when designing good planes is just not enough. this project is pretty crazy. IMO you pushed the cg back very far for a first iteration. I would have worked my way back a bit slower but if anyone can pull this off it's you. Excellent video as always. Cheers!
@2:22, thanks for displaying our website briefly in your video, much appreciated. It also reminds me that I really need to add pictures for all of the different sheets of foam. Cool project! I'm looking forward to the next video.
The pitch oscillations evident in the video indicate that a significant issue here is rate-limited servos. You can have the perfect control loop, but if your servos cannot keep up with the commands, you'll always end up with oscillatory response.
I don't think this is the case. It's a system, and the servo responses are a part of the dynamics of this system. Many real world rockets such as those flown by SpaceX are not stable, and also use slow actuators. These systems are very well controlled due to the system dynamics being very well understood. I would hazard that faster actuators, if not properly understood, will simply lead to higher frequency oscillations and problems. As other comments have put far better than I will, properly modeling craft dynamics is the solution.
@@KnowledgePerformance7 That's a fair point. Joe Barnard at BPS.Space has done a lot of interesting things in this arena, like vertical landing a model rocket like Falcon 9 does, and his channel is worth checking out. I think, however, that "slow" actuators like SpaceX uses are far more acceptable for large heavy systems. The smaller the vehicle, the more "twitchy" it is, and the faster the control systems have to operate. Also, the less stable, the faster a response is required. Watching this video makes it clear that the vehicle is responding VERY quickly, and you cannot hope to control it unless your actuator can go stop-to-stop in about the time that the vehicle executes a single oscillation. Those actuators in this video are WAY too slow for THAT. Dial down the aerodynamic instability (move the CG closer to the aero center), and it MIGHT work with the current actuators. So in summary, I think you're right to a point, but faster actuators will be required to handle this level of instability, even if the system dynamics are perfectly modeled.
No. Oscillations are often the result of closed loop feedback system which is unstable. This can be for example caused by delay in the closed loop resulting in the feedback signal being phase shifted to the point where it becomes additive at the input to the control system. One such cause of a delay is the interaction between the human operating the demand controls and the control system, with the human introducing a delay in the feedback loop. It can just be down to adjustment of the parameters in the control system. This control theory subject is not actually taught to students until they get to an undergraduate degree course at university in Engineering. So it it not surprising that many people don't understand it.
Hey Kavin, the most compelling thing about your channel for me is the identification and exploration of novel concepts. Can't wait for the next episode !.
Cool idea. You might want to increase the strength and speed of your servos. They're doing much more work in the unstable configuration than you would usually have. Also if there is any play in your hinge, you may need to tighten things up as any sort of play will have a magnified effect.
Just a point of clarification - aerobatic planes ARE NOT UNSTABLE. Some aerobatics planes, like the Pitts are less stable then the modern competition aerobatic planes like the Extras. It comes from how competition aerobatics is scored. There are no additional points for high Gs ore tighter loops or how fast you can roll. Aerobatics has a similar scoring concept to Gymnastics and diving. Judges start with a score of 10 and then take of 1/2 points and points for deviations from the ideal figure. So if a loop is NOT round the judges start taking of points. If a line isn't straight the judges start taking off points. Pitts are great fun (I have flown one) and they are awesome to learn tail dragging in, BUT the very short coupling (length of fuselage and wing span) makes them very twitchy. I have flown in an Extra 300L (2 seater) and it is amazing how stable they are. They don't twitch and jump about like a Pitts and for very clean aerobatics they score more points.
MS student in Aerospace Controls Engineering here. Just a little comment: instability does NOT make an airplane more maneuverable, that’s a super common myth. I recommend reading the X29 papers, where researchers expected a greatly maneuverable plane but found no benefit in make it more open loop unstable. An airplane is most maneuverable when it is neutrally stable, and making it more unstable only reduces control authority. Control surface effectiveness is the main factor affecting the overall maneuverability of the aircraft. As others have said, you should create a model of your aircraft and design a controller around it. Feel free to contact me if you have any questions/need some help. All the best!
Butts, S. L.; Hoover, A. D. (May 1989). "Flying Qualities Evaluation of the X-29A Research Aircraft". U.S. Air Force Flight Test Center. AFFTC-TR-89-08.
Nice to see you pick this idea up! I've been working on something similar for a while. I plan on fully modelling the flight mechanics & actuator dynamics before hand and using that model to set up a custom flight controller. It should be perfectly doable without all that, it'll just take a lot of time to tune the controller parameters right. Most importantly try not to make it too unstable since that'll require greater servo speed & response time. Another big problem is that you won't need a lot of elevator travel while the aileron travel is largely unaffected. So it might be a good idea to split the elevons into separate ailerons & elevator. Looking forward to the progress & discoveries you'll make along the way!
With my vast experience with betaflight and high preformance quads, and trying AP on these high preformance quads AP cannot handle or is setup to this task. iNav is a fork of BF and they race with iNav on planes, I know of no high preformance racing done with AP (the pixracer was a total failure in the racing scene). AP is not a low latency pid and filters loop, they are made for larger size, heavier and slower craft and autonomous, I seriously think you will fight the limits of AP slow loop for a high preformance plane, try iNav.
@@user-rs8zg8ey2b well, I won't be using ardupilot but fully self written code on a custom board. However, ardupilot is more than fine for this application as the loop time of the PID is multiple orders of magnitude above the response of the servo actuators, which are the limiting factor here. An ATmega328 would probably still be plenty fast enough for this task, and that is way slower than the ARM uCs in the pixhawk and other ardupilot FCs.
@@tyskStefan Well good to know your writing your own code. AP is NOT successful in any racing task, so imo is is not good enough for this application in this video and maybe not your application (not seeing your application, I dont know). Of course this is not racing, but needs quick response, unlike anything I have seen AP do. AP was a pain just to get a old school mini quad flying w/o oscillating, and it never flew "good". Good luck.
@@user-rs8zg8ey2b multiple reasons: first of all because I need dedicated sensor inputs for angle of attack vanes, pitot tube and the likes as well as a beefy onboard BEC to supply the servos with. Additionally I don't need the OSD chip and some of the other features provided by multirotor FCs, I hardly use those on my multirotors... But most importantly it saves me the headache of searching for a new board and writing a new config every time a board gets discontinued.
This is in essence a really cool control system problem. I'm not sure if you have a background in this, but in case you don't I have some tips. Key to designing a good controller for your aircraft is having a good model of the dynamics of your aircraft, because that way you can tune the controller in software first behind your desk without needing to go driving for every test. There's two main ways of getting a good software model. The first is by explicitly modeling the dynamics of the plane using standard equations. If you want I can share some resources on this from my bachelor courses on this. However modeling this fully is hard, there are always some dynamics that are going to be hard to capture. So that leads to the second approach, which is called system identification. Basically what you do is log the inputs you give the actuators and the response in terms of pitch, yaw and roll and feed that to a program that will infer the plane dynamics from that. Then you can tune your controller based on that. If you have any questions please let me know, always happy to give my 2 cents.
@@marthinwurer When I took Flight modelling and automatic control in college last semester, we used "Flight Stability and Automatic Control" by Robert C. Nelson. A bit dated and somewhat of a tough read but it's great for making mathematical models of aircraft dynamics. Then, making a control system of your own based on that model can be done, Anderson and Moore's "Linear Quadratic Methods" is great but reads like ancient greek and the subject matter is still half black magic to me.
@@Mrissecool Man I felt some relief when you said "subject matter is still half black magic to me" because as an aeronautic engineer I felt kind of bad for not completely grasping the subject, is good to know I´m not the only one. BTW the first book you mentioned, I love it!
This approach is used for adaptive controllers. First part entails system identification, then it can program the pids on the fly. However what he is suggesting is to use just the identification for creating a mathematical model, from which you can create a stable PID setting. This can be done in MATLAB once you have the stored the data. You need the transfer function from inputs to sensor readings (outputs). Typically this requires an iterative algorithm which over time converges to the correct model.
I think you could try increasing the moment of inertia in the pitch axis, so instability will onset slower and give the servos more time to react. Full sized relaxed stability aircraft or slightly unstable aircraft are flyable (albeit with difficulty) by humans, since it takes a while for deviations in the pitch axis to put the plane out of control.
Not quite the same as your study; but I won a fair number of soaring contests using a Hobie Hawk with a very rearward CG. In clear air any thermal would upset the sailplane outward; and by just adjusting the elevator & rudder trim to a preset turn against the upset, the Hawk would catch the thermal and start taking its ride up on the thermal elevator. When the Hawk got too far downwind, the trims were put back to neutral and the Hawk was again sent straight upwind hunting for another upset and thermal.
Looks like your controller is underdamped. If possible in your SW, tune your PID with less proportional gain (to reduce overcorrection), and more derivative gain (to anticipate overcorrection and dampen it). I also agree with those saying to start with a less drastic CG shift. Very interested to see how this develops - you definitely earned my sub.
I will probably use the autotune feature and slowly walk the CG back. I am not great at tuning PIDs, very curious to see (if it can stabilize it) if it follows your suggestion.
