Research On The Architecture And Self-management Optimization Strategies Design For Dependable Cyber Physical System

Cyber Physical Systems(CPSs)is an organic intelligent system with loose physical structures and rigorous logical organization.It involves various cross-disciplinary technologies such as computing,networking,control,physical processing,and application domain knowledge.Generally,the CPS applications should satisfy numbers of critical requirements,i.e.safety,dependability and timeliness.Meanwhile,CPS continuously generates so large amounts of complex data that no administrator can correctly understand the subsystems’ statues and promptly instruct their actions in time.Therefore,it is necessary to construct a self-management CPS(SCPS)to interact with the dynamic physical world autonomously and flexibly.Moreover,the SCPS also can automatically recover from various anomalies without significantly affecting the normal business.With the rapidly growing scale,the complexity and delay of SCPS are out of control;the uncertainty and the dependability issues of self-management decision-making are becoming more and more serious.It urgently needs to explore systematic design and dependable(self-)maintenance solutions to integrate the cross-disciplinary technologies.The physical world is a flexible and parallel system with full of uncertainties and randomness.To match the degree of parallelism,SCPS usually contains a large number of(heterogeneous)sensors and actuators.In order to flexibly interact with physical space in different scenarios,SCPS needs to organize different embedded subsystems on demand in different topologies.However,few models and implementations can well satisfy the requirements of dynamical behavior.Moreover,it needs to reproduce the causal relationship of events in the cyber space for decision making and failure diagnose.Aiming at these requirements,we propose an actor meta-model to describe the subsystems of SCPS,and model the interaction of subsystems with(stochastic)activity network.In order to tame the complexity,the composability & compositionality(C&C)constraints of the actor model are studied and the C&C of actor under the failure mode are improved.The composition schemes of runtime reorganization are summarized.Then the algebraic theory is employed to analyze the satisfiability rules of attributes and requirements of different compositions.To explore the Model Driven Engineering(MDE)based SCPS development,the Architecture Analysis and Design Language(AADL)is employed to implement and validate the actor model.Based on the integrated activity model implemented with AADL,formal model transformation methods are applied to transfrom the meta-model(a.k.a actor model)to analytical models,such as FTA,CTMC,automata.So that we can integrate the existing analytical tools and sysmatically validate SCPS model,study the co-effect between error behavior and normal behavior.Finally,aming at the requrirements of dynamic adaptation between architecture and strategy,this thesis firstly proposes the idea of architecture-strategic collaborative analysis.Then analyzes the flexibility and reliability of different solutions.The simulation resoults show that the hierarchical decentralization solution has the highest reliability and stability of behavior.Based on MDE evaluation results,we propose a hierarchical-decentralized SCPS architecture and a compositional self-management framework for local embedded systems.We also prove that the global complexity of the SCPS can be reduced by decoupling the architecture control logic and normal functional logic.To overcome the drawbacks of the global absolute time hypothesis,we propose a relative time based solution to guarantee distributed timing consensus and to support decentralized observations and actions.Aiming at the defect of the actor model on fault propagation hypothesis,a lightweight container isolation solution is designed to provide an ideal runtime environment for the actor and to block the fault propagation through shared resources.Moreover,a multi-level FDIR scheme is integrated into the container to limit fault propagation and to improve the self-healing ability of actors.The testing results should the self-healing solution can deal with both the software failures and the transient hardware failures.Based on the hierarchical-decentralized architecture,to further handle the uncertainty of prophetic decision-making,we proposes a contract-based self-management solution to coordinate the cooperation between distributed embedded subsystems.In this thesis,we model the decision sequences in the contract specification with directed acyclic activity network,and the topology of SCPS with a doubly weighted vertex-colored graph.The contract optimization is divided into three coherent stages according to the stages of feedback loop,and a multi-objective progressive contract optimization solution is proposed.In the contract making stage,the extended Dijkstra algorithm is applied to verify the practicability of contracts.At the advice refinement stage,the improved NSGA-II algorithm is employed to optimize the organization of the actors.At the decision processing stage,the composition schemes and decision specification are flexibly applied to guide the cooperation of actors.Furthermore,the processing time alignment algorithm is introduced to synchronize the progress of decision branches.It can reduce the fluctuation of both the reliability and processing time,and also improve the stability predictability and controllability of SCPS’s behavior.The simulation results reveal the main factors for behavioral stability and contract optimization,and show the effectiveness of the progressive optimization solution.Finally,the self-similar actor is designed to further simplify the complexity of runtime actor management.The self-adaptive system is built and tested on the real world embedded system.We also analyze and verify the efficiency of FDIR solution and message management of the container,as well as the efficiency of contract progressive optimization,and the reliability and stability of decision execution.The test results also show the effectiveness of the SCPS design solution,system optimization solution,and self-management solution,etc.The log shows our SCPS solutions can achieve impressive dependability when actors fail or even are misdiagnosed.In the last chapter,we summarize the challenges that complexity and uncertainty pose to SCPS design and maintenance.And two principles are presented for taming the complexity and systematic advices are given to overcome uncertainty in design and decision making.Finally,based on the inspiration during current design,we propose a conceptual solution by taking advantages of the model reusability between MDE and model@run.time,as well the similarity of methodology between the system design and decision specification making.The conceptual solution tries to build the self-evolution CPS by integrating the mathematics based validation,simulation based evaluation and data driven analysis through feedback loops.