Distributed Control-as-a-Service with Wireless Swarm Systems"

Size: px

Start display at page:

Download "Distributed Control-as-a-Service with Wireless Swarm Systems""

Gertrude Walsh
5 years ago
Views:

1 Distributed Control-as-a-Service with Wireless Swarm Systems" Prof. Rahul Mangharam Director, Real-Time & Embedded Systems Lab Dept. Electrical & Systems Engineering Dept. Computer & Information Science University of Pennsylvania 1

Discrete and Process Wireless Control 2020 2015 2010 2005 2010 2015 2020 Wireless

2 Discrete and Process Wireless Control Wireless Control/Actuation Wireless Monitoring Wireless Control/Actuation Roadmap from Siemens

3 Our Focus: Industrial Control Systems Natural gas processing plants Oil- refineries Paper pulp manufacturing $120 Billion/Year market PLC Architectures and Software are from the mid-1980s to early-1990

4 Software Issues with Industrial Automation Automotive assembly lines lose over $22,000/minute downtime Systems are rigid, difficult to maintain, operate and diagnose Goal: Plug-n-Play Wireless Automation Control Systems

Advantages of wireless control system architectures 1) Plug-n-Play

controller faults as re-connecting the logical I/O lines of a

2) Compositionality: Enables system evolution through logical

control systems. Suitable for emerging markets.

5 Advantages of wireless control system architectures 1) Plug-n-Play capabilities: Minimizes downtime with efficient recovery from controller faults as re-connecting the logical I/O lines of a wireless backup controllers is seamless. 2) Compositionality: Enables system evolution through logical expansion/ contraction of plants and controllers with composable control systems. Suitable for emerging markets. 3) Runtime adaptation: Control stability and performance are maintained in the presence of node, link and topological changes. 5

6 A Two Disruptive Approaches for Wireless Control Embedded Virtual Machines Runtime abstraction of wireless control Controller tasks migrate across physical nodes Robust to topological changes B Wireless Control Network Fully distributed in-network approach Low computational overhead and simple, static scheduling Enables plant & controller composition Plant s 1 a 1 a 2 s 2 s 3... s p a m v 1 v 2 v 5 v 4 v 3 v 6 v 9 v 7 v 8 v v 10 9 Controller Plant s 1 a 1 a 2 s 2 s 3... s p v 1 v 2 v 5 v 4 v 3 v 6 v 9 v 7 v 8 EVM Plant s 1 a 1 a 2 s 2 s 3... s p v 1 v 2 v 5 v 4 v 3 v 6 v 9 v 7 v 8 WCN a m v 9 v 10 a m v 9 v 10 Published in: IEEE WCPS 2009, RTAS 10, CDC 1 10, CDC 2 10, ACM TECS 11, 6 IEEE TAC 11, CDC 11, ACC 11, CDC 12, IPSN 12, JSAC 12, IISc 13

7 Embedded Virtual Machines For Robust Wireless Control

8 The Core Idea Virtual Task to Physical Resource Mapping V T1 V T2 V T3 V 4 Time t = t 0 N 1 N 2 N 3 V T1 V T2 V T3 V 4 Time t = t 1 N 1 N 2 N 3

9 Runtime Task Management Task Migrate Task instructions Stack Data Associate libraries Control/Schedule/Resource meta data Node 1 Task_Migrate(T2) Node 2 Current Task-set: T1, T2, T3 Current Task-set: T 1, T 2,, T 7

10 Example: When Routing Fails Migrate Primary Controller Actuator Backup Controller Sensors 10

11 Example: When Routing Fails Migrate What if we just used the backup and re-routed the control path? New Primary Controller Routing from Backup Controller to Actuator fails to meet Stability Constraint 11

12 Example: When Routing Fails Migrate Actuator Sensors Migrated Primary Controller Migrated Backup Controller 12

13 Embedded Virtual Machines Distributed runtime system that dynamically selects primarybackup sets of controllers to guarantee QoS given spatial and temporal constraints of the underlying wireless network. Algorithms to program virtual components and maintain functional and para-functional invariants across the system. Focus on controller reliability and fault tolerance

14 The Goal Maintain Functional Invariant Control Law Maintain Para-Functional Invariants Timeliness, reliability, fault-tolerance Control Stability Predictable outcomes in presence of controller / link failures For planned changes Graceful degradation without violating safety For unplanned changes Composability: for multiple plants at runtime Increase functionality or respond failure Adaptive Runtime Resource Re-appropriation Optimization for Dynamic changes in service/throughput Reliability Composition 14

15 EVM Design Flow Design Time with Simulink Design Time Control System Description (Matlab/Simulink, C ) Platform Independent FORTH Description (Common/Control FORTH) Platform Dependent Bin. Format VCM/NanoRK 15

