With cooling circuits in a passenger electric vehicle, the battery pack energy density is generally 140Wh/kg. The general-purpose IoT forces a landing/termination when the ion reserve is at a 10% level because the 10% level can't sustain lifting the craft, as shown in the video below. So, as shown in the video here, no protection or cooling circuits are needed where the lift lowers to a landing with a 10% battery level. The prototype, weighing 250 grams with a 35-minute endurance, has full HD video on the right. The production build after endurance optimization is pictured here. We use congruent mass distribution and mechanics as the extreme robotics build to have the same robotics characteristics as the extreme build.
Endurance Optimization
T-Motor 4004's data is saved in https://drive.google.com/file/d/1u8NkHY28v4AklgQdXVgva0rvAXevIcff/ ,
T-Motor 2806's data is saved in https://drive.google.com/file/d/1A3C4TgKU-0X7GzRMv7PgJEgXHwu2BZ2c/ ,
Sunnysky 4004 400kv's data is saved in https://drive.google.com/file/d/1pYOJN9Hi-93svuun-8ODcXMY-UGgvON8/ ,
Sunnysky 4004 300kv's data is saved in https://drive.google.com/file/d/1xwbcVjCucTQAQFk8mq_o6eJbOfekKUEO/ ,
Sunnysky 2806 400kv's data is saved in https://drive.google.com/file/d/1MKCFu733HeZotZHhqp3N-I6adnCU65xG/ ,
Sunnysky 2806 650kv's data is saved in https://drive.google.com/file/d/1xpaYDaShMJg4fl07ka_vFqt4RPkkexjF/ ,
HGLRC 1303.5's data is saved in https://drive.google.com/file/d/1F2D4R6p1V8eqaQ5DHQGh-GXK_b36V62K/ ,
Our prototype rotor's efficiency is equivalent to the 13.5-inch drone propeller's efficiency in terms of the same torque, RPM, and compound efficiency. So, the question is how to shift the prototype 4004-motor-13.5-inch-prop combination toward a more efficient combination in the efficiency trend curve.
According to https://www.researchgate.net/publication/339342177_Gemini_A_Compact_yet_Efficient_Bi-copter_UAV_for_Indoor_Applications , disk loading is reciprocally correlated to lifting power consumption efficiency. So, the Phython code of calculating the efficiency of the most efficient motor-propeller combinations for 249 gram lifting force to fit all manufacturer and experimented data is,
lifting_weight_per_watt = 14.7 * (propeller_span_x/13.5);
This formula's propeller_span_x is for blades with twisted and curved edges. This formula can be applied to heavier motors. For example, the 4004 kv300 motor has an efficiency of conservative estimation of 16.3g/W for 250g lifting force by the curve. And, the matching blade span width is 16.3/14.7*13.5 = 15 inches, which is close to the highest efficiency curve for the 4004 300kv motor.
The generalized formula can be applied to lighter, smaller motors, such as T-Motor AS2306 1500kV at 8.3 g/W efficiency and an 8-to-9-inch propeller. 14.7*8.5inch/13.5 = 9.25 g/W , which is close to the manufacturer's published efficiency of 8.3 g/W. For the HGLRC 1303.5 motor, the formula gives 14.7*3.67/13.5 = 3.9 g/W, also close to the manufacturer's published data.
The propeller_span_x is the diameter of 2-blade propellers with twisted and curved edges. For 3-blade propellers, because of utilizing the air column 3/2 times as often as a 2-blade propeller, the equivalent air column cross-section area should be considered 150%. So, the equivalent 2-blade diameter should be considered the square root of 1.5, which is 1.224 times the 3-blade propeller's diameter. And a 3-inch 3-blade propeller will be considered a 3.67-inch 2-blade propeller.
