In the past, mass production of mathematical models and column design correlations was hindered by the extensive calculations involved. With the event of high-speed and personal computers, this bottleneck has been eliminated. Flood gates have opened, and new mathematical models are pouring into the published literature at a record pace. Further growth in mathematical model production appears to be restricted only by the availability of persons willing to punch buttons on computer keyboards.
One would expect this state of art to be the heaven that column designers always dreamt of. Instead, it turned out to be the hell they always feared. Few could keep up with the large influx of mass produced mathematical models. Little is known about the limitations of each new correlation or design method. Our prediction methods turned into black boxes: key in numbers, print out results. But how reliable are these results?
About fifteen years ago I was applying a leading and very well known literature correlation as part of my graduate thesis. I punched in keys and the computer printed out results. For some reason the results did not look right. Upon investigation, I discovered the unbelievable: the correlation just did not work for my case. Not that the correlation was bad; over many years, it gained a very healthy reputation. It just happened that it had limitations, just like every other correlation. The limitations of this correlation were fairly well explored, but a quarter of a century after it was derived, I found the hard way that it had one more limitation which remained hidden over all these years,
Fortunately, I was only doing a graduate thesis and not basing a column design on this prediction. Fortunately, the correlation was simple enough to permit a person to identify the limitation. And fortunately, some gut feeling saved me from falling into the trap.
As we head into the twenty-first century, the above type of anecdote is becoming ancient history. The black box in the computer has taken over. Looking for correlation limitations today becomes like looking for a needle in an ever-growing haystack.
With the busy life style and the pressure to publish papers, the problem is becoming more acute. There are deadlines to meet, technical papers need to be produced, and there is no time to explore correlation limitations. Besides, who needs to look for limitations when a computerized regression analysis (performed, of course, by one of the best regression packages in the business) shows an excellent data fit? Does it really matter if a handful of points do not fit the correlation— even if this handful includes all the points for systems above atmospheric pressures? In real life, no one will know, unless the designer ends up with a column that does not work. And if the error is on the conservative side, no one will ever find out, because the column will work.
Data collection is another neglected child of the late twentieth century. There are so many data around that collecting them all (or even most of them) for the sake of deriving a correlation becomes painful, mundane, and an extremely unattractive exercise. Not to mention the labor involved in reading data off plots and in the arithmetic involved in ensuring that all the data points have been correctly entered. I challenge anyone to cite a more boring task than this. An economical way of dealing with the excess data problem is by using the "ignore it and hope it goes away" principle. It will suffice that the new correlation will fit a handful of data thrown at it. And if data from other sources do not agree, that just means there is something wrong with the other data.
What hope has the designer who sits at the other end of the rainbow and attempts to make use of the mathematical models and design correlations?
The purpose of this book is to bridge the gap between developers of design procedures and those who ultimately use them. Correlations and design methods are recommended only when their data base is wide, and their range of application and limitations are well understood. Rules of thumb are recommended over theoretical models provided they have been shown to be more reliable in predicting commercial column performance. Sound theoretical models that give unreliable or sparsely tested predictions were left outside the covers of this book.
Contrary to a popular belief, some distillation characteristics still cannot be satisfactorily predicted by correlation, regardless of the number of correlations available for their prediction. Data interpolation with the aid of an empirical procedure is probably the most reliable means of estimating these characteristics. The last two chapters of this book provide the designer with the data needed.
Throughout, the book emphasizes designing columns that "look right, ' and techniques that can help distinguish a design that makes sense from one that does not. Computers have provided distillation designers with speed, accuracy, and flexibility. Computers, however, still have a long way to go before—if ever—they are capable of replacing good engineering judgment.
Henry Z. Kister
Was this article helpful?