16 EVM Design Flow Domain Specific Language Interpreter Design Time Platform Independent Platform Dependent Bin. Format Control System Description (Matlab/Simulink, C ) FORTH Description (Common/Control FORTH) VCM/NanoRK 01 CONSTANT PORT \ Port assignments /bit-mask name bit-mask name 1 CONSTANT MOTOR 8 CONSTANT FAUCET 2 CONSTANT CLUTCH 16 CONSTANT DETERGENT 4 CONSTANT PUMP 32 CONSTANT LEVEL \ Device control : ON ( mask -- ) PORT C@ OR PORT C! ; : OFF ( mask -- ) INVERT PORT C@ AND PORT C! ; \ Timing functions : SECONDS ( n -- ) 0?DO 1000 MS LOOP ; : MINUTES ( n -- ) 60 * SECONDS ; : TILL-FULL ( -- ) BEGIN PORT C@ LEVEL AND UNTIL ; \ Wait till level switch is on \ Washing machine : ADD ( mask -- ) DUP ON 10 SECONDS OFF ; : DRAIN ( -- ) PUMP ON 3 MINUTES ; : AGITATE ( -- ) MOTOR ON 10 MINUTES MOTOR OFF ; : SPIN ( -- ) CLUTCH ON MOTOR ON 5 MINUTES MOTOR OFF CLUTCH OFF PUMP OFF ; : FILL-TUB ( -- ) FAUCET ON TILL-FULL FAUCET OFF ; \ Wash cycles : WASH ( -- ) FILL-TUB DETERGENT ADD AGITATE DRAIN ; : RINSE ( -- ) FILL-TUB AGITATE DRAIN ; : WASHER ( -- ) WASH SPIN RINSE SPIN ; \ Top-level control 16

17 EVM Design Flow Platform Dependent Binary Design Time Control System Description (Matlab/Simulink, C ) Platform Independent FORTH Description (Common/Control FORTH) Platform Dependent Bin. Format VCM/NanoRK Interp. Task VCM FFM Task VT 1... VT m virtual tasks Task 1 native tasks... Task n nanork 17

18 Embedded Virtual Machine Architecture Apps Motor Control Task Overload Detection Task State Migration Task Focus of EVM Work Kernel Hardware Reserves Reserves Reserves Real-Time Scheduler Task Management Reservations Time Sync RX Microcontroller Peripheral Drivers Network Management RT-Link RX Buffer TX Buffer Power Control Radio Adaptive Virtual Machine Runtime System Parametric Control Task Partitioning Scheduleability Analysis Software Attestation Algorithm Activation Protocol Adaptation Policy Negotiation Data Migration Online Fault Diagnosis nano-rk Sensor RTOS Runtime Parametric and Programmatic Control

19 Wireless Control Network A Simple Distributed Method for Control over Wireless Networks

20 Goal: Use multi-hop wireless networks for closed-loop control Move from open-loop monitoring to distributed control! For the past 40 years control architectures have been based on wired networks Sensors ( ) and Actuators ( ) are installed on a plant Communicate with controller ( ) over a wired network Plant Controller Plant Controller Wired Control System Wireless Control System

21 Route assignment is static Sensors Plant Controller Actuators

22 Control System s view of the Network Channel Plant Controller Plant Controller Channel Route assignment is staac Abstracts away system design to an ideal network Control problems impose strict delay requirements

23 Why we need Distributed Control? Plant Controller Problem I: Changes involve global reorganizaaon Requires significant sogware support (e.g., EVM, Etherware)

24 Networks for industrial automation systems are usually shared among several control loops! Plant Controller Communication substrate is shared! New Plant Problem II: New feedback loops might affect the exisang loops requires full schedule recalculaaon (problem at runame) It is necessary to provide a composable control scheme!

25 Multi-hop Control Networks: Architectures Sensors ( ) and Actuators ( ) are installed on a plant Communicate with controllers ( ) over a wireless network WCN Plant Controller Plant Out- of- network computation Distributed In- network computation

26 Outline: Control-as-a-Service A simple distributed method for control over wireless networks 1. Wireless Control Networks (WCNs) 2. Modeling 3. Synthesis of OpAmal WCNs 4. Robustness WCN 5. Case Study Plant

Standard Feedback Control Schemes Linear- Time Invariant model of the plant: Plant Controller Linear dynamic feedback controllers: Goal: Leverage the computaaonal

27 Standard Feedback Control Schemes Linear- Time Invariant model of the plant: Plant Controller Linear dynamic feedback controllers: Goal: Leverage the computaaonal capability in the network Each node acts as a local linear dynamical controller Resource constrained nodes Small states States of the nodes neighbors are considered as inputs!