The phython code of the calculation is
craft_construct_mass = 111.0; #change it for a particulr craft constructions endure time curve
propeller_span_x = 13.5; #change it to slide the rotor width with in a endure time curve
motor_propeller_mass = 62.5 * (propeller_span_x/13.5)**1.5;
battery_cells_weight = 249.0 - craft_construct_mass - motor_propeller_mass;
battery_energy = 0.171 * battery_cells_weight;
lifting_weight_per_watt = 14.7 * propeller_span_x/13.5;
power_tran_power_consume_rate = 249.0 / lifting_weight_per_watt;
total_craft_power_consume_rate = power_tran_power_consume_rate + 5.2;
endure_time_minutes = battery_energy / total_craft_power_consume_rate * 60.0;
print (endure_time_minutes);
The motor and propeller combined mass can be calculated by the Python3 code
motor_propeller_mass = 62.5 * (propeller_span_x/13.5)**1.5;
. This formula excludes the motor shaft and propeller hub. The coefficient 62.5 is calibrated to this experiment's measure. This formula is suitable for larger motors and smaller motors. For example, for the 4004 300kv motor of 50.5 grams, drone propellers of 15 inches are about 21 grams, which is readily available in the DJI Inspire 3 aftermarket. 62.5*(15/13.5)**1.5 = 73 grams = 50.5g + 22.5g. For the DJI Mavic Pro motor at 23.5 grams with a 5.3-gram, 8.3-inch propeller blade pair. 62.5*(8.3/13.5)**1.5=30g ~= 23.5g + 5g. For HGLRC 1303.5 motor, 62.5*(3.67/13.5)**1.5=8.86g ~= 6.5g + 1.9g (propeller). For 2306 motor with an 8.5-inch prop as optimal, 62.5*(8.5inch/13.5)**1.5=31.2g, close to actual available parts weights of 25.8g +5.3g=31.1g.
Finally, according to the endurance efficiency trend curve, endurance can be improved by changing the helicopter rotor span to be equivalent to an 11.7-inch drone propeller. The T-Motor 2806 kv400 motor fits the motor mass data points of the endurance curve, and a 12-inch propeller is close enough to the 11.7-inch calculation.
Unfortunately, attempting to use a 12-inch drone propeller with a twisted Sokolov airfoil and a 2806 motor failed. The propeller has 249 grams of thrust, similar to the original endurance build with a straight, symmetrical airfoil, at 100RPM above the designated 3300RPM. However, the gyroscopic effect required for single-rotor mechanics is too weak to have stable craft control, so, as shown in the pictures below, large ballasts at 1mm thickness were added.
The twisted Sokolov airfoil couldn't work even if tweaking rotor diameter to slightly less than optimal to gain gyroscopic effect and more optimized industry battery construction in the next video because the relationship between drag and attack angle is complex.
In the video below, the diameter is lowered to 9.5 inches, and motor changed to 2306. The RPM is designated to be 5000, and endurance is calculated to be better than 25 minutes. The lack of gyroscopic effect requires raising RPM to 7100 to barely maintain control; near the end of the video, the Li-Ion battery voltage sag destabilizes the gyroscopic effect even though the battery is not fully discharged. And actual endurance is 21 minutes. Ballast modification may not improve it beyond 30 minutes.The slanted Clark-Y airfoil, pictured above on the right, also suffers from the complex correlation between drag and lift of the twisted Sokolov airfoil. But what makes Clark-Y worse is that it is heavier than the straight, symmetrical prototype build. The advantage of the peak of the optimization curve by reduced blade and motor weights is largely canceled. In the picture above on the right, you can see a flat surface of the blade root area, giving a fixed, designated angle of attack of the blade when the flat surface is level. At 3500 RPM, the 77 throttle (collective pitch) point is lower than needed to give the designated angle of attack. About 90 throttle points to make the flat surface level. And yet, the RPM is still low on giving adequate gyroscopic effect. And, when RPM is raised, and the angle of attack lowered further, motor overheating occurs because the Clark-Y airfoil has a larger drag than the operating drag when the angle of attack is lower than the designated angle.
To refocus on straight symmetrical airfoil, according to NASA education page, the lift equation is as pictured on the right. A is the airfoil area, V is the airspeed, rho is air density, Cl is lift coefficient. Division by 2 should be discarded because our propeller has 2 blades that doubles the lift. Air density is about 1.2 kg/m^3 , V changes linearly with x position of the propeller origin from the main shaft axis with constant RPM 2640, angular speed 2640*2*pi/60 rad/s. Area da is 0.018 blade width multiplied by dx. So, the integration is for the quadratic of x from the axis origin to the blade tip at 0.236 m. The simple polynomial integration solution is the 3rd power of x multiplied by a third of the constant. To generate 249 grams of lift, the lift force equals 2.44 newtons. And according to the dashed curve, the theoretical lift is much higher than 2.44 newtons. The section of the blade from the axis to 127 mm out should be considered zero-lifting because the hub and bolts are not part of the airfoil out to 50 mm. And between 50 mm to 127 mm, Reynold's number is too low to produce lift, which can be felt as zero downwash when tethered hover in hand. Between 200 mm to 236 mm, the blade is blown downward with the wingtip vortex of the previous sweeping of the blades. The green curve is the corrected curve. The wing section between 200 mm to 236 mm from the origin axis out is considered non-lifting. At the integration point of 200 mm, 0.2 m, the total lift is the target of 2.44 newtons.
