Dual micro squirt aircraft project

[flightlead404]

After a couple years designing, building, and running a micro squirt install on my ‘86 Porsche 930 and getting considerably more HP I’ve decided I’m going to leverage this knowledge to improve my aircraft performance. This is a Lycoming IO-360 parallel valve clone with around 8:1 static compression (flat “boxer style” 4 cylinder oil cooled twin plugged engine). Turbo charging not in the initial scope but a second phase might include turbo normalizing. CAFE style racing a possibility.

Before I start DO NOT WASTE MY TIME WITH COMMENTS ABOUT RISK!

I already fly an experimental aircraft with an experimental engine and in the past have BEEN PAID to fly low level formation aerobatics on the worlds largest formation aerobatics team.

At night.

With pyro.

I understand risk.

I’m in the early requirements definition phase of the project but initially I’m looking at the following goals:
-Increased HP overall
-Reduction in HP drop off with altitude
-Increased fuel efficiency
-Reduction in complexity in the sense of not manually controlling mixture
-Switchable “best power” and “best economy” modes (roughly equivalent to running 200 f ROP or 45 f LOP adjusted for altitude)
-Built in safety with dual independent microsquirts and supporting components
-Knock detection possibly with automatic retard, at min visual warning

At this point I’m looking for general ideas, lessons learned from anyone who has travelled this road.

Engine currently has one impulse coupled magneto and one SDS ECU driven electronic ignition with VR pickup (I did not design), and mechanical CIS-like fuel injection.

[Rrrracer]

Your post was a breath of fresh air. I have nothing to offer, other than my thanks to you for being a badass.

[ArttuH]

Few random comments and questions:
-Do you have already planned how switching between the systems should happen? Manually or somehow automatically?
-Regarding redundancy it's naturally important to decide in erly phase what system parts you want to dublicate. Dublicating everything is probably the safest and simplest option if it's feasible. But even in that case you need to pay attention to ensure that the systems really are independent.
-Microsquirt doesn't have much support for knock detection. So if you really need it some MS3 based ECU would be a better choice.

[flightlead404]

Your post was a breath of fresh air. I have nothing to offer, other than my thanks to you for being a badass.

Lol thanks!

[R100RT]

Along the lines of what ArttuH has raised, the aspect of dual redundancy is worth considering further. A helpful question might be, envisioning all different aspects of being in flight or taking off and failure. What's the worst case scenario? That's probably what is best to envision to shape your choices - you're probably the best judge of that/ those conditions?
Having two ecu's along with a complete array of sensors and wiring would cover most anything and could be a pilot control switch over, or automatic. The exception would then fall to the fuel pump/ psi regulator/ ignition coils/ spark plugs. Pump and regulator could be duplicated, and the engine most likely has dual plugging?
For testing, regardless of above "full Monty" duplication, or if on a lessor level with clever "switch over" gear, I'd utilize a single Jim Stim extensively initially.
My vote would be to "Fly at it".

[flightlead404]

Few random comments and questions:
-Do you have already planned how switching between the systems should happen? Manually or somehow automatically?
-Regarding redundancy it's naturally important to decide in erly phase what system parts you want to dublicate. Dublicating everything is probably the safest and simplest option if it's feasible. But even in that case you need to pay attention to ensure that the systems really are independent.
-Microsquirt doesn't have much support for knock detection. So if you really need it some MS3 based ECU would be a better choice.

I'm fine with switching manually as long as its a single action, e.g. a single switch to throw, which could of course be multi-pole or control some logic that does other things. I'm fine with building some circuit to do that or programming a micro controller such as an Atmel (Arduino). Of course, that would also need some fault tolerance. If the switching happens automatically I'd want some visual indication. However, I'd like to explore avoiding the need to switch between ECUs completely. That is, design the system such that both ECUs are running all the time. That avoids a lot of issues, including issues around booting an ECU on an already running engine, as well as human factors.

In terms of duplication, I'd like to duplicate as much as practical. On the input side, I can run separate and independent sensors for crank position (dual VR sensors behind the flywheel which I'll machine to insert a couple of neodinium magnets at say 25 BTDC. No cam position sensor needed as I'll run wasted spark, but it would not be be too difficult to run something off the accessory case if I decided I wanted that. Dual MAP easy to tee off, dual ambient air pressure easy to tee off the static system, dual IAT likely no problem I think I can find space there. The big question is MAF, whether I have space to be able to "stack" two MAFs in the intake and if that would cause any problems. If not, I'll need a way to feed the same data to two ECU's.

