direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Page Content

Automated Evaluation of Vehicle Measurement Data with modern Pattern Recognition Methods

Third party funded project in cooperation with IAV GmbH

In the development of new vehicle components a huge amount of measurement data is recorded for validation. These data has to be evaluated automated for an effective data analysis.
In cooperation with the IAV GmbH a software is developed for an automated analysis of vehicle measurement data. The data evaluation works either signal or message based (CAN-Bus) and once configurated the configuration can be easily used on many datasets. The focus of the Chair of Electronic Measurement and Diagnostic Technology is the development of modern pattern recognition modules for error detection in the vehicle measurement data. These modules are integrated modular in the software and can be configured by XML-Files.

Development of evaluators for state recognition

The vehicle measurement data is evaluated automated on predefined states. Inside the software this will be managed by the so-called evaluators. For each sample point the evaluators reject a measure for appearance of the definded event. For the recognition of different states it is necessary to have the possibilty to describe these states and that accordant evaluators for recognition are available.
In the development and diagnosis of new vehicle components the measurement data is analysed on specified error states. For the definition of these error states there exist often only a few example datasets. For this reason the classical pattern recognition methods are difficult to use. In sequence to this, new methods have to be developed: on the one hand for describing the error states and on the other hand for an effective evaluation of the data according to these error states.

For recognition and description of states different approches are observed:

  • Description and recognition of states with

    • Java code
    • A verbal descriptive language in terms of analytical signal descriptions in combination with fuzzy logic methods

  • Allocation of evaluators for recognition of basis patterns (e.g. oscillations, step functions, etc.)
  • Searching of a defined signal pattern with dynamic time warping
  • Splitting of complex states in partial states
  • Connected evaluation over multiple channels in consideration of time correlation


Example:

A system state extended on three channels. Each channel can be processed by one special evaluator. The task of the first evaluator is the recognition of a sinus burst. The second detects a rising edge and the third evaluator reacts on a determined signal pattern. The information will be transformed by the evaluators and than deliverd to a further evaluator for summary. The result is a measure for the appearance of the described state.

Lupe

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions

Prof. Dr.-Ing. Clemens Gühmann
+49 30 314-29393
Building EN
Room EN539

Website

Henrik Isernhagen
+49 30 314-21168
Building EN
Room EN450

Website