I made a few adjustments to the board layout and also to the pcb-gcode setting and milled another circuit board. This one turned out nicely, even with a couple hiccups. First, the board wasn’t taped to the fixture exactly level, so a corner with nothing important didn’t mill through the copper (it was only the board outline). Then a phantom E-stop signal stopped the machine before it was done and had to restart from the beginning (I upped the debounce value to correct this). I thought it was another beverage coaster while watching it mill, but let it finish and it turned out fine. I ran the drill routine with a .035″ bit and everything but the screw terminal block and power jack fit. I was able to make the terminal block fit by sanding down the edges of the pins, but the power jack I’ll just use some wires and plug later.
Since the majority of components fit, I soldered this one up to give me an idea of what the next round of changes need to be. The main one is to increase the size of the pads to make it easier to solder and to increase the isolation path around them to prevent bridges to the ground plane. I’ll also have to change the through holes to the right size in the Eagle component library I made. At least the spacings were all right so everything still fit. Another change will be to the text ratio so the thin areas of copper don’t peel off when milled. They probably would have been fine if the second pass wasn’t done, but whatever.
Anyways, the CAN Main board only contains the basic circuit to run a 18F4580 micrcocontroller with the MCP2551 CAN transceiver; power from a 7805, 20MHz crystal oscillator, ICD plug, reset button, CAN status LED and then the rest of the pins are brought out to female headers to plug in a shield. I also included an area for the CAN bus termination resistor and three capacitors with jumpers, and a place to connect the cable shielding to the board’s ground. There is a jumper to bypass the 7805 regulator to use a 5V wall wart directly which are more common these days.
I finally was able to mill a circuit board using my CNC machine. No more acid etching, but I never had much luck with it anyways. After milling the table and fixture, I was held up by the lack of 1/8″ collet, which Bosch doesn’t make. I bought a nice collet set from Think & Tinker/PreciseBits for a modest price and it seems it was worth it. I also needed to buy a license for Mach3 to go beyond the 500 line code limit, since the code to etch the board was near 20,000 lines. To produce the G-code from an EagleCAD project, a nice script is available for free, http://www.pcbgcode.org/. It installs like any other ULP script for EagleCAD and comes with docs to get you going, plus forums and plenty of other websites with tutorials.
It was neat watching the CNC machine cut the traces and no broken bits or other catastrophes. Unfortunately, I didn’t take the time to properly zero the end of the 45 degree v-bit and cut too deep. This washed out the traces and pads, so it’s scrap. Before I realized this though, I tried fiddling with the settings for the pcb-gcode script and kept milling on the sheet of copper clad board and got a lot more bad traces.
The correct way to zero the end of the tool was to use a piece of paper between the board and the tool and to jog down at a very low speed (about 3% of max). While the tool is going down, move the paper back and forth till the tip catches it. Now zero the Z-axis in the Offsets tab of Mach3 in one of special G-code offset codes of your choice. Then measure the thickness of the paper with calipers (my piece was .0035″) and subtract it from the Z offset (press the “Save Work Offset” button to bring up a dialog box that allows you to edit the coordinates). This puts the Z-axis 0 coordinate on top of the circuit board without any pressure. The X & Y coordinates are set to where you want to start milling the board, and can be done by eyeballing it, just don’t re-zero the Z-axis. Now the machine is set to mill.
My settings for pcb-gcode script (not all are included, just what I changed):
Default = .0001″
Maximum = .0001″
Step Size = .005″
Etching Tool Size = .007″
Z High = .5″
Z Up = .1″
Z Down = .002″ (someone posted an excel file in the pcb-gcode forums to calc this, I set mine .001″ higher and it worked)
Drill Depth = .065″ (go a little deeper than the board thickness)
X = 0
Y = 0
Z = .5″ (don’t leave as 0 or the machine will drag the end of the drill bit across the board between tool changes)
XY = 10″/min
Z = 5″/min
Change any other options you need, then hit “Accept and make my board”. All the files generated will be placed in the project folder with the board. Check out the readme file in the docs folder for explanations, it was helpful. I’ll probably try setting the Maximum Isolation setting to something higher, so the slivers between close traces are removed. I’m sure I’ll update with more progress, or at least to show of stuff I made.
I’ll bet the term troubleshooting came from someone who had trouble, then started shooting. At least that’s what I would do if I still lived in an area where I could shoot something (damn neighbors). I’ve been trying for weeks to get the K-Node circuit to work with any of my other CAN capable boards, with no avail. First, I thought the bus had too much noise, and used the oscilloscope to find a extra 60 Hz wave causing mayhem. This was removed without success and then thought a shitty clock signal was to blame. It only slightly was inhibiting communication (it only caused a few errors, but not enough to totally block the message).
