Packet Priority Assignment for Wireless Control Systems of Multiple Physical Systems Wenchen Wang, Daniel Mosse, and Alessandro V. Papadopoulos May 8th, 2019 @ISORC, Valencia, Spain
Interconnected World Made of Computing Systems ! 2
IoT Connections 4.1 Billion Connections in 2024!!!
Photo-illustration: iStockphoto
404 CAR NOT FOUND Photo-illustration: iStockphoto
Wired Control System Environment Targets Controller Actuators Plant & Objectives Sensors
Wired Control System Environment Targets Controller Actuators Plant & Objectives Sensors
Wired Control System Environment Targets Controller Actuators Plant & Objectives Sensors
Wired Control System Environment Targets Controller Actuators Plant & Objectives Sensors
Wired Control System Environment Targets Controller Actuators Plant & Objectives Sensors
Wired Control System Environment Targets Controller Actuators Plant & Objectives Sensors Not easy to deploy and maintain
Wired Control System At most 4 robots can be Environment connected to 1 cabinet [Salman et al., Fog-IoT 2019] Targets Controller Actuators Plant & Objectives Sensors Not easy to deploy and maintain
Wireless Control System (WCS) Network Actuators Delay and Message Loss Targets Controller(s) & Objectives Plant(s) Environment Sensors Network
Major Challenges of WCS 10 5 Instability 20 PS 1 10 When the physical system is 0 unstable, the plant or the device 0 0.2 0.4 0.6 0.8 1 can be damaged and leads to time serious safety issues and 1 financial loss. PS 2 0 -1 0 0.2 0.4 0.6 0.8 1 Performance Degradation time 1 Induced additional error PS 3 0 Network-induced error -1 0 0.2 0.4 0.6 0.8 1 time
Major Challenges of WCS 10 5 Unstable Instability 20 PS 1 10 When the physical system is 0 unstable, the plant or the device 0 0.2 0.4 0.6 0.8 1 can be damaged and leads to time serious safety issues and 1 financial loss. PS 2 0 -1 0 0.2 0.4 0.6 0.8 1 Performance Degradation time 1 Induced additional error PS 3 0 Network-induced error -1 0 0.2 0.4 0.6 0.8 1 time
Major Challenges of WCS 10 5 Unstable Instability 20 PS 1 10 When the physical system is 0 unstable, the plant or the device 0 0.2 0.4 0.6 0.8 1 can be damaged and leads to time serious safety issues and 1 financial loss. PS 2 0 -1 0 0.2 0.4 0.6 0.8 1 Performance Degradation time 1 Induced additional error PS 3 0 Network-induced error -1 Degraded 0 0.2 0.4 0.6 0.8 1 Performance time
Problem Formulation Remote Controller Shared multi-hop network Di ff erent paths p_1 , p_2 , …, p_m Multi-hop Network Each path with delay D_j TDMA fixed topology Time-varying delivery ratio dr_j … N Physical Systems (PSs) Dynamic network reconfiguration in [Wang, RTNS 2018]
Di ff erent Requirements, Shared Network LO-Critical HI-Frequency Demand time HI-Critical HI-Frequency Demand time HI-Critical LO-Frequency Demand time
Di ff erent Requirements, Shared Network LO-Critical HI-Frequency Demand time HI-Critical HI-Frequency Demand time HI-Critical LO-Frequency Demand time
Di ff erent Requirements, Shared Network LO-Critical HI-Frequency Demand time HI-Critical HI-Frequency Demand time HI-Critical LO-Frequency Demand time
Di ff erent Requirements, Shared Network LO-Critical HI-Frequency Demand time HI-Critical HI-Frequency Demand time HI-Critical LO-Frequency Demand time
Problem Formulation The control is more or less di ffi cult based on Setpoint (or reference) tracking Nonlinearity of the controlled system Reliability of the communication path
Problem Formulation The control is more or less di ffi cult based on Requested Setpoint (or reference) tracking RCA Change Amount Nonlinearity of the controlled system Reliability of the communication path
Problem Formulation The control is more or less di ffi cult based on Requested Setpoint (or reference) tracking RCA Change Amount Nonlinearity of the controlled system RCD Reliability of the communication path Requested Change Duration
Problem Formulation The control is more or less di ffi cult based on Requested Setpoint (or reference) tracking RCA Change Amount Nonlinearity of the controlled system RCD Reliability of the communication path Requested Change Duration Objective: Minimize Control performance degradation Induced by the wireless realization Without redesigning the control system