That guy asking if its a Zagi was awesome. Brought back so may memories of my zagi's, hahah messing with brushed 400 motors and nicads. I miss those days. Great video love it!
I built a 23 degree forward swept wing many years ago, motivated by that Nasa plane also. They have they same beneficial drag reduction as a conventional rearward swept wing, but none of the downsides. Having the wing swept forward does do some interesting beneficial things to drag also- it gives positive feedback both because of more of the leading edge is ahead of the balance point and because of slight flexing of the wing. The upshot of the unstable "positive feedback"effects is that it takes MUCH less elevator deflection to change angle of attack. On my plane, it was so unstable and so quick to react it could literally do a zero radius 180, and was more or less unflyable until I calmed down the balance point.
So I think you will find that you will get the controller to keep things pseduo-stable when the angle of attack is low; but when it deviates too far, the controller won't be able to catch up in time and you'll get a deepening stall or dive. Rather like what you can see. What you could do to counteract this, is an aircraft where there is a lot of mass well in front, and well behind the centre of gravity. That would increase the rotational inertia around pitch axis, and slow down those pitch oscillations. you can still move the CG wherever you want but the inertia would be higher. much higher aircraft inertia around pitch axis and much lower inertia for elevons. v interesting project good luck!
Hey, that’s very cool. I’ve done this already with my tailsitter that I built some years ago. Never got it fly stable on high speeds though. Had used all symmetrical airfoil. Maybe you just put the motors on the LE and activate tailsitter mode. I can help you tuning the vertical hover perfect without flying. Then you have a backup if the horizontal flight isn’t stable. BR David
I bet it would be easy to stabilize that with up/down thrust vectoring the two motors, especially at low speed when the elevons don't have much authority.
I think this is going to be extremely hard without active AoA sensing. The flight controller can do it's best to damp oscillations, but this sort of instability is considerably worse than just oscillations. Lets say you want it stable at 5' AoA. There is an upset, and the aircraft tilts up to 6'. The gyro can detect that, and strongarm the controls to return the aircraft to the original orientation, but the direction of flight will change in that time, so the original orientation might have an AoA of 4'. If the controls then return to the 'neutral' setting (metastable at 5'), the aircraft will still find itself with a pitch down movement. If held perfectly at the same orientation the flight path will start to oscillate up and down. We want the aircraft to be stable (after computer control) with regards to AoA. In order for the computer to do a good job of this, it needs some good way of understanding where it is wanting to stabilise towards. Even if you manage to get the computer to infer the AoA from inertial measurements, a gust of wind will throw all that off. One design that I have been looking into is weather-vaning canards. The idea is that the canards are placed behind their axis, and are controlled either by a motor that outputs a certain torque for given control input, rather than setting the position, or a smaller control surface on the back of the canards. They effectively get an AoA term affecting their position automatically.
Also looks like air-speed is the cause of all the crashes as the airplane gets less and less stable. Aeroplane dynamics and the required response time is speed dependent. Can you put a speed-limit in Ardupilot while you tune it up? How good is Ardupilot's airspeed? Maybe add a Pitot tube to improve accuracy? Does Ardupilot have an internal model of the airplane it's flying? Would help a lot if that was tuned to match your flying wing. Even a 1-D or "homework level" model of the airplane can be enough to let you tune the flight controller in simulation and then can be used in-flight to hide most of a non-linear system from the PID controller. i.e. PID says I need "X" pitch torque, airplane model translates that to "Y" elevon based on air-speed. Or the instability could be simplified as a negative stiffness torsion spring for the PID loop, and the airplane model would set the stiffness and update PID gains at any measured speed.
For a flying wing, your best cruise L/D will be achieved when your CG is exactly on your CL at your optimal angle of attack. This puts the equilibrium point exactly on the undeformed wing and allows your nice airfoil to operate as intended. For example, a DAE-11 section achieves max L/D at about 6.5 degrees AoA. At this angle, your CL is at 34.7% chord. The airfoil has a neutral point of 25.6% chord, so your optimal static margin is -9.1% and you require extensive pitch stabilization.
I was thinking that too. Isnt having the CG either forward or aft of the CP going to force your elevators to generate some lift up or down to counteract the CG/CP moment? Thats just additional drag.
Might be too simplistic an approach but, how about setting up the plane so the CG is closer to the neutral point? Then as you get things tuned in Ardupilot you can shift the battery pack incrementally rearward until you get it the way you're looking for.
For someone like you...Flying a AR Wing PRO... have to be a boring walk around your own backyard... Dam man!! what a way to bring new excitement in to this hobby.. for sure I will be following you all the way to the conclusion on this particular project... "I'm sad to say that I don't know anything a bout Ardupilot...... but I have spend well over 4 years using iNav and since I don't do complex projects like you... I wouldn't try!!! I do hope to see what comes out of this!! I love that it almost look like a brutally unstable large Nano Goblin.... 😅
I does unstable design myself and I can say it is best design. If center of gravity is in front it generally required higher speed. If center of gravity is in back it fly nice in slow cruising.
Canards instead of elevators can offer a similar unstable but beneficial setup - they provide upforce at the front of the plane rather than downforce at the rear. This requires more fine control, but means no negative lift in stable flight, at any altitude. Upshot is the CoL can remain behind the CoM.
Great demos of divergent stability at 8:45, 8:52. 🛫🤣🛬🤣 Might be educational for viewers if the the topic of "convergent vs. divergent stability, and neutral stability" were covered in a future video.
Can the servos use the 400Hz refresh rate? I'm under the impression that you'd need special digital servos to take advantage of 400Hz refresh rates. I think helicopter tail rotor servos can use this rate but I don't think these types of servos are very common. Thanks for sharing your experiments with us. I've often wondered about unstable RC aircraft myself.
Just have to ask, why is this any better than a canard aircraft? Such as the Long-EZ and Velocity manned planes. Moving the CG behind the main wing Aerodynamic center is essentially doing the same thing as a canard configuration - it puts the CG between the aero centers of your lifting surfaces (in this case the wing and elevators) Cheers to the new idea and keep having fun!!! PS I still want to check out the code you and Daniel used for the ekranoplan height sensing, would save me a bunch of time!
It seems that drag reduction is a primary driver here. When CG is forward of the AC, a small amount of up-trim is needed to maintain level flight. That amount of up-trim results in drag. This is a stable configuration. Since gravity is always pulling the nose down, this up-trimmed configuration seems to work pretty well for landings too. So, in an effort to eliminate that built-in drag, it seems that the designer is moving the CG aft of the AC in order to eliminate the up-trim. There is more to it, I'm sure but I think that's one of the starting points for exploring this dynamic.
Another issue with canard configs is that the flow over the main wing gets heavily modified by the canard ahead of it. If the canard is for high maneuverability at high AoA, like on some modern fighters, then that's a great feature; but if efficiency is your primary goal, it's a bug.
In order for a canard airplane to not deep stall, the canard has to stall before the main wing. That means the main wing can never fully develop it's CLmax, so the wing is oversized to begin with. It also means that high lift devices like flaps are ineffective because you're still limited by your canard. As a result, a canard with a wing sized to meet a given takeoff performance will need a larger wing, and thus be slower and draggier at cruise compared to a conventional plane that can achieve the same takeoff performance with the use of flaps and smaller wings. To get around this you either need flaps on the canard that deploy with the main wing flaps to raise the CLmax, move the canards forward during landing and takeoff to achieve a greater moment to counteract the pitching moment of the flaps and high AOA main wing, or accept that if the plane stalls it will be unable to recover. Using flaps on the canard is problematic because that's where your elevators go and it also makes the canard lift curve steeper, degrading your static stability. Moving the canards forward is viable, and was done via adjustable sweep in the Beechcraft Starship, but it's heavy, complicated, and also steepens the canard lift moment curve (thanks to the longer lever arm) degrading static stability. Finally, accepting the risk of deep stall poses similar control system issues as presented in ThinkFlight's video, but hobbyist grade stall recovery software, stall sensors, and AoA sensors are far less developed than hobbyist control software to deal with aerodynamic instability, thanks to multirotors which are inherently unstable.
What about Prandl wings? Efficiency is not speed. If you fly slower and have no tip vortexes and proverse yaw and thus no vertical stabilizers the drag is reduced alot. Look into Albion Bowers Research with Nasa and the Flying like birds series he has on youtube. Birds are flying wings without vertical tails. Dont see many birds with spin departure problems right?
@@mortache is seem to make sense if u mean most optimal at least? Any reference? Anyways, too much proverse yaw is better than too little (in terms of stall spin). And how about pitch axis stability of prantdl? Is it 'autostabilizing' provided good trim (it likely seems). also do you know if prantdl is autostabilizing the pitch provided slightly bad trim (like say e186 should be, as airspeed raises or lowers the changed pitching moment then fixes the angle of attack)?
@@vidmasze i don't know a LOT about these wings, only that they hqve exceptionally high twist making the wingtips neutral or negative angle of attack. Something like 20-30 degrees
Sailplane pilots will regularly compete with the most aft CG possible because of the efficiency gains, but don't expect much. It takes downforce on the tail to counteract the CG being ahead of the center of lift. Given the location of the downforce(H-stab in a standard config) and the arm from the center of lift of the main wing to the center of lift of the H-stab, you can calculate how much down force the plane is working against, and then run that through the L/D and get an idea of how much thrust is lost for stability. Rutan canard designs take advantage of this as well by lifting with the equivalent of an elevator up front. Anyway, I'm guessing it will be hard to document any efficiency gains because they will be pretty small but I'm looking forward to the conclusion!