Anyways, turns out the CAN transceiver was screwed up beyond use. The MCP2551 chip is very sensitive to whatever, and fries when either of the bus wires are disturbed. I should have known, but didn’t. The previous post outlines my attempt at clearing up any extra noise on the bus. Then thought it could have been an inaccurate clock to the microcontroller. I bought some nice 20Mhz oscillators and replaced the crystal/caps on both the SpartaNode and K-Node boards, without success.
So before I threw both circuits, the scope, and all the other equipment I have out the window, I replaced the MCP2551 IC. Of course it worked right away. Guh, I suppose the lesson here is to make sure the CAN bus wires are secured very, very well and to absotively, not-arino hot swap the wires. I’ll pretend it wasn’t all for waste since the random, mysterious communication errors don’t show up during testing. Now that I can send a message between the K-Node and SpartaNode, I can run the CAN bus through my house and really test the system.
I made some long overdue progress on my Smart House project that uses CAN bus. Up till now, the physical bus were two scrap wires I twisted together and soldered on a couple pins for plugging into breadboards. This wasn’t going to be an option once it was time to run the actual wires through my house, since it will be more than a few feet. A search for a single twisted pair, around 18 AWG and unsheathed yielded nothing, unfortunately. I settled on some security cable, which is two conductor 22 AWG, shielded and sheathed, as well as readily available for cheap at any hardware or big box home store.
For a test bus, I randomly cut a piece 12 feet long and soldered on 120 ohm resistors to each end, then connected to the K-node and SpartaNode. Of course it didn’t work, and probing with an oscilloscope showed a nice 60 Hz wave causing interference. The wave was present with or without power on and also near or far from other wires. The interference wave’s amplitude varied from about 1 volt peak to peak, to over 3V when moved around and was superimposed on any CAN messages on the bus. I suppose Murphy’s electrical law was in effect; the one about a wire being an antenna when you don’t want it to be.
The good news is the book, CAN System Engineering, held the solution. By using three capacitors per end, all the 60 Hz wave was removed and a crisp CAN message was able to travel just like in the test setup. The values of the capacitors are the same for both ends and depend on the length of the bus. For mine, I needed a 68pF cap between the CAN High and Low wires, and two 220pF caps from each to ground. I soldered these to the board (but just realized I shouldn’t have, d’oh) and added a place to connect the cable shield. The two nodes are still showing a communication error, so the same CAN message is being repeated for a response, but the at least you can see it’s nice and clean looking. I suspect it is a software issue.
Dammit, I tried writing the best darn paper I could in hopes of scoring a free trip to Vegas on university money, but costs a lot of money to have you paper submitted to the conference being held there. Oh well, I guess I ended up learning something, if that’s worth anything.
So the class project I did was to make a 4-bit full adder using threshold gates and written in Verilog HDL so it can be loaded onto an FPGA. A threshold gate is sort of a model of a neuron cell from the brain. Basically, the binary inputs (1 or 0) are multiplied by individual weights (positive or negative integers) and summed. Then the gate output is “1” if the sum is greater than a threshold value, or “0” is less. The benefit of this is the ability to reduce a complex Boolean function to one gate (certain restrictions apply). To find the weights and threshold value Dertouzos’ method is used, which surprisingly was not googleable.
Even if a function can’t be reduced to a single gate, multiple threshold gates can be connected together to form a neural network or cascaded to still be realized. The sum function for a full adder can’t be realized as one, but the carry out can be. Luckily another method, called IMS (Implied Minterm Structure), does the trick and groups the minterms so Dertouzos’ method works. Using IMS, three threshold gates are needed to cover the sum minterms and then connected together with a forth.
Now that the number crunching is done, its time to code this into Verilog. Supposedly real numbers can be used, but the old version in the lab couldn’t, so I had to settle for integers. To get the code to compile, all of the threshold values needed to be rounded up and then the threshold inequality needs to be changed to greater or equal. Since each threshold gate has a different combo of weights and threshold, each one was written separately, then combined as a full adder module and finally into a 4-bit full adder. It even worked when loaded into an FPGA (Spartan variety). Below is the code so you can claim it as original for you’re next class project or perhaps for some nerdy shits and giggles.