On the output side, spark is easy as the engine is dual plugged. So one ECU can run the top plugs and one the bottom with completely separate coils and harnesses. Fuel is something I'm struggling with. Can I wire two ECUs to the injectors in parallel without some switching circuit? If so, does that mean the injector is "on" whenever at least one ECU is commanding it on, and off when both ECUs are commanding it off (i.e. the inputs are OR'd), which I think is OK. Can I do this simply by wiring them in parallel or will that cause issues? Do I need to build an OR logic circuit maybe that switches a FET that in turn switches the injector. Simple is better as long as its reliable. What are the failure modes here? If the failure mode is only that the ECU fails to turn the inject on, then ORing is OK. But if there's a failure more where the injector is never shut off then I need something else.

Remember, not everything absolutely MUST be duplicated. I already have only one throttle, throttle cable, one cam, one crank, one fuel pump, etc etc. If a MAF failure is highly unlikely, and/or there's some sort of "limp home" mode, then I'm ok with one MAF.

One outstanding question wrt switching is around the pre-takeoff checklist. A standard pre-takeoff check in a piston engine aircraft is the "mag check" ie alternately switching one ignition off then the other and making sure engine RPM stays within limits. I'd like to have a similar check where I can completely disable one ECU at a time. Are there issues booting an ECU on an already running engine? e.g. at start up does it sample MAP for example and use it as a baseline somewhere?

For knock detection I was thinking about the J&S Safeguard system but my knowledge about that is limited, I need to do more research.

Right now I'm in the stage of assessing "doability", questions like yours help me walk through the implementation in my head, so thanks for that. I need to make sure not only do I have answers to the question, but that I'm even asking the right questions at this point.

The next set of problems is how to tune this. I expect it to be a bit more complicated than tuning a car install, but I could be wrong. I also need to figure out how to switch between tunes, best power and economy cruise modes.

[flightlead404]

Along the lines of what ArttuH has raised, the aspect of dual redundancy is worth considering further. A helpful question might be, envisioning all different aspects of being in flight or taking off and failure. What's the worst case scenario? That's probably what is best to envision to shape your choices - you're probably the best judge of that/ those conditions?
Having two ecu's along with a complete array of sensors and wiring would cover most anything and could be a pilot control switch over, or automatic. The exception would then fall to the fuel pump/ psi regulator/ ignition coils/ spark plugs. Pump and regulator could be duplicated, and the engine most likely has dual plugging?
For testing, regardless of above "full Monty" duplication, or if on a lessor level with clever "switch over" gear, I'd utilize a single Jim Stim extensively initially.
My vote would be to "Fly at it".

Thanks.

Remember, it doesn't need to be completely fault tolerant, it just needs to be no worse than the current setup I already have only one fuel pump for example, and have thousands of hours in aircraft with a single fuel pump with only one failure I'm ok not duplicating that.

[ArttuH]

In past I have done some designing of redundant control systems. Based on my experience setting requirements for fault tolerance can be a slippery slope. If you try to cover all possible fault scenarios the system gets quickly very complex. And more complex it gets easier it is to miss some little details that can still make it vulnerable for single point failures regardless of all redundancy and other precautions. So I think that key to success is to start by going through different fault scenarios, trying to think their probability and severity as realistically as possible and then deciding which ones you want to cover and which ones you will leave as known risks.

Duplicating the fuel system or not is one good example of these decisions. With single fuel system you get a handfull of potential single point failures: pump, regulator, fuel line leaks, power supply of the injectors and pump and so on. On the other hand duplicating all this adds complexity quite significantly.

Then to technical side.

Duplicating the sensors should be pretty straightforward and simple as you said. Regarding MAF, unless you have a good reason to use it I would probably leave it out completely and just use speed-density control. But I don't have much experience about tuning for varying altitudes, maybe a MAF could help that. If the engine currently has a magneto distributor ignition one option could be keeping it and use another ECU controlling it and reading timing reference from it. And another ECU would get triggering directly from the crank. That way you would get two totally independent triggering systems. But again, at cost of added complexity.

