PENNSTATE 1 8 5 5 Project PS 5.1 Carefree Maneuvering Control Laws for Rotorcraft PI: Asst. Prof. Joseph F. Horn Tel: (814) 865 6434 Email: joehorn@psu.edu Graduate Student: Nilesh Sahani 2005 RCOE Program Review May 3, 2005 PENNSTATE 1 8 5 5 Background / Problem Statement Rotorcraft are constrained by multiple flight envelope limits. It is a major contributor to pilot workload. Carefree Maneuvering (CFM) systems allow us to avoid using conservative operational envelopes. Pilots can use maximum performance and maneuverability without sacrificing safety, reliability, handling qualities Can be achieved through cueing systems or AFCS Normal Pilot Cues: Out the window, instruments, acceleration, vibration, etc … Active Control Stick Pilot Pilot Force Technical Barriers Robust and efficient algorithms to predict onset of limits and control constraints to avoid the limits Develop systems that do not take away a pilot’s option to exceed a limit and do not annoy the pilot Avoid adding costly sensors or systems Need comprehensive systems that handle multiple limits and multiple control axes Integration of envelope cueing with envelope limiting AFCS Quantify the payoffs of CFM in terms of handling qualities, safety, reliability, maintainability Exploit potential use of CFM technology on UAV’s Aircraft Aircraft Sensors Sensors Variable Soft Stops Stick Position AFCS AFCS Inceptor Inceptor Constraints Constraints Prediction Algorithm: e.g. Neural Networks, On-board models Carefree Maneuvering (CFM) Control System Using Tactile Cueing PENNSTATE 1 8 5 5 Task Objectives: Develop limit detection and avoidance algorithms for peak responses and transient limits Develop an integrated CFM control architecture that features cueing and limiting in the AFCS Develop practical systems, demonstrate technologies in piloted simulation / flight test Apply CFM technology to UAV’s Approaches: Develop improved transient limit prediction algorithm and constraint calculation methods Cooperation with industry / academic partners (Sikorsky / Georgia Tech), in order to demonstrate systems in piloted simulations and flight tests Focus on comprehensive collective axis cueing and hub moment limiting systems Extend technologies to integrate with advanced AFCS designs / UAV flight controls Expected Results: Develop limit avoidance methods for advanced AFCS designs / UAV flight controls Evaluate integration of envelope protection system with UAV flight controls An assessment of the requirements for effective carefree maneuvering control systems Some measure of the potential payoffs of carefree maneuvering technology Outline of Technical Presentation PENNSTATE 1 8 5 5 Piloted Rotorcraft / Cueing Systems Review - Longitudinal hub moment limit avoidance algorithm Pilot cueing techniques Results Comparison of different techniques UAVs / Future Rotorcraft Overview of control architecture – Inner loop / outer loop structure Problem with implementing envelope protection system within the feedback loop Continuous torque limit protection for UAVs Generating constraints on outer loop command Results PENNSTATE Limit Avoidance Algorithm - Transient Limits 1 8 5 5 Stick Constraint Closest proximity of response to limit if no change in control input Sensitivit y - Transient peak due to unit step Offline trained neural networks Find dynamic trim values Stick constraint calculation Input from Aircraft Sensors Signal + g i ( x f 0 ) ylim,l , y lim,u ylim,u max Q x , t ~ t ulim,u max H x , t t Q( x , t ) ylim,l min Q x, t ~ t ulim,l max H x , t t Neural net for f i ( x s , t ) - ub x f : 1c , q t H ( x, t ) Neural net for H ( x s , t ) Increment in t Longitudinal Hub Moment Limiting Algorithm Output to Limit-Cue Arbitration and Tactile Cues Scan through time 1 s 1 xf e 1 s 1 + long MH PENNSTATE Background – Swoop Maneuver 1 8 5 5 Integrated Hub Moment Limit Exceedance Factor (IHMLF) Desired Performance Adequate Performance Time integral of hub moment above limit Sum of all green pained areas PENNSTATE Swoop Maneuver Results 1 8 5 5 Slight increase in maneuver time Does not inordinately restrict agility of the aircraft Reduction in integrated hub moment violations less fatigue wear Reduction in absolute peak hub moment reduced risk of catastrophic failure Pilot comments: softstop cue for hub moment objectionable during very aggressive maneuvers With Cue With Cue Wihtout Cue 26 26 24 Maneuver Time (s) 24 Maneuver Time (s) Wihtout Cue 22 20 18 16 14 12 10 0 5000 10000 15000 20000 25000 Integrated