Architecture of Intelligent Systems - 2 1 Plan • Analysis of a Physical system with simple control function • Complex system and Abstraction • Mapping of the physical system to the Intelligent system 2 Sensor Compa Refere rator nce Signal Conditioning Error signal Cold air Control signal Contr oller Correc tor Correction signal Actuator Physical system Room with A/C Controller 3 Sensor Sensor signal Contro ller Cold air Control signal Reference value or Control parameters Actuator Physical system - A/C Room 4 Se ns or Intelligent system Environment Ac tua tor Is IS a controller? 5 •IS is some sort of controller (or at least a controller and may be more) Se ns or Intelligent Controller system Environment Ac tua tor 6 •Can there be more than one sensor and one actuator in a system? •What other actions are needed in our IS as a result of multiple sensors? •Guess what type of sensors, actuators and controls would you like to have in our crazy car – 4 independent wheel drive car! •Would our simple controller be sufficient for our car? • There can be many sensors, actuators and controllers • Coordinate the actuators • The information from multiple sensors is to be analysed and control actions for each actuator are to be identified, control parameters and control actions are to be generated. • The model of the environment (world model) is to be generated. • Error handling information is to be incorporated Thus the system looks like World model Se nso rs Controller Sig coor nal dina Controller An tor Controller alys er Controller ? ? ? Environment Act uat ors Crazy car Sensors for indicating the direction and speed of each wheel User communication interface Task identification and its decomposition Control signal generation Coordinate the control signals at the actuators Model for the actions of the wheels Sensors for the proximity and speed of other cars and its own speed World model for taking decisions regarding collisions and collision avoidance Knowledge base and expert advice for taking decisions How should we build our Intelligent system? Do we need to pack all the features into a single monolithic program or should we build up a modular / layered product? Remember all complex programs are modular/layered - OS How many layers? Suggest three based on management model – execution level, coordination level, executive level. But could be as many as we like. Execution level coordination level Top management & Executive level Sensors Actua tors High level management Low level management Environment Middle level management Low Senso rs Physic al system Middle High Data flow/ response Commands Actu ators Human interventions (in red) Hu ma n ope rat ors Functional Architecture of Intelligent (control) system Centralised Vs Distributed Is it technology dependent? Low level Or execution level control manager Signal conditioning, sampling, Sensor holding, buffering, multiplexing and inform demultiplexing, queuing, scheduling of data acquisition & actuators, ation data supply to controllers, control parameter generation, error detection and send error information to appropriate units, signals to actua Algorithms in HW & SW Execution of control tors Multiple sensor information, multiple controllers Multiple copies of HW & SW Commn with middle level Human interve ntions Middle level Or coordination level control manager Coordination, Synchronisation, Commn timing signals, Prioritisation and with low sequencing, Monitoring sys health & processes, decision making, capability level assessment, fault assessment and analysis, system & environ model generation & updating, learning, Task analysis, Task decomposition, planning for execution, Resource management, algorithm development, Knowledge bases, logic resolution, Algorithm development, simulations Interface between the low level &high level, limited copies of HW & SW Commn with high level Human interve ntions High level Or organisational manager Goal assessment and generation, Commn Goal analysis, goal to task conversion, Strategising, with middle planning to achieve goals, consequence evaluation, decision level making, User I/F, feasibility analysis and capability assessment, learning knowledge bases, data mining, Single copy Human interven tions and commn with humans P h y s i c a l s Y s Sensors M u x Refe rence KB Signal Condi tioning Decision making unit (Int control) Load algos Contro llers DB Data Switch Commn Sys state & adaptive parameters Timer Actua tors KB Di git Info buffer/ ise storage/anal rs De Mux Err cor Low level control details Instructions, commands, algorithms Fault info analyser KB with middle level Human interve ntions From data switch KB Synchro niser Sched uler DSS Sim/mod Info Assessor Control Implementation supervisor From System Algos, commands, State instructions model Math model Algo KB Sim develo per From Fault info anal Designer Fault Super visor Sim KB Commn. with Control Exe Manager cutive (Task anal) Level Algo Planner Rep osito Plan Plan ry selec gener tor ator KB Plan repository Human inter vention Fault info Stra tegi ser Task generator / goal analyser Comma nds, & tasks Build model Goal generator KB and Data warehouse Capability/ feasi bility assessor System model Fault info Soln. opti miser Command Interpreter Identification of problems & solutions Commn. from Human operator Simulator DSS Decision on action Info gener ator Information formatter and presenter Info/ Messa ges to human operator End