I would suggest to play with pitch inertia. Extend the motors backwards and move the battery forward, while keeping the same target CoG. Second I thing a very specific PID (PIFF) settings are required to achieve stability.
It would be great to see a video where you explain what you do in terms of autotune, magfit etc... I have always wanted to try this and thought about moving the CG a very small distance and then running autotune, repeating the process over and over again.
Whenever I built a new wing I always flew in this configuration, tail heavy. Swept back wing does not require any vertical stabilizers too. I flew my EPP wings without fins quite often and they flew great, just had to keep a little speed up for landing. But that is neither here nor there.
The autopilot update is 400hz - and the Saturn rocket, which lifted the crew and the module to the space, and to the moon, without any major problems, had an update of 50hz - so the update speed is probably not a main problem. I certainly doubt that servo motors can respond that quickly
In your video you mention "normal wing elevons are trimmed up", I see this a lot on wing gliders, I have found out the cg on these planes are too far forward, requiring more up trim, slide the cg back until no more up trim is needed, this gives more pitch authority, easier to tip stall, so dial in less max up/down elevon deflection.
I really love this channel and can't wait for the next video, and I have one small suggestion. Can you bring your voice a bit up and some of the flying footage volume down? They are quite different in levels and we can't hear you properly.
You know, I need to do some research myself on this but I think that your standard servos that have like a 60deg/100ms slew rate *only* move that quickly when they're *already* moving. I think the rotational mass of the moving parts of the servo means you get a lot, lot less, like maybe 10deg/100ms in that first 100ms. Its fine if you're going from +90deg to -90deg in one shot, but for stabilization, when you're constantly changing the direction of the servo, you get an order of magnitude less rotation per unit time than the servo's listed specifications. I'll have to take one apart and tap off the position potentiometer and see just how slowly servos move from a dead stop. Obviously it will be even worse when you're making it actually change directions. If you're making a 30 degree change in aileron position and (all else being equal) it actually takes 500ms instead of the 50ms the servo specs say it will, then its no wonder the flight controller and/or IMU have so much trouble stabilizing anything.
I appreciate your experiment as your arguments seem aerodynamically valid. As you stated, wings could be reduced in size while achieving better L/D ratios by having the CG further back. My calculations led to similar results, but I then stopped my research as the instability issue seemed too risky to me. IMO it depends on the weight and distance between CG and NP that needs to be compensated by the control surfaces. I mean, having the CG behind the NP means that changes in pitch PLUS the weight moment need to be managed through the control surfaces. On the other side, you want to keep the control inputs as small as possible, as they are augmented by the CG position. A linear expo is not bringing you close to the goal, I guess.... I'm very curious about the proceedings of this study. Looking forward to the next videos
You will need to use helicopter tail rotor servos with the fastest throw rates you can buy. Your cycle rates are getting bottlenecked if you are not running those specialty servos.
View at 6:19 caught me off guard. Up to this point I'd assumed this was one of your classic designs with a single pusher motor. But then an awe-haw moment, two motors to eliminate motor torque from this instability experiment? My excuse for overlooking this detail is the rearward facing camera was not centred on the longitudinal axis. The biggest hurtle with this project to learning to tune and fly an instable plane during the the launch phase of flight, which in itself is more unstable than other phases of normal flight. My suggestion would be to start (in)stability testing and tuning after stable level flight is achieved. A simple way to accomplish this is to add a servo that drops a weight attached under the planes forward fuselage with an optional “remove before flight” streamer. (to reduce fall speed, and to make it easier to locate later) I noticed that this UAV doesn't have a true airspeed indicator? Without true airspeed, Ardupilot is not able to calculate/remove windspeed from a GPS derived groundspeed! Hint: while you may be flighting in light winds, the plane is not always flying parallel to the ground, so ground speed would be just as unstable as the UAV. 🤯 No real IFR test pilot would attempt to fly an unstable aircraft using only an artificial horizon and vertical speed without airspeed as a cross reference. BUT this is the task you are requesting ArduPilot to attempt. Personally I'd go a different route to installing pilot tube for airspeed as it is likely to undergo a RUD (rapid unplanned disassembly) during repeated durability testing. An "Angle of Attack" (AOA) sensor would provide much quicker response rate and more accurate data. Pilot tubes work best when in a near-zero AOA environment, otherwise they report inconsistent performance data. Another reason to delay adding a pilot tube. An AOA typically install on the forward side of the fuselage, so should be protected during UAV durability testing. FYI: all fighter jets rely on AOA to ensure stay within the engineered flight performance box, particularly during high-alpha, and/or high-G manoeuvres . While I haven't seen a good AOA sensor marketed for using on small UAV's, it should be possible construct a useful AOA sensor that would be compatible with the Matek flight controller. One option would be to use a resistive potentiometer, like that found in a servo, and feed a voltage to the FC's spare analog to digital input. An optical rotary encoder could be an alliterative, but would not be FC compatible Another option, would be to use a diametric magnet (magnetized across it diameter) and a hall-effect sensor to track position to a fine degree. Other methods could measure pressure differential on the wing, but this would be a complicated project itself, thus would become a distraction. BTW: with a working AOA, I suspect even Nicholas's dRehmFlight software could tame this unstable flying wing. It should also be possible to estimate true airspeed based on an aircrafts AOA and motor power level. This would need calibrating, but could be a technique that prove useful enough to include on future aerodynamic explorations. A couple thought that could be follow-ons to explore once stable efficient flight is achieved.
Move the center of gravity forward but also add kennard wings that pivot and a higher aspect ratio. The kennard wings use positive control input. Also the angle of attack can increase approaching stall speed.
When I build a plane with ardupilot many years ago we did a lot of hard ware in the loop simulation with x plane and the ardupilot. That worked well and saved us many crashes, as the thing flew ok first go.
A canard wing up front will soften the pinching up and down, thus possibly make the autopilot time to catch up! Without having to dial in up elevons, which is a draggy solution! But I do like Herr Günter's suggestion!
With an aircraft that unstable, you need an angle of attack probe with sensor for it to work. Then your stick on your controller (fore and aft) for elevator control becomes an angle of attack "setter" and provides input for the controller. Then set your control system to maintain that angle of attack that the stick is asking. A really simple analogy would be zero stick equal zero angle of attack and full aft stick equals +13 degrees assuming wing stalls at 15degrees. Full forward stick equals -13 degrees angle of attack. That would be the basis of a simple fly-by-wire flight control system.
I finished a similar project with a very short plane (forward swept), to make PID work correctly you need very high values, just as much as your servos can handle. good luck!
I was thinking, perhaps move the motor in front of the wing and a bit outwards so the airflow pass through the control surface, perhaps it would help the FC stabilize it more effectively
Yes, agility is the main goal of instability. The fighter jets using it are not concerned about efficiency. Yes, flying around with elevators down (like flaps) would give more lift, but also more drag (like... flaps). This would be a hard road even with a conventional plane layout, but a plank style wing is really asking for problems. You'd at least want the motors in front of the control surfaces for any hope of controlling it at low speed (or swivel the motors maybe). I think the less work the elevator has to do the better, so having the CG very close or right on top of the aerodynamic center, and a plane design (airfoil etc) that naturally produces low pitch moment and high pitch inertia is best. The perfect example is a sailplane with long tail and small elevator. The worst would be where you have to constantly stick a control surface into the airflow. There is a full-size plane called Celera 500 that claims to be the holy grail of efficiency, would be interesting to see model-scale testing of those concepts (stretched egg looking fuselage and super skinny wings).
Great video! It was very informative and well presented. It would be great to see more videos on XFLR5 and aircraft design in the future. It will be a great help for those who are interested in aircraft design and simulation. Keep up the good work!
If you add an elevator about 2 feet behind the plane, it will allow you to play with the center of gravity without losing stability. I know you want a flying wing, but the elevator doesn’t have to be bigger than 10% of the plane width if you mount it back far enough
Great video... Suggestion,. adding a lot of extra wieght would counter balance the extreme lift? I.e. up verses down... It's ok to have a highly efficient aircraft that only works (is stable) with a huge amount of extra weight... It would serve as a good cargo plane or super large commercial airplane, that could carry much more passengers, 2x the amount, meaning less flights overall and less carbon emissions...
Wow! I just say the idea of changing the CG in flight.. what a nice concept... also what about using a bungee cord!!!! I use it all the time for my large birds.
Get the fastest servos you can buy and add the a large servo arm that would convert small movements at the servo to large movements at the ailevons. Thats how you will get the fastest response possible.
