Vortex generators

Subaru 2.5 liter — totally stock

Installation details….. Subaru engine into aircraft.



    False rumors:

There is a huge amount of misinformation about auto conversions. 

"Auto conversions are expensive". I have $6200 into mine. My install could be duplicated for less than $5000. 

"Auto conversions greatly increase aircraft build time". I believe this is often true, but this 2.5 liter engine required absolutely no airframe mods. Even the dip stick and oil fill locations are identical to Lycoming. I strongly believe in using the stock wiring, which greatly reduces install time. It took me six years to build my aircraft, which is slightly less than average. 

"Auto engines weren't designed for high rpm, high horsepower use". It's true that continuous high power use is unusual for the engine. However, there are hundreds of these engines used in such applications. I've never been able to find a single incidence of failure related to workload. What sounds logical and significant is not supported by empirical data. No workload related failures.


This page describes the changes needed to install any 1995 thru 1998 Subaru auto engine into an aircraft. Specifically a 2.5 liter engine. The goal is to use stock components and minimize the changes. Thus taking advantage of the reliability of the new auto systems. Exceptional auto reliability in recent history is partially due to the engine control module (ECM) programming. It monitors the accuracy of many of the sensors. For example, the ECM recognizes if a critical component like the crank sensor fails. It then uses the cam angle sensor as a replacement. This is not an easy programming task. If you research it, you will find that all non-stock engine controllers don't have this important safety feature. 

When I started installing my engine I was unable to find accurate info on how to modify the engine for aircraft use.  These newer engines have OBDII firmware which is more sophisticated than previous engine control systems. I was fortunate to have access to an oscilloscope, a Subaru expert, and a number of vehicles. I was able to fly without error conditions early in the development stage. I ended up with only one fault code would pop up during taxiing. I would read the code, verify it was insignificant fault, then reset the ECM.  The ECM thinks it is driving down the road at 37 mph. I've detailed the few changes needed to accomplish this. The mods are easy to put in place. 

Special thanks to Brian Oakes. There isn't a single design characteristic I didn't run by him. He always was able to provide alternatives and feedback. He has an immense knowledge of Subaru engines.


I burn 7.4 gph auto fuel at 4200 engine rpm at 175 mph cruise. Gross weight 1854#, 2.0 sq. ft. drag,1145# empty weight. My Cozy IV departs the runway at exactly 2000 ft with 2 heavy occupants at sea level. Climb is 900 fpm at 100 mph (1000 fpm if I fly alone). It looks like my performance numbers are very similar to an 0320 Lycoming.

I have gained another very substantial benefit from the auto conversion. The slower turning prop and greater spark control yields fantastic descent rate when needed. This is very important safety advantage in a canard aircraft. I can do short base and final similar to 30 degree flaps on my Cessna 150.

You will not find a quieter, smoother operating, more fuel efficient combination.

ECM (Engine Control Module) modifications:

The ECM expects certain inputs that aircraft builders don't have available. Inputs such as the transmission status, EGR solenoid, and vehicle speed aren't applicable to the aircraft. If you just disconnect the wires from these devices, the ECM recognizes these as faults. It then turns on the "check engine" light and stores the fault code in ECM memory. You don't want to be flying around with a check engine light on, as it might be on for a truly significant reason. So I had to find ways to fake out certain inputs. These changes described below.

It appears that auto manufacturers tend to add more and more input requirements for each model year. Newer ones now look at fuel temperature, power steering fluid level, etc. This can make our conversion process more difficult.


Changes needed for 1995 thru 1998 year ECM:

1) Tell the ECM you have a manual transmission. Since your engine most likely came with automatic tranny, you need to remove wire #2 from the mass air flow sensor (this connects to pin 47 of ECM). You also need to supply a ground wire to pin 81 of the ECM. Auto tranny wiring harnesses don't have a pin at that location, so I just stole one from an unused wire. While I was at it, I also snipped the wires from pins 61, 79, and 80 since manual transmission vehicles have no connection at those pins. Note, the transmission code still faults occasionally. I have yet to completely eliminate it.

2) Pin 83 of the ECM needs to see a square wave form that goes from less than 0vdc to more than 5vdc. This circuit is really the only significant change you need make. It prevents the ECM from going into "limp mode" 50 seconds after you apply full throttle. (Rumor has it that this limp mode is not an issue with newer engines). One way to accomplish this is to use a 555 timer, a couple resistors and a couple of capacitors. You can look up instructions on how to wire this on the net. I suspect there are even simpler solutions, and the mod in step 1 above may negate the need. 

3) You need to duplicate the ignition switch wiring. It needs to see at least 5v at pin 82 of ECM in order to start the engine. I just wired the 12v from the ignition "accessory" terminal to that pin. This tells the ECM that the tranny is safely in "park" position.

Engine modifications:

egr changes

The EGR valve interferes with my engine cowl. So I just removed it along with the supply tube. I then mounted an aluminum plate over the opening to seal off the intake manifold. I placed a large ball bearing behind the tube fitting to seal the hole.  I left the controlling solenoid in place and wired, as the ECM checks for the voltage drop. Optionally I could have removed solenoid and installed a high watt resistor. See pic below.

