AFR Control

[William Boulton]

Hi,

I'd like to add a bit to the current debate on AFR control with O2
sensors. I was planning to build my own ECU and did a lot of research
over the last three years or so and found getting good information on O2
sensors very difficult but I think I may have some interesting stuff for
you to think about. I was able to talk to an R&D engineer at Bosch for a
short time and learned enough to realize that it was not going to be
easy and I would have to settle for a limited range of operation and
only fair accuracy. Here goes.

Basically all lambda sensors are the same except for the connections and
mountings and they can read residual O2 well away from stoichiometric.
The trick is to interpret the readings from the sensor and that requires
considerable computing power if it is to be done  accurately. There are
two problem areas with the sensor output 1) the shoulder at lambda < 0.9
and 2) the toe at lambda > 1.1. The shape and level of the shoulder is
highly temperature dependant, is non-linear and also subject to gas
speed and some other (unknown by me) influences. The signal level on the
toe drops into the area of noise generated by half cell potentials in
the engine and requires DSP filtering to extract meaningful data.

Now for some good news! It seems that it is possible to obtain an
INDICATION of mixture between lambda 0.85 and 1.15 or there abouts. I
have made a simple led bargraph display set for a range 0-1 volt and it
was useful and cheap. I think the covered range should be all that is
needed for most tuning applications. I believe that  setting up a
permanent closed loop system to operate at other than stoichiometric
(14.7:1) is not realistic for the small operator as there are too many
variables to consider unless you happen to have access to a FREE engine
dyno and a good 4-gas. The temperature dependance and variable lag time
could prove difficult to overcome.

There are at least three basic connection schemes for lambda sensors,
two common and the last difficult to find and expensive. The single wire
type uses chassis signal return and is really only useful for
determining "too rich", "too lean" and "maybe". The reliability of
reading a 0.0 to 1.0 volt high impedance signal in a very noisy 12 volt
system is highly suspect. The second type is the three wire unit used
commonly by Ford (HEGO) is more useful since the temperature can be
estimated from the resistance of the heater element which has a positive
temperature coefficient. It still suffers from having a chassis signal
return. The heater element draws 40+ watts and will heat the unit to
some 400+ degrees C so it can be mounted further from the exhaust port
and still give good results. The last type, sometimes referred to as a
"wide range lambda sensor" has four wires, a special shield and is
expensive (about $250.00) and rare. This unit uses wire signal return so
the noise floor problem can be substantially reduced. It still needs
complex processing to get really good. I believe this unit is used in
the USA (in some states) on vehicles which have an ultra-lean highway
cruise mode.

On the subject of lead poisoning, all lambda sensors are subject to lead
contamination. There is no avoiding it and there is no special type
which is not. In the previous paragraph, I mentioned the special shield
on the four wire type. Rather than the shield being vented on all sides,
this unit has a single hole (small) in the end of the cap and "sniffs"
the gas as it goes past. This reduces the rate at which lead
accumulation takes place, so selective mounting may be a way of
similarly reducing the effect on other units.

BTW. Both Bosch & Motec do highly accurate, wide range lambda test units
which are very impressive to watch in operation with a usable range of
0.6 to 1.75. As you might expect, they are expensive (>$3000.00). These
units are essential when developing the full fuel and ignition maps and
are always used in conjunction with a good four-gas analyser. Relying on
just the O2 reading without CO2, CO and HC values can lead to erroneous
assumptions, particularly on highly modified engines.

I have come to the conclusion that using a lambda sensor for tuning on
an engine that is basically un-modified is quite feasible if all the
appropriate measures are taken. If best possible results are to be
achieved, then ENGINE dyno time (6 hours +) and a good 4-gas are
essential, as is a knowledgeable dyno operator. That all adds up to big
bickies. I watched a friend dial in a relatively simple unit using ear and
an O2 sensor bar graph and it took months before he was happy with its'
performance over the entire range and that unit had only 16 or 20
speed/load sites to map.

Here's another "I believe". Closed loop was intended to maintain engine
tune over the life of the motor to compensate to wear and general
operational degradation in the hands of Joe Bloggs as well as satisfy the
legislators. By all means use a lambda sensor to tune the system but
then get rid of it. Subject to normal maintenance, the system should not
need resetting until the engine has done a lot of hard work. I've seen
the telemetry from 5L group A race engines (650bhp @ 7500) maintaining
lambda 0.9 over the entire rev range using TPS/RPM based maps. The only
time the map requires updating is when modifications are performed, not
even on full re-builds which are done every one or two meetings.

There it is for what it's worth. Someone may be able to find out more
than I have. I hope so.

Thanks

Bill Boulton <william at premier.apana.org.au>