The problem with the chip is how to interface it to an existing control. Very interesting though, thanks for sharing
The problem with the chip is how to interface it to an existing control. Very interesting though, thanks for sharing
well maniacs like Brian are not afraid to roll their own!
I am kind of interested in getting a lower power ram for my system, it presently uses about 6A when the pump motor is on, and the solenoid draws 1A when engaged. I bought a very husky drive on the theory that a electric motor/gear drive would burn out under load. There must be something intermediate though.
This is a great site that offers components that would make building your own doable: http://www.pololu.com/
They offer just about everything that a home brewer might need.. with the hydraulics being the exeption. It is very affordable stuff as well. Here is a video that shows a self balancing robot that uses cascaded PID control, very similar to what an autopilot would need: http://www.youtube.com/watch?feature...&v=DmoShkJclLI
Here is a youtube of my AP design in action on Elise, an Express 27 on a gusty run through the slot, using all 9 axis of data to maintain course: https://www.youtube.com/watch?v=EWObyejTb3A
No good shots in a heavy quartering seas, my videos of a SF to Kauai solo trip, under AP control, were lost when a gear bag leaked swimming back to the boat in Hanalei Bay. I have several beta units on a variety of boats, one out cruising, a few on the bay.
Brian, what kind of servo algorithm are you using? PID, or with some feedforward, or coupling of pitch to anticipated yaw?
For example, in a quartering sea, as a wave approaches the stern will rise first (pitch) and then heel a bit (roll), and then the boat will yaw, say consistently toward the side the wave is on. So a human will anticipate the yaw and counter steer as he feels the pitch change. I believe that this is what is going on with the better pilots like NKE. But most manufacturers don't talk about what they are doing or even how they are sensing. The Nexus compass has MEMS type accellerometers in it, but they don't say what kind, and it is not clear that all the data gets sent over the bus to their AP control head, which is a pretty old design.
Edit: I did a little poking around on the web, it looks like Kalman filters are the done thing for this kind of control problem. But how to build the matrices? Maybe that is what is going on inside my electronic compass, which can give a good heading output even when the boat is heeled over or pitching down a wave. I think the AP is pretty crude though, only a PI with a feedforward term for when you are way off.
If one had lots of time, you could log all with wind data, attitude data, BSP etc. on different points of sail and sea states with a good driver, and 'teach' the algorithm what to do that way. A neural net, I think.
Last edited by pogen; 08-22-2013 at 11:21 AM.
Hi Dave,
A few responses:
Compass: For mems sensors without a gimbled platform one has to maintain a computed artificial horizon in which to translate the magnetometer sensor 3 axis data. That is typically done with the accelerometers.
Filtering: Yes, I use Kalman filtering as do others. The accelerometers are particularly noisy and sensitive to high frequency vibration. For example motor vibration can create offsets in computed results. A fun thing to try, set your heading on a calm day under sail, then turn on the engine and see if an error is introduced. I have done this with a couple of commercial units and both had apparent errors introduced with engine noise, in the range of 5 degrees. I have seen both offsets and course wandering increase in noisy environments.
Sea State: Yes, I use all the sensors to compensate for boat attitude and sea state. I could discuss the guts of the algorithms over a beer somewhere. It kind of is the secret sauce of the design.
Data Capture: Yes, I capture all the sensor data and NMEA, when attached, on a PC. I have hours of recorded data. This has been used extensively to perfect results on the water. I attached an example plot that shows an RM X5 recovering from a step change in course (20 degrees). The motor pulse data can be ignored.
HI Dave,
I am looking for an optical wave estimator that can look out 100 yards and estimate sea state. Got any good optical concepts ?