Wireless Control Systems - from theory to a tool chain, Mikael Björkbom, Aalto University

• 1. Mikael BjrkbomWireless Sensor and Actuator Networks for Measurement and Control Phase IIWireless Control Systems - from theory to a tool chainAalto UniversityDepartment of Communications and Networking Control Engineering GroupKTHRadio Communication Systems GroupAutomatic Control Group Royal Institute of Technology KTH

2. Wireless Automation: ControlCommunication affects control performance-> Control should be robust to problems in the networkRoyal Institute of Technology KTH 3. Nordite WISA ProjectQuality of serviceRequirements forcurrent control algorithmsData fusion Increase jitter marginPID Controller tuningand tolerance to errorsNew control algorithmsWireless automation systems Increase robustness Coexistence protocolsPerformance ofDecrease jitterMulti-path routing (mesh)current wireless networksSynchronizationRoyal Institute of Technology KTH 4. Workpackages WP1: Reliable and secure communication protocols for wireless automation WP2: Communication constrained reliable control WP3: Implementation of WiSA toolchain WP4: Project managementAalto KTHRoyal Institute of Technology KTH 5. WISA Phase I & IIWISA Phase I WISA Phase II Control, data fusion and networking algorithms, testbeds and simulation tools Control and Wireless Design data fusion networking toolsCross-layer designTool chainRoyal Institute of Technology KTH 6. Results: Toolchains PiccSIM Simulation of wireless control systems WirelessTools Planning of wireless network schedule PROSE Node and simulated network Royal Institute of Technology KTH 7. WP 1: Reliable and secure communication T1.1. Interference avoidance and dynamic spectrummanagement Time and frequency domain methods Adaptive frequency hopping T1.2. Reliable networking SIRP, Antenna switching Tools for scheduling T1.3. Sensor and network monitoring, fault detection,and fault recovery Fault detection part is partly missing Fault recovery: Code dissemination toolRoyal Institute of Technology KTH 8. WP2: Communication constrained reliable control T2.1. Communication-aware data fusion and control New data fusion schemes Network jitter aware PID tuning rules T2.2. Control structures, architectures and scalability Impact of MAC on control and data fusion were analyzed Tuning of PID controllers for distributed MIMO systems T2.3. Adaptive and robust control Delay adaptive Internal Model Control based tuning Network performance adaptive controller Royal Institute of Technology KTH 9. WP3: Implementation of WiSA Tool Chain T3.1. Automated implementation of routing protocols This was not accomplished! There is no automation in the development of routing protocols PROSE tool for hardware in the loop simulation T3.2. Automated control algorithm implementation Part of PiccSIM T3.3. Design tools and interfaces for the WiSA tool chain Part of PiccSIM T3.4. Demonstrator development Several demo sessions were arrange (including NORDITE workshop)Royal Institute of Technology KTH 10. WP 1/T1.1-T1.2: Results Objective: wireless sensor nodes should be able tocommunicate in a reliable fashion despite bad channelconditions (interference, fading). We aim at improving reliability by means of: Interference Avoidance through Dynamic Spectrum Access Frequency Hopping Channel Coding Dynamic Spectrum AccessFrequency HoppingChannel CodingAntenna SwitchingRELIABILITYHybrid ARQ Receiver diversityRoyal Institute of Technology KTH 11. WP 1/T1.1: Dynamic Spectrum Access An Example: Experimental Comparison of DSA schemes: Spectrum Holes in the Time domain Performance of DSA in the time domain depends heavily on channel conditions: Energy increased of up to 5 times forhigh interference! DSA in the frequency domain (channel selection) requires larger energy for spectrum sensing but allows to avoid interference: By selecting the communication channeleffects of interferenceSpectrum Holes in the Frequency domain can be mitigated Royal Institute of Technology KTH 12. WP 1/T1.2: Spatial diversity TABLE I CHANNEL PARAMETERS Tap CHANNEL 1CHANNEL 2RelativeRelative tapRelative tap Relative taptap delayamplitude delay [ns] amplitude[ns][ns] [ns]100 0-0.12 20 -0.9 20 -0.63 30 -2.6 50 -2.94 40 -3.5 100-5.85100 -6.7 150-8.76300 -17.9200-11.6 Time Diversity Approaches for 0.1km/h and 1km/h 1 0.950.9 Packet Delivery Ratio 0.85 Pure Time Diversity (0.1km/h) Elektrobits: Channel Emulator PropSIM-c20.8 Piggybacking (0.1km/h) Switch if No Acknowledgement (0.1km/h) 0.75 Piggybacking (1km/h) Pure Time Diversity (1km/h)0.7 Switch if No Acknowlodgement (1km/h)Multiple receiving antennas: 0.65 26% increase in packet delivery ratio0.6-90-85 -80-75-70 -65Mean RSSI (dBm)12 Royal Institute of Technology KTH 13. WP 1/T1.2: