Automotive Demonstrator "TUC DriveCloud"
The TUC DriveCloud is a Big Data storage and analysis plattform targeted at automotive test applications. The main functionalities are:
- Structured management of sensors and vehicles
- Standardized input and output interfaces (abstraction of the actual physical storage)
- Test drive management, recording and live view of collect sensor data
- Automatic labeling of test drives
Architecture
The central system of the TUCDriveCloud is a structured database for managing and recording test drive data. This cloud database can be accessed through standardized APIs for input and output as illustrated in the following figure:
Another part of the cloud application is the website which runs on a web server at Chemnitz University of Technology and provides a comprehensive user interface for the input and output of cloud data. Via this web interface, test vehicles can be added to the cloud and test drives can be configured for them. A test drive is based on a so-called ¨Drive Setup", which contains a set of sensors for a test vehicle. With this drive setup, the test vehicle can then undertake any number of test drives, the data of which is stored in the TUCDriveCloud grouped by drive. The data can already be stored directly in the cloud database during the drive via a token-protected API. If this is not possible due to a missing internet connection, the data is stored locally and written to the cloud database at a later point in time [1]. The test drive data can be visualized and evaluated on the website, as shown in the following screenshot:
Storage model
A fundamental goal of the TUC DriveCloud platform is to abstract the storage model. This means that users and developers can add to or retrieve sensor data from the cloud exclusively via the public interfaces (ReST-API). A direct access to the underlying physical storage is not possible. This allows to select a suitabke storage model depending on the sensor data to be stored. Additionally, users do not need to be aware of this or have knowledge of the storage system. This approach also simplifies the extension of the platform with new sensor types. The standard storage for scalar sensor data streams, i.e. sensors that provide a single data value for a time value (e.g. speed, GPS longitude, GPS latitude) is a relational database. In addition, at the present time, image sequences (stored internally as HDF5) can also be stored. A storage of video streams in the form of mp4 files is also possible. Additionally, support for Lidar data is currently under development. The described storage abstraction is illustrated in the following figure: