Text

Team: Westermo + RISE+IMFA+UNIVAQ+JKU
Challenge: Combined architecture modelling and analysis in the eclipse tooling framework

Context

  • During the AIDOaRt project VCE has been working in collaboration with several solution providers to develop a tool-chain architecture to address the challenges posed by the Use Cases inside AIDOaRt. In this challenge, we extend the previous work to improve previously identified limitations.

Goals

  • The goal is the more robust integration of different components in the overall architecture. Concerning the modeling recommendations, we propose a Docker-based tool that integrates two provided solutions, i.e., MER and MORGAN. In such a way, the VCE engineers can rely on an easy-to-use modeling environment equipped with recommendations, aiming at facilitating the specification of SysML models
Team: VCE + DT
Challenge: Keptn for FMI

Context

  • VCE and MDU have developed some small scale FMI-supported simulation artefacts in the context of the AIDOaRt project. VCE has goals in AIDOaRt to introduce and evaluate DevOps principles in the context of their system architecture management/definition. The aim is to create a continuous and automated simulation execution, parameterization, optimization, and evaluation of FMI models in a completely cloud-based implementation.

Goals

  • Implement a cloud-based DevOps pipeline
    Execute FMI-simulations iteratively
    Optimise parameters based on some heuristics
    Present results to engineers to compare variant configurations.


Team: VCE+ AVL
Challenge: Abstraction gap in industrial co-simulation

Context

Industrial Model-Based Systems Engineering (MBSE) has recently adopted several emerging technologies in the simulation domain. Of particular note is co-simulation, often enabled by the Functional Mockup Interface (FMI) standard. In this challenge, we investigate FMI-based co-simulation with SysML models acting as architecture definitions of eventual simulations. When dealing with this context, we have identified several challenges from a industrial perspective: (1) The abstraction gap between SysML and simulation (although bridged by FMI) creates difficulties in integration. (2) It is difficult to understand when the technology can be used effectively, for example considering the high uncertainty of early systems design where FMI-simulation often is employed. (3) Lack of a standard method of integrating such technologies with commonly used workflows.

Goals

The goal is to investigate how the use of the newly formed SSP standard might alleviate the identified issues with FMI, and work towards a prototype implementation to evaluate the standard.


Team: WMO,RISE,COPADO,ABO,SOFT,MDU

Challenge: Anomaly detection and Seasonal quality metrics

Context

  • In the nightly testing at Westermo, cyber-physical systems run the latest version of the software. The nightly testing is driven by a test framework and runs on physical servers. These servers, like all servers, use varying amounts of memory, CPU, network, etc. These quality metrics are routinely recorded, and for this hackathon challenge we will bring one month of data sampled every 30 seconds for a month, on 19 different systems. This data set has been released on Github https://github.com/westermo/test-system-performance-dataset

Goals

  • The goal of the challenge is to do anomaly detection on this data. Can we detect when nightly testing has failed to start? Can we identify memory leaks? Can we find the root cause to why test systems were rebooted?

Team: ABI, UNISS

Challenge: Formal Verification of Neural Networks: A “Step Zero” Approach for Vehicle Detection

Context

  • This challenge is a continuation of a previous challenge from the 2nd hackathon. Generally, the idea is to implement an automation system that will assist in architecture modeling of a thermal management system at Volvo CE. In this regard it is also of interest to understand the best way to model such a system within the scope of Volvo CE.

Goals

  • In this challenge, we are interested in automate the creation of the layout as shown in  the Figure above. There are numerous variants of each component (component of the TMS, battery, dc/dc, etc,.). Each machine has their own variants of components such as the battery, dc/dc converter. The goal is to optimize the layout based on coolant flow and pressure properties of each component in the coolant loop.