Hub Moment Limit Exceedance Factor (ft-lb-s) 22 20 18 16 14 12 10 20000 25000 30000 35000 Absolute Peak Hub Moment (ft-lb) 40000 PENNSTATE Command Shaping for Limit Protection 1 8 5 5 Pure automatic limit protection Upper Limit 4 x 10 H Hub Moment, M (ft-lb) Lower Limit Does not allow pilot to override limit Useful for evaluating the effectiveness of the prediction algorithm only Gives a standard for comparing other cueing techniques 2 + - 0 -2 220 222 Time (sec) 224 226 1 Control Axis, u yp lim 0.5 0 230 232 Time234 (sec) 4 x 10 2 0 -2 220 1 -0.5 -1 228 0.5 + + 225 230 235 236 δ crit SS 238 eff - PENNSTATE Command Shaping for Limit Protection 1 8 5 5 Deadband Upper Limit Allows pilot to override the limit Can be implemented using stick shakers Provides easier upgrade than softstop Lower Limit (ft-lb) 4 yp lim - 296 0.5 0 298 4 x 10 Time300 (sec) 2 H -2 (ft-lb) 0 1 Control Axis, u + 2 Hub Moment, M Hub Moment, M H x 10 302 304 306 308 + + 0 -2 300 0.7 -0.5 0.6 -1 0.5 Time (sec) 310 305 310 312 δ crit SS 314 eff - PENNSTATE Advanced Cueing Methods (proposed by our colleagues at Georgia Tech) 1 8 5 5 Frequency Distribution Sidestick momentum can inadvertently climb over a softstop cue Can result in transient limit violation Technique: No change to softstop Fleeting command restraint Technique: Low frequency component for pilot cueing High frequency component for autonomous protection 4 x 10 2 + 268 270 1 yp lim - 4 x 10 H -1 0 -1.5 -2 -2 -2.5 1 Control Axis, u 4 x 10 Hub Moment, M (ft-lb) H Hub Moment, M (ft-lb) Dynamic Overshoot Compensation 2 Time (sec) 272 274 269 269.5 0 276 278 270 280 270.5 271 Time (sec) 282 δ -2 0.5 0.5 0 0 -0.5 -0.5 -0.5 -1 -1 -0.6 268 -0.4 270 272 274 276 278 280 282 284 284 + + 286 286 eff crit SS - PENNSTATE Performance Evaluation 1 8 5 5 Pure Autonomous Deadband Dynamic Overshoot Compensation Frequency Content Distribution Same underlying longitudinal hub moment limit avoidance algorithm (developed by PSU) Summary (ft-lb-s) Thousands Limit Protection Approaches Softstop cueing provided better results compared to stick shakers Overshoot/ Freq. Distribution (ft-lb) Thousands Constraint calculation (PSU) + Pilot cueing (GT) = Smallest maneuver time + Fewer limit violations Agility + Component Life Max Integrated Hub Moment Limit Violation Extent of 2nd & 3rd Quartiles Min Pure Over- Freq. Dead- Pure Pure Visual shoot Dist band Tactile Auto GT Smaller maneuver time Fewer limit violations better pilot-in-the-loop performance Best results using frequency distribution “Steadier” softstop of Freq. Dist. was noticed and preferred 20 18 16 14 12 10 8 6 4 2 0 38 36 34 32 30 28 26 24 22 GT PSU PSU 16.3 16.5 16.8 18.1 18.3 18.5 Average Maneuver Time (sec) Absolute Peak Hub Moment Conclusions from Piloted Simulation PENNSTATE 1 8 5 5 Algorithm was reasonably effective in predicting the constraints Neural Network - Predicts critical value of the future response over some time interval Sensitivity for constraint calculation Constraints moved too quickly for the pilot to react efficiently to a pure soft stop cue Frequency distribution of constraints provided best results Low frequency component – softstop cue High frequency component – Autonomous protection Pilot was able to interact with the softstop much more efficiently Maneuvers consistently resulted in smaller maneuver time and fewer limit violations Satisfactory pilot in the loop performance – no major pilot objections Limit Avoidance for UAVs / Future Rotorcraft PENNSTATE 1 8 5 5 Future rotorcraft / UAV are likely to feature model following / inversion controllers Structural limit protection desirable To save weight on future designs (particularly on large aircraft) To protect highly agile vehicles with high bandwidth flight controls Achieve limit protection with constraints in desired response (not in feedback loop) UAV Velocity / Angle Commands ACAH / RCAH Active Control Stick Pilot Pilot Control Trajectory Tracking Limit Avoidance Desired Response Tracking Controller Constraints on Desired Response Limit Prediction Aircraft Sensors Limit Parameter to Command Dynamics Problems with Constraints in Feedback Loop PENNSTATE 1 8 5 5 c Saturation Inside Feedback Path Saturation Outside Feedback Path cmd cmd Command Filter lim + - PD Error + Dynamics + q lim Inversion Controller Aircraft c Command Filter + - PD Error + Dynamics Command Filter States Constraint on input command q lim + Inversion Controller Aircraft PENNSTATE Inner Loop / Outer Loop Controller 1 8 5 5 acc. command