David M. Koenig, who had a 27 year career in process control and analysis for Corning, Inc. before he retired, wrote his relevant experience in this field into his new book “Practical Control Engineering: Guide for Engineers, Managers, and Practitioners” (2009), making it very different from other books on this topic.

In the end of chapter eight of this book, after introducing the basic concepts of stochastic process disturbance, he added a very insightful comment section “Comments on Stochastic Disturbances and Difficulties of Control” to this whole chapter:

“How well can a process be controlled when it is subject to white noise only? This is an interesting question because many statisticians will immediately throw up their hands and make some condescending comment suggesting that control engineers should keep their hands off processes subject to white noise because any attempts to control such a process only causes troubles.

Part of that answer, minus the condescension, is correct, at least in my opinion. Consider the case where you are the controller and you observe samples of the process output whose average has been satisfactorily close to set point and that suffers only from white noise disturbances. Should you make an adjustment to the control output upon observing a sample of the process output that is not on set point? If the average of the process output is indeed nearly at the set point then any deviation, if it is really white or unautocorrelated, will be completely independent of the previous value of the control output and it will have no impact on subsequent disturbances. Therefore, if you should react to such a deviation, you would be wasting your time because the next observation will contain another deviation that has nothing to do with the previous deviation on which you acted. You, in fact, may make things worse.

Figure 1. White noise with standard deviation 0.103

…Consider the Foreword Figure 1 above where the process output is subject to white noise (whose standard deviation is 0.103), the process is first-order with a time constant of 10.0 time units and a PI controller, conservatively tuned, is active. Note the activity of the control output as the noise on the process output feeds through the controller. At 50 time units, the set point is stepped and the controller satisfactorily drives the average value of the process output to the new set point. However, the standard deviation of the process output about the set point is 0.115. So, the controller has amplified the noise. Ha! The statistician is smirking. A feedback controller cannot decrease the standard deviation of the white noise riding on the process output. At best it can keep the average on set point. The catch is that in most industrial situations one needs the controller actively watching and controlling the process in case there are set-point changes and in case some non-white noise disturbance appears. To quote a famous control engineer, “life is not white noise.”…”

In chapter eleven of his book, David once again touched upon the topic of “Control of White Noise”:

“… Statisticians consistently claim that processes subject to white noise should not be controlled because the act of control amplifies the white noise riding on the process variable. The logic (which we have already touched on in earlier chapters, see the foreword of this report) goes something like this. Consider the case where you are the controller and you are responsible for making control adjustments based on a stream of samples coming at you at the rate of, say, one per minute. Assume that you know that a sample is deviating from the target solely because of white noise. Therefore, the deviation of the ith sample is completely unautocorrelated with the deviation of the i -1th sample and will be completely unautocorrelated with the i + 1 th sample. Consequently, it would be useless to make a control adjustment. If you did make an adjustment based on the ith sample's deviation, it would likely make subsequent deviations larger. On the other hand, if you knew the deviation of the ith sample was the result of a sudden offset that would persist if you did nothing, then you would likely make an adjustment.

I certainly agree with this logic but there are some realities on the industrial manufacturing floor where automatic feedback control of process variables subject to white noise is unfortunately necessary, especially when a load disturbance comes through the process or when there is a need to change the set point.”

转载本文请联系原作者获取授权，同时请注明本文来自房松科学网博客。
链接地址：https://m.sciencenet.cn/blog-286797-596490.html

上一篇：Some underrated quotes
下一篇：Selected English readings for recitation: Robert M. Pirsig