Visible to the public File preview

Spatial variation of temperature

EEG BP

Tackling New Frontiers in Modeling and Analysis of Cyber Physical Systems
EKG SpO2 Base Station Motion Sensor

Ayan Banerjee and Sandeep K.S. Gupta IMPACT Lab, CIDSE Arizona State University x = cos(θ )
.

y = sin(θ )

.

θ = w

.

Cyber-Physical System Model
Computing System Computing Requirements
Reliability Accuracy Throughput Latency Safety

Physical System Continuous physical process Actuator Physical Requirements

1. 2. 3. 4. 5.

Control Algorithm

Control Algorithm

Sensor Unintended Side Effects

1. Safety 2. Energy efficiency Dynamic contexts 3. Low carbon Random footprint

Multi-dimensional Partial Differential Equations

Processes

Cyber-Physical Interactions (CPI) SpatioTemporal Simplifying Assumptions
Linearity Time Invariance

Aggregate Effects
Low dimension Determinism

Dynamic Contexts
Independence of Computation & Physics Ignore emergent behavior

Challenges to CPS Modeling
Model based safety analysis and verification under context driven spatio-temporal aggregate cyber-physical interactions Hybrid models of CPI, theoretical and simulation based analysis, and automated synthesis of implementation from models

• Linearity assumptions are invalid
– Non-linear control systems

• Spatio-temporal considerations
– Solution of partial differential equations

• Dynamic contexts affect CPI
– Unified modeling of physical and cyber events

• Emergent behavior
– Theoretically unpredictable

Emergent Behavior
CPSes are complex systems - Emergent behavior: patterns arising out of a multiplicity of relatively simple interactions
Emergence cannot be predicted

Approximation of Emergence is necessary

-

Example: Multi drug interaction

Drug concentration for multi-channel infusion
50 45 40 35

Single Drug:
∂d (r, t) + ∇(u d (r, t)) = ∇(D(r)∇d (r, t)) + Γ(r )(d B (t) − d (r, t)) − λ d (r, t) ∂t

Infusion sites

Y - Axis

d – drug concentration at a distance r at time t, D and λ are constants Multi Drug Approximation:

30 25 20 15 10 5 0

∂d1 (r, t) + ∇(u d1 (r, t)) = ∇(D(r)∇d1 (r, t)) + Γ(r )(d B (t) − d (r, t)) − λ (d1 (r, t) + d 2 (r, t)) ∂t ∂d1 (r, t) + ∇(u d1 (r, t)) = ∇(D(r)∇d1 (r, t)) + Γ(r )(d B (t) − d (r, t)) − λ (d1 (r, t) + d 2 (r, t)) ∂t

Time = 10s 500s 100s

Simultaneous solution of free boundary problems

0

5

10

15

20

25

30

35

40

45

50

X - Axis

CPS Modeling Course
• Defining CPSes
– Study systems models of CPSes in different domains

• Control Systems
– Focus on non-linear time variant analysis (Lyapunov)

• Abstract Mathematics
– Theory of real numbers, vector spaces, convexity theory

• Differential Equations
– Exact and finite error solutions of non-linear partial differential equations

• Formal Methods
– Stochastic hybrid automata and reachability theory

• Theory of Emergence

Solutions and Tools
• Spatio-Temporal Hybrid Automata
– Hybrid model checking for CPS
• Applied to drug infusion systems

• BAND-Aide: Simulation analysis of CPI
– Applied on body sensor networks and data centers

• Evaluation of CPI under Dynamic contexts
– Applied on wearable infusion pumps

• Health-Dev: Safety assured automatic code generation for healthcare systems
Effective characterization of CPI in some mathematical form

Thank You