28 Linear Iterative Scheme Each node maintains its (possibly vector) state Transmits state exactly once in each step (per frame) Updates state using linear iteraave strategy Example: z 5 = 1 z 2 = 2 v 2 v 5 z 4 = 0.2 v 4 v 3 z 3 = -2 Plant v 6 z 6 = -4.3 v 8 z 8 = 3.2 Initial state

29 Linear Iterative Scheme Each node maintains its (possibly vector) state Transmits state exactly once in each step (per frame) Updates state using linear iteraave strategy Example: Transmit slots z 2 = 2 z 5 = 1 v 2 v 5 z 4 = 0.2 v 4 v 3 z 3 = -2 v 4 informed about its neighbors states v 4 updates its state v 6 z 6 = -4.3 v 8 z 8 = 3.2 Slot Initial 3: 2: 5: 1: 4: 6: v 25 v 64 8 state 3 transmits

30 WCN Modeling Each node maintains state Node state update procedure: From neighbors From sensors Actuator update procedure: WCN From actuator s neighbors Plant Each node acts as part of a dynamical compensator

31 WCN Modeling Network acts as a linear dynamical compensator WCN Structural constraints: Only elements corresponding to exisang links (link weights) are allowed to be non- zero Plant

32 WCN modeling: Closing the loop Overall system state: plant network Closed- loop system: WCN Matrices W, G, H are structured Sparsity constraints imposed by topology! Plant

33 Advantages: Simple & Powerful Low overhead Easily incorporated into exisang wireless networks (e.g., ISA100.11a or wirelesshart) Simple scheduling Each node needs to transmit only once per frame StaAc (conflict- free) schedule No rouang! Easily deals with mulaple sensing/actuaaon points

34 Advantages: Compositionality Adding new control loops is easy! Does not require any communicaaon schedule recalculaaon WCN configuraaons can be combined Stable configuration Plant 1 Plant 2 Plant k

35 Outline A simple distributed method for control over wireless networks 1. Wireless Control Networks (WCNs) 2. Modeling 3. Synthesis of OpAmal WCNs 4. Robustness WCN 5. Case Study Plant

36 Robustness to Link Failures Example u[k] x[k+1]=αx[k]+u[k], y[k]=x[k] y[k] g v 2 w 21 v 1 h w 12 Maximal message drop probability which guarantees MSS, α=2

37 Robustness to Link Failures Example WCN with observer style updates u[k] x[k+1]=αx[k]+u[k], y[k]=x[k] y[k] g v 2 w 21 v 1 h w 12 For α=2, maximal message drop probability which guarantees MSS p max 21% < 25% Approaching theoretical limit for robustness with centralized controllers!

38 Outline A simple distributed method for control over wireless networks 1. Wireless Control Networks (WCNs) 2. Modeling 3. Synthesis of OpAmal WCNs 4. Robustness WCN 5. Case Study Plant

39 WCN demo: Distillation column process control DisAllaAon column control Plant conanuous- Ame model contains 8 states, 4 inputs, 4 outputs DisAllaAon column structure s 3 a 1 a 3 v 4 v 3 v 1 s 1 a 2 s 4 v 2 a 4 s 2 System configuration

40 WCN demo: Distillation column process control DisAllaAon column control Plant model contains 8 states, 4 inputs, 4 outputs WCN contains 4 nodes Stable configuration (obtained after plant discretization): s 3 nodeà node a 1 a 3 sensorà node v 4 v 3 v 1 s 1 a 2 s 4 v 2 nodeà actuator a 4 s 2 Network topology

41 WCN demo: Distillation column process control Process- in- the- loop test- bed Scenario I: v 1 turned OFF/ON

42 WCN demo: Distillation column process control Process- in- the- loop test- bed Scenario II: OpAmal control

43 Winners of the Honeywell Users Group 2011 Industrial Wireless Competition PRECISE Ph.D. Student: Miroslav Pajic. Embedded Sys MS Student: Harsh Jain

Controller-as-a-Service with Wireless Control

Requirements Plant Dynamics Network Topology

44 Controller-as-a-Service with Wireless Control Networks Intrusion Detection Level Monitoring Requirements Plant Dynamics Network Topology Network Synthesis Communication schedule OpAmal Control Robustness Embedding of exisang controllers Wireless Control Network ConfiguraAon RunAme AdaptaAon

45 Thank You

46 Demo 46

Dependable Wireless Control

Dependable Wireless Control through Cyber-Physical Co-Design Chenyang Lu Cyber-Physical Systems Laboratory Department of Computer Science and Engineering Wireless for Process Automa1on Emerson 5.9+ billion