The calculation points to the solution of improving the fuel efficiency by a winglet to block vortex turbulence. However, as tried in crafting the ballasts for 12 inch twisted Sokolov airfoil, any substantial winglet can only be safely made as original factory part, not as a after-market modification.
The overall, compound power-tran efficiency has 2 peaks with some systems, as shown in the video, and we use the first peaks of motor-propeller combinations for the highest efficiency (more than double the fuel efficiency).
Ultimately, the ideal solution for the endurance of 36 minutes is to use a 2806 motor running at 3300 RPM with a straight, not twisted, wing to simulate a 12-inch twisted Sokolov setup's RPM and torque (the blue curves, not cyan, in the first 2 graphs of this section). And 193 mm blades will have the same angular momentum at 13.1 grams as the prototype build. The prototype build has an angular momentum of 1/12 x (14.6+5)/1000 x 0.473^2 x 2640/60*6.28 = 0.1, and the ideal solution will have 1/12 x (13.1 + 5)/1000 x 0.438^2 x 3300/60*6.28 = 0.1. The ideal compound efficiency will be worsened at most in the second order, 14.7 g/W x (439mm/473mm)^2 = 12.7 g/W. The ideal symmetrical airfoil chord length will be 18mm x (2640/3300)^2 x (473/439)^3 = 14.4 mm. However, this ultimate airfoil blade is not available in the product market. This ultimate endurance is 0.171*(249-111-(42.2+13.1)) / (249/(12.7*(12/14))) * 60 = 37 minutes.
The production endurance build uses the available 193mm blades, which have the same 18 mm chord length as the prototype 210mm blades. When reducing the production of straight, symmetrical blade longitudinally to test endurances of different motor-propeller combinations, as shown in the video recorded on the right, the compound efficiency deteriorates with inverse blade diameter proportion to the second order and worse, rather than a simple inverse proportion of the twisted Sokolov setup. The build with 20g less batttery mass is expected to have 35 * (55 / 75) * 0.8(high C rating) = 21 minutes endurance. The trimmed blades impair efficiency in the second and half order (degree) as this python3 code, 35 * (55 / 75) * ((155*2+53)/(211*2+52))**2.5
=13.17 minutes of endurance (the full HD recording is absent but there is a lot of high energy consumption climbing/diving/flipping).
Trimming the straight blades without shortening the airfoil chord length shifts the wing toward an unfavorable aspect ratio and impairs efficiency more than the second order.
The result is that the production endurance build has the same endurance as the prototype. As shown in the video on the right at RPM 3200, angular momentum 1/12 x (13.4 + 5)/1000 x 0.438^2 x 3240/60*6.28 = 0.1. The endurance was considered 35 minutes due to a false launch of 2 minutes in the beginning while full HD recording was rolling, and the build collective pitch was not set to zero at zero points, then the hovering was adjusted to 18 points pitch, and changes of RPM in the middle from 3240 to 3140 and during the adjusting, and the extra climbing in the first 7 minutes.
Change From Prototype
The prototype GensAce900 airsoft rifle battery is of thin-layer construction at about 25C discharge rating, with capacity measured between 949 and 998mAh at the first charge cycle, depending on how balanced the cells are when depleted, as shown in the below pictures.
The below-left picture was with Tattu1000 cells, measurement taken after about 1 dozen charge cycles after this pack was used for the video "World record endurance". The pack gave 870mAh electricity during the mentioned world record mission. The protrusion of lower-right cell tab is characteristic of the Tattu1000 packs. At the first charge cycle, the Tattu1000 cells have identical charge intake, between 949 and 998mAh, as shown below on the right.
The pictures together mean that the pack forced a landing with about 90mAh reservation, or 10 percent.
So, overall, the GensAce900mah/Tattu1000mah airsoft rifle battery has a capacity of 970mah. In a 4-S configuration, the endurance time of 35 minutes means power for cruising with HD video is 0.87Ah / (35min/60min) * (4 x 3.7V) = 22.2W. Each cell weighs 18.9 grams. Energy density 0.97Ah x 3.7V / 0.0189kg = 190Wh/kg. Total battery energy is 0.97Ah x 4 x 3.7V = 14.4Wh. Useable energy is 14.4Wh*90% = 12.9Wh.