cmd cmd Command Filter rcmd VD cmd Inner Loop Commands Pos. / Vel. command Notch Filter PID Error Dynamics Inversion Controller Inner Loop Aircraft Sensors ANN Controller implemented for GENHEL Inner Loop Commands Outer Loop Commands Vcmd cmd cmd h cmd PI Error Dynamics cmd cmd rcmd VD cmd Outer loop – trajectory commands Inner loop – velocity / angle commands Maneuver – for aggressive maneuvers Switch To Command Maneuver switch for command switching Filter cmd cmd rcmd VD cmd Challenges: From Sensors Ensure continuity of constraints with command switching Closed loop stability due to saturation constraints in feedback path Continuous Torque Limit PENNSTATE 1 8 5 5 Command Filter VDcmd VD cmd lim VD Q Q Aircraft Qlim Torque Limit 1 s 1 Q pred Qm Envelope Constraint Calculation Torque Prediction (Neural Net) Measured Torque Dynamic Trim algorithm Offline trained neural network for future response prediction Sensitivity to command input for constraint calculation Constraint on descent velocity corresponding to transmission torque limit Constraint on Inner loop command Cont. Torque Limit Evaluation – Inner Loop PENNSTATE 1 8 5 5 Envelope protection off Torque exceeds the limit Envelope protection on Steady state torque stays close to the limit Constraint on descent velocity (Inner loop) command Saturation limit inside feedback loop when using altitude (outer loop) commands Moving Constraints out of Feedback Loop PENNSTATE 1 8 5 5 Move outside feedback loop wlim + w - wlim w e Why: ulim u C (s ) G (s ) y C (s ) + e - + D C ( sI A) 1 B + G (s ) + D Controller dynamics : x Ax Be ; u Cx De e w y ; D0 If controller output limit is ulim , then controller input limit is, - + y Affects closed loop stability Pilot cueing of command input 1 Example: P/PI Controllers P: {A=0 B=0 C=0 D=Kp} elim D 1 ulim Cx + ulim Command input limit, wlim elim y D 1 ulim Cx y PI: {A=0 B=1 C=Ki D=Kp} Heave Axis: Inner Loop Command – Descent Velocity Outer Loop Command – Altitude (Using Proportional controller) Cont. Torque Limit Evaluation – Outer Loop PENNSTATE 1 8 5 5 Flight Simulation Future Upgrades PENNSTATE 1 8 5 5 Honeywell 2-Axis Active Sticks Donated by Boeing - Philadelphia Bell 206 Simulation Cockpit Donated by Bell-Textron Accomplishments PENNSTATE 1 8 5 5 2004 Accomplishments Continued collaboration with Georgia Tech. – Evaluated different pilot cueing techniques Demonstrated in piloted simulation that accurate prediction algorithm and appropriate pilot cueing techniques do result in smaller maneuver time and reduced limit violations Integrated hub moment limiting constraints with adaptive model following controller Integrated continuous torque limiting algorithm with inner loop /outer loop type controller – useful for autonomous UAVs Continued to produce publications Planned 2005 Accomplishments Integrate Active Control Stick with simulator at PSU Implement CFM systems on real-time simulator at PSU Continue development of CFM integrated with adaptive model following controller for UAV’s or manned aircraft Further assessment of CFM impact on reduction of loads and handling qualities PENNSTATE Schedule and Milestones 1 8 5 5 Tasks • Hub moment limit detection and avoidance system • Develop comprehensive collective axis cueing system • Demonstrate in piloted simulation at Sikorsky • Simplify algorithms for practical flight system • Develop integrated CFM cueing and control laws • Piloted simulation / HQ evaluation of integrated CFM • Integrate with adaptive control • Quantitative assessment of benefits of CFM • Develop CFM for UAV’s • Sahani MS Degree • Sahani PhD Degree 2001 2002 2003 2004 2005 Completed Short Term Long Term PENNSTATE 1 8 5 5 Technology Transfer Activities: • Continued collaboration with Georgia Tech RCOE in simulator work, have discussed potential transfer to their UAV work • Publications - AIAA Journal of CIC, AIAA Journal of GCD (accepted), AHS Journal (under review), AIAA GNC Conference, AHS Forum • Briefings to Boeing, Sikorsky, Lockheed Leveraging or Attracting Other Resources or Programs: • Collaboration with Prof. Asok Ray on ARO MURI. Task “Damage Mitigating Control for Rotorcraft”. Incorporating CFM technology into that work. • Active control sticks donated by Boeing Recommendations at the 2004 Review: • Presentation too long, needs better focus • Talk directly to PIs from HACT project • Review team questioned usefulness of structural load limiting • Resolve export control issue when dealing with DLR Actions Taken: • Addressed