SENSOR NETWORK SERVICE INFRASTRUCTURE FOR REAL- TIME BUSINESS INTELLIGENCE

Transcription

1 SENSOR NETWORK SERVICE INFRASTRUCTURE FOR REAL- TIME BUSINESS INTELLIGENCE A. Musa and Y. Yusuf Institute of Logistics and Operations Management University of Central Lancashire, Preston

2 2 Our interest in RFID dates back to 2005 Initial interest was in supply chain applications of RFID Focus was on industrial case studies Our interest has gradually shifted to Generalized sensor data sources and their integration Service development and deployment How edge sensor data is used in supply chain operations to derive business intelligence (BI) and how the BI is transformed into concrete actions at the edge How sensor data may be used to control supply chains in real-time Applications of system dynamics and control theory

3 3 Review RFID products and services from several vendors Three RFID standards organizations (ISO, EPCglobal and DASH7) were surveyed 10 vendors were surveyed Assess the level of adoption of other sensor types beyond RFID DASH-7 (ISO ) MEMS sensors (accelerometer, gyro and pressure) Light sensor Temperature sensor GPS and other location technologies wireless connectivity (active RFID, WiFi, Bluetooth, GSM/GPRS)

4 4 Review reference architectures and stacks for sensor network deployment EPCglobal Microsoft BizTalk RFID SAP auto-id DASH-7 Identify some knowledge gaps in deriving business intelligence from sensor data Focus on and seek to contribute to addressing one of the identified gaps

5 A generic stack for device deployment 5

6 6 Efficient and effective enterprise system scalability and decomposition of business logic between the backend and the enterprise edge For instance, a large retailer that uses RFID across its network on most of its merchandize might require an annual throughput rate of up to 60 billion items When replicated across retailers and supply chains, this has the potential to put severe stress on network resources Performing process logic on the mobile thin client, at the enterprise edge, reduces communication costs and computational overheads at the backend

7 7 Scalability In order for this to be realized, there is a need for algorithms that scale sufficiently well in terms of bandwidth, energy and computational power requirements with respect to client topology Products such as DASH-7-compliant thin clients can communicate peer-to-peer and execute simple logic locally DASH-7 products are likely to become cheaper and much more widespread in the future; much as local processing capacities of thin devices are likely to increase, and their energy budgets shrink, in the medium term

8 8 Event-based communication can relieve bandwidth requirements and improve operational efficiency Optimal models and automatic (or even semi-automatic) systems for handling exceptions in real-time are needed For example, if there is a sudden general or specific breakpoint in the supply chain, how is the chain able to reorganize itself in real-time so as to minimize the negative consequences of the break? An efficient, but effective, control system for event-based management systems in supply chains is desired What other data sources beyond sensors exist or are needed for system control

9 9 Multidirectional decision flow If decisions are taken at the strategic or tactical levels of the supply chain to address an identified break, how are these decisions communicated and turned into concrete actions at the operational levels of the chain or enterprise in real-time? If actions are not taken quickly at the operational level to resolve identified system deficiencies or failures, then the spirit of urgent data acquisition at the enterprise edge and its transmission to managerial levels will be defeated Supply chains are best in acquiring edge data; they are good in deriving intelligence from data; but they are poor in turning intelligence into action at the operational level in real time Beyond sensor data, ontological data sources are needed for this task (source, user, shared, and application ontologies)

10 10 There are groups focussing on optimization in supply chain management in the areas of Inventory decision and policy development Time compression Measures to counter Forrester effects (demand de-amplification) Supply chain design and integration International supply chain management Aspects of risk management Our approach differs from these strands

11 11 The rest of the presentation describes our approach and the modelling issues we have been considering Our aim is to build an optimal closed-loop MIMO control system for supply chains with data from the enterprise s frontline In the current study, data come only from sensors at the edge No ontological or contextual information is being used yet Supply chains are dominated by open-loop (feedforward) controls that rely on dashboard reporting Open-loop systems may serve to improve reference tracking performance but they re not enough for supply chain management

