Quote:
Originally Posted by RobertSmalls
How many channels is your datalogger, and are there any vacant channels? I'd like to be able to record the current in and out of my hybrid battery, and maybe some other things like engine RPM (from which to infer what gear I'm in) and air intake temperature.
Does your model distinguish between braking, deceleration fuel cutoff, rough roads, and ordinary rolling resistance?
|
1) I have to check to be absolutely sure, but I believe I left at least 4 analog inputs free. I made as many through-hole pads as possible on the remaining unused pins, and the ADC is not at all used, so you can definitely add what you'd like.
2) There's (almost) no need to for a second rpm sensor. You can measure rpm from the fuel injection signal, although the "almost" case is that when you're in engine braking mode you no longer have a direct measurement of engine speed.
However, by knowing the gear ratios, you can always observe the engine speed based on car speed, except in cases where you've shifted while engine braking. Even then, just as soon as you reapply the throttle, the observer will catch up and switch to the appropriate gear.
3) My model just estimates thrust force, without any attempt to know where the force is coming from. My assumption is that any positive thrust force on a moving car is from the engine pushing. (Note that this is not the case for a car at rest on a slope. You can have a positive thrust force due to the brakes.) I only calculate efficiency at these times.
Unfortunately, we cannot directly measure the difference between rough roads and smooth ones. This is within the capability of the observer, but in order to have that level of precision you would have to have much more accurate sensors. Specifically, you'd need a danged good elevation profile for your road. Also, a throttle position sensor would help.
The problem comes from the fact that the observer tries to filter noise based on the model it expects. If you need a strong filter, which is the case when your measurements are noisy, then you have trouble reacting to these kind of perturbations. If, however, your model is quite good and your sensors are quite precise, you can extend the model along those lines. So it would be mathematically trivial to extend the model to 18 dimensions and include an estimation of C_rr (rolling resistance coefficient) that can adapt very rapidly, but as a consequence it would be harder to tune it so that you had reliable readings.
I'm not sure if I'm explaining this well, but it's sort of a give and take. A few years ago, my 17-dimension model couldn't have worked due to inaccuracy in sensors (specifically the accelerometer and timers). Now that we have better sensors, we can go further into detail in the model.
It would be interesting to see (and is indeed something I hope to do with a new model we're developing for an ASME conference) how well these parameters can be estimated. In theory, you have no need for a roll-down test, as this is already included in the model, but in reality the observer might not be stable enough to both rapidly converge to the real drag values (which is what you're asking for when you want to be able to detect changes in road quality), and yet still converge to useful force estimations.