder model is trained with input datasets of device batteries of normal function to map an input dataset onto an input dataset that is reconstructed as identically as possible. When a reconstruction error is found, a rule-based criticality value is determined as a function of one or more characteristic-based criticality values. An error type is also signaled.

Inventors

Alexandr Savinov
Parameswaran Krishnan
Calvin Pfob
Christian Simonis
Karthikeyan Ramachandran
Raimund Kaiser
Tobias Huelsing

Assignees

ROBERT BOSCH GMBH

Dates

Publication Date: 20260505
Application Date: 20230301
Priority Date: 20220302

Claims (8)

1 . A method for monitoring a device battery ( 41 ) for a presence of an anomaly in a operation of the device battery ( 41 ), the method comprising: providing (S 1 ) temporal operational variable profiles (F) of operational variables for the device battery ( 41 ); determining (S 2 ), via a computer, an input dataset (X) of operational characteristics for a historical evaluation period as a function of the temporal operational variable profiles (F); using (S 3 , S 4 ), via the computer, an anomaly detection model ( 10 ) with an encoder/decoder model to determine a reconstruction error (e) for the input dataset (X), wherein the encoder/decoder model is trained with input datasets (X) of device batteries ( 41 ) of normal function in order to map an input dataset (X) onto an input dataset (X′) that is reconstructed as identically as possible; when a reconstruction error (e) is found to be above a specified anomaly threshold (S 5 ), determining (S 6 , S 7 ), via the computer, at least one rule-based criticality value (r) as a function of one or more characteristic-based criticality values (h i ), which depend on or correspond to a reconstruction error (e i ) based on a characteristic, and at least one predetermined rule, wherein the at least one predetermined rule considers one or more of the characteristic-based criticality values (h i ) and specifies a criterion for a presence of an error mode, which depends on the one or more characteristic-based criticality values (h i ); and signaling (S 9 , S 10 ), via the computer, a corresponding error type as a function of the at least one rule-based criticality value (r).
2 . The method according to claim 1 , wherein the rule-based criticality value (r) is determined from one or more characteristic-based criticality values (h i ) specified by the corresponding rule, and wherein the characteristic-based criticality values (h i ) correspond to normalized characteristic-based reconstruction errors (e i ).
3 . The method according to claim 2 , wherein the characteristic-based criticality value (h i ) for an operational characteristic is determined from the input dataset by specifying an upper and lower reconstruction error threshold ( e i max , e i min ) and an anticipated value for a characteristic-based reconstruction error (e i ) of the characteristic-based reconstruction errors (e i ) of a relevant operational characteristic, and wherein the characteristic-based reconstruction error (e i ) is normalized to the reconstruction error thresholds and the anticipated value.
4 . The method according to claim 3 , wherein the upper and lower reconstruction error threshold ( e i max , e i min ) is determined as specified quantile values of the characteristic-based reconstruction errors (e i ) from an evaluation of specified training datasets for the anomaly detection model ( 10 ) and the anticipated value is determined as the mean value of the characteristic-based reconstruction errors (e i ).
5 . The method according to claim 1 , wherein the encoder/decoder model comprises a data-based encoder model ( 11 ), which is configured as a dimension-reducing deep neural network or PCA model, and a data-based decoder model ( 12 ), which is configured as a dimension-extending deep neural network, wherein the encoder/decoder model is or becomes trained with training datasets that represent the input datasets for a properly functioning device battery ( 41 ).
6 . The method according to claim 1 , wherein an anomaly is detected when a reconstruction error (e) is detected above the specified anomaly threshold, wherein the frequency of performing the anomaly monitoring is determined in order to find one of a fraction of anomalies detected with respect to all anomalies of the anomaly monitoring.
7 . A computer configured to monitor a device battery ( 41 ) for a presence of an anomaly in an operation of the device battery ( 41 ), by: obtaining (S 1 ) temporal operational variable profiles (F) of operational variables for the device battery ( 41 ); determining (S 2 ) an input dataset (X) of operational characteristics for a historical evaluation period as a function of the temporal operational variable profiles (F); using (S 3 , S 4 ) an anomaly detection model ( 10 ) with an encoder/decoder model to determine a reconstruction error (e) for the input dataset (X), wherein the encoder/decoder model is trained with input datasets (X) of device batteries ( 41 ) of normal function in order to map an input dataset (X) onto an input dataset (X′) that is reconstructed as identically as possible; when a reconstruction error (e) is found to be above a specified anomaly threshold (S 5 ), determining (S 6 , S 7 ) at least one rule-based criticality value (r) as a function of one or more characteristic-based criticality values (h i ), which depend on or correspond to a reconstruction error (e i ) based on a characteristic, and at least one predetermined rule, wherein the at least one predetermined rule considers one or more of the characteristic-based criticality values (h i ) and specifies a criterion for a presence of an error mode, which depends on the one or more characteristic-based criticality values (h i ); and signaling (S 9 , S 10 ) a corresponding error type as a function of the at least one rule-based criticality value (r).
8 . A non-transitory, computer-readable storage medium containing instructions which, when executed by the computer, cause the computer to monitor a device battery ( 41 ) for a presence of an anomaly in an operation of the device battery ( 41 ), by: obtaining (S 1 ) temporal operational variable profiles (F) of operational variables for the device battery ( 41 ); determining (S 2 ) an input dataset (X) of operational characteristics for a historical evaluation period as a function of the temporal operational variable profiles (F); using (S 3 , S 4 ) an anomaly detection model ( 10 ) with an encoder/decoder model to determine a reconstruction error (e) for the input dataset (X), wherein the encoder/decoder model is trained with input datasets (X) of device batteries ( 41 ) of normal function in order to map an input dataset) onto an input dataset that is reconstructed as identically as possible; when a reconstruction error (e) is found to be above a specified anomaly threshold (S 5 ), determining (S 6 , S 7 ) at least one rule-based criticality value (r) as a function of one or more characteristic-based criticality values (h i ), which depend on or correspond to a reconstruction error (e i ) based on a characteristic, and at least one predetermined rule, wherein the at least one predetermined rule considers one or more of the characteristic-based criticality values (h i ) and specifies a criterion for a presence of an error mode, which depends on the one or more characteristic-based criticality values (e i ); and signaling (S 9 , S 10 ) a corresponding error type as a function of the at least one rule-based criticality value (r).