The endurance builds, prototype and production builds, dive in these videos,
, which dives the same way as the other builds for adrenaline diving. It just needs extra labor to build. Compared to the general purpose build, the extra 15 grams of battery weight are shaved from (sorted by increasing labor) GPS (5.5 grams absent), the motor(5.9 grams saving), and ESC power system (4.1 grams saving). The HD camera system weighs the same as the general purpose's video system. The weight gain of the long distance video transmission and building custom battery pack is canceled out by lighter shaft and lighter rotor blades. And the custom battery pack itself needs most of the extra labor.
The prototype fuselage is weighed in the pictures here,
And the production endurance build's takeoff weight is calculated as the following python3 code,
246.3 - 1.7 + 1.2 - 3 - 5.6 - 55.4 + (65.4*4/3+3) - 12.4 + 7.4 - 2.2 + 1 - 8 + 4.7 - 49 + 41.4 - 17 + 13.6 - 0.3 - 0.3 - 0.9 >>> 250 g.
Carbon Fiber Shaft
It saves 4.8-(1.8-0.1)=3.1 grams.
When using this carbon fiber shaft, the Kevlar insert is 2 and half strands. Using 3 strands resulted in too tight a fit, resulting in a bent collar after pulling out the shaft, as pictured below on the right with the tilted collar. Also, 3 strands require very, very strong hands to insert the shaft. Different batches of the carbon fiber shaft need different balancing orientations.
When measured from the rotor hub bolt hole to the bottom, the carbon fiber main shaft is 3.5mm shorter than the original build's titanium shaft. In the following picture, the rotor hub bolt holes don't appear aligned but are correctly aligned with the upper (left) bore hole on the 10 cm mark. The tape measure is concaved, so the heavier titanium shaft sinks to the bottom, and the carbon fiber shaft is closer to the camera lens. The camera lens is directly above the 18cm mark, so the hub bolt-hole image is distorted in the picture.
When using the production Blade Fusion 180s fuselage: The shortened main shaft is not a problem because the v950 shaft is too long and needs to be trimmed at the bottom and still has a 3mm protrusion. The carbon fiber shaft has the right length without excess protrusion.About every 10 sales of the carbon fiber shaft has off-center bolt boring manufacturing error, as in the picture comparison here.
T-Motor Shaft Work Surface Finishing
When using T- motor, regardless of the shaft used, the bell rim only needs scraping once, as pictured on the right, because there is no slope of the bore opening like what the Sunnysky motors have. Battery Pack Build
Use cuticle scissors/cutter and soldering iron at 350C to remove old wiring first, then the GensAce900 pack needs to be trimmed. The rotary tool cutting disk is commonly made of insulating ceramics, and we use it to cut out the first cell with the exposed perforation clamped tab. Cut the glass fiber plate itself so that the cell tabs remain soldered to the terminal. Do not cleave the cell tabs and then attempt to solder the cell tabs because the cobalt alloy does not weld with other metals, even with specialty soldering fluxes such as aluminum flux.
The Tattu1000 2S cells have the same dimensions with a solid terminal glass fiber plate weighing 0.3 grams heavier per cell. The following illustrations use the Tattu1000. In the pictures, you can see the exposed lower right cell tab that is the middle pole of each pack. This is different from the GensAce900 pack shown in the 250 gram weight breakdown page, where the cell tab is behind in the canal of the carved glass fiber plate.
To start the build, the 2 packs are staggered by 1 inch with a temporary taping. The 22AWG charging wires are bent to the opposite poles and tied to the protruding top pack for the right side of the craft.
After soldering the 5 wires, the additional middle wire is tapped then soldered to the top positive pole of the lower pack. The yellow terminal is what cascades the 2 packs. After the temporary taping are removed, the permanent taping adds 0.47 grams of 35cm clear tape. The XT30 plug is 0.8 grams, 9 soldering joins weight 0.18 grams, and charging wire set weighs 2.4 grams. The Tattu1000 yields a 76.6+2.4+0.8+0.26+0.18=76.6+3.64=80.2 gram pack; The GensAce900 yields a 75.4+3.64=79 gram pack.
ESC Build Weight Reduction
From left to right, DYS 16A, LittleBee 20A, Blade Thrust 20A.
