Hello,
Perhaps, someone on the list could help me determine which path to take in a project I am working on. Here is my problem: My company manufactures a device that produces two sinewaves in quadrature that are used to indicate position. The device is called an encoder and is used in machines of all sorts. The benefits of this encoder include that fact that it is very difficult to break compared to most other kinds of encoders because it does not use glass or other materials that are fragile or can be affected by the oil or solvents that might be used in an industrial environment. The reason is we use magnetics. By testing the phase of the two sinusoids you can decode 1 of 4 possible positions increasing the resolution of each electrical cycle by 4. The encoder is rotary and produces 16 (or more) electrical cycles per mechanical cycle. This gives us 16*4 positions per mechanical cycle. Now, we want to produce an interpolator that will increase the resolution of the encoder a thousand times. We can do this by using the sinusoids and deriving the atan2 between then. Now comes the problem. Because of the magnetics and the sensors that read them, the sinusoids are not perfectly in quadrature. In fact, each of the 16 electrical cycles per mechanical cycle varies slightly in phase and velocity: up to +/-3%. Proper filtering, gain and offset correction and RBF interpolation can do a pretty good job of cleaning up most of my problems but there is still a little bit of wiggle left. I have found by taking the electrical cycles apart and training them individually that I can make any one of the electrical cycles perfect, and this would be a great solution if, in operation, I knew where I was in the series of electrical cycles that make up each mechanical cycle. It seems to me that I could use a classifier to tell me which cycle I am in. Each product we make currently is set up and mated for life with a single interpolator. If I can train the individual interpolator how to discern between each of the 16 electrical cycles, I could produce a very high quality device. I have read many many articles on classification and tried their algorithms out but none seems to produce the result I need, or, at least, well enough. I am wondering if anyone here has the experience to perhaps suggest at path that could help me solve this problem. Thank you in advance. And I apologize if this is the wrong list for this question. Don Morgan ------------------------------------------------------------------------------ October Webinars: Code for Performance Free Intel webinars can help you accelerate application performance. Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from the latest Intel processors and coprocessors. See abstracts and register > http://pubads.g.doubleclick.net/gampad/clk?id=60133471&iu=/4140/ostg.clktrk _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