Spatial diversity Antenna switching and receiver selection diversity 10 packets/s Dual Antenna System Receivers Array (4)Bridge 23.3m, Receiver sensitivity = -94 dBm 10.90.8PACKET DELIVERY RATIO0.70.60.50.40.3 Receiver Array0.20.1 012345 67 8Links (1-8)Royal Institute of Technology KTH 14. Performance of Multi-Channel MAC Protocols Performance of G-McMAC analyzed and compared to other existingprotocols G-McMAC outperforms other protocols with respect to delayregardless of the used parameters G-McMAC achieves the highest throughput in many cases. Royal Institute of Technology KTH 15. Time-Synchronization in Multi-Channel WSN Multi-Channel Time-Synchronization (MCTS) protocol Time-synchronization Critical for many WSN applications, e.g. control Enables efficient communications and deterministic operation Multiple channels can be used simultaneously in order to minimize convergence timeRoyal Institute of Technology KTH 16. WP 2/T2.1 communication aware data fusion and controlControl over WirelessHART networks PData stream characteristics:WirelessHART network Slotted time Minimum transmission delay Time-varying latency, loss CMany analysis tools and control design techniques, but no perfect match theory most complete for linear-quadratic controlHere: explore sampling interval as interface parameter in co-design. Royal Institute of Technology KTH 17. Realiable real-time challengeMeeting hard deadlines on unreliable multi-hop networkMaximize deadline-constrained reliability (the timelythroughput)ii Royal Institute of Technology KTH 18. WISA-II solutionsFocusing on WirelessHART-compliant solutionsNew theory, algorithms and software for networkscheduling minimize multi-source data collection delay maximize deadline-constrained reliability for unicast joint routing and transmission schedulingLimits of performance, rules of thumb, and optimalalgorithms Royal Institute of Technology KTH 19. WP 1/T1.2 : ConvergecastGiven: sensors with single packet to send at time zeroFind: schedule that delivers all packets to sink (in an optimal fashion)Key operation WirelessHARTssensing-compution-actuation cycle: Royal Institute of Technology KTH 20. WP 1/T1.2 : ConvergecastOptimal convergecast on treesProposition. The minimum evacuation time for a wireless HARTnetwork with tree topology is max(N, 2Nmax-1) timeslots, whereNmax is the number of nodes in the largest subtree.Also here, we can characterize the channel-latency tradeoff.Efficient (O(N2)) time-optimal policies, channel-limited case slightlyharder.Royal Institute of Technology KTH 21. WP 1/T1.2 : ConvergecastIf links are unreliable, then the complete operation might fail.Observation. If links fail with probability pl, convergecast fails withprobability (1-pl)S where S=# transmissions in the schedule.For line with N nodes, S=N(N+1)/2Schedule quickly becomes unreliable!Several simple ways of improving reliability of a given schedule duplicating each slot, repeating schedule, Need methods for quantifying the resulting latency distributions. Royal Institute of Technology KTH 22. Optimal co-designUnderstanding what controllers need, and what network can provideKey result: optimal co-design is modular, can be computed efficientlydeadline-constrained maximum reliability and control under lossoptimal parameters found by sweeping over sampling interval Royal Institute of Technology KTH 23. WirelessHART toolsKey features: Powerful network editor Interactive scheduler Integrated schedule optimizer Reliability analysis Matlab/Simulink integration Multiple superframe support Sensors, actuators, relaysRoyal Institute of Technology KTH 24. WP 2/T2.1 Communication aware data fusion and control Tune the controller s.t. stable even with varying delay One proposed method: Extended plant PID tuning Step experimentFilteringExtended plant response1G f (s) G(s)1 sT nf Filter design0 (t) maxTf f max , n AMIGO designon extended plant,tuning rules Measurement filter design based on the network delays1 3 max , n 1 Tf (1 n )/ 21 n max , n 1.3 n n 1 Royal Institute of Technology KTH 25. WP 2/T2.3. Adaptive and robust control Network congestion causes packet drops Adjust control speed and required communication rate Maintain good network QoS Keep packet drop at 3 %0.08 400.07 20 0.060.05 00 200400 600800 1000 1200QoS0.04 Time [s]0.03 60.02 4 h [s]0.01 200 0200 400 600800 1000 12000 200400 600800 1000 1200Time [s] Time [s]Royal Institute of Technology KTH 26. WP 3: WISA Toolchains, PiccSIM PiccSIM Control simulation in Simulink Network simulation in ns-2 Graphical user interfaces for network design Data-based modeling tools, controller design and tuningGUI Automatic code generation, and code reusability All in one tool Released as open-source toresearchers