On output side ignition looks pretty straightforward. As you said one ECU can control one set of plugs and another one the other set. For "mag check" function you can simply have a power selection switch for the ECUs, A / B / both. Switching the ECU off and on while the engine is running shouldn't be an issue. You just get unnecessary afterstart enrichment when switching on but that shouldn't be a big deal. Alternatively you can keep both ECUs active all the time and just switch power of ignition systems.

Connecting the ECUs parallel to single set of injectors should work, I think. Effectively you will get combination of control pulses from both ECUs. So the trick is to keep the ECUs completely synchronized, otherwise you may get up to double amount of the fuel if the control pulses aren't aligned. This may require using a cam sensor for sequential injection, I'm not sure if injection events are accurately timed in batch injection mode. Good side of this arrangement is that then you could keep both ECUs fully active all the time and dropping off another ECU shouldn't cause any hiccups except losing another set of spark plugs.

Microsquirt supports knock detection with external knock detector as you said. So that way you can do automatic timing retard, for example. But how well the detection works will depend completely on that external detector.

[flightlead404]

...setting requirements for fault tolerance can be a slippery slope. If you try to cover all possible fault scenarios the system gets quickly very complex. And more complex it gets easier it is to miss some little details that can still make it vulnerable for single point failures regardless of all redundancy and other precautions. So I think that key to success is to start by going through different fault scenarios, trying to think their probability and severity as realistically as possible and then deciding which ones you want to cover and which ones you will leave as known risks.

Duplicating the fuel system or not is one good example of these decisions. With single fuel system you get a handfull of potential single point failures: pump, regulator, fuel line leaks, power supply of the injectors and pump and so on. On the other hand duplicating all this adds complexity quite significantly.

Agreed. As mentioned, I don't need to be completely free of single-point failures. I already have several potential single point failure modes. I just need to be no worse than before.

Real world example the "dual magneto" used on O-360 engines in several Mooney and Cessna aircraft was two magnetos in a single box driven by a single gear on the accessory case. It contained several single-point failure modes but (wisely or not) was approved by the FAA in certified aircraft.

I have a single fuel pump, filter, line, selector, single cam, crank, oil pump, oil filter, throttle cable, lever, cable etc etc etc.

Then to technical side.

Duplicating the sensors should be pretty straightforward and simple as you said. Regarding MAF, unless you have a good reason to use it I would probably leave it out completely and just use speed-density control. But I don't have much experience about tuning for varying altitudes, maybe a MAF could help that. If the engine currently has a magneto distributor ignition one option could be keeping it and use another ECU controlling it and reading timing reference from it. And another ECU would get triggering directly from the crank. That way you would get two totally independent triggering systems. But again, at cost of added complexity.

It appears commercial EFI uses speed/density so that's probably the way to go.

On output side ignition looks pretty straightforward. As you said one ECU can control one set of plugs and another one the other set. For "mag check" function you can simply have a power selection switch for the ECUs, A / B / both. Switching the ECU off and on while the engine is running shouldn't be an issue. You just get unnecessary afterstart enrichment when switching on but that shouldn't be a big deal. Alternatively you can keep both ECUs active all the time and just switch power of ignition systems.

Connecting the ECUs parallel to single set of injectors should work, I think. Effectively you will get combination of control pulses from both ECUs. So the trick is to keep the ECUs completely synchronized, otherwise you may get up to double amount of the fuel if the control pulses aren't aligned. This may require using a cam sensor for sequential injection, I'm not sure if injection events are accurately timed in batch injection mode. Good side of this arrangement is that then you could keep both ECUs fully active all the time and dropping off another ECU shouldn't cause any hiccups except losing another set of spark plugs.

Microsquirt supports knock detection with external knock detector as you said. So that way you can do automatic timing retard, for example. But how well the detection works will depend completely on that external detector.

If the after start enrichment only causes a momentary rich stumble then that's fine, not a problem.

I was planning on a separate sensor baro adjustment anyway. I can probably double up the TPS on the butterfly valve easily enough if I decide I need to do that.

If the ECUs are running the same tune and the sensors are calibrated close enough, I don't see a way for them to become significantly out of "sync" except in some sort of failure mode. If they are just a teeny bit out of sync I expect it to mean a slightly wider pulse aka slightly rich.

