Structural Optimization Of Supervisory Controllers In Automated Manufacturing Systems

In the background of Industry 4.0,industrial production is nowadays driven by the global competition and the fast adaptation of production to the ever-changing market requests.Traditional flow production systems hardly tackle such modern challenges and are gradually updated to automated manufacturing systems(AMSs)to stay competitive.Such transformation of industrial manufacturing requires digitalization and exploitation of potentials of new technologies.Consequently,it attracts many researchers and engineers to model,analyze,and control AMSs.Usually,an AMS is an integration of computer-controlled machine devices,buffers,robots,automated guided vehicles,and other material-handling equipments.One obvious benefit of such systems is that they can accomplish complex assignments and produce qualified products faster and more efficiently under well-defined controls.AMSs can be regarded as a typical kind of resource allocation systems.They are characterized by the high degree of resource sharing by concurrent processes,for the sake of high flexibility and low cost.Thus,it brings the first notorious logical problem,i.e.,deadlocks,because all involved concurrent processes would compete for the limited resources.This leads to the development of supervisory control techniques(SCTs),to synthesize supervisors that enforce the system to be live.Meanwhile,the synthesis of a supervisor may suffer from computational and structural complexity,leading to an oversized one that may hardly be implemented.This brings another critical control problem,i.e.,supervisor simplification.So far,most SCTs are developed for general AMSs which do not have complex process structures and resource usage,while only a little work deals with complex systems.Supervisor simplification is also intuitively and naturally associated with structural and/or reachability analysis.In fact,supervisor simplification based on structural analysis is theoretically cumbersome and that based on reachability graph is only suitable for small scale systems.Therefore,liveness analysis,liveness enforcement,and supervisor simplification for complex systems remain to be explored.Most AMSs can be modeled by discrete event systems(DESs)as the discrete event-driven property of DESs can describe the state transfer of AMSs.As a powerful mathematical tool in DESs,Petri nets(PNs)have been extensively used to model,analyze,and control AMSs.They can provide not only the abundant structural information but also the rigorous mathematical expressivity.This thesis focuses on supervisor synthesis and its simplification for complex AMSs using PNs as models.The main results are shown as follows.First,for a kind of AMSs represented by Hierarchical Augmented Marked Graphs(HAMGs),this work studies their liveness analysis and liveness enforcement.First,based on such a PN structure,it proposes a novel characterization of siphons,i.e.,Type I and Type II.This characterization of siphons is further proved to be a crucial and essential description for both synchronization and flexibility.Then,event circular-wait is proposed to describe the occurrence of deadlocks in HAMGs.Its counterpart is resource circular-wait in flexibility.It shows that the former is a more general and accurate characterization of deadlocks.The liveness of HAMGs can be attributed to the absence of undermarked siphons.Finally,a unified liveness-enforcing supervisory control is established based on new characterization of siphons with the aid of generalized mutual exclusion constraints(GMECs).Supervisor simplification based on inequality analysis is provided for structure simplicity.Second,for a kind of AMSs with time-varying manufacturing mode,represented by Timevarying Systems of Sequential Systems with Shared Resources,this work studies the implementation of time-varying specifications and liveness-enforcing specifications.It proves that time-varying specifications and liveness-enforcing specifications can be enforced independently in such systems.For the former,monitors with self-loops are designed.For the latter,it provides a comprehensive and comparative study on invariant-based linear constraints in supervisory control of time-varying systems.First,generalized linear constraints(GLCs)which contain both marking and firing vectors are utilized to solve state/event separation problem,in contrast with GMECs which solve state forbidden problem.They are denoted as event-and state-based control respectively.Then,weighted event-and state-based control methods are proposed.Weight coefficients are introduced to design both GLCs and GMECs,to enhance their specifications' expressivity capability in these respective approaches.Above linear constraints can be implemented via place invariants(P-invariant)in PNs.Finally,the approach shows that the weighted state-based control has an obvious advantage in improving permissiveness for the systems whose processes can accommodate more tokens.Third,for the supervisor simplification problem in AMSs,this work begins to explore the relationship between the supervisor simplification and the system structure,i.e.,the nature of structure-dependence-independence of supervisor simplification.It regards the supervisor simplification as an independent work.When dealing with livness-enforcing supervisor simplification,there are two typical methods,i.e.,elementary-siphon-based technique and the inequality analysis.The former is based on structure analysis while the latter is on algebra analysis.It discovers that the former is a special case in the latter in supervisor simplification,i.e.,the dependent siphons in the former correspond to dependent inequalities in the latter,but the inverse is not true.When extending to specification simplification,the P-invariant-based method is developed from structural perspective.Nevertheless,it demonstrates that the P-invariant-based method can be explained and replaced by inequality analysis.As a consequence,it demonstrates that supervisor and specification simplification can be independent from structural information.Meanwhile,inequality analysis is considered as the simple and general,convenient and efficient supervisor simplification method in general PNs.Fourth,for the supervisor simplification problem in AMSs,this work further studies the unified supervisor simplification technique in general PNs.Static and dynamic partitions are proposed,so as to reduce the size of supervisor by partitioning inequalities,i.e.,GMECs,into redundant and necessary ones.The former considers the specifications themselves and exhibits higher efficiency;while the latter takes into account both specifications and reachable markings and achieves more economic supervisors.In an incremental manner,it develops Type I supervisor simplification method which has a compromised efficiency and complexity.In an integrated manner,it proposes Type II supervisor simplification method that further improves efficiency and eases computational complexity.All techniques proposed here can be realized in an algebraic manner.Furthermore,such approach provides a general framework to unify simplification methodologies based on structural analysis.Comparison and discussion as well as illustrations are presented for the sake of clarification as well as demonstration.