UNISI Team. UNISI Team - Expertise

Transcription

1 Control Alberto Bemporad (prof.) Davide Barcelli (student) Daniele Bernardini (PhD student) Marta Capiluppi (postdoc) Giulio Ripaccioli (PhD student) XXXXX (postdoc) Communications Andrea Abrardo (prof.) Alessandro Mecocci (prof.) Lapo Balucanti (PhD student) Marco Belleschi (PhD student) Cesare Carretti (PhD student) Operations research Alessandro Agnetis (prof.) Marco Pranzo (prof.) UNISI Team Project manager Ilaria Sbragi 8 UNISI Team - Expertise Model Predictive Control (): theory, algorithms, tools Hybrid systems: modeling, analysis, control, optimization, tools Resource allocation for wireless networks (ad-hoc & sensor networks) Combinatorial optimization for planning and scheduling problems Experimental activities on data processing using WSNs 9

2 of Large-Scale Constrained Linear Systems Cost function: full matrices Q,R,P N 1 X k=0 h x 0 k Qx k + u 0 k Ru i k + x 0 N Px N Process model: coupled dynamics x k+1 = Ax k + Bu k, x R n, u R m Centralized approach Constraints: u min u k u max Drawback All the n components of x(t) must be transmitted to a central unit to solve a (large) QP, and m signals must be transmitted back to the corresponding actuators Idea: replace a centralized algorithm with m simpler decentralized algorithms, one for each actuator (Alessio, Bemporad, ECC 2007) (Alessio, Bemporad, ACC 08) 10 Decentralized Model Assume matrices A,B have certain number of negligible components (dynamical sub-systems are partially decoupled) a 11 a a 1j a b 1n b 1k b 1j 0 a 21 a 22 0 a 2k 0 0 b b b m A = 0 a k2 0 a kk 0 a kn B = 0 b a j1 a j2 0 0 a jj 0 k2 0 b kk 0 b km b j1 0 0 b jk b jj b jm a n1 0 0 a nk 0 a nn 0 b n2 0 b nk 0 b nm Let m be the number of decentralized control actions we want to design Define m (possibly overlapping ) submodels u1 u2 u3 u4 u5 u6 x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 A 1 = a 11 a 12 a 13 a 21 a 22 a 23 B 1 = a 31 a 32 a 33 b 11 b 12 b 21 b 22 b 31 b 32 a 33 a 34 a 35 b 32 b 33 A 2 = a 43 a 44 a 45 B 2 = b 42 b 43 a 53 a 54 a 55 b 52 b 53,, etc. 11

3 Control over Networks: Pros Decentralized requires architectures where different components of the control system communicate over (possibly wireless) network links Real-time optimization Pros: - delocalization of the controllers - often reduced installation and maintenance costs - flexible and easily reconfigurable 12 Control over Networks: Cons Closing the loop over wireless communication networks introduces additional problems: Cons: - Data loss (packet drop) - Data loss (total loss of connectivity) Real-time optimization - Communication delays - Jitter - Wireless sensors are usually powered by batteries, so energy consumption must be minimized New challenges in decentralized control over networks! 13

4 Energy-Aware Control over WSNs Problem: battery consumption Radio chip is the dominant power hog. Control strategy should keep the radio off (both TX & RX) as much as possible! (Bernardini, Bemporad, CDC 08) Real-time optimization Tradeoff: closed-loop performance vs. transmission rate Idea: transmit measurements only when necessary, according to the threshold logic: where ŷ(k) is a prediction of y(k) computed by the controller and received by the sensor node 14 Energy-aware control Intuitive sketch Note: a non-measurement is also a measurement! (assuming all packets are received) 15

5 Energy-Aware Control over WSNs Note: Wireless communication protocols require a minimum frame size (e.g. 248 bits in ZigBee) Transmitting a (small) set of measurements costs (almost) as much as transmitting a single measurement When the controller receives the measurement, it computes a new set of M predictions and transmits them to the node In the ideal case of no disturbances, i.e. the overall transmission rate is 1/M y(k) =ŷ(k) k, Compute control action via explicit Model Predictive Control: - can handle constraints, - closed-loop predictions easily computed and transmitted - can be formulated to achieve robust control by explicitly taking into account the error induced by the threshold logic 16 Energy-Aware Control over WSNs radio use performance Controller performance Tx. Rate Radio savings: -55.4% Standard % Performance loss: +1.58% Energy-Aware % 17

6 Sensor placement and network topology design (Abrardo et al.) Maximize the network lifetime Minimize the applicationspecific total cost, given a fixed number of sensor nodes in a region with a certain coverage requirement Optimize routing Optimize nodes redundancy 18 Protocols and distributed processing for state estimation and monitoring (Abrardo et al.) Distributed optimization algorithms over IEEE (in collaboration with KTH) Communication only between neighbors or with routing Distributed approach vs. centralized approach Use of network synchronization techniques for battery saving 19

7 Protocols and distributed processing for state estimation and monitoring (Abrardo et al.) Use of optimized clustering to improve network efficiency and data sampling following real values distribution Trade-off between values precision and energy efficiency Network self-configuration and self-healing Synchronization inter-cluster and intra-cluster 20 Experimental activities at UNISI on WSNs 21

8 WSNs for agriculture Humidity Temperature Light Soil Moisture Leaf wetness Wind GOALS: Prevent presence of fungi and parasites Automate irrigation process 22 WSNs for localization and tracking Accelerometers and pyro-electric (PIR) sensors used to detect position of humans (indoor) Monitoring movements in the house Detect patient falling down and send alarms 23

9 WSN for architecturual heritage Sensor fusion of humidity, temperature, vibration measurements Send alarm messages in case of danger Surveillance and touristic informations Report generation (e.g. while monitoring cracks) Data gathering to decide intervention priorities 24 WSNs for vehicle detection Telos + sensor board Magnetic field sampled at 64Hz Estimation algorithm provide vehicle detection and vehicle velocity Packets received in Matlab through Java interface Sensor board developed at the Automatic Control Lab (Siena) based on Honeywell magnetoresistive sensors 25

10 Wireless Automation Experiments WSN Radio Control Access Point TCP/IP Lab Process Process PC 26 Experimental results: Robustness w.r.t. packet drop Reliable WSN (3.7% max loss) Unreliable WSN (69.7% loss) 27

11 Demo Application in Wireless Automation steady-state temperature position Hybrid problem: 2 binary inputs (lamps) 1 continuous input (speed) PWL state function heating=f(position) Outputs: temp, position Sampling = 4Hz Objective: track position and temperature references while enforcing safety constraints 28