[R100RT]

Having a thought regarding running dual enabled Micro's that "share" the burden on control. Much like closely matched pneumatic pressure regulators might gain a resonant fight as they overcorrect each other, probably needing to have your ego control nullified to avoid a similar condition would be helpful, perhaps critical. Not sure you're going to gain 100% identical behaviour, one to the other even given same MSQ, & duplicated sensors. Perhaps wrong here, but I now suggest a dual Jim stim test - which more fairly replicates dual sensors etc. $0.02 offered.

[racingmini_mtl]

Just off the top of my head, if the ECU doesn't have an RPM input it won't inject fuel or fire a coil. So switching the crank sensor between ECUs might be something to look into since there wouldn't be any possible conflict between them on fuelling and spark. I haven't thought about what else could be impacted though.

Jean

[flightlead404]

Having a thought regarding running dual enabled Micro's that "share" the burden on control. Much like closely matched pneumatic pressure regulators might gain a resonant fight as they overcorrect each other, probably needing to have your ego control nullified to avoid a similar condition would be helpful, perhaps critical. Not sure you're going to gain 100% identical behaviour, one to the other even given same MSQ, & duplicated sensors. Perhaps wrong here, but I now suggest a dual Jim stim test - which more fairly replicates dual sensors etc. $0.02 offered.

Thank you

I could definitely see a situation running in closed loop mode where due to some tiny difference the injector pulses are not completely aligned, giving me a wider than expected pulse, in turn resulting in a richer mixture than the AFR table dictates, resulting in shorter pulses form each ECU, resulting in either or both of a slightly narrower than normal pulse from each ECU but the total pulse being on target. Each ECU would be commanding a leaner mixture, but the total would be on target. I don't see how it could result in increasingly worse aligned pulses or mixture. Thoughts on this?

Honestly I thought about this and don't really see a way for some mixture oscillation to become increasingly divergent. That would require one or both of i) a condition that causes increasingly out of sync injector pulses, and/or ii) max EGO control percentage being quite large. In that event, the action would be a pilot-controlled disabling of one ECU which would return mixture to nominal. I see a much more likely scenario to be that each ECU would simply lean very slightly to account for the slightly larger injector pulse. That said, I don't plan on running closed loop, although I may need to tune closed loop for a short while, which I could do on one controller while running my current ignition on the other side then duplicate.

However, this has made me think a bit about what would cause injector pulses to be out of sync. The only cause I think of is a difference between when the crank position sensors trigger. They wouldn't give a significantly different RPM, and therefore wider or narrower pulse, but they could give a slightly different indication of TDC and therefore when the pulse starts. My plan was to have two VR sensors mounted on the nose of the crank case. There's a boss machined into the case there and its an "engineered dimension" with tolerances in the .001's of an inch. I was going to CAD up and CNC a mount for the VR sensors, then have the back of the flywheel machined to accept a small neodinium magnet. Lets say tolerance stack-up results in an unintended offset of 2mm or roughly 0.08" (which I think is larger than reality). If the VR sensor/magnets are 100mm from the center of the flywheel, the circumference of the magnets travel is ~314mm. One degree is 0.873mm so at 2mm they would be off ~2.3 degrees. At 3,000 redline the flywheel is turning 18 degrees per millisecond, so we're talking about 0.128 millisecond difference. IDK what the pulse width would be, so I don't know what percentage that is or if its significant. Anyone hazard a guess?

I need to come up with a procedure to verify and set the sensor angle before TDCs to minimize any differences here. Maybe I can do that cranking the engine over with my Oscope connected to the sensors, maybe I can get that level of accuracy with manual static measurements. IDK

Other than EGO control, what other feedback loops might occur that could cause some sort of oscillation, particularly divergent oscillations?

[flightlead404]

Just off the top of my head, if the ECU doesn't have an RPM input it won't inject fuel or fire a coil. So switching the crank sensor between ECUs might be something to look into since there wouldn't be any possible conflict between them on fuelling and spark. I haven't thought about what else could be impacted though.

Jean

Excellent idea. The ECU would be powered up and ready to go, but wouldn't be firing.

I still like the idea of having both running all the time, but this is something to keep in the back of my mind.