module threshold1(f, A, B, C); output f; input A,B,C; reg f; parameter integer w1 = 1, w2 = 1, w3 = 1, t = 1; integer out; always@(A or B or C) begin out = A*w1 + B*w2 + C*w3; if (out >= t) f = 1'b1; else f = 1'b0; end endmodule module threshold2(f, A, B, C); output f; input A,B,C; reg f; parameter integer w1 = 1, w2 = 1, w3 = 1, t = 2; integer out; always@(A or B or C) begin out = A*w1 + B*w2 + C*w3; if (out >= t) f = 1'b1; else f = 1'b0; end endmodule module threshold3(f, A, B, C); output f; input A,B,C; reg f; parameter integer w1 = 1, w2 = 1, w3 = 1, t = 3; integer out; always@(A or B or C) begin out = A*w1 + B*w2 + C*w3; if (out >= t) f = 1'b1; else f = 1'b0; end endmodule module threshold4(f, A, B, C); output f; input A,B,C; reg f; parameter integer w1 = 1, w2 = -1, w3 = 1, t = 1; integer out; always@(A or B or C) begin out = A*w1 + B*w2 + C*w3; if (out >= t) f = 1'b1; else f = 1'b0; end endmodule module threshold5(f, A, B, C); output f; input A,B,C; reg f; parameter integer w1 = 1, w2 = 1, w3 = 1, t = 2; integer out; always@(A or B or C) begin out = A*w1 + B*w2 + C*w3; if (out >= t) f = 1'b1; else f = 1'b0; end endmodule module fulladder (sum, c_out, A, B, c_in); output sum, c_out; input A, B, c_in; wire x, y, z; threshold1 n1 (x, A, B, c_in); threshold2 n2 (y, A, B, c_in); threshold3 n3 (z, A, B, c_in); threshold4 n4 (sum, x, y, z); threshold5 n5 (c_out, A, B, c_in); endmodule module full_4bit_adder (sum, c_out, A, B, c_in); output [3:0] sum; output c_out; input [3:0] A; input [3:0] B; input c_in; wire x, y, z; fulladder a1 (sum, x, A, B, c_in); fulladder a2 (sum, y, A, B, x); fulladder a3 (sum, z, A, B, y); fulladder a4 (sum, c_out, A, B, z); endmodule
I hope my neighbors don’t mind me milling things late at night. The vacuum is the loudest over the router and stepper motors, so they’ll probably just think I like my house really clean. To kick off my CNC adventures, I’m making a fixture to hold PCBs for isolation routing.
It’s nothing fancy, it mainly helped me understand the CAM portion making stuff. I ended up not using the Sketchup model because of added learning involved to use a script or something, but it can be done. Instead I used CAD to draw the square with the rounded out corner reliefs, then used the CAM software CamBam to setup the stock size and generate G-code. My worst problem was trying to make the coordinate systems jive between the CamBam and my machine. In standard drawing coordinates, the +x goes to the right, while my machine’s +x moves left. This was corrected with an option to set a separate machining origin, and also using offsets in the machines software.
I made the pocket for the PCB slightly larger than the width and length of the raw stock and to a depth of .060″ to match the thickness. The corners were over cut to allow the PCB’s square corners to fit. Next I’ll be drawing up some various clamps to go over the edges to press down. By using some double sided tape underneath, the PCB should be a little higher to give the clamps something to press into. I’m also going to add two locator pins so the fixture can be taken off and put back on in the same place. Built into G-code are places to store fixture locations and makes having multiple fixtures even easier. I milled the main pocket and corners separate and both programs aligned perfectly, Yay!
Yay, three times is the charm. My CNC machine finally milled it’s table flat without motors stalling or random nuts coming loose. Since the second attempt, I tweaked the bearing adjustments on the Y-axis and upped the current to all the motors for more torque. The motors got slightly hotter, but not so much that I couldn’t touch them. I also ran the table cutting program a few times without a tool to work out any other kinks.
Cutting the table seemed to take out the inaccuracies from building and you can see in the corners where the high sides were cut shallower than the low side. When I glued the table together I didn’t notice it was twisted slightly on one end. Luckily it could be compensated for in the end.
The run wasn’t without its problems though. Being so pleased with the dust extractor I built, I got a new filter for the shop vac for maximum performance. It was too much for the all-purpose bucket, and the vacuum imploded it slightly and let all the wood dust through to the vacuum. Once the vacuum filter clogged it quit working and the dust went everywhere. Oh well, I bought a new bucket and some extra filtering material and it seems to be working again.
So now I can start making stuff on it, but since I’ve been putting my energy into building it, I’m not sure how to do that. I’ll probably start by making a fixture and hold-down clamps for circuit boards and plastic sheet. I’ve already read some webpages that I’ll try and emulate, so I just need to put it all together.