Well, back in the late 70s, I built a series or flying wings to figure the minimum reflex required to achieve stability. I found that a swept back wing required far less reflex than a plank wing. I didn't take it to a final conclusion though. I just cut my tests short at zero reflex at the root, tapering to 25% chord reflex at the tip. The tip chord was 60% of the root chord. So that left me with 9% of the total wing area reflexed compared to the standard 25% for plank wings of the day. That was as far as I cared to push the envelope on this before building a larger RC wing, I'm guessing I could have gone further before reaching pitch instability though. The way I developed the reflex was pretty basic. I used a fairly high cambered thin profile. I tool a straight line from the nose point through the reflex point on the mean camber line, projecting that to the trailing edge. Then just reflexed the mean camber line in a smooth curve to meet the TE point which was naturally above the root reference line. So it would be fair to say that this graphic rule of thumb solution was not just reflex but incorporated washout to the tip as well. I didn't know any analytical aerodynamics at the time, but I suppose it had a kind of vague similarity to the Prandtl wing in that to unloads the tip. Anyway, I just thought it might be of some interest, see it here : drive.google.com/file/d/1s-Z7ob6zHEiKgaB_0w8HHgwYFyc8hrrZ/view?usp=share_link I think that if I built this again, I'd just do elevons tapering to put most of the control surface at the tips. As it was, I think the inboard elevators weren't all that effective except by altering mean cMo to effect stability.
My diagnosis would be that the controller needs stronger damping of angular acceleration and/or velocity on the pitch axis. I've never worked on a controller for real-life aircraft so I might be wrong.
I just stumbled on your channel, and have never seen this particular method of building. When I last built EPP models for slope soaring, Goop skim coats and fiber tape was the goto method for combat and rough duty models. My old timer/newb question...Where to get propyl acetate?
I'm guessing that because the plane is very short from front to back, it has a low pitching inertia. This results in pitch-changes occurring more rapidly, as a result of which the stabilization system needs to react to pitch-changes more rapidly. Pitching inertia could be increased in either of two ways: by increasing the size of the plane; or by giving the plane swept wings.
When I was flying, I put the center of gravity ahead by 1/4 inches. My brother-in-law put it 1/2 inches in the back! His fingers were a lot faster than mine, he could make an airplane fly backwards!
No one can fly an unstable aircraft without using normal G force for gain,or using an AOA sensor. So my friends and I are using angle sensor as AOA sensor to fly model fighters ,use AOA gain to canard or elevator.If you use gyro only, it will not work.
I’m not entirely certain, but I think the pitch control may be reversed for this aircraft. We might be tempted to think the elevons are producing lift (up or down) on the elevon surfaces, but I don’t that’s the right mental model for this configuration. I think deflecting the elevons one way or the other changes the amount of lift being produced on the main wing. And because the aerodynamic center of the main wing is ahead of the CG rather than behind it, the pitch controls are reversed from normal. So deflecting the elevons downward produces an upward pitch moment, and deflecting them upward produces a downward pitch moment.
ha ha - sometimes controller introduced frustration kicks in, part of the control problem comes from using linear controllers in non linear ranges - to nonlinearise the control PIDs- one needs to be able to something like gain scheduling (a lookup table with velocity vs PID values - (in pitch)) based on airspeed. Flying under manual control controlling the +/- Flaperons with a side slider you should be able to get to a stable takeoff/ landing setting and a cruise / high speed setting. (I have used similar in manual RC planes with really low wing loading to get them behind the power curve for controllable short landings - the sort of aircraft which otherwise lock into ground effect and never want to touch down.).
Maybe its better to use a Canard-Configuration This is more efficent, because there is no Need for negative Lift on the Stabiliser. You can trim those Planes stable for initial calibrating your Flightcontroller Settings. Than you can optimise the CoG for best Efficency without totally loss of Stability.
could it also be that plane (sort of) lacks airspeed to be controllable well enough? would a catapult or wheels help? did you already some do some Autotune or manual adjustment of PIDs already? another interesting topic will be how to do stall recovery with "bad" CG... movable CG or have some quad-copter or small heli-copter style motor(s) for recovery?
You need to do a similar investigation into Prandtl NASA style wings that are finless and have the "tip" vortices moved inboard. Allegedly they are the most efficient wing for a given weight (not given span).
how controlled is your motor mounting and considerations for center(s) of thrust? edit: i.e. addition/reduction of thrust can cause pitching, yawing, and rolling. so your wing isn't only designed for a specific weight, speed, and AoA combo, it is also designed for a specific thrust (and the accompanying pitch, yaw, and roll induced by said specific thrust)
Would this perhaps work better with thrust vectoring? Just a guess, but wouldn't the effectiveness of the elevon's change dramatically and unpredictably at higher angles of attack as the air flow begins to separate. Making tuning the autopilots output next to impossible without a wind tunnel. At least with thrust vectoring the pitching moment they would apply at a given power% and angle would be more predictable, particularly if they were in front of the wing. Though I suspect their effectiveness would still change depending on the airspeed because the blades are fixed pitch.
I was gathering from the video that the point of the instability was so that the elevons could have negative reflex and increase lift at slow speeds. That effect would be negated if thrust vectoring was used instead.
@@FlyXenonRC That effect could be amplified further if the elevons weren't constantly oscillating to provide roll authority. and instead acted more like self-adjusting flaps that only responded to pitch. In any case using thrust vectoring or differential thrust could provide more pitch stability.
Back in the day....I would occasionally fly a model with the c.g. way out of whack, further rearward than your line indicates from center on this craft. It was always a handful to gain altitude, circle, and back down to terra firma. But not nearly as squirrely as that flight controller trying to wrestle your wing into the air. have you tried to manually fly the aircraft?
Try launching it upside down. I built an unstable control line flying wing that alway dove for the ground and until it was launched upside down it was unflyable. It floated around upside down until it built up speed, became unstable again and flipped over at high speed becoming stable again. When the speed bled off it was another flip and landing upside down. I should have shown it off more. It was wild and taught me something. I'm still not sure what.
Major isssue is airflow over the wing and control surfaces. with the wing infront of the props the airflow over the wing is determined by the speed of the wing through the air and at low speeds and high angles of attack there is little authority. put a control surface behind the wing and you may solve the issue,
Explain me one thing please. If the autopilot calculates 400 times a second but the servo get new positions only 50 times a second, do the 400 hz make sense?
hm control engineering was a long time ago but out of the top of my head. Even on 50 Hz, letting the control loop run at 400Hz, it could improve still help for noise from input sensors. It could still help as reading of inputs and acting on them is not necessarily close together when on the same frequency (no idea how the code is setup). And if you don't adjust gain values then it could still have a fortunate effect too but that could be easily fixed by putting in the correct gains. But actually the servos are doing 400hz now too 9:30 . ;)
not that I'm qualified to answer but yes and no. In some ways no because as you're thinking, the servos can only move at 50hz..but a 400hz sample rate allows much finer analysis of what's ahppening to teh craft and as this is r&d I'd assume that's valuable in itself. In analagous areas having high quality data that you then downsample to get from in this case 400hz sampling to 50hz action can still be useful as at least you have a better picture of what's happening to base your 50 actiosn per second on if that makes sense. ANyway, I wondered the same thing...hope TF replies to your question ;-)
Great content as always. How would you feel if a manned plane was tail heavy though? Would you get in it? Would love some more content on your design process at some point, particularly airfoil selection. Why one or the other, how do you decide etc.
Have you thought about moving the center of gravity inflight, with a servo? This would allow to start your plane in a stable configuration.
Kind of like Concorde and the trim fuel tank in the tail? Boeing has an airplane with fuel tanks in the horizontal stabilizer too....
@@231pilot yes similar to similar to that. it would allow to fly high in the air and testing of the unstable configuration away from the ground.
nice idea to shift weight, esp to ease the exploration for finding what's the good-enough spot for CG
@@231pilot Also the F22 uses air bladders to push fuel forward to counteract the CG change when dropping heavy bombs.
Clever
when designing good planes is just not enough. this project is pretty crazy. IMO you pushed the cg back very far for a first iteration. I would have worked my way back a bit slower but if anyone can pull this off it's you. Excellent video as always. Cheers!
Exactly. "Crawl before you walk."
He is building Porco Rosso plane! Hard to get from the ground but excellent in the sky.
@2:22, thanks for displaying our website briefly in your video, much appreciated. It also reminds me that I really need to add pictures for all of the different sheets of foam.
Cool project! I'm looking forward to the next video.
I can only imagine that you climaxed when he increased the quantity to 100.
I wish I had $4000 in foam money.
The pitch oscillations evident in the video indicate that a significant issue here is rate-limited servos. You can have the perfect control loop, but if your servos cannot keep up with the commands, you'll always end up with oscillatory response.
I don't think this is the case. It's a system, and the servo responses are a part of the dynamics of this system. Many real world rockets such as those flown by SpaceX are not stable, and also use slow actuators. These systems are very well controlled due to the system dynamics being very well understood. I would hazard that faster actuators, if not properly understood, will simply lead to higher frequency oscillations and problems.
As other comments have put far better than I will, properly modeling craft dynamics is the solution.
Oscillations can also occur in a perfect "instant" system. Looks more like bad pid tune than rate limiting.