Description

BACKGROUND OF THE INVENTION The invention relates to device batteries for use in technical devices as well as to methods for detecting an anomaly in device batteries. The invention further relates to a method for assessing a criticality of an anomaly in a device battery. The invention relates to systems in which a plurality of device batteries are monitored by a device-external central processing unit. The invention further relates to methods for detecting an anomaly in a device battery and determining its criticality. The supply of energy to network-independently operated electrical devices and machines, such as electrically drivable motor vehicles, as a rule takes place by means of device batteries or vehicle batteries. The latter supply electrical energy for operating the devices. Device batteries degrade over their service life and as a function of their load or usage. This so-called aging leads to a continuously decreasing maximum power or storage capacity. The aging state corresponds to a measure for indicating the aging of energy stores. In accordance with the convention, a new device battery can have a 100% aging state (regarding its capacity, SOH-C) which increasingly decreases over the course of its service life. A degree of aging of the device battery (change in the aging state over time) depends on an individual load on the device battery, i.e., in the case of vehicle batteries of motor vehicles, on the usage behavior of a driver, external ambient conditions and on the type of vehicle battery. In order to monitor device batteries from a plurality of devices, operating value data is typically continuously captured and, as operating value profiles, are transmitted in block fashion to an in-device central processing unit. In the case of device batteries having a plurality of battery cells, the operating values can be captured at the cell level and transmitted to the central processing unit, in particular in compressed form. To evaluate the operating value data, in particular to determine aging states in models based on differential equations, the operating value data is scanned with a comparatively high temporal resolution (scanning frequencies) of, for example, between 1 and 100 Hz and an aging state is determined therefrom using a time integration method. In addition to age-based degradation, device batteries can experience errors due to a variety of causes that can result in faster aging or sudden failure of the device battery. These failures and errors of device batteries and individual battery cells are often discernible in advance by changes in battery behavior and can be detected as a result. In order to increase the acceptance of device batteries, it is necessary to ensure their safety, durability, performance, and reliable operation. For this purpose, it is important to carefully monitor battery performance. This is usually done using anomaly detection methods, which can detect deviations from normal operation of the device batteries and can also determine the criticality and/or cause of the abnormalities. Existing methods focus on an estimation of the state of aging or a prediction of the state of aging, which, however, depends significantly on a recognition of the behavior of the type of battery in question. However, if there is no suitable model for mapping the battery behavior, a detection of an anomaly in the battery behavior is difficult. SUMMARY OF THE INVENTION According to the invention, there is provided a method for detecting an error type in a device battery for a technical device as well as a corresponding apparatus Further configurations are specified in the dependent claims. According to a first aspect, a method for monitoring a device battery for the presence of an anomaly in the battery operation is provided, having the following steps: providing temporal operational variable profiles of operational variables for the device battery;determining an input dataset of operational characteristics for a historical evaluation period;using an anomaly detection model with an encoder/decoder model to determine a reconstruction error for the input dataset, wherein the encoder/decoder model is trained with input datasets of device batteries of normal function in order to map an input dataset onto an input dataset that is reconstructed as identically as possible;When a reconstruction error is found to be above a specified anomaly threshold, determining at least one rule-based criticality value as a function of one or more characteristic-based criticality values, which depend on or correspond to a reconstruction error based on a characteristic, and at least one predetermined rule, wherein the at least one predetermined rule considers one or more of the characteristic-based criticality values and specifies a criterion for the presence of an error mode, which depends on the one or more characteristic-based criticality values;signaling the corresponding error type as a function of the at le