[R100RT]

Agreed on the relatively low potential for an aggravated "over correction" situation to develop between MS of fuel being delivered between the parallel ecu's - however I've found things frequently turn out the way they turn out, as the tires hit the pavement - or in your case wings soar you to the skies. So in that case, testing becomes critical and highly beneficial. The closed loop vulnerability seemed an obvious potential, but I'm not actually seeing anything else of concern.
Regarding sequential fuel injection, I have a jaded opinion on not using it (and some supporting experience with my 2 cylinder boxer revealing a high sensitivity to the fuel charge crank angle (based on non sequential changing that crank angle point with every start up). If your engine has a similar sensitivity to point of injection then the differences found between starts might be enough to introduce some irregularity. I'm not familiar with how well the "Sequence Batch Fire" works. I added a cam sensor to gain sequential and solve those issues on my boxer.
But your mention of a "procedure to verify and set the sensor angle before TDCs to minimize any differences" , JimStim testing along with a fake flywheel turning in your lathe or drill press and having the VR's and magnets mounted might be easier on the engine. Especially if there needs to be some slight shimming or massage of one unit components to the other to gain absolute concurrent readings. Although the engine with plugs out etc. is the ultimate proving ground, the other approach allows you to get up to and above anticipated rpm's to fully prove the method.
Good luck on the project, photos required too

[bobxyz]

Starting off, personally, I'd rather play in traffic during rush hour than go up in a MicroSquirt controlled experimental aircraft, but I'm risk adverse.

For the injectors, those absolutely need to be switched with a relay (or toggle switch) since the MS FET drivers can fail full-on (search for "dead VND5N07").

The MS only lightly loads the WBO2 sensor (if any), the MAF sensor, the MAP sensors and the TPS sensor, so those sensors can be shared by both ECUs.
The MS heavily loads the coolant and air temperature sensors, so those need to be duplicated (or the load resistors need to be removed from one MS).

The MS2/MicroSquirt doesn't include knock sensing, and the simple marketplace boards may be more of a rpm/engine noise sensor, than an accurate knock sensor. The MS3 knock board (if available?) does knock sensing over a narrow window of rotation around the spark time, which is _much_ better than the simple boards.

Due to slight component variations and electrical noise, the sensor values seen by each MS won't be identical. This means that the 2 MSes won't have identical responses. If I were doing a redundant racecar, I'd probably add an arduino-class processor that compares the 2 MS outputs and turns on a miscompare light when they're significantly different.

Have you thought about how you'll do the initial tuning? For a car, it's no big deal if it takes a while to incrementally get the tune setup. If the tune is off, you coast to the side of the road, adjust a few settings, and try again. In an aircraft, the stakes are a lot higher since you need full reliable power on initial takeoff.

For automotive use, the MS firmware has been refined over many years and many installs. For aircraft use, you'll be actively using more of the table entries due to big&fast altitude&temperature changes. This may not be well refined in the MS. Likewise, for automotive use, initial startup is only at normal ground elevations. I don't know how well the MS startup algorithms work if you need to do a restart at high altitude and cold temperature.

[flightlead404]

For the injectors, those absolutely need to be switched with a relay (or toggle switch) since the MS FET drivers can fail full-on (search for "dead VND5N07").

Ok will check it out. I was thinking I could do a solid state switch using FETs but can you think of any issues running an MS with the INJ simply disconnected ?

The MS only lightly loads the WBO2 sensor (if any), the MAF sensor, the MAP sensors and the TPS sensor, so those sensors can be shared by both ECUs.
The MS heavily loads the coolant and air temperature sensors, so those need to be duplicated (or the load resistors need to be removed from one MS).

I would duplicate all for fault tolerance. It’s looking like speed/density is the way to go so no MAF.

The MS2/MicroSquirt doesn't include knock sensing, and the simple marketplace boards may be more of a rpm/engine noise sensor, than an accurate knock sensor. The MS3 knock board (if available?) does knock sensing over a narrow window of rotation around the spark time, which is _much_ better than the simple boards.

I was originally thinking J&S but im reconsidering whether it’s needed at all

Due to slight component variations and electrical noise, the sensor values seen by each MS won't be identical. This means that the 2 MSes won't have identical responses. If I were doing a redundant racecar, I'd probably add an arduino-class processor that compares the 2 MS outputs and turns on a miscompare light when they're significantly different.

Interesting idea. From a strictly timing perspective I could take input from one IGN and one INJ from each MS and compare, and warn if greater than some percentage of rotation or pulse width. Not too complicated but for full checking it would be all INJ and IGN, a bit more fiddly.

