App settings:

Hella Gutmann

Motivation

Vehicle diagnostics uses error codes and sensor values to identify potentially defective components in the vehicle. At present, vehicle diagnosis in a workshop requires comprehensive knowledge of the vehicle and is associated with a considerable amount of work. This is made more difficult by the fact that the advancing electrification of the automobile leads to a steady increase in the complexity of vehicle diagnosis.

About the project

The aim of the pilot project is to develop and test methods for the automated detection of potentially defective components in the vehicle with the aid of artificial intelligence. In particular, suitable methods for data pre-processing and training of machine learning models with the help of vehicle data are to be developed and validated. Furthermore, common machine learning models will be evaluated with respect to their applicability for vehicle diagnosis and integrated into a prediction model for defective components. Automotive experts validate the prediction model and transfer it to a demonstrator for AI-supported vehicle diagnostics. In addition, a testing framework for the continuous integration of new data using active learning approaches will be developed. Apart from the demonstrator and the testing framework, this pilot project will also entail a comprehensive collection of best practices for data processing and the development and validation of machine learning models in product creation.

The implementation

The following steps are planned for the realization of the project:

  • Data Cleaning: Analysis and preparation of the database by automotive experts and data engineers. The main focus is on the exploration of the raw data, the review of the processed data by automotive experts and the implementation of procedures for data cleansing.

  • Extract-Transform-Load (ETL) pipeline: Development of an ETL pipeline to integrate additional data sources. Possible data sources are, for example, invoice data and logging data.

  • Model development: Evaluation, selection and integration of common machine learning models for predicting potentially defective components in the context of a vehicle diagnosis. The result is a new aggregated prediction model for vehicle diagnostics.

  • Model validation: Validation of the developed prediction model by domain experts from the automotive sector. The feedback of the domain experts is used for step-by-step optimization of the prediction model.

  • Testing Framework: Development of a framework for the administration of the prediction model and the database used. This framework should enable the creation of reports on the model and the database used as well as the integration of new data through active learning.