Problem Formulation The control is more or less di ffi cult based on Requested Setpoint (or reference) tracking RCA Change Amount Nonlinearity of the controlled system RCD Reliability of the communication path Requested Change Duration Objective: Minimize T trans 1 ∑ ∥ y W i ( t ) − y WL ( t ) ∥ 2 Control performance degradation RMSE i = i T trans Induced by the wireless realization t =0 Without redesigning the control system
Problem Formulation The control is more or less di ffi cult based on Requested Setpoint (or reference) tracking RCA Change Amount Nonlinearity of the controlled system RCD Reliability of the communication path Requested Change Duration Wired Objective: Minimize T trans 1 ∑ ∥ y W i ( t ) − y WL ( t ) ∥ 2 Control performance degradation RMSE i = i T trans Induced by the wireless realization t =0 Without redesigning the control system
Problem Formulation The control is more or less di ffi cult based on Requested Setpoint (or reference) tracking RCA Change Amount Nonlinearity of the controlled system RCD Reliability of the communication path Requested Change Duration Wired Wireless Objective: Minimize T trans 1 ∑ ∥ y W i ( t ) − y WL ( t ) ∥ 2 Control performance degradation RMSE i = i T trans Induced by the wireless realization t =0 Without redesigning the control system
Problem Formulation The control is more or less di ffi cult based on Requested Setpoint (or reference) tracking RCA Change Amount Nonlinearity of the controlled system RCD Reliability of the communication path Requested Change Duration Wired Wireless Objective: Minimize T trans 1 ∑ ∥ y W i ( t ) − y WL ( t ) ∥ 2 Control performance degradation RMSE i = i T trans Induced by the wireless realization t =0 Without redesigning the control system Transient
Problem Formulation The control is more or less di ffi cult based on Requested Setpoint (or reference) tracking RCA Change Amount Nonlinearity of the controlled system RCD Reliability of the communication path Requested Change Duration Wired Wireless Objective: Minimize T trans 1 ∑ ∥ y W i ( t ) − y WL ( t ) ∥ 2 Control performance degradation RMSE i = i T trans Induced by the wireless realization t =0 Without redesigning the control system Transient ∀ i ∈ Physical Systems
Solution: Dynamic Packet Priority Assignment Priority Remote Assignment Controller Priority Assignment Path Selection Static heuristic (baseline) Dynamic heuristic PID Dynamic heuristic Multi-hop Network Path Selection Network Path Quality Determination …
Static Heuristic - Baseline Priority Remote O ffl ine analysis Assignment Controller Assuming no packet loss For di ff erent requested changes in Path Selection demand For all PS i compute T sim 1 ∑ ∥ r i ( j ) − y i ( j ) ∥ 2 rRMSE i ( T sim ) = Multi-hop T sim Network j =0 Assign the priorities that minimise the average rRMSE i ( T sim ) Do not change the priorities online …
Dynamic Heuristic Priority Remote At every time t, for all PS i compute Assignment Controller t 1 ∑ ∥ r i ( j ) − y i ( j ) ∥ 2 rRMSE i ( t ) = Path Selection t j =0 Sort the PS i by rRMSE i ( t ) Multi-hop Assign the highest priority to the PS with Network highest value of rRMSE i ( t ) …
PID Dynamic Heuristic Priority Remote We define the tracking error as Assignment Controller e i ( t ) = | r i ( t ) − y i ( t ) | Path Selection The priority for every PS i is computed as π i ( t ) = K P ( e i ( t ) + λ e i ( t ) ) + K D ( e i ( t ) − e i ( t − 1)) t ∑ t Multi-hop i =1 Network strange formula, isn’t it? It is a PID controller ! …
PID Dynamic Heuristic Priority Remote We define the tracking error as Assignment Controller e i ( t ) = | r i ( t ) − y i ( t ) | Path Selection The priority for every PS i is computed as π i ( t ) = K P ( e i ( t ) + λ e i ( t ) ) + K D ( e i ( t ) − e i ( t − 1)) t ∑ t Multi-hop i =1 Network strange formula, isn’t it? It is a PID controller ! …
The Path Quality Model: PQModel Priority Remote After we determine the priority of the Assignment Controller measurement packets Includes Path Selection Network delay Network reliability We compute the path quality for all the paths as Multi-hop Network PQ = D net + α n loss Δ csp …
The Path Quality Model: PQModel Priority Remote After we determine the priority of the Assignment Controller measurement packets Includes Path Selection Network delay Network reliability We compute the path quality for all the paths as Multi-hop Network End-to-end delay PQ = D net + α n loss Δ csp …
Recommend
More recommend