in presentation development • Had discussions and exchange of information with P. Einthoven at Boeing • Addressed in following slide • No longer planning collaboration with DLR Relevance of Structural Load Limiting Work PENNSTATE 1 8 5 5 Last review the question was posed “Who Cares?” RAH-66 Hub moment limiting was implemented in RAH-66 by limiting pitch acceleration commands. Seemed to work adequately but … Constraints not as accurate our system Constraints were inside feedback loop for outer loop controls – should avoid this if possible. NASA Heavy Lift Rotorcraft Systems Effort to reduce weight Automatic Load Limiting / Hub Moment Control identified as a desirable technology for heavy lift rotorcraft Weight savings could be achieved using better loads estimates to remove conservatism – or actively limit loads to remove conservatism HACT Program Discussion with P. Einthoven at Boeing Philadelphia. He identified the need to more rigorously study closed-loop stability. PENNSTATE Overview of Accomplishments 2001-2005 1 8 5 5 Response to step input Comprehensive Collective Axis Limit Avoidance Includes transmission torque limits, engine torque limits, rotor RPM limits, OEI and autorotation limits Collaborated with Sikorsky – Evaluated in piloted simulation Longitudinal Hub Moment Limit Avoidance Developed new algorithm that provided accurate constraints over entire flight envelope Evaluated in piloted simulations at Georgia Tech. Evaluated different pilot cueing technique Demonstrated in pilot sim. : CFM results in smaller maneuver time and reduced limit violations Integration of CFM with Model Following Controller (MFC) Integrated long. hub moment limit protection with MFC Integrated torque limiting with inner loop/ outer loop controller architecture Publications Conference: 6 Journal: 2 published + 1 accepted + 2 under review f (x, ped , col col ) f (x, ped , col col ) f lim col col 0 2 col Limited Parameter Peak Response Critical Quasi-Steady Response Critical 1 Q col * col Integrated Response Q + Qlim - Qcorrected col Neural Net UH-60A Time Q predicted Normal Conditions Qlim min Qtranslim Qenglim + OEI Autorotation Qlim QOEI K omg * 0 - Qactual Qlim K omg * 0 Normal Pilot Cues: + Out the window, instruments, acceleration, vibration, etc … e 1 s 1 - eqs Pilot Pilot Force Active Control Stick Aircraft Aircraft Sensors Sensors Stick Position AFCS AFCS Prediction Algorithm: e.g. Neural Networks, On-board models Stick Constraint Closest proximity of response to limit if no change in control input Sensitivity - Transient peak due to unit step Offline trained neural networks Find dynamic trim values Stick constraint calculation Input from Aircraft Sensors Signal ylim ,l , ylim,u Q( x , t ) - 1 xf e s 1 ub x f : 1c , q ylim,l min Q x, t ~ t ulim,l max H x , t t Neural net for f i ( x s , t ) ylim,u max Q x , t ~ t ulim,u max H x , t t + g i ( x f 0 ) t Neural net for H ( x s , t ) H ( x, t ) 1 s 1 - Increment in t + long MH Longitudinal Hub Moment Limiting Algorithm Output to Limit-Cue Arbitration and Tactile Cues Scan through time With Cue Wihtout Cue 26 (ft-lb-s) Thousands 24 Maneuver Time (s) 22 20 18 16 14 12 10 0 5000 10000 15000 20000 25000 Integrated Hub Moment Limit Exceedance Factor (ft-lb-s) 20 18 16 14 12 10 8 6 4 2 0 Max Integrated Hub Moment Limit Violation Extent of 2nd & 3rd Quartiles Min Pure Over- Freq. Dead- Pure Pure Visual shoot Dist band Tactile Auto GT No Limit Protection With Limit Protection Safety (PV or ILM) (ft-lb) Thousands Inceptor Inceptor Constraints Constraints Performance (T) Variable Soft Stops 38 36 34 32 30 28 26 24 22 GT PSU PSU 16.3 16.5 16.8 18.1 18.3 18.5 Average Maneuver Time (sec) Absolute Peak Hub Moment Future Path PENNSTATE 1 8 5 5 Additional Basic Research • Potential for more research on basic stability issues of load limiting and bio-mechanical issues of tactile cueing. • Investigate use for vibratory loads limiting – need comprehensive analysis code (RCAS) • Incorporate CFM technologies within a damage mitigating control / life extending control architecture Transition to Applications / Applied Research • Near-term potential for application to UAV systems • Georgia Tech implementing their controller on Boeing Maverick UAV (uses R-22 airframe) – potential application for mast bumping / torque limiting • Long term application on heavy lift rotorcraft – use CFM methods to save weight. • There needs to be further study of certification issues for active control sticks and tactile cueing systems.