Saturday, December 30, 2017

Rewiring a Hoverboard wheel



Hoverboard wheel are 350 Watt brushless hub motors.  That's the same type of motor used in electric bikes and scooters.  I figured what the heck, let's see if this thing will chooch.  I bought a couple of cheap brushless motor drivers (QS-909 JYQD_V7.3) from amazon and ebay.  Sure enough the motor moves, if you get the connections right.  








There's a couple of things to watch out for.  
First off, there is no heat sink on the mosfets.  BE FREEKIN CAREFUL there!  the body of the Low side mosfets are grounded, and the body of the High side mosfets are connected to 36V DC.  So if you put a heat sink on both, you'll short the power source, POOF goes the fuse if you're lucky!  Spark goes the battery if you aren't!  




Problem 2:  there isn't any documentation for the motor controller IC: 
JY01
B636G001DG3





Best possible circuit trace I could make:




Traces on the front of motor controller.



B


Back of board. NOTE: the image is mirrored, as if looking through the front of the board.  

Both
ALL traces drawn on the front of the board.  

Mystery pin:  The second pin on the top right of the IC is tied to ground with a 10k resistor.  
I have no idea what that pin does.


Leave me a comment if you find this helpful or can help solve the "Mystery of the mystery pin"






some other images:



Monday, December 18, 2017

Real world Google Cloud Vision API results

Google Cloud Vision API is pretty powerful.  But it needs more classifiers before I can use it in my work.  
The customer is a scrapyard that wants to get the most amount from every recycled part.  The cashier needs to be able to identify every part.  Not many people, if any, can do that.  So maybe Google Cloud Vision can identify the auto part for them.  Maybe it cam filter BMW parts from Ford part. 
I was amazed that the Vision API recognized that it was looking at car parts.  However, the results as of 20171107 (November 11, 2017) were not good enough to solve this problem.
I think the image set has a lot of cars in it, but not very many disassembled cars.  


The scrapyard would be happy to collect a library of images to teach a new set of classifiers.  Unfortunately, I don't know how to use TensorFlow to do this.  If you know how to teach classifiers in TensorFlow leave me a comment or send me an email.