About

MoDeS3 stands for Model-based Demonstrator for Smart and Safe Systems. The main goal is to demonstrate the many cool and innovative ways in which open source modeling tools can be used for systems development in the age of Internet-of-Things.

Our case study is a railway system: users can control trains arbitrarily as long as it is not dangerous. Accidents and dangerous situations are detected using sensors embedded into the track: they sense the trespass of the trains and send this information to the controllers. It is important to note that this is just local information, so we have to ensure that it will be shared between the components. We employ six BeagleBone Black (BBB) embedded computers to run the safety logic, configured as a distributed system, where each BBB is responsible for some track sections.

Overview of MoDeS3

As you can imagine, the real software engineering challenge is how to develop the distributed safety logic. We use the open source Yakindu Statecharts tool to design the software components. It is much easier than manual programming, as it provides a code generator to produce C or Java code. The Gamma framework supports the construction of the distributed logic from the components and it also provides formal verification, testing and runtime verification capabilities.

The Gamma framework supercharges the expressive power of Yakindu models, we have developed custom validation and verification rules. For this purpose, we used the open source VIATRA Query engine. Our VIATRA Queries can be used to analyse the well-formedness of the models, for example to check if the state chart is deterministic and complete. This turned-out to be a useful feature as many design time errors were found, well before deployment and debugging even began!

In addition to the validators, the Gamma framework also provides model transformations to generate formal models from the state chart models. These formal models, together with associated model checking tools such as UPPAAL, are used to check the deadlock freedom and reachability of the states in the model. In the future, we plan to work more in this direction, in order to automatically analyse other properties such as the fault-tolerance property. We developed the model transformations using the VIATRA framework.

To make distributed systems work in practice, we need communication channels. For this project, we are going to use MQTT for its simplicity, reliability and flexibility. The open source Eclipse Paho framework helps us in establishing and maintaining the communication between the components.

In order to provide system-wide safety guarantees, we plan to build an additional layer on top of reliable communication channels, to facilitate runtime monitors. These smart components will evaluate the behaviour of the local components, and run locally on the PRU 32-bit microcontrollers of the BBBs. They will analyse if the communication works correctly and there is no problem within the controller itself.

Results and education

To augment local monitors, the overall system status will be monitored using computer vision techniques. For this purpose, we attached a camera to a stage above the tracks that will observe the movement of the trains. The video stream is processed by OpenCV, a state-of-the-art open source computer vision library. We have implemented train recognition algorithms to detect the position of the trains. We are now integrating deep learning techniques to recognize the trains without using markers. Our innovative solution combines the intelligence of neural networks with the logic and expressive power of graph models.

The combination of local monitoring data and computer vision data will be aggregated on the system level and processed using complex event processing (CEP). The role of this high-level monitoring technique is to integrate multiple monitoring data sources and make sure that if the distributed safety logic does not work correctly, this additional level of logic can still bring the system to a safe state. Our system-level safety framework will be built using an open source complex event processing engine called VIATRA-CEP.

Videos related to the project are available on YouTube.

The MoDeS3 project  is supported by the MTA-BME Lendület Cyber-Physical Systems Research Group, IncQuery Labs Ltd. and Quanopt Ltd.