OT: How O2 sensors work, narrowband and wideband Options

[lapuwali]

Groot asked how these sensors differed in another thread, and I was in too much of a hurry to answer him then. Here's a more complete answer.

Both sensors rely on what's called a Nernst cell, invented in the late 19th century. It was found that certain ceramic insulators (zirconium oxide being popular), when heated to 675dF, become electrical conductors of a very useful type. When one side of the cell is exposed to gas A and the other side of the cell is exposed to gas B, oxygen ions will flow across the cell in proportion to the difference in oxygen concentration between sides of the cell. Since ion flow is electric current flow, a small voltage is developed across the cell, again in proportion to the difference in O2 concentration.

Bosch used this to make the first narrowband sensors in the 1970s, using air as the reference gas. There's a fixed curve that defines O2 levels in the exhaust gases with AFR, so this device gave a simple voltage to AFR meter. However, one serious downside is that the cell temperature has a very strong effect on how the cell responds to O2. The first sensors simply used the ambient temp of the exhaust gases to heat up the sensor. EGT varies quite a lot as the engine runs, based on load as well as AFR, so these first unheated sensors weren't all that accurate, and they also took awhile to heat up, so the first few seconds (50-60 seconds) of engine operation was unregulated by the sensor. By the 1980s, powered heaters were used to decrease this warmup time to under 15 seconds. However, the heaters themselves were unregulated, so accuracy wasn't improved, just cold-start response.

In the 1990s, research done over the past several decades on lean-burn engines approached production. However, the O2 sensors then in use couldn't measure these lean mixtures with any accuracy. Two things were done to fix this. One, a regulated heater was used to keep the cell temp in the very narrow range. This substantially improved accuracy, but the cell response curve made measuring mixtures leaner than 15:1 (or richer than 14:1) very difficult. The voltage difference between 15:1 and, say, 17:1 was very, very small, and required an extremely accurate (and therefore expensive) amplifier stage to measure the difference.

Another quirk of the Nernst cell was that if you APPLIED a voltage to the cell, oxygen ions were pumped across it, reversing the normal process. By using two cells, one generating voltage and the other pumping ions, a feedback circuit could be set up to keep the O2 concentration at exactly 0.45mV between the cells. If more O2 appeared on the exhaust side, more ions flowed, so a higher voltage was generated. This was detected, and a higher voltage applied to the pump cell to maintain the balance. This correction voltage was much more linear across a much wider range of O2 concentrations than the single cell approach, which widened the voltage differences between "interesting" AFRs, and thus removed the need for the expensive amplifiers. With current sensors, AFRs of roughly 22:1 to 11:1 can be accurately measured. By varying the pump to measuring voltage feedback curves, you could also calibrate a generic sensor to operate with any fuel (natural gas, ethanol, methanol, which all have different "ideal" AFRs), which fit well with the movement to alternative fuels.

The cost of this new techonology was that the sensor system became a lot more complicated. Whereas before, with the narrowband unit, you simply had a voltage that corresponded with an O2 measurement (so all you needed was the sensor and a cheap voltmeter), you now had to run a tightly regulated heater, and you had to run the pump voltage in a feedback loop with the measuring voltage, then take the difference to get the AFR. There are proven analog hardware circuits to run this feedback loop, but recalibration requires hardware changes. Doing the feedback loop in software allows on-the-fly recalibration (if you switch sensors), and allows for "flex-fuel" operation.