SMART-TProjectOverviewKurtD.GuentherAS&MDryden

11,SMART-T Briefing to OSMA SAS - July 19, 2004,Click to edit the title text format,Click to edit the outline text format,Second Outline Level,Third Outline Level,Fourth Outline Level,Fifth Outline Level,Sixth Outline Level,Seventh Outline Level,Eighth Outline Level,Ninth Outline Level,SMART-T Project Overview,Kurt D. Guenther,AS&M / Dryden Flight Research Center,July 19, 2004,SMART-T Objectives,SMART-T:,Strategic Methodologies for Autonomous and Robust Technology Testing,The goal of SMART-T is to address V&V issues of adaptive control systems, including neural networks,Develop methodologies to design and validate NN controllers,Develop tools and methodologies to support the eventual certification of adaptive systems,Coordinate the community at-large,This technology development effort seeks to build the confidence,that is needed to intelligently design, test and safely fly,adaptive flight controllers.,Where SMART-T Fits In,Research Effort,Tools,Sensitivity Tool,Confidence Tool,Neural Network Evaluator,ANCT Tool,Flight Test &,Eval,F-15 Gen II IFCS, C-17 Gen II IFCS,UAVs,Confidence Tool,Neural Network Evaluator,Tools,Flight Test,Simulation,Toolset,Methods,Methods,(applied to SW life-cycle: design, development, test),Generic Guide,F-15 Guide,C-17 Guide,Research Effort (cont),Confidence Tool for IFCS,Control,Law,Inverse,Plant,Model,Pilot Inputs,Commanded,State,Neural,Network,Measurements,Filter,Confidence,Tool,The Confidence Tool, based on a Bayesian approach, provides a,Measure of how well the neural network is performing at the moment,Control,Augmentation,NN Weights,Sensitivity Analysis for IFCS,Control,Law,Inverse,Plant,Model,Pilot Inputs,Commanded,State,Neural,Network,Measurements,Filter,The Sensitivity Analysis provides a Measure of Stability,in the sense of Lyaponov 2nd Method for Nonlinear Systems,Control,Augmentation,Sensitivity,Analysis,Sensitivity Tool,NN sensitivity tool provides verification of,Lyapunov,stability bounds by perturbation of the gains and noise parameters.,All current axis learning parameters are robust to gain and noise in,Sigma Pi NN, except yaw axis,Tool implementation completed for:,SHL non-ITAR.,SHL and Sigma Pi with VCAS controller designs.,Automated Neural Controller Test Tool (ANCT),Developed under grant to Case Western Reserve U.,Developed to automate,Lyapunov,boundary estimation using,sens,. tool,Rich GUI for the test engineer to vary inputs, automate Monte Carlo,sim, set performance criteria,analyze,outputs,Interacts with,Simulink,model, automatically populates GUI with internal variables,ANCT Test Generation,NN Evaluator,Developed by Institute for Scientific Research (ISR), Fairmont, WV,“Fault indicators” derived from inspection of the - NN Weighting adaptation law,discrete poles,weight norms,set-point error norms,Lyapunov stability criteria,Lyapunov rate stability criteria,Implemented in F-15 IFCS ARTS II computer,