AICloud: Development of an AI-based virtual sensor using a cloud computing platform
AICloud: Development of an AI-based virtual sensor using a cloud computing platform
AICloud: Development of an AI-based virtual sensor using a cloud computing platform
As part of the AICloud project, the institute is researching innovative AI-based sensor algorithms using a cloud platform together with EFS TechHub GmbH. The aim of the project is to implement a cloud environment, known as the SDK/AICloud, in which the data processing and computationally intensive training of artificial neural networks can be carried out. A virtual sensor for a car is implemented as a showcase. It replaces a real hardware sensor by synthesising a trained, i.e. data-based software sensor using measurement data from real sensors and AI methods.
The DLR's AI-For-Mobility (AFM) research vehicle serves as a test platform. The vertical wheel force, i.e. the force between a tyre and a contact patch, is selected as the variable to be estimated by the virtual sensor. Knowledge of this force is essential for vehicle dynamics control systems for both manual and automated driving. This physical quantity cannot be measured by a real sensor with reasonable effort. Its estimation is a challenging task, which is tackled here with the help of a trained recurrent neural network (RNN). The trained RNN network can then be loaded back into the AFM, where it can be used as a virtual sensor to estimate the vertical wheel force.
In order to generate training data, the AFM is measured on a so-called four-post test rig. Here, the vehicle is placed on four hydraulically driven posts, each of which is controlled so that it follows predefined position signals. In this way, the posts under the four wheels stimulate the vertical movement of the vehicle. In this way, driving on precisely defined road profiles can be simulated reproducibly, with the dynamic wheel loads being measured by load cell sensors.
RNN Training
The widely used TensorFlow framework with its Keras library is used to implement the recurrent neural networks. This is an open source library for deep learning in the Python programming language. With Keras, it is possible to implement most of the deep learning architectures currently in use. The RNNs used for the virtual wheel load sensor, as well as their Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) architectures, are implemented in Keras. An RNN has numerous degrees of freedom or hyperparameters that specify the model and must be taken into account in the design process. These include the following parameters (see figure below):
The RNN training must be carried out for each hyperparameter set in order to subsequently evaluate the performance of the respective model. Due to the rapidly growing computing capacities required for this, hyperparameter optimisation is carried out efficiently in a high-performance cloud environment. The AICloud platform presented in more detail below is used for this purpose. The following figure shows the interactive OpenSearch dashboard visualisation of the grid search results over the hyperparameter space of the RNN trainings. Each line represents a training configuration of the grid with associated validation metric (mean absolute error and percentage fit). The best configuration is highlighted in red.
Results
The RNN architecture determined from the hyperparameter optimisation can be used to obtain good estimation results for the entire test data set. The following figure shows an example of the comparison between the estimated value of the learnt virtual sensor and the true vertical wheel force measured on the four-post test rig. Obviously, the highly dynamic wheel load can be estimated very well and without significant phase delay.
Managing and analysing the large amount of data measured with the AFM presents some common challenges. To overcome these challenges, as the data
uploaded and saved
Researched
Processed and analysed
Visualised
Collaborated
Stored in a legally compliant manner
we have developed a universal, cloud-based data management platform on which the training and evaluation of the virtual wheel load sensor takes place. The data is stored in the cloud together with metadata that makes it easy for the user to search and allocate.
Daten Management Platform
This project has designed and developed the Superb Data Kraken (also known as SDK), a general purpose, multi-functional data management platform based on existing open source solutions where possible and customised solutions otherwise. It offers a versatile and rich set of tools to easily handle the most common data management tasks. It offers all the necessary functions, including data and metadata storage, data processing, search functions via sensor data, metadata and training data, as well as a data science environment and visualisation tools for test, training and metadata. Throughout the platform, the management of user rights and the separation of data between different use cases is guaranteed.
The core components of the project have been published under the Apache 2.0 licence and can be found on GitHub.
To date, several use cases and projects are already utilising the versatile SDK platform on a daily basis, from industrial sectors to automotive, transport and advertising. Among these, the AICloud project is an important flagship project, as it combines big data with machine learning:
a metadata model was developed that standardises the input of descriptive metadata to enable precise search, data retrieval and visualisation of the sensor data recorded with the AFM vehicle.
a workflow was established to enrich the metadata with analysis results from the data.
Training of the virtual sensor with hyperparameter tuning was performed within the platform's data science environment, utilising cloud computing resources for significant acceleration.
the virtual sensor was then deployed in the cloud and its predictions were validated against the wheel load measurement
This work was funded by the Bavarian State Ministry of Economic Affairs, Regional Development and Energy (StMWi, project AlCloud: Data management system using the example of virtual sensors, funding reference DlK0150/02).
Kontakt
Dr. Jonathan Brembeck
Abteilungsleiter
Deutsches Zentrum für Luft- und Raumfahrt e.V.
Institut für Fahrzeugkonzepte
Fahrzeug Systemdynamik und Regelungstechnik
Münchener Straße 20, 82234 Oberpfaffenhofen-Wessling