@@KnowledgePerformance7 That's a fair point. Joe Barnard at BPS.Space has done a lot of interesting things in this arena, like vertical landing a model rocket like Falcon 9 does, and his channel is worth checking out. I think, however, that "slow" actuators like SpaceX uses are far more acceptable for large heavy systems. The smaller the vehicle, the more "twitchy" it is, and the faster the control systems have to operate. Also, the less stable, the faster a response is required. Watching this video makes it clear that the vehicle is responding VERY quickly, and you cannot hope to control it unless your actuator can go stop-to-stop in about the time that the vehicle executes a single oscillation. Those actuators in this video are WAY too slow for THAT. Dial down the aerodynamic instability (move the CG closer to the aero center), and it MIGHT work with the current actuators.
So in summary, I think you're right to a point, but faster actuators will be required to handle this level of instability, even if the system dynamics are perfectly modeled.
Agree that the smaller the model, the faster the response loop has to be. Would stepper motors give a faster response?? Or perhaps linear actuators?
No. Oscillations are often the result of closed loop feedback system which is unstable. This can be for example caused by delay in the closed loop resulting in the feedback signal being phase shifted to the point where it becomes additive at the input to the control system.
One such cause of a delay is the interaction between the human operating the demand controls and the control system, with the human introducing a delay in the feedback loop.
It can just be down to adjustment of the parameters in the control system.
This control theory subject is not actually taught to students until they get to an undergraduate degree course at university in Engineering. So it it not surprising that many people don't understand it.
Hey Kavin, the most compelling thing about your channel for me is the identification and exploration of novel concepts. Can't wait for the next episode !.
Cool idea. You might want to increase the strength and speed of your servos. They're doing much more work in the unstable configuration than you would usually have. Also if there is any play in your hinge, you may need to tighten things up as any sort of play will have a magnified effect.
Just a point of clarification - aerobatic planes ARE NOT UNSTABLE.
Some aerobatics planes, like the Pitts are less stable then the modern competition aerobatic planes like the Extras. It comes from how competition aerobatics is scored. There are no additional points for high Gs ore tighter loops or how fast you can roll.
Aerobatics has a similar scoring concept to Gymnastics and diving. Judges start with a score of 10 and then take of 1/2 points and points for deviations from the ideal figure. So if a loop is NOT round the judges start taking of points. If a line isn't straight the judges start taking off points.
Pitts are great fun (I have flown one) and they are awesome to learn tail dragging in, BUT the very short coupling (length of fuselage and wing span) makes them very twitchy. I have flown in an Extra 300L (2 seater) and it is amazing how stable they are. They don't twitch and jump about like a Pitts and for very clean aerobatics they score more points.
MS student in Aerospace Controls Engineering here. Just a little comment: instability does NOT make an airplane more maneuverable, that’s a super common myth. I recommend reading the X29 papers, where researchers expected a greatly maneuverable plane but found no benefit in make it more open loop unstable. An airplane is most maneuverable when it is neutrally stable, and making it more unstable only reduces control authority. Control surface effectiveness is the main factor affecting the overall maneuverability of the aircraft.
As others have said, you should create a model of your aircraft and design a controller around it. Feel free to contact me if you have any questions/need some help. All the best!
Butts, S. L.; Hoover, A. D. (May 1989). "Flying Qualities Evaluation of the X-29A Research Aircraft". U.S. Air Force Flight Test Center. AFFTC-TR-89-08.
Nice to see you pick this idea up! I've been working on something similar for a while. I plan on fully modelling the flight mechanics & actuator dynamics before hand and using that model to set up a custom flight controller. It should be perfectly doable without all that, it'll just take a lot of time to tune the controller parameters right. Most importantly try not to make it too unstable since that'll require greater servo speed & response time. Another big problem is that you won't need a lot of elevator travel while the aileron travel is largely unaffected. So it might be a good idea to split the elevons into separate ailerons & elevator.
Looking forward to the progress & discoveries you'll make along the way!
With my vast experience with betaflight and high preformance quads, and trying AP on these high preformance quads AP cannot handle or is setup to this task. iNav is a fork of BF and they race with iNav on planes, I know of no high preformance racing done with AP (the pixracer was a total failure in the racing scene). AP is not a low latency pid and filters loop, they are made for larger size, heavier and slower craft and autonomous, I seriously think you will fight the limits of AP slow loop for a high preformance plane, try iNav.
@@user-rs8zg8ey2b well, I won't be using ardupilot but fully self written code on a custom board.
However, ardupilot is more than fine for this application as the loop time of the PID is multiple orders of magnitude above the response of the servo actuators, which are the limiting factor here. An ATmega328 would probably still be plenty fast enough for this task, and that is way slower than the ARM uCs in the pixhawk and other ardupilot FCs.
@@tyskStefan Well good to know your writing your own code. AP is NOT successful in any racing task, so imo is is not good enough for this application in this video and maybe not your application (not seeing your application, I dont know). Of course this is not racing, but needs quick response, unlike anything I have seen AP do. AP was a pain just to get a old school mini quad flying w/o oscillating, and it never flew "good". Good luck.
@@tyskStefan Curious why a custom board, because F4 FC are less than $20 nowadays? What do you need different?
@@user-rs8zg8ey2b multiple reasons: first of all because I need dedicated sensor inputs for angle of attack vanes, pitot tube and the likes as well as a beefy onboard BEC to supply the servos with. Additionally I don't need the OSD chip and some of the other features provided by multirotor FCs, I hardly use those on my multirotors... But most importantly it saves me the headache of searching for a new board and writing a new config every time a board gets discontinued.
This is in essence a really cool control system problem. I'm not sure if you have a background in this, but in case you don't I have some tips.
Key to designing a good controller for your aircraft is having a good model of the dynamics of your aircraft, because that way you can tune the controller in software first behind your desk without needing to go driving for every test.
There's two main ways of getting a good software model. The first is by explicitly modeling the dynamics of the plane using standard equations. If you want I can share some resources on this from my bachelor courses on this. However modeling this fully is hard, there are always some dynamics that are going to be hard to capture.
So that leads to the second approach, which is called system identification. Basically what you do is log the inputs you give the actuators and the response in terms of pitch, yaw and roll and feed that to a program that will infer the plane dynamics from that. Then you can tune your controller based on that.
If you have any questions please let me know, always happy to give my 2 cents.
I'd love to hear about what textbooks you used to learn this stuff in college!
I'd like to hear which textbooks you learned this in as well.
@@marthinwurer When I took Flight modelling and automatic control in college last semester, we used "Flight Stability and Automatic Control" by Robert C. Nelson. A bit dated and somewhat of a tough read but it's great for making mathematical models of aircraft dynamics. Then, making a control system of your own based on that model can be done, Anderson and Moore's "Linear Quadratic Methods" is great but reads like ancient greek and the subject matter is still half black magic to me.
@@Mrissecool Man I felt some relief when you said "subject matter is still half black magic to me" because as an aeronautic engineer I felt kind of bad for not completely grasping the subject, is good to know I´m not the only one. BTW the first book you mentioned, I love it!
This approach is used for adaptive controllers. First part entails system identification, then it can program the pids on the fly. However what he is suggesting is to use just the identification for creating a mathematical model, from which you can create a stable PID setting. This can be done in MATLAB once you have the stored the data. You need the transfer function from inputs to sensor readings (outputs). Typically this requires an iterative algorithm which over time converges to the correct model.
I think you could try increasing the moment of inertia in the pitch axis, so instability will onset slower and give the servos more time to react. Full sized relaxed stability aircraft or slightly unstable aircraft are flyable (albeit with difficulty) by humans, since it takes a while for deviations in the pitch axis to put the plane out of control.
Not quite the same as your study; but I won a fair number of soaring contests using a Hobie Hawk with a very rearward CG. In clear air any thermal would upset the sailplane outward; and by just adjusting the elevator & rudder trim to a preset turn against the upset, the Hawk would catch the thermal and start taking its ride up on the thermal elevator. When the Hawk got too far downwind, the trims were put back to neutral and the Hawk was again sent straight upwind hunting for another upset and thermal.
Looks like your controller is underdamped.
If possible in your SW, tune your PID with less proportional gain (to reduce overcorrection), and more derivative gain (to anticipate overcorrection and dampen it).
I also agree with those saying to start with a less drastic CG shift.
Very interested to see how this develops - you definitely earned my sub.
I will probably use the autotune feature and slowly walk the CG back. I am not great at tuning PIDs, very curious to see (if it can stabilize it) if it follows your suggestion.
That guy asking if its a Zagi was awesome. Brought back so may memories of my zagi's, hahah messing with brushed 400 motors and nicads. I miss those days. Great video love it!
I built a 23 degree forward swept wing many years ago, motivated by that Nasa plane also. They have they same beneficial drag reduction as a conventional rearward swept wing, but none of the downsides. Having the wing swept forward does do some interesting beneficial things to drag also- it gives positive feedback both because of more of the leading edge is ahead of the balance point and because of slight flexing of the wing. The upshot of the unstable "positive feedback"effects is that it takes MUCH less elevator deflection to change angle of attack. On my plane, it was so unstable and so quick to react it could literally do a zero radius 180, and was more or less unflyable until I calmed down the balance point.
