File preview
Tackling New Frontiers in Modeling and Analysis of Cyber-Physical Systems
IMPACT Lab (impact.asu.edu), School of Computing Informatics and Decision Systems Engineering, Arizona State University
Cyber-Physical Systems New Frontiers in CPS Research
1. Spatio-Temporal Interactions: Effect of CPI is felt over time and space More than one independent dimension. Increases complexity of solutions. Example: Drug concentration for insulin infusion can vary over both space and time having different physiological consequences.
Infusion Sites
Effects of Dynamic Contexts
Context dependent environmental changes affect CPS operation Example: Wireless connectivity is affected by environmental changes due to mobility Two models of mobility: 1.Random Way Point—popularly used 2.Levy Walk—recently found to match human mobility patterns
Lessons Learnt: Space and time dynamics may not be decoupled
Complex PDE models with coupling between space and time dependent variables
Lessons Learnt: Hybrid models have limited support for random physical events Unified modeling of events in computing and physical domain
Educational Outcomes and Course Outline
Discretization of space leads to approximation errors
Traditional solutions such as Finite difference time domain have inaccuracies
Cyber-Physical Interactions (CPI)- Interaction of the software with the physical environment
Example: Artificial Pancreas Controller receives feedback from glucose monitor, and administers infusion through pump to control blood glucose level.. CPI: Non-linear spatio temporal Insulin glucose interaction along with transport delays
Requires exact solution of PDEs (often infeasible)
2. Emergent Behavior: Patterns arising out of a multiplicity of simple interactions Theoretically it can not be predicted. However, approximations exist. Example: Simultaneous multi-drug infusion in chemotherapy induces higher rate of cell death than sequential infusion of two drugs.
Drug concentration
Identifying Characteristics of CPSes
Content: Study examples such as artificial pancreas, data centers, ultrasound control systems Outcomes: Students will be able to associate system models to specific examples and use established solutions for the given problem.
Understanding Mathematical Implications of CPS Characteristics
Content: Advance topics in spatio-temporal modeling, PDE solution techniques for free boundary assumptions, non-linear control systems, Lyapunov techniques Outcomes: Students will be well versed in advanced theories of real numbers, non-linear control systems and PDE solving techniques and will apply them to CPS problems
Challenges in CPS Research
CPS Research Goals:
Combining Models of Computation and Physical Dynamics
Content: Hybrid systems, hierarchical formal methods, reachability analysis and model checking Outcomes: Students will be able to apply hybrid modeling to a CPS example and capture emergent behavior through relevant approximations.
Lessons Learnt:
No closed form solutions under Free Boundary Conditions Computation with error bounds are used to estimate CPI.
Prediction of emergent behavior increases complexity
3. Non-Linear Interaction: Physical environment is inherently non-linear Example: Glucose insulin interaction is non-linear in nature Non-linearity caused due to interaction of insulin generated from the pancreas and blood glucose digested by the liver.
Guaranteeing Requirements in CPSes
Content: Automated code generation, Requirements verification through static analysis, hardware implementation of physical environment as a test-bed for CPSes Outcomes: Students will use tools and techniques developed at IMPACT Lab such as Health-Dev, BAND-Aide, to explore different aspects of automated implementation of CPSes.
Experience with Experimental Validation
Content: Design of experiments, Study protocols, consent drafting, IRB issues Outcomes: Students will get a first hand experience with clinical studies and associated protocols through the ongoing collaboration of IMPACT lab with St. Luke’s Hospital. Course Webpage: http://impact.asu.edu/CPSCourseOutline.html
Challenges: 1.How to design reliable software that reaps benefit from the physical environment to meet real time requirements? 2. How to ensure safety, security, and long term sustainable operation of software under dynamic context driven CPI with emergent behavior? 3. How to assure that the implementation of CPS software meets the requirements in real deployments? 4.How to enable experimental validation of CPS software in a simulated deployment which matches desired aspects of the real world?
Lessons Learnt: Often difficult to find solutions to Non-linear systems
Example: Lyapunov functions for most non-linear problems cannot be obtained
Contributors and Acknowledgements
IMPACT Members: Faculty: Sandeep K.S. Gupta, Research Faculty: Georgios Varsamopoulos, Post Doc: Ayan Banerjee, Graduate Students: Priyanka Bagade, , Joseph Milazzo, Zahra Abbassi, Madhurima Pore, Joydeep Banerjee. Funding: NSF CNS #0831544 & IIS #1116385, Intel Collaborators: Food and Drugs Administrations, St. Lukes Hospital
Advanced topics in non-linear control systems Diversion from linearity assumptions
IMPACT Lab (impact.asu.edu), School of Computing Informatics and Decision Systems Engineering, Arizona State University
Cyber-Physical Systems New Frontiers in CPS Research
1. Spatio-Temporal Interactions: Effect of CPI is felt over time and space More than one independent dimension. Increases complexity of solutions. Example: Drug concentration for insulin infusion can vary over both space and time having different physiological consequences.
