Scheduling of the FlexRay static segment for robust controller integration Stefano Longo, Tingli Su, Guido Herrmann Senior Member, IEEE, Phil Barber and Uwe Gerlinger Abstract—In this paper, we consider the optimal and robust communication scheduling of the static part of the automotive communication protocol FlexRay from the system dynamic and control viewpoint. In general, given a plant and a controller, we design a communication policy between ‘nodes’ (or ECUs) that minimizes the performance degradation introduced by the communication bus, using an H∞ framework. The resulting combinatorial optimization is performed by a fast versatile algorithm that can guarantee schedule calculation close to the optimum and that can easily handle the constraints imposed by FlexRay (such as bandwidth demands on selected ECUs). The theoretical results are validated using commercial FlexRay simulation products. I. I NTRODUCTION The number of computer-based functions embedded in today’s vehicles like adaptive cruise control, traction control, stabilization control and active safety systems has increased exponentially. These functions, implemented in Electronic Control Units (ECUs), are not stand-alone in the sense that ECUs need to exchange information. For example, the vehicle velocity estimated by an engine controller node or a wheel sensor node needs to be known by the steering node in order to control its stiffness or the windscreen wiper node in order to control its speed [1]. To give an idea of the complexity, it is sufficient to consider that a modern vehicle can have up to 70 ECUs that exchange up to 2500 signals [1]. Information is exchanged through a shared communication medium (bus) that inevitably place constraints on the amount of communication supported (finite bandwidth). Signals need to be time multiplexed according to some scheduling mechanism. It is clear that, with the increasing complexity, traditional methods for scheduling are no longer adequate. The scheduling optimization method we propose can be applied in general to any networked system under a Time- Triggered (TT) protocol. In a TT protocol, as opposed to Event-Triggered (ET) protocols, the exchange of information is solely dependent on the progression in time and therefore allows pre-planning of a particular periodic communication sequence at the design stage. Such protocols are for example those used by TTCAN or FlexRay. Although our method is more general, we concentrate here on the special case of scheduling of the static (TT) part of FlexRay. The reason being the fact that FlexRay is likely to become soon the de facto standard for automotive communication and a sys- tematic method for optimizing its communication has not yet been proposed. TT communication is one of the main features of FlexRay that makes it better suited to modern automotive applications compared to the old CAN. S. Longo is with the Faculty of Electrical and Electronic Engineering, Im- perial College London, London, UK. s.longo@imperial.ac.uk T. Su and G. Herrmann are with the Faculty of Mechan- ical Engineering, University of Bristol, Bristol, UK. {mexts, g.herrmann}@bristol.ac.uk P. Barber is with Jaguar and Land Rover Research, W/2/021 Engineering Centre, Whitley, Coventry, UK. U. Gerlinger is with Vector GB Limited, Rhodium, Central Boulevard, Blythe Valley Park, Solihull, Birmingham, UK. It is important to point out here that the optimal and robust scheduling we are proposing takes into account the dynamics of the control system. For instance, a number of methods for optimal scheduling of the static segment of FlexRay can be found in [2]–[5]. In those cases, the optimization is performed in terms of minimizing message jitter, synchronizing signal-to-task information passing and maximizing the number of free slots (for reduced bus load- ing). All of this is perfectly valid and useful from the network optimization point of view. However, the dynamics of the control system are completely ignored. The timing of message exchange in the network (especially sensor and actuator signals) is crucial for the performance of the overall closed-loop system dynamics. What we propose instead is a method that schedule messages in order to optimize the performance of the control system rather than the network itself. Hence, our definition of ‘optimal scheduling’ in this case is different from the one used in the literature. The design of optimal and/or robust schedules presented so far [6]–[9] is possible only for a discrete-time model of the plant. In robust control problems, it is more natural to consider a disturbance as a continuous-time signal since the controlled system evolves in continuous time. Hence, it is preferable to calculate the norms of the continuous-time systems. On the other hand, the scheduling mechanism is inherently a discrete-time process because the connections to sensors, actuators and demand signals switch according to the communication sequence. The overall closed-loop dynamics are therefore those of a hybrid system made up of a continuous plant and discrete scheduler and controller. We start the analysis from a continuous-time system model connected to a given discrete-time controller via a network with communication constraints modeled as discrete switch- ings. The continuous plant with Zero-Order-Hold (ZOH) and sampler form a periodic system with period equal to the sampling time. The lifting technique for continuous- time periodic systems [10], [11] is used to ‘lift’ continuous- time signals into discrete-time signals. We will call this the continuous-time lifting to distinguish it from the discrete-time lifting used for periodic discrete systems [11]. Hence, the H ∞ -norm of the networked control system can be calculated. This cost permits subsequent optimization of the schedule (which is the communication sequence) that turns out to be a combinatorial problem. We exploit the features of the Particle Swarm Optimization (PSO) algorithm, re-adapted for a discrete-variable search space, to solve this problem. The PSO-based algorithm exploits the benefits of partitioning the problem into smaller sub-problems resulting in being faster and more effective. Theoretical results are validated via experiments performed on the FlexRay simulation tools provided by Vector Informatik GmbH. Our setup consists of eight ECUs, connected by a FlexRay bus, performing a realistic vehicle control algorithm in a real-time simulation environment. Results show how na¨ ıve scheduling is not optimal and how our method for bus optimization gives a clear overall performance improvement. 2011 IEEE International Conference on Control Applications (CCA) Part of 2011 IEEE Multi-Conference on Systems and Control Denver, CO, USA. September 28-30, 2011 978-1-4577-1063-6/11/$26.00 ©2011 IEEE 1487