wsn.tkk.fi/en/software/piccsimControl designRoyal Institute of Technology KTH 27. WP 3: PiccSIMRoyal Institute of Technology KTH 28. WP3: Using Field data in SimulationsLight MachineryHeavy Machinery Simulation: Crane Control 60100 90 9080 50 80 70Packet Delvery Ratio (%) 70 4060FreeSpace Packet drop [%]Packet drop [%] 60 Light Machinery 50 3050 Medium Machinery 40Heavy Machinery 40 2030 30 2020 10 10100 0 1 2 3 4 56 7 8 1234 5 67 8012 3Link [#]Link [#]Mobility Models10000 1 10000 1400 120000.550000.5 50000.5200 0.5 10000 0000 0 00Good Bad Good Bad Good Bad Good Bad 4001 200 110001 2001 SINK 2000.5 100 0.5500 0.5 1000.5 NODES0 0000 0 00Good Bad Good Bad Good Bad Good Bad20M 20001 200140001 200 1 20M 10000.5 1000.520000.5 100 0.5 0 0 000 000 Good Bad Good BadGood Bad Good Bad 40001 2001 10014000 1 20000.5 1000.5 10M500.5 20000.5 0 0 000 000 Good Bad Good BadGood Bad Good Bad40M A Gilbert-Elliot packet drop model is implemented on PiccSIM Each link model consists of the state residence times and the packetdrop probabilities for each stateResidence timePacket drop probability Royal Institute of Technology KTH 29. ID = 1Data N Data N 0 T 1WP3: Automatic Code Generation Node _ KF Process Simulation -> Implementation and testing on real ProcesshardwareAD 0DA 6 Generic node block in PiccSIM library DA 7Radio timestamp 1 Make implementation in blockRadio recv 1 Radio send 1 Simulink blocks, Matlab code... Process_interfacedo { ... } while Radio blocks for communication between nodesSynchronize with Ns -2 Matlab Real-Time WorkshopTimestamps NodeSend to N 1 T 1 Target Language Compiler (TLC) uData N 2 T 3 ID = 0 Generate code from Simulink block Interface node Wrapper main file for Sensinode node hardware Other wrappers can easily be implementedRoyal Institute of Technology KTH 30. WP3: Automatic Code Generation 1 u -2 .5 /290 /2 .5AD 0Bias 2 3u +0 .42 .5 / 0 . 8 1Gain 1Radio recv 1LD: 33SaturationBias1DA 6Gain 2 Signal Specification Radio send 1ConstantRate Transition4 2 U ~ = U /z doubleDelays2 .5 2Radio timestamp 1 DA 7 DetectData Type Conversion Tapped DelayAdd Gain 3 ChangeRoyal Institute of Technology KTH 31. PROSE Hardware in network simulation Test hardware with simulatednetwork Wireless protocol Testing, debugging Logging all activities Controllable channel conditions Royal Institute of Technology KTH 32. PROSE communication detailsRoyal Institute of Technology KTH 33. Collaborationbetween research groups Researcher visits: Lasse Eriksson @ KTH, 5/2006 and 6/2007 Researcher visits: Mikael Bjrkbom @ KTH, 5/2009 One day visits from KTH to Aalto Joint publicationsbetween research groups and industry Active participation of the industry in the steering board meetings (e.g. simulation testbed demo attracted kerstrms (Sweden) to travel to Helsinki) Tomorrow PiccSIM demo at ABB, Sweden Joint workshop on Standards and research challenges for industrial wireless control with industrial partners in Stockholm, Sweden 4th of March 2008 Royal Institute of Technology KTH 34. Information dissemination Results summary (2008-2010) 5 (+1) Ph.D theses 3 Masters theses 5 Bachelor theses 8 Journals papers 38 Conference papers Seminar presentations and invited talks: DoD/TEKES workshop in Washington 11 - 12 March 2008 Rutgers/HIIT Workshop on Spontaneous Networks in Rutgers 5-9 May, 2008 Third International Summer School on Applications of WSN and Wireless Sensing in the Future Internet (SenZations) in Slovenia 1 - 5 September 2008 8th Scandinavian Workshop on Wireless Adhoc Networks (Adhoc 08) May 7-8, 2008 Johannesberg Estate Sensinode research seminar, Vuokatti, Finland, 16th of September 2008 Lecture on Reliable WSNs at Prairie View Texas A&M, 15th of October 2009 Presentation at Scandinavian Electronics Event, 14.4.2010 Royal Institute of Technology KTH 35. Wireless Sensor Systems group at Aalto Started in 2008 to collaborate in the field of WSS Made possible by WiSA project Aalto University Workshop on Wireless Sensor Systems 2010 Currently 4 projects, multiple departments, about 15researchers Research fields Network Management Wireless Automation (Gensen, RELA, RIWA) Indoor Situation Awarenes (WISM II) Structural Health Monitoring (ISMO)Royal Institute of Technology KTH 36. Final thoughts Nordic cooperation Closeby, initial visits Still videconference more convenient NORDITE program Nordic cooperation good Basic research oriented Industry involvement Only interest group Less feedback than in industrially financed projectsRoyal Institute of Technology KTH 37. Contact information:Mikael BjrkbomAalto UniversitySchool of Electrical EngineeringDept. of Automation and Systems TechnologyP.O.Box 15500FI-00076 AALTOFinlandTel. +358 9 470 25213Email: [email protected]://wsn.tkk.fi Royal Institute of Technology KTH