The ESC build is modified from the original IoT build with all stranded 20AWG wires replaced with lighter solid 24AWG enameled wires to save 2 and a half grams. The solid wires have slightly larger equivalent cross section when the gauge numbers are the same, so the solid wires can have a slightly larger gauge number.
The 16 Amp BlHeli products earlier than year 2016 are not to be used because either the dampening can not be disabled or PI loop not available.
20 Amp BlHeli ESCs in the market don't have such problem. Further more, all 20 Amp and 16 Amp BlHeli (not BlHeli_s nor BlHeli_32) have the exact same weight as shown in the above weighing. Blade Thrust brand has navy blue PCB, DYS uses green, and LittleBee uses black. BlHeli_s or BlHeli_32 are not suitable because neither has the PI loop to provide constant stable RPMs.
For the tail motor, in theory, the 3S 6A ESC can save another 0.9 grams as shown in the picture on the right, but this alternative can only function properly when maneuver is small and smooth. The startup power with 3S is too weak even though the output of the 6A ESC is adequate, as the video below on the left. And this problem persists even after startup boosting is changed to maximum in the ESC's configuration.
HD-FPV Camera
Four different powering options have been tested, similar to the FPV camera setup section of the high-performance build post https://nocomputerbutphone.blogspot.com/2018/08/converged-drone-at-edge-of-space.html.
- The MP1584 chip-based buck converters, such as the Bluesky 5V power supply, are the worst for analog video applications.
- The second worst noise interfering with FPV feed is with 12-16V direct battery power supply to the Caddx Turtle cameras (v1, v2, and Baby) in the following video; if you watch the original feed video file https://drive.google.com/file/d/1nL07CPVMDrWu_w7SFtraSYZqxUejrDJr (download it then watch it on a video player instead of watching it online with an online player because online video servers process out the noises to save their internet bandwidth).
- The FPV feed is best clear of interference lines when VTX's 5V aux power filters out low-frequency ripples. However, the market trends move away from VTX aux power and toward pass-thru powering.
- The solution is to share power with servos of the base build, the same as the high-performance build, which has a Wolfwhoop MP2315-chip-based buck converter power supply.
With an extra alternative 10uH inductor, the FPV feed has the video ripple noise level between direct battery power and VTX's aux power in this solution. The ripple noise can not be completely eliminated with a 10uH inductor LC filter because the ripple frequency is similar to or lower than the 10uH inductor working frequency. The ripple noise can not be eliminated with an electrolyte audio A/C capacitor, either, because the ripple frequency, around 1MHz, is much higher than the electrolyte capacitor's audio working frequency.
As pictured on the right, a scrap plastic strut fixture is needed to hold the camera board vertically. Use kevlar strings or enameled wires to tie the board to the strut and craft body. And then, CA glues the fixture arm.Prepare the SD card by sticking it onto a tape with a mounting square, then tear off half of the unused tape. After installing the SD card, fasten the mounting square onto the steel plate of the camera board.
Do not drill holes on the sides of the main frame wall to fasten the camera board because that will weaken the structure and damage the camera wire connectors during a crash. The experiment shows a complete cleavage of the upper main frame when the left side wall is weakened with fastening holes.
The alternative Eachine VTX03 has AUX power at 360mA, as shown in the specification on the right. It can give the best FPV feed with Caddx Baby Turtle. However, as seen in the following specification comparison, the alternative Runcam Split has a much larger power consumption than the VTX03 can support.
Runcam Split series tapping AUX power of VTX03 results in video freeze with high G maneuver power surges/dips. One incidence of such video freeze is below on the left during punching, where the drone is lost from a considerable altitude.
A second video freeze incident with Split is above the right. If you watch the video to the end, the frozen last frame stays indefinitely after the drone touches the ground.
Fasteners
Velcro strips use are split in the middle to cut the weight to half to bring down the craft under 250 grams.
Notice that the 250g weighing uses alternative grips(1.5g extra), alternative servo arrangements (1.2g+ 0.4g esc tabs extra but 0.6g solder reduction), CC3D+RX wire(0.6 reduction), alternative tail ESC(0.9g reduction), alternative tail boom(0.4g reduction), alternative RX kit wires (0.3 grams extra), and alternative tail motor(0.3 grams reduction). Net weight gain 0.6 grams.
