Supervisory control synthesis for large-scale infrastructural systems

Abstract

In the coming decades, numerous locks and bridges in the Netherlands have to be renovated or replaced, as they reach their end of life cycle or have capacity problems. In the past, these infrastructural systems have been engineered, built, and maintained on a project basis, resulting in a large variety of solutions to the same engineering problem. Furthermore, more and more functionality of these infrastructural systems is being automated. Therefore, Rijkswaterstaat, part of the Dutch Ministry of Infrastructure and Water Management, is seeking methods for modularization and standardization to increase quality, increase evolvability, and decrease life-cycle costs of future infrastructural systems. The control system of these infrastructural objects is identified as one of the subsystems having a significant impact on the availability and reliability of the object. The aim of this thesis is to show that Supervisory Control Theory (SCT) as initiated by Ramadge-Wonham is very suitable for designing control systems for these large infrastructural systems. SCT is a research area focused on developing model-based techniques for systems for which the uncontrolled behavior of the components is given and the task of the engineer is to restrict this behavior such that the system only exhibits desired behavior. SCT allows the engineer to specify what the system should do, not how the system should achieve this. Therefore, this synthesis-based engineering method may contribute to the aim of Rijkswaterstaat of increasing the quality and evolvability of control systems. In the past, applying supervisory control synthesis to industrial-size systems was often limited by the computational power, both in time and in memory. Recent developments from the literature as well as the characteristics of infrastructural systems are utilized in this thesis to synthesize supervisors for these systems. The thesis has the following four main contributions. First, monolithic synthesis of a basic lock, namely Lock III located at Tilburg, the Netherlands, showed not only that supervisory control synthesis is possible for these systems, but also that the synthesized supervisor does not impose any additional restrictions on the system besides the formulated requirements, i.e., the requirements together with the plant models form a maximally permissive, controllable, and nonblocking supervisor. Sufficient model properties are proposed to verify whether a given set of plant and requirement models can already act as a supervisor. If so, synthesis can be skipped, saving computational resources. Second, in case synthesis still needs to be performed, several divide-and-conquer synthesis techniques exist, such as modular, decentralized, and multilevel synthesis. The latter one closely resembles decomposition of a system, what is often done by engineers to cope with the size of the system. This thesis proposes a method to transform a given set of plant and requirement models into a multilevel system needed as input for multilevel synthesis. The method uses Dependency Structure Matrices and clustering algorithms to identify such a multilevel structure. This allows for easy application of multilevel synthesis to infrastructural systems, and therefore opens the possibility to synthesize supervisors for large infrastructural systems. Third, several divide-and-conquer methods like multilevel synthesis ease the computation of local supervisors, but no longer guarantee that the complete system is nonblocking. A coordinator needs to be synthesized to resolve any conflict between supervisors. A compositional coordinator synthesis method is developed to synthesize such a coordinator efficiently. The method deploys known abstractions, like for example bisimulation semantics, synthesis equivalence, and variable unfolding, to simplify the models until a single one is obtained. Based on this abstracted model, a coordinator is synthesized with monolithic synthesis. Finally, this coordinator is refined back to the original model to resolve the conflicts between the synthesized supervisors. Fourth, several case studies with infrastructural systems have shown that the formulation of models has impact on the applicability and computational gain of the previously mentioned contributions. As modeling is an art, obtaining the ‘right’ model formulation of the system together with its requirements is nontrivial. Several modeling guidelines are proposed in this thesis, based on experience. Proposing these guidelines contributes to bridging the gap between theoretical research and industrial application of supervisory control synthesis. In 2017, 30 years of supervisory control theory of Ramadge-Wonham have been celebrated. Research presented in this thesis shows that the maturity level of this theory has reached a critical point where it is ready to be fully embraced by industry, as supervisors can be synthesized for large-scale infrastructural systems and not just for academic toy problems. A challenge lies ahead to support the industry in adapting synthesis-based engineering methods for their supervisory control systems.