So I think you will find that you will get the controller to keep things pseduo-stable when the angle of attack is low; but when it deviates too far, the controller won't be able to catch up in time and you'll get a deepening stall or dive. Rather like what you can see. What you could do to counteract this, is an aircraft where there is a lot of mass well in front, and well behind the centre of gravity. That would increase the rotational inertia around pitch axis, and slow down those pitch oscillations. you can still move the CG wherever you want but the inertia would be higher. much higher aircraft inertia around pitch axis and much lower inertia for elevons. v interesting project good luck!
Awesome video! How did you know the location of the neutral point?
Hey, that’s very cool. I’ve done this already with my tailsitter that I built some years ago. Never got it fly stable on high speeds though. Had used all symmetrical airfoil. Maybe you just put the motors on the LE and activate tailsitter mode. I can help you tuning the vertical hover perfect without flying. Then you have a backup if the horizontal flight isn’t stable. BR David
😅😮😅
I believe moving the CG to the neutral point instead of rearwards would have a better effect for what you are trying to do
I bet it would be easy to stabilize that with up/down thrust vectoring the two motors, especially at low speed when the elevons don't have much authority.
I think this is going to be extremely hard without active AoA sensing. The flight controller can do it's best to damp oscillations, but this sort of instability is considerably worse than just oscillations.
Lets say you want it stable at 5' AoA. There is an upset, and the aircraft tilts up to 6'. The gyro can detect that, and strongarm the controls to return the aircraft to the original orientation, but the direction of flight will change in that time, so the original orientation might have an AoA of 4'. If the controls then return to the 'neutral' setting (metastable at 5'), the aircraft will still find itself with a pitch down movement. If held perfectly at the same orientation the flight path will start to oscillate up and down. We want the aircraft to be stable (after computer control) with regards to AoA. In order for the computer to do a good job of this, it needs some good way of understanding where it is wanting to stabilise towards.
Even if you manage to get the computer to infer the AoA from inertial measurements, a gust of wind will throw all that off.
One design that I have been looking into is weather-vaning canards. The idea is that the canards are placed behind their axis, and are controlled either by a motor that outputs a certain torque for given control input, rather than setting the position, or a smaller control surface on the back of the canards. They effectively get an AoA term affecting their position automatically.
Also looks like air-speed is the cause of all the crashes as the airplane gets less and less stable. Aeroplane dynamics and the required response time is speed dependent. Can you put a speed-limit in Ardupilot while you tune it up? How good is Ardupilot's airspeed? Maybe add a Pitot tube to improve accuracy?
Does Ardupilot have an internal model of the airplane it's flying? Would help a lot if that was tuned to match your flying wing. Even a 1-D or "homework level" model of the airplane can be enough to let you tune the flight controller in simulation and then can be used in-flight to hide most of a non-linear system from the PID controller. i.e. PID says I need "X" pitch torque, airplane model translates that to "Y" elevon based on air-speed. Or the instability could be simplified as a negative stiffness torsion spring for the PID loop, and the airplane model would set the stiffness and update PID gains at any measured speed.
looking forward to part 2
For a flying wing, your best cruise L/D will be achieved when your CG is exactly on your CL at your optimal angle of attack. This puts the equilibrium point exactly on the undeformed wing and allows your nice airfoil to operate as intended. For example, a DAE-11 section achieves max L/D at about 6.5 degrees AoA. At this angle, your CL is at 34.7% chord. The airfoil has a neutral point of 25.6% chord, so your optimal static margin is -9.1% and you require extensive pitch stabilization.
I was thinking that too. Isnt having the CG either forward or aft of the CP going to force your elevators to generate some lift up or down to counteract the CG/CP moment? Thats just additional drag.
Might be too simplistic an approach but, how about setting up the plane so the CG is closer to the neutral point? Then as you get things tuned in Ardupilot you can shift the battery pack incrementally rearward until you get it the way you're looking for.
For someone like you...Flying a AR Wing PRO... have to be a boring walk around your own backyard... Dam man!! what a way to bring new excitement in to this hobby.. for sure I will be following you all the way to the conclusion on this particular project... "I'm sad to say that I don't know anything a bout Ardupilot...... but I have spend well over 4 years using iNav and since I don't do complex projects like you... I wouldn't try!!! I do hope to see what comes out of this!! I love that it almost look like a brutally unstable large Nano Goblin.... 😅
I does unstable design myself and I can say it is best design. If center of gravity is in front it generally required higher speed. If center of gravity is in back it fly nice in slow cruising.
Canards instead of elevators can offer a similar unstable but beneficial setup - they provide upforce at the front of the plane rather than downforce at the rear. This requires more fine control, but means no negative lift in stable flight, at any altitude. Upshot is the CoL can remain behind the CoM.
Great demos of divergent stability at 8:45, 8:52. 🛫🤣🛬🤣
Might be educational for viewers if the the topic of "convergent vs. divergent stability, and neutral stability" were covered in a future video.
This seems like the kind of project that is perfect for simulation to help get all the little details figured out.
Can the servos use the 400Hz refresh rate? I'm under the impression that you'd need special digital servos to take advantage of 400Hz refresh rates. I think helicopter tail rotor servos can use this rate but I don't think these types of servos are very common.
Thanks for sharing your experiments with us. I've often wondered about unstable RC aircraft myself.
Cheaper, or analog servos can not. (typically will emit an audio buzz if pushed too many cycles). Digital servos are mare capable.
Extra fast tail-rotor servos will be worth the cost, for development. One less thing that might be the problem.
Just have to ask, why is this any better than a canard aircraft? Such as the Long-EZ and Velocity manned planes. Moving the CG behind the main wing Aerodynamic center is essentially doing the same thing as a canard configuration - it puts the CG between the aero centers of your lifting surfaces (in this case the wing and elevators)
Cheers to the new idea and keep having fun!!!
PS I still want to check out the code you and Daniel used for the ekranoplan height sensing, would save me a bunch of time!
It seems that drag reduction is a primary driver here. When CG is forward of the AC, a small amount of up-trim is needed to maintain level flight. That amount of up-trim results in drag. This is a stable configuration. Since gravity is always pulling the nose down, this up-trimmed configuration seems to work pretty well for landings too.
So, in an effort to eliminate that built-in drag, it seems that the designer is moving the CG aft of the AC in order to eliminate the up-trim. There is more to it, I'm sure but I think that's one of the starting points for exploring this dynamic.
Another issue with canard configs is that the flow over the main wing gets heavily modified by the canard ahead of it. If the canard is for high maneuverability at high AoA, like on some modern fighters, then that's a great feature; but if efficiency is your primary goal, it's a bug.
In order for a canard airplane to not deep stall, the canard has to stall before the main wing. That means the main wing can never fully develop it's CLmax, so the wing is oversized to begin with. It also means that high lift devices like flaps are ineffective because you're still limited by your canard. As a result, a canard with a wing sized to meet a given takeoff performance will need a larger wing, and thus be slower and draggier at cruise compared to a conventional plane that can achieve the same takeoff performance with the use of flaps and smaller wings. To get around this you either need flaps on the canard that deploy with the main wing flaps to raise the CLmax, move the canards forward during landing and takeoff to achieve a greater moment to counteract the pitching moment of the flaps and high AOA main wing, or accept that if the plane stalls it will be unable to recover.
Using flaps on the canard is problematic because that's where your elevators go and it also makes the canard lift curve steeper, degrading your static stability. Moving the canards forward is viable, and was done via adjustable sweep in the Beechcraft Starship, but it's heavy, complicated, and also steepens the canard lift moment curve (thanks to the longer lever arm) degrading static stability. Finally, accepting the risk of deep stall poses similar control system issues as presented in ThinkFlight's video, but hobbyist grade stall recovery software, stall sensors, and AoA sensors are far less developed than hobbyist control software to deal with aerodynamic instability, thanks to multirotors which are inherently unstable.
What about Prandl wings? Efficiency is not speed. If you fly slower and have no tip vortexes and proverse yaw and thus no vertical stabilizers the drag is reduced alot. Look into Albion Bowers Research with Nasa and the Flying like birds series he has on youtube. Birds are flying wings without vertical tails. Dont see many birds with spin departure problems right?
Kavin, PROVERSE YAW! Prandtl wings are one to explore.
But they can finely tune their wing twist according to angle of attack. A Prandtl wing is only good at a very specific angle of attack from what I saw
@@mortache is seem to make sense if u mean most optimal at least? Any reference? Anyways, too much proverse yaw is better than too little (in terms of stall spin). And how about pitch axis stability of prantdl? Is it 'autostabilizing' provided good trim (it likely seems). also do you know if prantdl is autostabilizing the pitch provided slightly bad trim (like say e186 should be, as airspeed raises or lowers the changed pitching moment then fixes the angle of attack)?
@@vidmasze i don't know a LOT about these wings, only that they hqve exceptionally high twist making the wingtips neutral or negative angle of attack. Something like 20-30 degrees
There's an old saying about aircraft the nose heavy aircraft can be fixed the tail heavy aircraft only flies once!
