It sounds like you have your ULN2803 wired up incorrectly.
Here is the pinout from the datasheet:
And here is an individual darlington from the IC:
You can see there is no Vcc pin - this is because the outputs are open collector. This means you connect your load (relay coil) in between the output and Vcc. This is just the same as you would have done with your BJT setup.
You can use a diode to protect against inductive kickback as with the single transistor, but the ULN2803 has integrated diodes for this purpose which you can use instead. The diodes anodes are connected to each output, and the COM pin is the common cathode connection for these diodes (so you can connect this to the Vcc to put the diode across the relay coil)
Effectively each output should look something like this when setup correctly:
The input can be driven from the micro output directly (the micro would be where the DS89C4x0 is in the diagram) or use an open drain with pullup resistor as in the circuit above. Notice how the COM pin is connected to +5V to put the internal diode across the relay coil.
EDIT - I notice you have edited your question to change the +5V from pin 18 to pin 10 - I assume this was a typo and it was like this to start with.
In this case, and judging from the picture it does appear that things are wired correctly as Russell mentions.
It's hard to know what might be causing your issue without more data - what do you mean when you say it runs for 5 seconds? What is it doing during this time? How often are the relays switching? What are they switching? Does whatever is being switched share power lines with the micro?
If you have a scope, then posting a capture of the ULN2803 outputs and the +5V line would probably help.