Caution
Ben,
With proportional-only control you shouldn't notice any difference in operation when you change to dependent gains. The differences come into effect when you have I and D terms.
As Ron says, it is absolutely critical to have the instruction execution rate match the update rate entered in the PID instruction. The instruction assumes that the specified period (1 sec in your case), no more and no less, has elapsed between executions. It uses that assumption to calculate the magnitude of the I and D terms.
With a proportional-only setup, you will always have an error except when the bias value is exactly right for the load conditions. The amount of error is inversely proportional to the P gain.
When introducing integral action, I would recommend starting with a fairly slow repeat rate (1 or 2 minutes?), leave the bias in, and observe what happens. The 10-20 deg error should slowly reduce. Increasing the repeat rate (smaller numbers) will reduce it faster. Reduce the bias in steps to observe the effectiveness of your tuning. Note that the P (or controller) gain influences the integral term and you may end up reducing it.
Instead of using integral action, you could approach the system as Peter has suggested and calculate the bias term rather than leaving it constant. (I'm not sure what he was disagreeing with in his post since the only difference I can see is he guesses the bias using MathCad and a look-up table, whereas I plucked 50% out of the air). Whether you calculate the bias from setpoint or loading or both is your call. When used this way, the bias term is more correctly referred to as feed-forward.
Ben,
With proportional-only control you shouldn't notice any difference in operation when you change to dependent gains. The differences come into effect when you have I and D terms.
As Ron says, it is absolutely critical to have the instruction execution rate match the update rate entered in the PID instruction. The instruction assumes that the specified period (1 sec in your case), no more and no less, has elapsed between executions. It uses that assumption to calculate the magnitude of the I and D terms.
With a proportional-only setup, you will always have an error except when the bias value is exactly right for the load conditions. The amount of error is inversely proportional to the P gain.
When introducing integral action, I would recommend starting with a fairly slow repeat rate (1 or 2 minutes?), leave the bias in, and observe what happens. The 10-20 deg error should slowly reduce. Increasing the repeat rate (smaller numbers) will reduce it faster. Reduce the bias in steps to observe the effectiveness of your tuning. Note that the P (or controller) gain influences the integral term and you may end up reducing it.
Instead of using integral action, you could approach the system as Peter has suggested and calculate the bias term rather than leaving it constant. (I'm not sure what he was disagreeing with in his post since the only difference I can see is he guesses the bias using MathCad and a look-up table, whereas I plucked 50% out of the air). Whether you calculate the bias from setpoint or loading or both is your call. When used this way, the bias term is more correctly referred to as feed-forward.