because there are a lot of intakes for Subaru's that have a different diameter than the stock intake... does that really mess the car up because the MAF throws off the AFR's?
yes. when buying an aftermaket intake make sure the intake diameter is the same as stock. Injen CAI is the most widely used with no poroblems that I have heard of
65mm is what you want... a larger diameter is only required if u go with a BIG turbo upgrade and pushing in the ballpark of >450whp (if i remember correctly) since the MAF voltage will max out at that flow with a 65mm intake
Yes and no. When running closed loop the ECU will read the O2 feedback and compare it to what it should be (usually 14.7:1). If it's not because something is off (like the MAF) it will enter corrections in the short term and soon long term fuel trim tables. These corrections are then applied to open loop as well. So the ECU _will_ adjust for it, though I would wager a guess that it's not a complete 'fix'. Maybe SS or other tuner can chime in and explain what would not be corrected by the fuel trim changes.
scott was the one who asked me to go with the 70mm intake with my new setup... hopefully he will chime in on this
I do think that if you intentionally change the intake to a larger one you should adjust the MAF scalings for it, but curious as to just how much damage an aftermarket intake that's slightly off can do.
im curious too... my research area involves adaptive systems and i believe the learning system in the ECU follows a simple LMS approach that tries to minimize a cost function given certain constraints... such systems have the drawback of converging to local minima, which do exist in most cases when the cost function has a large number of variables... hence you have to start the learning process close to the global minimum of the cost function... with a bigger MAF thats not scaled, you might start the process far from the global minimum and hence may never converge to it... anyways, im not sure what learning algorithm the ECU uses... so my understanding may be completely off... i have sent this link to SS... hopefully he will chime in soon
If I'm understanding you correctly, it would be an issue when you have multiple parameters in the function... but in this case it only needs to look at the O2 feedback, the target O2, and the calculation to make a correction is at that point extremely basic. It's not attempting to specifically correct the MAF signal, but rather make an adjustment to the whole system; it should do the same whether the error was introduced by injector scaling or fuel pressure or whatnot. But like you, I can't do more than guess as to what really goes on in there. PS: got anywhere I can read up on LMS? I googled it briefly and it looks like it's related to linear regression...
"Bad" intakes can mess things up... It's a reasonably long discussion, but remember, the car does closed-loop fueling control. There are short-term and long-term fuel trims. The long-term are still used at WOT/open-loop. So if the car has leaned "bad" trims, they are still there at WOT/open loop. "Big-Mafs" like the APS 70, the "Siegel Racing Big Maf" and the Perrin Big Maf are bigger on purpose... they allow more power to be made before hitting 5.0 volts on the MAF sensor. SS
I have a fujita CAI, the cruise and the check engine light keeps coming on. I tried to reset the ECU but eventually it will come back on. Is there anything i can do as to modifications to get rid of this, and fix the problem at the same time? or do I just have to buy another intake (if so..what are some good ones that won't cause this problem)
LMS = Least Mean Square. It attempts to minimize the mean square error (cost function). It can be applied to a variety of optimization problems, including linear regression, where you are trying to find the line that fits the data best by minimizing the residual squared errors. The most common and simplest method to solve this problem is the method of steepest descent, where you determine the gradient of the cost function at the current point, and then move in the opposite direction of the gradient for a 'suitable' step size. You can think of the cost function as a bowl. You start somewhere on the surface of the bowl. Then you move towards the bottom of the bowl, which lies in the opposite direction of the gradient from the current point. The process repeats indefinitely. There are more complicated methods that I doubt are necessary to delve into right now. However, most of these methods assume (1) no local minima and (2) a convex surface. In most cases neither of these are true when a large number of variables are involved. Now there are 3 types of parameters in such a learning algorithm: 1) constants (eg: entered into various tables when the ECU is programmed. In case of linear regression, its the set of given data points) 2) input signals (eg: reading from the sensors. In linear regression, this is the sum or the squared errors from the current line fit.) 3) variables (eg: stuff that the algorithm can alter to minimize the cost function. In linear regression, this is the slope and y-intercept of the current line fit.) If you simply change to a larger intake, the ECU does not fix all variables. It simply tries to move to the minimum of the cost function from the current point. The situation is worse if the input signals are not reliable, especially if the algorithm does not know which input is erroneous. disclaimer: im trying to explain this in the simplest possible sense. Im not sure which exact algorithm is being used in the ECU, not which variables are changed. Moose: if you would like to read about this more, I could send you some references
I know the AEM intake is a "bad" intake and is slightly larger. This being said, while monitoring my AFR's with the wideand it really does not go anywhere close to leaning out...further in open loop it does not lean out either. This could be due to some form of turulance induced by the bend...really not sure, but it has not had any adverse effects on my AFR's but it does idle hunt some. My stock intake did a little of that too though. My general theory on intakes has always been to make sure that whatever air is needed is available...on the mustangs, they went with HUGE 90 mm units. Was this needed? Probably not, but it was there if the engine demanded it. You do want the ability when going with a larger intake to be able to scale the values so that the ECU is getting the amount of air it thinks it is though. This is where tuning comes in...street tuner, seigel tuned, whatever, it is a good idea. But, always better to be able to flow more than is needed than not enough. I mean think about it...
Yes the MAF is located on the large part of the intake...the whole intake is huge!! The purpose is max flow. Engines...turbo, NA, Supercharged all can take in some amazing amounts of air. It is one big air pump...the thought process is less restriction in, less restriction out = more hp. I think this is one of the reasons people go rotated on the wrx no? You are no longer limited in the size of the turbo inlet hose right? On an ideal set you I would go slightly larger than the intake of the turbo...on turbos that are larger like the 3" you would want a slightly larger intake with a velocity stack. This allows max air flow coming in and removes all restrictions....again just my experience from the V-8 world. Same reason you port the heads, throttle body, intake, etc...more air through the engine. I mentioned this in another thread, but there was a study done where there were some seriously free flowing ported heads, intake and exhaust puton an engine. They dynoed before at 5-7 psi higher and made less hp than after with the exact same compression ratio and less boost...50 whp just from making things flow better with less psi! It was a supercharger, but same principles apply I would think. I would surmise that there are some serious gains to be made with some good quality port and polish work on the heads, intake and exhaust...but it has to be all of them or there is no gain as it will bottle neck.
I've heard good things about K&N. Course stock is already pretty good, just don't get a lot of noise.