Wire harness design questions

[markmakeitso]

I'm guessing not many people here are involved with small batch design and construction of harnesses.

On the last J1939 250 kbps automotive harness design that went out the manufacturer substituted some Deutsch junction connectors which effectively turned the bus into a star. Arm lengths were also pretty bad: 13, 19, 37, 108, and 120", and 60" to a connector that joins another harness under our customer's design control. Presumably that one will have several nodes and a terminating resistor. Adding to the difficulty is that the 19" arm is to an ECU with built in 60 ohm resistor, and the 13" arm also has a node with built in resistor, which was developed with the understanding that the resistor was optional, which hasn't been the case just yet.

I complained a bit, and came up with a compromise design that does some doubling back inside the harness to maintain a decent bus topology. Now it has the terminated ECU at one bus end, 30" stub, 4" branch to the other terminated node, 60" branch, 101" branch and then connector to customer's harness. Lengths between the branches range from 12" to 30", plus at least 50" for the customer connection. Not super thrilled about the remaining branch lengths, but it's better. With some more cajoling we could maybe double back and cut that 101" down. The 60" branch is not going to get shorter, as that would require 2 trips through a cannon connector. All of this is run through shielded twisted pair wiring, bound up with the general wiring (24 and 12 VDC primarily).

Sorry for the long setup. Functional questions:

-barring rigorous testing, how detrimental are the long stub and branches, and the superfluous terminated node?

-any suggestions on making junctions for each branch? In and out of the shielded twisted wire each time is quite a roadblock for small scale manufacturing, especially to keep single end connection for the shield intact.

-is it worth looking into other robustness measures, such as split termination? That is assuming we ever have that opportunity.

Deutsch would love for us to go full tilt on their 3 pin setup, but that's as unlikely as a standalone CAN bus harness. I went through the TI CAN bus document here: http://www.ti.com/lit/an/slla270/slla270.pdf and have a slightly better understanding of the technical reasons for following the specifications. Nothing I can really point to as signal of doom however.

Thanks,
Mark

[Hooovahh]

The routing of a CAN a bus and it's wires can be a difficult thing.  On the surface it is easy, just follow the approved design spec for CAN which as a summary is the following:

• Create a bus which is linear where nodes attach to the bus, and no node is farther than 15cm away from the bus.
• Attatch a 120ohm resistor between CAN high and low on the first and the last node of the bus.

Does a CAN bus suddently fall appart if these recommendations aren't followed?  Not always.  These requirements are to help shield against noise and reflection on the lines.  If you are in an environment with no noise, then maybe you don't need to follow these rules so closly.

We make testers were we can test up to 12 devices at once.  Some times we are forced to put all of these devices on one CAN bus, leading to high bus load over 90% at 500K baud.  Each UUT gets its own 10 foot harness.  So this means we may have 12 nodes, each with a 10 foot harness, going to a single point where the CAN port is.  In this design we only have one terminating resistor and it is at the CAN port of the tester.  So we broke both of the CAN spec rules.  But in our test environment, which deal with 480VAC power supplies, shaker tables, and environmental conditions with humidity and temperature, we didn't have a problem.  But I was very nervious about the design choices made by others, until we had lots of data showing the robustness.

For us there wasn't much alternative.  Either have 12 different CAN ports, which increase cost and system complexity, or we could have extra wires so the CAN bus would go out and come back from each UUT which would mean more cost in harnesses, and it would mean all harnesses need to be connected, even when a UUT isn't to complete the bus.

[markmakeitso]

Hi Hooovahh, thanks for the response.

Your situation sounds a bit similar to mine, with less than ideal environmental conditions. Over here we're also weighing a potentially hefty price increase per manufactured harness, and the benefits of a compliant CAN bus are fairly nebulous to many people involved. That'll likely be the case, right up until the engine doesn't run correctly. Then we'll be troubleshooting in the field, indubitably 2 states away from home. I like travelling, but usually to get away from work.

There's a chance the 2nd terminated node will either lose the resistor or be eliminated completely, which would be nice. Losing the star arrangement also helps me sleep at night. Branch lengths are a continuing problem, but that might be getting better in the future. For now I'll push for bus design consideration to be larger part of harness design, and incorporate some error frame checking on installed units as part of our quality checks. Would be nice if we weren't knowingly violating every single design guideline though.

If you have some functional suggestions about the physical branching connections, or know of some alternate methods of checking bus characteristics I'd be interested. I recently started a VI that does some analysis of signal levels using a USB-204 from Measurement Computing: http://www.mccdaq.com/usb-data-acquisition/USB-204.aspx

Nearly fast enough for what I had in mind, although differential measurements would be nice. Right now it just checks for samples in the proper dominant and recessive voltage windows, which is a small start.

Mark