Sailplane pilots will regularly compete with the most aft CG possible because of the efficiency gains, but don't expect much. It takes downforce on the tail to counteract the CG being ahead of the center of lift. Given the location of the downforce(H-stab in a standard config) and the arm from the center of lift of the main wing to the center of lift of the H-stab, you can calculate how much down force the plane is working against, and then run that through the L/D and get an idea of how much thrust is lost for stability. Rutan canard designs take advantage of this as well by lifting with the equivalent of an elevator up front. Anyway, I'm guessing it will be hard to document any efficiency gains because they will be pretty small but I'm looking forward to the conclusion!
I would suggest to play with pitch inertia. Extend the motors backwards and move the battery forward, while keeping the same target CoG. Second I thing a very specific PID (PIFF) settings are required to achieve stability.
How do you find the aerodynamic neutral point?
It would be great to see a video where you explain what you do in terms of autotune, magfit etc... I have always wanted to try this and thought about moving the CG a very small distance and then running autotune, repeating the process over and over again.
Whenever I built a new wing I always flew in this configuration, tail heavy.
Swept back wing does not require any vertical stabilizers too. I flew my EPP wings without fins quite often and they flew great, just had to keep a little speed up for landing. But that is neither here nor there.
Windmills mill grains into flour, drive machines, or pump water. Wind TURBINES generate electricity :)
The autopilot update is 400hz - and the Saturn rocket, which lifted the crew and the module to the space, and to the moon, without any major problems, had an update of 50hz - so the update speed is probably not a main problem. I certainly doubt that servo motors can respond that quickly
In your video you mention "normal wing elevons are trimmed up", I see this a lot on wing gliders, I have found out the cg on these planes are too far forward, requiring more up trim, slide the cg back until no more up trim is needed, this gives more pitch authority, easier to tip stall, so dial in less max up/down elevon deflection.
Honestly I am very interested in the fabrication process, I wish you showed it for every airplane
I really love this channel and can't wait for the next video, and I have one small suggestion. Can you bring your voice a bit up and some of the flying footage volume down? They are quite different in levels and we can't hear you properly.
It would be interesting to see this done with wing-warping instead of moving flight controls to increase the efficiency even more.
You know, I need to do some research myself on this but I think that your standard servos that have like a 60deg/100ms slew rate *only* move that quickly when they're *already* moving. I think the rotational mass of the moving parts of the servo means you get a lot, lot less, like maybe 10deg/100ms in that first 100ms. Its fine if you're going from +90deg to -90deg in one shot, but for stabilization, when you're constantly changing the direction of the servo, you get an order of magnitude less rotation per unit time than the servo's listed specifications.
I'll have to take one apart and tap off the position potentiometer and see just how slowly servos move from a dead stop. Obviously it will be even worse when you're making it actually change directions.
If you're making a 30 degree change in aileron position and (all else being equal) it actually takes 500ms instead of the 50ms the servo specs say it will, then its no wonder the flight controller and/or IMU have so much trouble stabilizing anything.
I appreciate your experiment as your arguments seem aerodynamically valid. As you stated, wings could be reduced in size while achieving better L/D ratios by having the CG further back. My calculations led to similar results, but I then stopped my research as the instability issue seemed too risky to me.
IMO it depends on the weight and distance between CG and NP that needs to be compensated by the control surfaces. I mean, having the CG behind the NP means that changes in pitch PLUS the weight moment need to be managed through the control surfaces. On the other side, you want to keep the control inputs as small as possible, as they are augmented by the CG position. A linear expo is not bringing you close to the goal, I guess....
I'm very curious about the proceedings of this study. Looking forward to the next videos
You will need to use helicopter tail rotor servos with the fastest throw rates you can buy. Your cycle rates are getting bottlenecked if you are not running those specialty servos.
I love your boat experiments!
View at 6:19 caught me off guard. Up to this point I'd assumed this was one of your classic designs with a single pusher motor. But then an awe-haw moment, two motors to eliminate motor torque from this instability experiment? My excuse for overlooking this detail is the rearward facing camera was not centred on the longitudinal axis.
The biggest hurtle with this project to learning to tune and fly an instable plane during the the launch phase of flight, which in itself is more unstable than other phases of normal flight.
My suggestion would be to start (in)stability testing and tuning after stable level flight is achieved. A simple way to accomplish this is to add a servo that drops a weight attached under the planes forward fuselage with an optional “remove before flight” streamer. (to reduce fall speed, and to make it easier to locate later)
I noticed that this UAV doesn't have a true airspeed indicator? Without true airspeed, Ardupilot is not able to calculate/remove windspeed from a GPS derived groundspeed! Hint: while you may be flighting in light winds, the plane is not always flying parallel to the ground, so ground speed would be just as unstable as the UAV. 🤯
No real IFR test pilot would attempt to fly an unstable aircraft using only an artificial horizon and vertical speed without airspeed as a cross reference. BUT this is the task you are requesting ArduPilot to attempt.
Personally I'd go a different route to installing pilot tube for airspeed as it is likely to undergo a RUD (rapid unplanned disassembly) during repeated durability testing.
An "Angle of Attack" (AOA) sensor would provide much quicker response rate and more accurate data. Pilot tubes work best when in a near-zero AOA environment, otherwise they report inconsistent performance data. Another reason to delay adding a pilot tube. An AOA typically install on the forward side of the fuselage, so should be protected during UAV durability testing.
FYI: all fighter jets rely on AOA to ensure stay within the engineered flight performance box, particularly during high-alpha, and/or high-G manoeuvres .
While I haven't seen a good AOA sensor marketed for using on small UAV's, it should be possible construct a useful AOA sensor that would be compatible with the Matek flight controller. One option would be to use a resistive potentiometer, like that found in a servo, and feed a voltage to the FC's spare analog to digital input. An optical rotary encoder could be an alliterative, but would not be FC compatible
Another option, would be to use a diametric magnet (magnetized across it diameter) and a hall-effect sensor to track position to a fine degree. Other methods could measure pressure differential on the wing, but this would be a complicated project itself, thus would become a distraction.
BTW: with a working AOA, I suspect even Nicholas's dRehmFlight software could tame this unstable flying wing. It should also be possible to estimate true airspeed based on an aircrafts AOA and motor power level. This would need calibrating, but could be a technique that prove useful enough to include on future aerodynamic explorations. A couple thought that could be follow-ons to explore once stable efficient flight is achieved.
Move the center of gravity forward but also add kennard wings that pivot and a higher aspect ratio. The kennard wings use positive control input. Also the angle of attack can increase approaching stall speed.
When I build a plane with ardupilot many years ago we did a lot of hard ware in the loop simulation with x plane and the ardupilot. That worked well and saved us many crashes, as the thing flew ok first go.
Do you think the design of a B-2 bomber can be improved?
A canard wing up front will soften the pinching up and down, thus possibly make the autopilot time to catch up! Without having to dial in up elevons, which is a draggy solution! But I do like Herr Günter's suggestion!
Very interesting. Following closely. Keep up the good work
With an aircraft that unstable, you need an angle of attack probe with sensor for it to work. Then your stick on your controller (fore and aft) for elevator control becomes an angle of attack "setter" and provides input for the controller. Then set your control system to maintain that angle of attack that the stick is asking. A really simple analogy would be zero stick equal zero angle of attack and full aft stick equals +13 degrees assuming wing stalls at 15degrees. Full forward stick equals -13 degrees angle of attack. That would be the basis of a simple fly-by-wire flight control system.
I finished a similar project with a very short plane (forward swept), to make PID work correctly you need very high values, just as much as your servos can handle. good luck!
Love this thinking to push the boundaries further.
Loving your videos man!
I would add a dihedral v-shape to the wing and add little winglets to stabilise the tips, but maybe that wouldn't help at all.
Nothing is "unstable" if designed correctly, it just rides the edge of perfect efficiency! 😱🤪🤣😁👍👍🇺🇸
How do you know where the neutral point is?
Computer predictions
I was thinking, perhaps move the motor in front of the wing and a bit outwards so the airflow pass through the control surface, perhaps it would help the FC stabilize it more effectively
Yes, agility is the main goal of instability. The fighter jets using it are not concerned about efficiency. Yes, flying around with elevators down (like flaps) would give more lift, but also more drag (like... flaps).
This would be a hard road even with a conventional plane layout, but a plank style wing is really asking for problems. You'd at least want the motors in front of the control surfaces for any hope of controlling it at low speed (or swivel the motors maybe).
I think the less work the elevator has to do the better, so having the CG very close or right on top of the aerodynamic center, and a plane design (airfoil etc) that naturally produces low pitch moment and high pitch inertia is best. The perfect example is a sailplane with long tail and small elevator. The worst would be where you have to constantly stick a control surface into the airflow.
There is a full-size plane called Celera 500 that claims to be the holy grail of efficiency, would be interesting to see model-scale testing of those concepts (stretched egg looking fuselage and super skinny wings).
If you like the Celera you will enjoy my manned design I'm testing in subscale.
Thanks as always for the great content!!
Great video! It was very informative and well presented. It would be great to see more videos on XFLR5 and aircraft design in the future. It will be a great help for those who are interested in aircraft design and simulation. Keep up the good work!