Team: ABI, INT,AIT

Challenge: Adoption of AI and ML techniques for image processing in the automotive context – Update

Context

  • The Abinsula case study presents a virtual rear-view mirror scenario in which multiple cooperative cameras are used to capture the context outside the vehicle, by means of AI-based technology. AI is a recognized innovative technology, but it is still far from being applied in real safety-critical applications, as well as cameras are far from completely replacing the mirrors in a vehicle. This is something allowed only in concept cars and small productions that do not apply the same regulations of large productions.

Goals

  • One of the main challenges we address in this use case, is the replacement of car mirrors with cameras. Therefore, we want to study the adoption of AI and ML techniques for image processing to detect and to signal possible hazards, considering their possible implementation on FPGA boards.

Team:Alstom,MDU,AVL

Challenge: Integration to the Test Platform

Context

  • In the context of the BT UCS2 (Automatic Parameterization of Thermal Model), the development led to exploration of the feasibility of training the Reinforcement Learning and Symbolic Regression learning techniques for parameterizing the thermal parameters of the reduced order thermal model. The training procedure utilises both, simulation set up and experimental data for the learning. The challenge remains the integration/transposition of the AI trained thermal parameters in the physical test bench set-up.

Goals

  • Electric motor control systems on a test bench set-up typically run at the scale of hundreds of microseconds and contain multiple inherent forms of delays, e.g., calculation time of the controller hardware or the modulation scheme of the power electronic converter. Such delays slow down the learning process of learning agents significantly, if implemented online. Due to this real-time constraint, deploying an offline-learned algorithm on the control system hardware is rather pragmatic. Deployment of ML aided thermal model fitting parameters to the control system hardware and to measure the accuracy of predicted temperature on the test platform is the goal

Team: PRO,UOC,ITI,ACO

Challenge: Monitoring and modelling a SPMP (integration)

Context

  • In the port environment, it is necessary to implement a monitoring system for all the elements of a Smart Port Monitoring Platform (SPMP) in order to detect problems that may arise as early as possible and automatically, and also to model the entire infrastructure to anticipate future problems due to overload or lack of resources.
    The main objective of this hackathon is to address the integration of the different partial solutions that have been implemented throughout the project.

Goals

  • Once the tools have been developed for the different tasks necessary for monitoring the different elements of a smart port, where the challenge is to find the problems as soon as possible and, on the other hand, to have a model that simulates the behaviour of the port infrastructure in order to be able to anticipate problems. It is necessary to move on to the integration of these tools so that they can be used in a way that is intuitive and simple for the end user.As in the past, this challenge is composed of several sub-challenges and therefore we will need more time for the presentation of its results (14 minutes). These sub-challenges are the following:

Team: HIB,MDU

Challenge: AI for requirements cross-pollination

Context

  • During the course of AIDOaRt, HI Iberia has worked on a tool for requirements analysis (informal codename HIBRA - HI iBeria Requirements Analyzer) in connection to its requirement HIB_R02 that promotes the use of AI to analyse the current requirements workflow at HI Iberia. The baseline workflow is: the product manager writes requirements as tasks/cars on a Trello board that is accessible by the team. The text is natural language and written in Spanish. With the system in place, a watchdog reader processes incoming requirements and sends them through a NLU pipeline.

Goals

  • We have a working prototype but we've reached the limits of what our dataset can provide us. We'd like to enlist more partners to hand us requirements oftheir own for us to further train the system and also test its benefits and performance. Our system expects requirements in Spanish but we can adapt to a corpora of requirements in English. We expect to be able to perform basic tests on these requirements and generate category and developer suggestions for other team members.

Team: AVL,TUG

Challenge: Model-Based Testing of an Advanced Driver Assistance System - Update

Context

  • Advanced Driver Assistance Systems (ADAS) are systems that are capable of handling specific driving tasks such as highway driving. Those tasks are complex, and mistakes can incur high costs and in the worst case human casualties. Consequently, a lot of effort in the development of ADAS has to be spent on ensuring the safety of the system.


Goals

  • Currently, testing heavily depends on the quality of the logical scenarios. Typically, scenarios describe a very specific set of trajectories, assuming they are archetypical for a wide range of driving situations. The goal of this challenge is to explore methods for assessing the robustness of ADAS agents with respect to the logical scenarios in which they have been tested.

