MPC in automotive industry

Perhaps a bit of discussion would help.. Is this really a new idea?? It sounds a bit like fuzzy-logic control to me.

I know very little about the automotive control industry other than it is extremely well developed and more advanced than most realise. Having been around the process industry for a long time, my new car is way too clever for my liking ;-)

Cameron:-)

Cameron

MPC is not a new idea, it is a well known optimal control method with many advantages (e.g. deals with input/output constraints directly). I don't know much about fuzzy control; the biggest difference I suppose is the presence of an (approximate) model of the process to be controlled, that can be used for predictions and thus for more effective control -- as opposed to the "rule of thumb-y" nature of fuzzy "models"

The problem with MPC is that it normally requires a lot of processing power online. Our solution eradicates this problem, thus MPC can be applied to high sampling rate processes like engine control

All is fine in theory but we lack experience of how control things really work in the modern "smart" cars:

• is there some centralized CPU handling all the control loops, or each loop is a separate chip or perhaps something inbetween? (a few boxes)

• how do the various controllers cooperate? (e.g. they shouldn't be taking competing actions!)

• who are the manufacturers of such hardware?

If you have any URLs or names, or if you can offer some advice I'd appreciate it a lot!

thanks nikos

I went through a similar exercise a couple of years ago (we were marketing distributed parallel-computing embedded-systems technology to the automotive industry). You can do a lot of research on-line; but in my experience there's probably no substitute (if you can afford it) to employing an automotive market specialist to provide the necessary industry contacts and to slog around the country/europe visiting potential clients. Otherwise, you just have to get to know the key people in the industry by networking with the people who are in it. I'm not sure whether my marketing contacts and automatic-control contacts at Ford, Jaguar, Ricardo, etc. would still be current. You should probably consider approaching the top motor-racing R&D teams (who are always pushing the limit), as well as production vehicle manufacturers and their suppliers. One of the problems we found was that, apart from Ford (who have a research centre at Dunton in Essex), all the other manufacturers have their R&D (not to mention their production) outside the UK. Also, don't forget the defence arena: I got the impression that much of the R&D on integrated power-train control and control of multi-body vehicle dynamics is for large military vehicles. WRT control of multi-body dynamic systems - the Inst.MC have co-sponsored a series of annual conferences on this topic; e.g. - another possible source of leads.

Kelvin B. Hales Kelvin Hales Associates Limited Consulting Process Control Engineers Web:

thanks Kevin

this is the classic "chicken and egg" problem: we can't really approach automotive companies directly without a prototype, and we can't do a prototype without having some inside contacts in the first place!

Not to mention that contacting a company like say Jaguar is not a trivial issue: who exactly do we talk to?

that's why I thought a post in usenet could offer a back door so to speak

do you have anything in mind and how much do they cost? (did I mention we operate on a shoe-string? :)

thanks nikos

But the "rule of thumb-y" nature of fuzzy models does not imply inaccurate output. Do you have any research comparing the performance of the two?

-Will Dwinnell

i don't want to start a holy war between mpc/fuzzy :) as i said i only have a layman's knowledge of fuzzy systems using a "real" model allows future forecasts beyond the next sampling period, that's all.

but i guess whether we were doing fuzzy or mpc controllers we would be at the same position wrt knowledge of the automotive controller industry: at a loss!

nikos wrote: "i don't want to start a holy war between mpc/fuzzy :) as i said i only have a layman's knowledge of fuzzy systems"

My point was simply that, while I agree that fuzzy logic systems are (often, anyway) "rule-of-thumb-y", as you say, that is not related to how well fuzzy logic systems perform.

nikos wrote: 'using a "real" model allows future forecasts beyond the next sampling period, that's all.'

There is nothing about fuzzy logic which precludes predicting beyond the next sampling period.

-Will Dwinnell

From what I know of modern cars, including my own - a Subaru, the internal workings are divided into separate and distinct sub-systems, each with their own CPU, communicating with each other where necessary using a peer-peer Controller Area Network (CAN) chipset. A Google search for "car can bus" gets some good hits.

These sub-systems include EMS, ABS, digital dashboard, airbags, climate control, transmission systems, roll control, etc. etc.

I have first-hand experience in some of this:

1. The gearbox has been driving me nuts - the fuzzy-logic controller in the transmission system changes the gear shift points and "learns" as you drive... Mine's in Granny Mode ;-)
2. The climate control system would blow cold air at me in the middle of winter, so the garage techs ran their diagnostics and re-booted the processor - apparently it's a software issue and they are waiting on the next firmware rev. :-(

..bring back the old days!

Cameron:-)

i found a lot of relevant things to search for after a lot of googling like RTOS ECU DSP ASIC CAN AUTOSAR etc

so it must have some neural network somewhere in there too apart for the fuzzy logic?

or bring on better controllers, we can help there :)

nikos

Cameron Dorrough wrote: "The gearbox has been driving me nuts - the fuzzy-logic controller in the transmission system changes the gear shift points and "learns" as you drive... Mine's in Granny Mode ;-) "

Do you have any references regarding the fuzzy logic controlled transmission learning? I have been trying to find out about this. All of the literature I've come across about fuzzy logic in cars has described "static" fuzzy logic systems. A number of automotive systems have been engineered with fuzzy logic control, and I have heard non-technical people make this claim about learning, especially with regard to automatic transmissions. Although fuzzy logic systems can be constructed to adapt over time ("learn"), most deployed fuzzy systems do not.

-Will Dwinnell

...

I'm puzzled by that. If a manual shift, what does the controller do? If automatic, what does it learn?

I has a Plymouth that would choke up, lose power, and make the catalytic converter glow red hot after ten or fifteen miles of running if I had started the engine with my foot on the gas. Turning the ignition off and back on while running town the turnpike, so resetting the computer, was a fix. Chrysler had known about the problem. Like a dummy, I gave them the fix without charge.

Jerry

After a little googling:

" The timing is determined by a computer, based on the driver's operation of the accelerator pedal, brake pedal, steering wheel and automatic transmission select lever. Then, the electromagnetic solenoid-operated valve is activated. This valve activates another valve called the "spool" and shifts the distribution of hydraulic pressure."

The above Japanese doesn't say anything specifically about fuzzy-logic, but the "based on driver's operation.." bit sure sounds like one to me and it definitely changes every time I drive. I live in a hilly area and for the same speed and throttle the auto transmission kicks down at a different point on the hill every time.

I have been told by Subaru to give the car a good bit of stick every so often to reset the transmission - and it seems to work. Don't get me wrong - they are an amazing car in all respects and this is about the only thing I can find wrong with it: The darn thing is smarter than me! ;-)

Cameron:-)

Like we don't!?

Suggest continuing correspondence by e-mail.

Kelvin B. Hales Kelvin Hales Associates Limited Consulting Process Control Engineers Web: