Trouble with Feedback Only Control

Feed-forward provides better outcome

The Rio Olympics is coming up, and Olympic boxing provides a great, real-world example of the differences between feed-forward control and feed-back control. Consider the pictures of these boxers. On the left, the boxer used feed-forward. He stepped back from the punch and put his arms up to protect his head. Obviously a smart boxer.

The boxer in the blue on the right is getting some feedback – very NEGATIVE – feedback. So why do we rely on feedback control in our control systems and control strategy? Suppose that a product tank is being filled. You’d like it full and avoid a spill. Here is a comparison of what typical systems tell you.

1. On the far left, the "Alarm Light" column, you don’t know if the tank if full, empty or what. You only know there is a problem when the tank is overflowing.  Some alarm systems will give you an alarm value when you are close to overflowing – say 90% full. This can be good as the operator could react to a problem. This can also be bad. Suppose that the operator turns off the fill valve at 95% full. The tank isn’t filling, but there is still a warning alarm. Now, the operators have a full tanks and a lot of warning alarms that should be ignored. An alarm indicator isn’t very useful for trouble-shooting or after-the-fact analysis.
2. In the "Percent Full" column, are given a bit more information - the value of the tank percent full. This is an indication of the current level, but you don’t know where you’ve been, don’t know the rate of increase or what has happened before, etc. Is it trending up or trending down? With no historical view, no post issue analysis can be done.
3. A tank percent full with "Historical Trend Package" gives better information. In the graph, it looks like the trend is going up. There are still unanswered questions - is it going to stay at 90%, go down, or go up? It is much better than an alarm indicator light, and better than a static %-full value, but still this is not really telling you where you’re going. The operator may shut off the fill valve, by looking at the trend thinking it is going to overflow.
4. A trend and forecast tells you where you’ve been, and a forecast of where you’re going. Forecasts can come from a linear regression of the entire trend line, a linear regression of the last few points, or for a best-in-class forecast, a multi-variable model (e.g. fill flow rate and the empty flow rate).

Smart boxers look for multiple signals from their opponents and move out of the way of punches before they are landed. Smart control systems engineers know that multi-variable feed-forward signal can keep your process from getting punched.