If you add an elevator about 2 feet behind the plane, it will allow you to play with the center of gravity without losing stability. I know you want a flying wing, but the elevator doesn’t have to be bigger than 10% of the plane width if you mount it back far enough
Great video... Suggestion,. adding a lot of extra wieght would counter balance the extreme lift? I.e. up verses down... It's ok to have a highly efficient aircraft that only works (is stable) with a huge amount of extra weight... It would serve as a good cargo plane or super large commercial airplane, that could carry much more passengers, 2x the amount, meaning less flights overall and less carbon emissions...
Wow! I just say the idea of changing the CG in flight.. what a nice concept... also what about using a bungee cord!!!! I use it all the time for my large birds.
"Currently unstable, and not handling it well."
Sums me up pretty well too lol.
Get the fastest servos you can buy and add the a large servo arm that would convert small movements at the servo to large movements at the ailevons. Thats how you will get the fastest response possible.
Excited for your conclusion
The Dassault Mirage 2000 also had elevons and a CG behind the center of lift, so there's proof that it is in possible.
Well, back in the late 70s, I built a series or flying wings to figure the minimum reflex required to achieve stability. I found that a swept back wing required far less reflex than a plank wing. I didn't take it to a final conclusion though. I just cut my tests short at zero reflex at the root, tapering to 25% chord reflex at the tip. The tip chord was 60% of the root chord. So that left me with 9% of the total wing area reflexed compared to the standard 25% for plank wings of the day. That was as far as I cared to push the envelope on this before building a larger RC wing, I'm guessing I could have gone further before reaching pitch instability though.
The way I developed the reflex was pretty basic. I used a fairly high cambered thin profile. I tool a straight line from the nose point through the reflex point on the mean camber line, projecting that to the trailing edge. Then just reflexed the mean camber line in a smooth curve to meet the TE point which was naturally above the root reference line. So it would be fair to say that this graphic rule of thumb solution was not just reflex but incorporated washout to the tip as well.
I didn't know any analytical aerodynamics at the time, but I suppose it had a kind of vague similarity to the Prandtl wing in that to unloads the tip.
Anyway, I just thought it might be of some interest, see it here :
drive.google.com/file/d/1s-Z7ob6zHEiKgaB_0w8HHgwYFyc8hrrZ/view?usp=share_link
I think that if I built this again, I'd just do elevons tapering to put most of the control surface at the tips.
As it was, I think the inboard elevators weren't all that effective except by altering mean cMo to effect stability.
My diagnosis would be that the controller needs stronger damping of angular acceleration and/or velocity on the pitch axis. I've never worked on a controller for real-life aircraft so I might be wrong.
This was a great video. Would like one of how to choose NACA profiles and implement them.
Where is part 2?
Coming
I just stumbled on your channel, and have never seen this particular method of building. When I last built EPP models for slope soaring, Goop skim coats and fiber tape was the goto method for combat and rough duty models. My old timer/newb question...Where to get propyl acetate?
I'm guessing that because the plane is very short from front to back, it has a low pitching inertia. This results in pitch-changes occurring more rapidly, as a result of which the stabilization system needs to react to pitch-changes more rapidly.
Pitching inertia could be increased in either of two ways: by increasing the size of the plane; or by giving the plane swept wings.
Science ! ! Awesome technical stuff. Great building skills.
This looks like UT. Where is it?
More Sweep angle might soften the pitch jerk... perhaps.? Not sure but will ve looking forward to the next iteration.
Love it. Bravo. Keen for the results
When I was flying, I put the center of gravity ahead by 1/4 inches. My brother-in-law put it 1/2 inches in the back! His fingers were a lot faster than mine, he could make an airplane fly backwards!
No one can fly an unstable aircraft without using normal G force for gain,or using an AOA sensor. So my friends and I are using angle sensor as AOA sensor to fly model fighters ,use AOA gain to canard or elevator.If you use gyro only, it will not work.
I’m not entirely certain, but I think the pitch control may be reversed for this aircraft.
We might be tempted to think the elevons are producing lift (up or down) on the elevon surfaces, but I don’t that’s the right mental model for this configuration.
I think deflecting the elevons one way or the other changes the amount of lift being produced on the main wing. And because the aerodynamic center of the main wing is ahead of the CG rather than behind it, the pitch controls are reversed from normal. So deflecting the elevons downward produces an upward pitch moment, and deflecting them upward produces a downward pitch moment.
ha ha - sometimes controller introduced frustration kicks in, part of the control problem comes from using linear controllers in non linear ranges - to nonlinearise the control PIDs- one needs to be able to something like gain scheduling (a lookup table with velocity vs PID values - (in pitch)) based on airspeed.
Flying under manual control controlling the +/- Flaperons with a side slider you should be able to get to a stable takeoff/ landing setting and a cruise / high speed setting. (I have used similar in manual RC planes with really low wing loading to get them behind the power curve for controllable short landings - the sort of aircraft which otherwise lock into ground effect and never want to touch down.).
Another interesting video
Thank you for all the great plane content
Maybe its better to use a Canard-Configuration This is more efficent, because there is no Need for negative Lift on the Stabiliser. You can trim those Planes stable for initial calibrating your Flightcontroller Settings. Than you can optimise the CoG for best Efficency without totally loss of Stability.
could it also be that plane (sort of) lacks airspeed to be controllable well enough? would a catapult or wheels help? did you already some do some Autotune or manual adjustment of PIDs already?
another interesting topic will be how to do stall recovery with "bad" CG... movable CG or have some quad-copter or small heli-copter style motor(s) for recovery?
You need to do a similar investigation into Prandtl NASA style wings that are finless and have the "tip" vortices moved inboard. Allegedly they are the most efficient wing for a given weight (not given span).
Holy grail is LAMINAR FLOW.
Lots of ways to accomplish this. Like to see a video on a laminar flow wing.
how controlled is your motor mounting and considerations for center(s) of thrust? edit: i.e. addition/reduction of thrust can cause pitching, yawing, and rolling. so your wing isn't only designed for a specific weight, speed, and AoA combo, it is also designed for a specific thrust (and the accompanying pitch, yaw, and roll induced by said specific thrust)
Would this perhaps work better with thrust vectoring? Just a guess, but wouldn't the effectiveness of the elevon's change dramatically and unpredictably at higher angles of attack as the air flow begins to separate. Making tuning the autopilots output next to impossible without a wind tunnel. At least with thrust vectoring the pitching moment they would apply at a given power% and angle would be more predictable, particularly if they were in front of the wing. Though I suspect their effectiveness would still change depending on the airspeed because the blades are fixed pitch.
I was gathering from the video that the point of the instability was so that the elevons could have negative reflex and increase lift at slow speeds. That effect would be negated if thrust vectoring was used instead.
@@FlyXenonRC That effect could be amplified further if the elevons weren't constantly oscillating to provide roll authority. and instead acted more like self-adjusting flaps that only responded to pitch. In any case using thrust vectoring or differential thrust could provide more pitch stability.
Back in the day....I would occasionally fly a model with the c.g. way out of whack, further rearward than your line indicates from center on this craft. It was always a handful to gain altitude, circle, and back down to terra firma. But not nearly as squirrely as that flight controller trying to wrestle your wing into the air. have you tried to manually fly the aircraft?
Balancing on the center of lift is most efficient, but it's twitchy it pitch. Reflex airfoils are more stable but less efficient
When you say "neutral point" are you referring to the center of pressure?
Try launching it upside down. I built an unstable control line flying wing that alway dove for the ground and until it was launched upside down it was unflyable. It floated around upside down until it built up speed, became unstable again and flipped over at high speed becoming stable again. When the speed bled off it was another flip and landing upside down.
I should have shown it off more. It was wild and taught me something. I'm still not sure what.
Major isssue is airflow over the wing and control surfaces. with the wing infront of the props the airflow over the wing is determined by the speed of the wing through the air and at low speeds and high angles of attack there is little authority. put a control surface behind the wing and you may solve the issue,
Explain me one thing please. If the autopilot calculates 400 times a second but the servo get new positions only 50 times a second, do the 400 hz make sense?
hm control engineering was a long time ago but out of the top of my head. Even on 50 Hz, letting the control loop run at 400Hz, it could improve still help for noise from input sensors. It could still help as reading of inputs and acting on them is not necessarily close together when on the same frequency (no idea how the code is setup). And if you don't adjust gain values then it could still have a fortunate effect too but that could be easily fixed by putting in the correct gains.
But actually the servos are doing 400hz now too 9:30 . ;)
not that I'm qualified to answer but yes and no. In some ways no because as you're thinking, the servos can only move at 50hz..but a 400hz sample rate allows much finer analysis of what's ahppening to teh craft and as this is r&d I'd assume that's valuable in itself. In analagous areas having high quality data that you then downsample to get from in this case 400hz sampling to 50hz action can still be useful as at least you have a better picture of what's happening to base your 50 actiosn per second on if that makes sense. ANyway, I wondered the same thing...hope TF replies to your question ;-)
You'd need servos capable of updating at 400Hz. I think these sorts of servos are available but I don't think they're very common.
Great content as always. How would you feel if a manned plane was tail heavy though? Would you get in it?
Would love some more content on your design process at some point, particularly airfoil selection. Why one or the other, how do you decide etc.