Felix Richter

Temporal and Spatial Clustering for a Parking Prediction Service

It has been estimated that in urban scenarios up to 30% of the traffic is due to vehicles looking for a free parking space. Thanks to recent technological evolutions, it is now possible to have at least a partial coverage of real-time data of parking space availability, and some preliminary mobile services are able to guide drivers towards free parking spaces. Nevertheless, the integration of this data within car navigators is challenging, mainly because (I) current In-Vehicle Telematic systems are not connected, and (II) they have strong limitations in terms of storage capabilities. To overcome these issues, in this paper we present a back-end based approach to learn historical models of parking availability per street. These compact models can then be easily stored on the map in the vehicle. In particular, we investigate the trade-off between the granularity level of the detailed spatial and temporal representation of parking space availability vs. The achievable prediction accuracy, using different spatio-temporal clustering strategies. The proposed solution is evaluated using five months of parking availability data, publicly available from the project Spark, based in San Francisco. Results show that clustering can reduce the needed storage up to 99%, still having an accuracy of around 70% in the predictions.

Automatic Defect Detection by One-Class Classification on Raw Vehicle Sensor Data

The next step in the automotive industry is the automatic detection of a defect in the vehicle behavior in addition to the current analysis of failure codes or costumer complaints. The idea of learning the normality by one-class classification is applied to the identification of an exemplary defect. Different neural network topologies for time series prediction are realized where the quality of the forecast indicates the strength of abnormality. It is compared how the detection possibilities of a concrete defect changes when the model is trained with different data extractions. A distinction is made between data from complete rides and filtered data, containing only the situations where the defect is visible. It can be shown that a generalization is possible.

Automatic Defect Detection by Classifying Aggregated Vehicular Behavior

Detecting defects is a major task for all complex products, as automobiles. Current symptoms are the failure codes a vehicle produces and the complaints of a customer. An important part on the defect detection is the vehicular behavior. This paper highlights the analysis of vehicular data as a new symptom in the customer service process. The proposed concept combines the necessary preprocessing of vehicular data, especially the feature-based aggregation of this data, with the analysis on different sets of features for detecting a defect. In the modeling part a Support Vector Machine classifier is trained on single observed situations in the vehicular behavior and a Decision Tree is used to abstract the model output to a trip decision. The evaluation states a detection quality of 0.9418 as the F1-score.

Automatic Root Cause Analysis by Integrating Heterogeneous Data Sources

This paper proposes a concept for automated root cause analysis, which integrates heterogeneous data sources and works in near real-time, in order to overcome the time-delay between failure occurrence and diagnosis. Such sources are (a) vehicle data, transmitted online to a backend and (b) customer service data comprising all historical diagnosed failures of a vehicle fleet and the performed repair actions. This approach focusses on the harmonization of the different granularity of the data sources, by abstracting them in a unified representation. The vehicle behavior is recorded by raw signal aggregations. These aggregations are representing the vehicle behavior in a respective time period. At discrete moments in time these aggregations are transmitted to a backend in order to build a history of the vehicle behavior. Each workshop session is used to link the historic vehicle behavior to the customer service data. The result is a root cause database. An automatic root cause analysis can be carried out by comparing the data collected for an ego-vehicle, the vehicle the failure situation occurred, with the root cause database. On the other hand, the customer service data can be analyzed by an occurred failure code and filtered by comparing the vehicle behavior. The most valid root cause is detected by weighting the patterns described above.