12 12

13 13 Closed-loop controllers have the following advantages over open-loop: Disturbance rejection (eg, unmeasured friction) Guaranteed performance even with model uncertainties, when Model structure does not match perfectly the real process and Model parameters are not known precisely Unstable processes can be stabilized Reduced sensitivity to parameter variations Improved reference tracking performance (especially when combined with open-loop)

14 14 We re using state-space representation (SSR) because It provides a convenient and compact way to model and analyze systems with multiple inputs and outputs Unlike the frequency domain approach, the use of SSR is not limited to systems with linear components and zero initial conditions Unobservable poles are not present in the transfer function realization of a SSR

15 15 Stability Stability for nonlinear systems that take input is an input-to-state stability (ISS) This combines Lyapunov stability and a kind of BIBO (bounded-input boundedoutput) stability Controllability, observability, detectability Supply chains are detectable and observable, but are they actually controllable? Calculative opportunism transaction cost economics

16 16 Choice of controller PID controllers are often general enough Control specifications Stability, ensuring that poles of the TF satisfy Re[ ]<-1, rather than just Re[ ]<0 Rise time, peak overshoot, settling time, quarter-decay Performance assessment (we re using integrated tracking errors) Robustness: controller properties should not change much when applied to a system slightly different from the one used for synthesis

17 17 System identification and robustness We re using both offline and online (adaptive*) model identification methods. See later Choice of nominal parameters Robustness of SISO controls are relatively straightforward (gain, phase margin and amplitude margin), but MIMO controls are quite hard to robustify Our MIMO control will have robustness qualities decided by us (see constraints below)

18 18 Constraints The control system must perform properly in the presence of input and state constraints; the controller should not send signals that can t be followed by the supply chain team We are investigating the applicability of model predictive controls (MPCs) and antiwind up systems to supply chain dynamics. See modelling strategies later

19 19 Dealing with nonlinearity Supply chain processes, like other multiechelon setups, exhibit strong nonlinear dynamics In control theory it is sometimes possible to linearize and apply linear techniques, but We wish to devise from scratch the means to the nonlinear system (feedback linearization, backstepping, sliding mode control, trajectory linearization* control) These approaches are still based on Lyapunov s results pertaining to linear cases They often disregard the inner dynamics of the system*

20 20 Differential geometry? This has been widely used as a tool for extending well-known linear control theories to the nonlinear case, as well as demonstrating the complexities that make non-linear cases a more challenging problem We aren t currently considering this issue

21 21 Centralized or decentralized control Use of single or multiple controllers Can supply chains be directed effectively by single controllers? No! Supply chains operate over large geographical area and at various managerial levels Agents in decentralized controls can interact using communication channels and coordinate their actions But our current effort is focussed on single controller scenarios This must be followed eventually by multiple controllers

22 22 We ll briefly review the control strategies we have considered and elaborate on the ones we have adopted. Adaptive control Using on-line identification of the process parameters, or modification of controller gains, and hence ensuring strong robustness Hierarchical/networked control Arranging devices and guidance software in a hierarchical tree Intelligent control AI approaches like neural networks, Bayesian probability, fuzzy logic, machine learning, evolutionary computation and genetic algorithms

23 23 Optimal control The control signal optimizes a certain cost index For example, in the case of a supply chain, we may think of the jet thrusts needed to bring the supply chain to the desired trajectory that consume the least amount of resources Two optimal control design methods that often guarantee closed-loop stability are model predictive control (MPC) and linear-quadratic-gaussian control (LQG) Together with PID controllers, MPC systems are the most widely used approaches in process control

24 24 Robust control Deals explicitly with uncertainty in its approach to controller design A modern example of a robust control technique is H-infinity loop-shaping (Duncan McFarlane and Keith Glover) Robust methods aim to achieve robust performance and/or stability in the presence of small modelling errors

25 25 Stochastic control Deals with control design with uncertainty in the model It is assumed that there exist random noise and disturbances in the model and the controller, and the control design must take into account these random variations At this experimental stage we re focussing on adaptive control only. Work on robust and stochastic controller types will be conducted in the future

26 Thank you 26