Throttle spacer

The throttle body was the only other component which interfered with the engine cowl, so I just built a triangle shaped spacer (Yellow item in pic) that rotates it to a horizontal position. I added an extra gasket, as it is essential to have absolutely no intake leaks.

My propeller speed reduction unit (psru) requires a return line to the oil pan, so I brazed half of a 1/4npt pipe coupling to the pan near the top. Make sure you fill pan with water and verify your weld will not leak.

I also cut the oil fill tube, rotated it to a better position, then glassed it back together. I drilled part way into the tube at the joint to make sure the glue had something to grab onto. Don't be tempted to glue this with pvc glue, it will fail eventually.

Vacuum hose routing:

Two of the error codes I saw initially were related to the vacuum hose plumbing. The hose going from the evaporative canister to the "Air pressure solenoid" (above pic) is necessary. It lets the ECM know that the fuel cap is installed and the tank is under pressure. So I just routed that hose to my fuel tank vent, since it sees pressure during flight.




I elected to make use of the oxygen sensors supplied with engine. I rolled my own muffler out of cold roll steel (crs) and inserted a ss baffle which separates the inlet from the outlet. This baffle must be ss as it gets extremely hot and rapidly degrades if not. The neat thing is that this muffler results in the two oxygen sensors having different output values, a requirement of the OBDII system. Ideally, the whole muffler should be ss. Look at my 3.0 build details for info on making SS muffler.

Psru modifications:

I'm using a Ross planetary gear reduction unit having a 1.85:1 ratio. This allows the prop to spin relatively slow at high engine rpm. There were a few reports of psru failures. All of the failures were related to the input shaft moving fore and aft. This causes a noticeable rattle at idle rpm. The movement can fatigue the clutch plate sheet material. It also causes the flywheel pilot bearing to degrade after only 40 hours of operation.

I never had any failures, but I did observe pilot bearing wear and loss of bearing lubrication. While investigating input shaft end clearances, I realized that differential heat expansion of the steel and aluminum components could allow the clearances to change substantially during normal operation. I changed the design to eliminate this variation and to reduce the input shaft end clearance. The input shaft can no longer impact the face of the pilot bearing.

Input shaft

Note that the thrust bearing does not see prop thrust loads, only input shaft thrust loads. Thickness I describe were ideal for my unit, your unit may be different. I used 2 thrust washers .092" thick, and one custom alum washer .02" thick. The aluminum one mates with the face of the sun gear. Sorry, none of these are shown in the picture.


Cad plot of cage cap

The cage cap assembly effectively serves as a spring. It allows the input shaft to have zero end clearance, yet not bind. On occasions when the input shaft pushes against the cage cap, the cap can flex approximately .005", whereupon it touches the output shaft. This is pretty difficult to explain. It works, but it's possible that a simple thrust washer could accomplish the same.

The thrust bearing gets plenty of lubrication. I have over 150 hours on this unit since design change.

Radiator design:

It's so difficult to maintain a perspective about these things. The first thing to keep in mind is that heat transfer with coolant is substantially more efficient than an air cooled engine. Why is that?

 The specific heat of water is 1.0, meaning it is extremely effective at grabbing and releasing heat. Air and oil are insulators in comparison. Oil has a specific heat of 4/10. So oil is 40% as effective. Air is even worse at .24. The conclusion? It should be easy to use the same ducting as the Lycoming engines.

So how big does my radiator need to be? Well, I started looking for this info, and found it to be described ambiguously. It was then that I realized I really don't care. Just fill the area behind the stock Naca duct with an aluminum radiator. If it's too big, I can just replace the radiator in the future with a smaller one. Custom aluminum radiators only cost $300, so why try to push the limit during initial flight testing? The weight penalty is trivial.

So I ended up with a radiator that measures 17"x 7.25"x 3" deep. That's 369 cubic inches of cooling fins for a 165 hp engine. 2.2 cubic inches of fins per hp. The measurements above are the fin area, not the total radiator size. I had multiple radiators made to allow various tests. One by Griffin Radiator, two by Macs Radiator. http://www.macsradiator.com/home.shtml 

I definitely prefer Macs. Price was much more reasonable. Below is the cad drawing of my latest radiator. The vent tube allows air to exit the radiator. It would not be needed if one of my large beaded hose ports was higher. I just happened to have engine mount tubing that prevented moving either of those two higher.

Rad cad plot

How did I duct the radiator? I just extended the walls of the stock Naca duct to meet my radiator. So basically, the radiator is 5 inches behind the firewall, beneath the engine. I then made my own rubber gasket material to seal the radiator tight to the ducts. No leaks.

In theory it's important to have outlet air ducting to reaccelerate the air and dump it  back to the air stream. I made some temporary ducting to test this concept. It negatively affected my cooling (oil temp) so I got rid of it. Currently I allow the outlet air to wander around the engine compartment, thus cooling the oil pan, then follow the contour of the engine covers and exit at the base of my propeller. As a result, I don't need an oil cooler (less drag, complexity). My oil temperature follows coolant temperature then peaks at 237F. Well within normal operating temp.