Have you thought about how you'll do the initial tuning?

Lots of ground testing, high speed taxiing, flying in ground effect etc. Possibly even run on a test stand.

[bobxyz]

I was thinking I could do a solid state switch using FETs but can you think of any issues running an MS with the INJ simply disconnected ?

I think you'll be OK with a relay or switch for the injectors. You could switch either the control signal from the MS, or power to the Injector. You don't want to use a solid state switch because the injector behavior (flyback voltage handling) depends on the driver FET in the MS, and adding another FET in an external switch may not work, or may be unreliable. When you switch on or off, there may be a partial squirt and the resulting misfire, but this is probably OK. (On the ignition side, you want to switch the crank input on/off so that there is never an partial early spark.)

Interesting idea. From a strictly timing perspective I could take input from one IGN and one INJ from each MS and compare, and warn if greater than some percentage of rotation or pulse width. Not too complicated but for full checking it would be all INJ and IGN, a bit more fiddly.

Checking a single INJ and a single IGN would cover most of the MS electronics and sensors, i.e. if a sensor goes bad, the INJ or IGN won't match.

Can you describe the current fueling/injector setup in more detail? Is there a way to retain the existing setup and add the MS as a second setup, with some way of switching between them?

Initially, for the ignition side, you might be able to replace one of the magnetos with a crank sensor and a MS spark-only setup. You'd be able to select Magneto or MS, but not both simultaneously. Once this is working reliably, you could add a 2nd spark-only MS, or could try to add fueling control to the first MS.

[flightlead404]

I was thinking I could do a solid state switch using FETs but can you think of any issues running an MS with the INJ simply disconnected ?

I think you'll be OK with a relay or switch for the injectors. You could switch either the control signal from the MS, or power to the Injector. You don't want to use a solid state switch because the injector behavior (flyback voltage handling) depends on the driver FET in the MS, and adding another FET in an external switch may not work, or may be unreliable. When you switch on or off, there may be a partial squirt and the resulting misfire, but this is probably OK. (On the ignition side, you want to switch the crank input on/off so that there is never an partial early spark.)

I was planning on switching the INJ power, leaving the signal from the MS direct

Can you describe the current fueling/injector setup in more detail? Is there a way to retain the existing setup and add the MS as a second setup, with some way of switching between them?

Initially, for the ignition side, you might be able to replace one of the magnetos with a crank sensor and a MS spark-only setup. You'd be able to select Magneto or MS, but not both simultaneously. Once this is working reliably, you could add a 2nd spark-only MS, or could try to add fueling control to the first MS.

Currently I have a single impulse coupled mag driving the top plugs. The bottom plugs are automotive driven by an SDS electronic ignition. This uses a VR sensor and magnets machined into the rear of the flywheel as well as a MAP and BARO sensor. I'm pretty sure its set to 0 BTDC until cranking then 25 BTDC until the manifold pressure goes above some threshold (can't remember) then 28 BTDC. Pretty simple.

Current fuel injection is a low pressure system similar to the Porssche CIS but simpler. Its the Bendix RSA-5. Essentially there's a throttle body with a venturi pressure measuring device that determines total flow for the engine, then there's a spider or flow divider that distributes the flow to the 4 injectors that continually spray. There's a manual mixture lever in the cockpit to adjust flow.

I could certainly replace one or the other side of the ignition with the MS, and in fact that was my plan for initial setup for spark. I don't think I could leave the old fuel injection in place and switch between them.

[Laminar]

What's the impact if the spark plugs fire at different times? If you have two Microsquirts and during a transient one of them fires a few degrees before or after the other, does the engine run poorly? If one Microsquirt dies and stops sparking its plugs, does the engine still run?

[flightlead404]

What's the impact if the spark plugs fire at different times? If you have two Microsquirts and during a transient one of them fires a few degrees before or after the other, does the engine run poorly? If one Microsquirt dies and stops sparking its plugs, does the engine still run?

Next to nothing. As long as the first to fire plug isn't too advanced its just not an issue.

With the traditional dual mags, or with my current EI they mostly fire at different times. With my current setup the EI is firing well before the mag typically. Turning off the mag has no effect, turning off EI results in a 150 to 300 rpm drop depending on conditions.