Maintaining optimal airfoil angle
The well-known helicopter dynamic says that cruising enhances lift. So, RPM needs to be lowered for the world record endurance to keep the cruising airfoil at 6.2 degrees at 100/200 throttle level. This is done by the transmitter RPM tuning knob to lower throttle 5 points from 85 to 80. The knob is multiplexed into throttle channel in the transmitter.
Alternative Swashplate Weight Reduction
The BSR swashplate weighs 1.5 grams less than Blade230s's. But the ball link EFLH1151 for the BSR swashplate is too short when pairing with Oxy2's DFC arm. The solution is the Align250 aluminum DFC arms. With Align250's DFC pins, the DFC arm and pin weigh 2.2g, 2.2g-(0.3g Oxy2 DFC arm + 1.2g Oxy2 DFC pins)=0.7g heavier. So, this alternative setup saves 1.5g-0.7g=0.8g. The Align300's DFC arm is identical to the Align250's DFC arm, same length, the same bearings, and the same ball socket. The ball socket diameter of the Align DFC arm kit is 3.5mm, too tight for BSR's swash's 3.6mm balls.
When the alternative Align DFC arm is used, the arm is about 2mm longer, and the swash is overall lowered by 3mm, including the mandatory lowering of 1mm with the shorter carbon fiber main shaft. To avoid the swashplate "twitch" and colliding with the shaft collar during the F411 flight computer startup, we need to use the alternative flight computer that doesn't have the twitch problem. So, overall the alternative Align DFC arm needs the EFLH 1151 ball socket and M1.4 threaded rod, such as the Lynx Blade 180 CFX DFC arm kit rod. It also needs the M2 0.2mm flange ring, such as the ones in the Align 250 grip kit, a 1mm nylon spacer, and 2 M2 0.3mm metal washers. The overall spacing is 1.8mm for each of the 2 DFC arms.
The combined 3.6mm spacing is the width difference between Align250's rotor hub and Blade180's. When using the 1.4mm rod of the Lynx kit, 2-5mm of the rods need to be trimmed out. The flange ring is needed so the spacer only presses onto the inner tube of the DFC arm's bearing, not rub on the DFC arm itself. The metal washers are needed to sandwich the nylon washer-spacer to prevent deforming the nylon. The plastic Align DFC arms are not suitable here because its 1.4mm rod is one-way installed into the DFC arm. As shown in the picture on the right, I tried to turn and pull the 1.4mm rod out with pliers to no avail. It left a dent on the hexagonal part of the 1.4mm rod.
In summary, the dimensions of all component options are in the following table. And the reference component diagram beneath the table.
Tail Weight Reduction Alternatives
The original Lynx Blade 230 tail boom is 255 mm long. The replacement 8x7x500mm carbon fiber tube from ebay xzw791 can produce 2 250mm booms. The weight of the replacement is expected to be 4.8g x 250mm / 255mm = 4.7 grams. But the actual weighing has 17.6g / 4 = 4.4 grams. So, the replacement has 0.4 grams weight reduction.
Motor Stator Rotation Stopper M3 Setscrew Substitute
 |  |
Main Rotor Alternative Weight Reduction
The alternative Align250 carbon-fiber blades are 0.2 grams lighter than Oxy2 nylon blades, and the blade root thickness fits the aluminum Align250 grips. The Align250 aluminum grips are incompatible with Oxy2 blades due to the narrower jaw width. In the picture below on the left, the upper blade has a pronounced sagging due to a previous experimental test with Align aluminum grip. In that experiment, the blade root was sanded thinner, which softened the material surface. A rapid pitch-up flip at 1:08 in the video below on the right caused a tail strike. Also, the test showed that generic translucent 18-lb zip ties in the tail are not strong enough and break with the tail strike. The part listing's zip tie is stronger than the generic product's nominal tensile strength of 18 lb.
The sagging and tail strike signifies that a blade surface should not be reworked outside the factory to force compatibility.
To prepare the motor for gear-less direct drive, first remove the setscrew(s) of the shaft and clamp the shaft down with a wrench socket spacing beneath the motor bell until the shaft is flush with the motor's top plane. Use a 1/2-inch wrench socket with the 4004 motors. Then, CA-glue a #6 flat-head 1/2 inch machine screw as a pusher on the clamp. Clamp the shaft down the second time with the push rod assembly, as shown in the picture, to remove the stock shaft of the motor. The main shaft is not pressure-fit to the motor but slightly thinner than the original motor shaft. The video here shows that the short bell shaft hole amplifies the thinned diameter, resulting in wobbling.