Measuring air flow:

I aggravated the cooling by flying with a compression leak. I caused the leak when I deliberately overheated the engine many times during my taxi test period. I guess you'd have to be familiar with my testing background to appreciate this test approach. Anyway, this was a great way to test cooling ideas. I used a barometric pressure sensor to measure the pressure differential across the radiator. This was accomplished by placing a small diameter tube on both sides of the radiator. I then had my computer measure the pressure difference between the tubes. I found that I would often have the air flowing in the wrong direction immediately after take off. But if I just leveled off or dipped my nose for a few seconds, the air would start flowing the correct direction and the pressure differential would increase. All of this would coincide with a drop in coolant temperature during high air flow periods.

Eventually I added some vortex generators ahead of the Naca, and simple exhaust augment tubes to increase air flow. The vortex generators made a big difference in air flow across the radiator. They consist of two pieces of aluminum angle 1.5” x 1.5” x 3” long. They form a 40 degree included angle. The front edges separated by 3.8”, the rear edges separated by 6.5”. Looks like this / \ when viewed from beneath plane. Basically, they create a little whirl wind which allows air to bust thru the dead air under the plane.

Yet the compression leak still dominated cooling. I finally decided I'd tested enough ideas, fixed the compression leak and my system is perfect now. If I'm at an air show on hot day (90F), my coolant can rise to 212f to 217f during long taxi periods. No biggie. However, it always cools off during climb. I can climb at max power from sea level to 12k ft with temperature constant at 199F. So basically, I never have to give any consideration to coolant temperature. In fact, it's pretty clear I have an opportunity to reduce the aircraft drag by reducing Naca air flow. I should be able to end up with a lower drag area than most canards, as I also don't need the cowl protrusions that are stock.













Trapped air in coolant:

The 2.5 liter Subaru is sensitive to air bubbles in the coolant. It only takes about 1/2 cup of air in the engine block to cause heat transfer problems. This engine is particularly sensitive. Trapped air is easily 10 times more significant than any other parameter. It's conclusive that trapped air causes local boiling in the block. This affects the entire cooling sys and rapidly degrades heat transfer.

During initial ground testing I did multiple high speed taxi runs. I would see the coolant temp elevate after each run. It would get to the boiling point after 4 runs down the 5000 ft runway. If I continued, the cooling sys would boil over. I deliberately did this many times. If I aborted any further runs, then after shutting the engine down I could hear it gurgling. This was the fluid boiling internally.

All of these problems were caused by a small quantity of air trapped in the block at the highest point. Eventually,  I added a small diameter tube to this block high point. This allows all the air to exit the block and move to the small reservoir under the radiator cap. Unfortunately, all of my deliberate overheating caused the heads to warp. This allowed compression gasses to flow into the cooling sys during high power settings. My computer detected this problem. 3 seconds after applying full throttle, the coolant pressure would rise to 24 psi. It would then slowly drop 5 seconds later.

So trapped air causes head warp, which causes air to enter the cooling system. It was pretty amusing that at the same time I discovered this entrained air sensitivity in the 2.5 engine, so too did the auto dealers. Head warp caused by customers changing their coolant is now the number one warranty item with this engine. All caused by an engine block that was not plumbed to dynamically remove air from the high point. Subaru has since changed their cooling system design. 

It's essential that the 2.5 liter coolant crossover tube atop the engine be drilled and tapped. This allows user to add a small tube from there to the coolant reservoir. Any air inside the engine block then automatically purges. Proof of effective purging is that I can now drain all the coolant from the entire sys, then refill.  Every drop can be refilled without hesitation. Before adding this air purge, I would end up with a few cups of fluid that I could not get back in to the system. Also, I now can't get the engine to gurgle after a hot shut down.


Keys to liquid cooled engines:

-         Design your cooling system to automatically remove trapped air from the various components. Called a dynamic air vent. Don’t assume air will flow with coolant. Trapped air in any engine causes local boiling. This suddenly causes power settings and temperatures that were manageable, to become a problem.


-      Design your cooling system with 2 cups of air under the radiator cap. This is a big safety advantage, for bizarre reason. With 2 cups of air under cap, the system pressure will never exceed 7 psi. This allows you to use your pressure sensor as a great way to evaluate the condition of your cooling system. If you didn’t have that air cushion under cap, your pressure would routinely jump to cap pressure (22psi on my system). That would render your pressure sensor useless for evaluation purposes.


-         Make sure your cooling system uses pressure sensor, temp sensor, and analog fluid level sensor. The combination of those sensors makes the pilot highly informed about cooling system status.


-         Use facts to make your decisions, not speculations. Convert speculations to facts.

Two vortex generators. This viewed from rear of plane. You can see the beginning of the NACA duct in foreground. The vortex generators are located in the middle of the landing brake.