At ,technically, Six-Sigma is more than just a quality measurement - It is a valuable tool that helps to develop process viability and set realistic specifications while optimizing yields and quality. Our Six-Sigma methodology enables us to extract the maximum amount of information from a minimum amount of trials, dramatically decreasing development time and resource consumption while enhancing process/product efficiency and quality. The enhanced understanding of the process and the models generated from our methods provide us with the capability to rapidly and efficiently scale-up.
The following paper is designed to give a general overview of the use of Six-Sigma in the Chemical Industry. It provides four examples of how ,technically, utilizes Six-Sigma for:
For more information, contact us.
Most of us have heard of the term "6-Sigma" and most of us understand that it has something to do with ultra tight quality specifications. We can also view it as a rather large kit of statistical tools, some of which work spectacularly well in the chemical business. The purpose of this paper is to demonstrate, with real examples, the use of a few of these tools.
Every effort is being made to keep the use of 'statistician speak' to a minimum; however, we will need to discuss statistical software. There are a lot of good software packages available, each with its own pros and cons. The author’s favorite is a package called Minitab, which is used in the examples outlined in this document.
When the chemistry of a process has been selected, the next step usually is to optimize process conditions, raw material ratios, etc. This can be done very efficiently with a designed experiment. Our favorite design for this use is the K8 which is an eight experiment, two level array. The eight experiments are based upon a process or reaction that is already being used, but may not be well-defined or optimized. The operators familiar with the process indicate the variables they think are important to the outcome of the process. Typically, these variables include reaction times, temperatures, stoichiometric ratios, purities of raw materials, and the like. The actual values for these variables, as they are used in the existing procedure, are designated as the "controls". In a "two-level array", values for each of the important variables are assigned above and below the control values, typically being some "± percent" relative to the controls. These quantities are used in the experimental array, the "Level One" quantities being the values below the control, and the "Level Two" quantities being the values above the control. For example, if the control temperature in a reaction is 80 °C, we might run it at 75 °C ("Level One") and 85 °C ("Level Two") in the K-8 model.
In this demonstration, we will be looking at the effect of varying the raw material quantities on the yield of an Ullmann reaction, in which aromatic halides are coupled through the action of a catalyst.
K8 Layout for Ullmann Reagent Study
The final step is to crunch this data using statistical analysis software such as Minitab. The following graphical output is one of several types of output analysis possible, and is called a Main Effects Plot. The slope and direction of the resulting lines in the plot indicate the degree of importance of a particular input variable on the desired outcome. In our example, yield in grams of the desired product is plotted as a function of the two different levels of each reagent used in the reaction. The effect of the catalyst on the yield of product is stunning: using low levels of catalyst in the reaction (designated by the left hash mark along the catalyst axis) leads to low yield of product. Using higher amounts of catalyst (as indicated by the right hash mark along the catalyst axis) leads to higher yields of desired product. The data also indicates that the effect the catalyst has on the outcome is at the > 95% confidence level. Thus, we can be certain that the amount of catalyst used in the reaction will have a large influence on subsequent yield. The Main Effects Plot also shows that increasing amounts of amine used as reagent lead to higher yields of desired product. To a lesser extent, increasing amounts of KOH result in increased yield of product, and decreasing amounts of chelating agent increase the yield of target product.
A process for a solution polymer is being run in the pilot plant as an initial supply but the customer is anxious to have it scaled up. The customer had requested a viscosity range of 180 to 220 cp. The viscosities are:
The average viscosity is 201 cp with a standard deviation of 16.5 cp. Let's take a look at a histogram of this data and give some thought to the specifications.
The good news is that the
process is very well centered at 200 cp and that the data spread appears to fit
the classical bell curve. The bad news is that setting the tolerance
range at the desired 180 to 220 cp
means that about 25% of the production is going to be out of specification
(blending is not feasible). The
Before we leave this example, we
should spend a few minutes on the subject of process capability. This refers to the width of the histogram of any process
output parameter. The narrower the
distribution, the more capable the process.
capability = Cp = tolerance range/(6* sigma)
process, Cp = (220-180)/(6*16.5) = .40
The Six Sigma school of thought defines a good process as one having a Cp of 2.0. In the case described, it is clear that the existing Standard Deviation of the process, 16.5, is too large and should be tightened up. Option 3 becomes an attractive alternative to Options 1 and 2.
A plant running a Fries Rearrangement was experiencing both yield fluctuations and variable amounts of an undesired impurity in the product. Three of the raw materials were questioned. The first two, AlCl3 and PCl5, were added in slightly variable quantities because the operators disliked working with partial drums. They always added complete drums and recorded the weights added. The third raw material, an ester, had variance in its assay.
The technique used (called multi-vari) was almost like a designed experiment, except that data had already been acquired in 37 previous plant runs. The data had to be categorized. For example, any AlCl3 quantity above the mean value used in the 37 runs was assigned as a "Level 2" input value. Any quantity below the mean was assigned as a "Level 1" input value. This assignment of "Level 1" and "Level 2" inputs was repeated for PCl5 quantity and Ester purity assays, as well. The results are shown in the following Main Effects Plots. (See the discussion in Example 1 of this paper, entitled "Using Designed Experiments for Process Development", for definitions of the terms used for input values and for a discussion of interpreting Main Effects Plots.)
Increasing quantities of AlCl3 and PCl5 tend to increase the yield and the impurity. Increasing the assayed purity of the Ester improves the yield and decreases the impurity in the finished product.
A simple blending operation was being performed in a coating plant in which PVA was being dissolved in warm water in a mixing tank. The resulting fluid was passed through a fine filter in the pipeline to the coating machine. The difficulty was that the pressure drop across the filter fluctuated severely and unpredictably. A four experiment, two level design called an L4 was performed.
The graphical results show that
this is an interactive process. At
the lower temperature setting used in the experiment (refer to the solid line
labeled 60 C), either PVA type can be used and still result in a
manageable pressure drop across the filter.
However, at the higher temperature setting used (refer to the dashed line
labeled 80 C), only PVA type 1 can be used.
Do you require any process development or custom synthesis?
Do you have a troubled process or sick plant?