Team: TEK,UNIVAQ,UCAN

Challenge: Design choice exploration/verification (continued)

  • Agile process and Electric/Electronic Architecture of a vehicle for professional applications.
    The use case deals with diagnostics and prognostics of power electronics

Goals

  • Design choices verification (TEK/UNIVAQ/UCAN)
    Before the implementation; on models and/or with rapid prototyping; environment: simulated/emulated
    Verification of the models: do they cover the requirements? do they "work"?
    Verify, at design time, the adequacy (the functional aspects, as well as the response versus the resources) of the target components that the system architect has in mind to map/realize the design of the system.

Team: TEK,AST

Challenge: Runtime Verification

Context

  • The scenario TEK_UCS_02 “Run-time verification” is related to software testing. TEK searches for a solution that, in a semi-automatic way, is capable of defining and executing the tests, as well as interpreting the tests results.

 

Goals

  • TEK use case scenario requires automatic generation of test code. The AST tool, devmate, uses technologies for equivalence-based testing, to produce a test model from which we can generate test code. Other features of devmate include the user interface to handle the input and output definition of the functions under test, a compatible parser, a plugin architecture for a suitable IDE and data handling of equivalence classes and test data, necessary for the proper fulfilment of the requirements.

Team: TEK,ROTECH

Challenge: Operating Life Monitoring:

Context

  • Services in an environment partly simulated.
    Measured data collected and pre-processed (cleaning, filtering, features extraction) to obtain “monitoring data”.
    Data transferred to the remote computing and data storage aggregator whose resources are available to run a full capabilities PHM system

Goals

  • Bridger: provides the communication between the On-board Platform and Remote Platform. It introduces the features of secure communication using an encryption and decryption algorithm.
    Data Aggregator: Provides to help them make better decisions, improve process efficiency and finally, understand performance of the Platform. Consistent evolution, and the usability of the data plays a vital role too.

Team: CAM,BUT,ABO

Challenge: Introducing load to the radar processing chain to test Power-aware radar configuration concept

Context

  • CAMEA traffic monitoring systems (video/radar) are often standalone. The radar sensor is placed in a hermetically sealed box and thus it lacks active cooling capability. As we are not modifying existing HW, the concept would be tuning sensor configuration to keep its quality with parallel power consumption reduction

Goals

  • Within the UC, we are mainly targeting low-power requirements using an AI guided configuration and setup. This is more about tuning specific parts of existing configuration rather than generating complete configuration from scratch. Such systems can be then deployed to the field with possibility of autonomous operation with e.g. battery supply or solar power. Heat dissipation should be reduced as well. To provide reliable test results, introduction of artificial or real load (in terms of detections - both pointcloud and tracking) to the system is necessary

Team: CSY

Challenge: Game of Proof : CLEARSY

Context

  • During last summer, we developed a deep learning model able to classify proof obligations in function of their complexity and the strength to apply to automatically prove them. This model reached an accuracy of 80 %. We demonstrated that a RNN model can manage B-written hypotheses and goals and extract patterns.
    UCS2 is about the part after the automatic proving, when the developer is supposed to manually solve the PO. When a model has already been worked on, there are manual demonstrations that are stored and that can be retried. If the PO is unchanged, the old demonstration can be reused. If it has slightly changed, it must be redone. Sometimes it’s just one command that must be added or removed. This has to be done manually and takes time.

Goals

  • Treat the data to reduce their volume and keep the good things, select a model to train in order to produce manual proof commands.

Team: AVL,AIT

Challenge: Parameter Space Reduction in TCV Use Case

Context

  • Hardware-in-the-loop (HIL) or vehicle test execution is very expensive
    Possible parameter space for generating ADAS/AV test cases is in general huge

Goals

  • Find new approaches for shrinking the parameter search space
    The approaches should be capable to generalize for different scenarios