Pulp and Paper Canada


June 1, 2000
By Pulp & Paper Canada

As many control engineers and operational supervisors have discovered, taking an elementary statistical approach in using operating data to control the quality of pulp and paper products, at various s…

As many control engineers and operational supervisors have discovered, taking an elementary statistical approach in using operating data to control the quality of pulp and paper products, at various stages in the manufacturing process, is far less effective than textbooks suggest.

One reason is that variables which are often considered to be independent, and which are normally measured, are actually interrelated in poorly understood ways. Today, most mills have stored reams of historical operating data in electronic form, which is difficult to analyze, because of the interrelationships, the sheer volume and the inevitable random errors in measurement.


This month we take a look at FactNet, a somewhat unconventional tool to assist in analyzing mill data and to predict the results of operating changes to improve control systems. FactNet has also been used as a “software sensor”, in scenarios where available operating data are used on-line to calculate values that cannot be measured continuously by available sensors, or where available sensors would be prohibitively expensive to install and maintain.


FactNet is software that can run under Windows 95, 98 or NT to assist people with some process and statistical knowledge to analyze data, and to develop models for prediction, process control and related purposes. It is useful for statistical quality control and various other measures aimed at optimizing control of mill systems

The two key features of FactNet are that the mathematical basis is “Factor Analysis” and that the developers have built a variety of clever graphical tools into the software to help our limited human brains visualize and analyze the data and the results of the analysis.

The principle underlying Factor Analysis is that where there are a number of somewhat interrelated variables available to describe a system, there are a relatively small number of “Common Factors” that are the key to the relationships between input data and outputs. The software will search for these factors, and the user can apply his or her process knowledge to influence the factors selected. The graphical tools within FactNet assist the user in testing different hypotheses and in selecting the best model for solving the problem at hand.

For example, in a batch kraft digester the production rate, black liquor charge, white liquor charge, effective alkali, ramp time, rejects charge and steam use are routinely measured, and it is desired to predict the kappa number, viscosity and shive content of the pulp that would be produced after blowing the digester. The effectiveness of traditional regression analysis is limited by the excessive expensive to apply independent changes to the input data to determine the output data, since this would force the production of off-spec pulp.

In a practical case, FactNet analysis of historical data showed that 85% of the variations in the three pulp characteristics of interest could be explained by three common factors. These factors are, typically, variables that cannot be measured directly in any practical way, so the traditional variables are still used as the basis for control.

With the ability to predict the effect of the input variables on pulp characteristics, the operator and/or control system can manipulate the operating conditions in the digester to ensure on-spec pulp production.


FactNet can read Microsoft Excel files, or simple text (ASCII) files, where the values are separated by commas. Today, almost any electronically recorded data can easily be manipulated to one of these formats. Log sheet data could of course be typed into Excel, if required. Typically, a few hundred sets of observations would be used for analysis, but the software has the capacity of up to 64 000 observations, presuming that the computer one uses has sufficient memory.

FactNet includes tools for conditioning input data, which give the user the ability to automatically ignore outliers, or to manually exclude specific sets of observations. Various ways of treating missing data and outliers can be selected, to facilitate using the less-than-ideal data that one has to normally work with.


The results and data interrelationships are displayed as traditional X-Y plots, histograms, and a unique “correlation matrix,” where individual correlations between all possible combinations of variables are shown by color. This makes the strong correlations stand out visually, and a simple mouse click shows an instant X-Y plot of the values, with graphs of predicted and measured values superimposed where relevant. I find this a great help in searching for useful correlations. The FactNet approach relies so heavily on color and the ability to instantly switch between graphs that it is not practical to show this page. The only way of seeing it is to download the free demo version of FactNet and run the associated sample problems.


Most readers of this magazine who are interested in data analysis will already have the necessary computer and supporting software (Windows 95 or later). The principal prerequisite for using FactNet effectively is a knowledge of statistics as it applies to mill operations. FactNet is not a program for those completely new to data analysis, and the manual assumes basic statistical knowledge.

Training courses are offered by PacSim, either regularly at their offices or by arrangement at mill locations. In the latter case, the course can be based on local mill data, and aimed at an immediate mill problem. While a course is always great help in becoming competent with any complex tool, I do not see it as necessary for FactNet.

Any new user will need some time experimenting first with the training examples, and then with his own data, to become competent in using this powerful software package. Once a FactNet model is developed, it can be used by almost anyone, and I found the correlation matrix particularly helpful.


A working copy of FactNet can be downloaded from PacSim’s Web site, at http://www.pacsim.com/ Download.asp. The demo is fully functional, except that you cannot save your work. Demo datasets are included, and the program can be run by using the on-line help; there is no need for a manual. The downloaded file had a size of about 5 Mb, and installation of the software on our computer was straightforward.

The PacSim home page at http://www.pacsim.com/ gives general information on the software.

Print this page


Leave a Reply

Your email address will not be published. Required fields are marked *