Infusion Sites
Effects of Dynamic Contexts
Context dependent environmental changes affect CPS operation Example: Wireless connectivity is affected by environmental changes due to mobility Two models of mobility: 1.Random Way Point—popularly used 2.Levy Walk—recently found to match human mobility patterns
Lessons Learnt: Space and time dynamics may not be decoupled
Complex PDE models with coupling between space and time dependent variables
Lessons Learnt: Hybrid models have limited support for random physical events Unified modeling of events in computing and physical domain
Educational Outcomes and Course Outline
Discretization of space leads to approximation errors
Traditional solutions such as Finite difference time domain have inaccuracies
Cyber-Physical Interactions (CPI)- Interaction of the software with the physical environment
Example: Artificial Pancreas Controller receives feedback from glucose monitor, and administers infusion through pump to control blood glucose level.. CPI: Non-linear spatio temporal Insulin glucose interaction along with transport delays
Requires exact solution of PDEs (often infeasible)
2. Emergent Behavior: Patterns arising out of a multiplicity of simple interactions Theoretically it can not be predicted. However, approximations exist. Example: Simultaneous multi-drug infusion in chemotherapy induces higher rate of cell death than sequential infusion of two drugs.
Drug concentration
Identifying Characteristics of CPSes
Content: Study examples such as artificial pancreas, data centers, ultrasound control systems Outcomes: Students will be able to associate system models to specific examples and use established solutions for the given problem.
Understanding Mathematical Implications of CPS Characteristics
Content: Advance topics in spatio-temporal modeling, PDE solution techniques for free boundary assumptions, non-linear control systems, Lyapunov techniques Outcomes: Students will be well versed in advanced theories of real numbers, non-linear control systems and PDE solving techniques and will apply them to CPS problems
Challenges in CPS Research
CPS Research Goals:
Combining Models of Computation and Physical Dynamics
Content: Hybrid systems, hierarchical formal methods, reachability analysis and model checking Outcomes: Students will be able to apply hybrid modeling to a CPS example and capture emergent behavior through relevant approximations.
Lessons Learnt:
No closed form solutions under Free Boundary Conditions Computation with error bounds are used to estimate CPI.
Prediction of emergent behavior increases complexity
3. Non-Linear Interaction: Physical environment is inherently non-linear Example: Glucose insulin interaction is non-linear in nature Non-linearity caused due to interaction of insulin generated from the pancreas and blood glucose digested by the liver.
Guaranteeing Requirements in CPSes
Content: Automated code generation, Requirements verification through static analysis, hardware implementation of physical environment as a test-bed for CPSes Outcomes: Students will use tools and techniques developed at IMPACT Lab such as Health-Dev, BAND-Aide, to explore different aspects of automated implementation of CPSes.
Experience with Experimental Validation
Content: Design of experiments, Study protocols, consent drafting, IRB issues Outcomes: Students will get a first hand experience with clinical studies and associated protocols through the ongoing collaboration of IMPACT lab with St. Luke’s Hospital. Course Webpage: http://impact.asu.edu/CPSCourseOutline.html
Challenges: 1.How to design reliable software that reaps benefit from the physical environment to meet real time requirements? 2. How to ensure safety, security, and long term sustainable operation of software under dynamic context driven CPI with emergent behavior? 3. How to assure that the implementation of CPS software meets the requirements in real deployments? 4.How to enable experimental validation of CPS software in a simulated deployment which matches desired aspects of the real world?
Lessons Learnt: Often difficult to find solutions to Non-linear systems
Example: Lyapunov functions for most non-linear problems cannot be obtained
Contributors and Acknowledgements
IMPACT Members: Faculty: Sandeep K.S. Gupta, Research Faculty: Georgios Varsamopoulos, Post Doc: Ayan Banerjee, Graduate Students: Priyanka Bagade, , Joseph Milazzo, Zahra Abbassi, Madhurima Pore, Joydeep Banerjee. Funding: NSF CNS #0831544 & IIS #1116385, Intel Collaborators: Food and Drugs Administrations, St. Lukes Hospital
Advanced topics in non-linear control systems Diversion from linearity assumptions