Cleveland, Ohio
Seminar Led By Dr. Harold Haller

(Where do we start)

The partitioning variability program (SVA©) is especially effective on efficient data collection plans. With these plans one can quantify accurately where the variability is coming from in order to guide problem solving efforts.

Problem: Where are the sources of viscosity variability? It was agreed to study 3 areas: test, shift-shift and day-day. Originally a plan of 66 samples was developed which kept all variables independent in order that analysis of variance techniques could be used. Due to the high cost of sampling and testing SVA© was used to analyze the results from only 31 samples showing clearly that day-day variation was causing the problem. (Result: 50% cost reduction while providing a more timely solution.)

Topics to be covered include:
Partitioning variability to determine where process improvement efforts should be directed. Setting up efficient sampling and testing plans for cost effectiveness. Analyzing cost effective data collection plans.

(What affects what and how)

The Haller & Company’s correlation program (MC©) penetrates difficult problems easily and effectively. For example:

Problem: Improve quality and productivity of annealing furnace. Using typical multiple regression techniques it was concluded 6 process variables had no effect, other variables must be considered. Using MC© which makes it easy to create variables to better represent mechanistic effects and to find odd-ball results which distort the true correlation if not removed, effects and conditions were found which improved quality 50%, productivity 20%.

Topics to be covered include:
Analyzing results from either an experimental design or historical data. Avoiding fortuitous correlations and being misled. Quantifying the best correlation using appropriate transformations. Handling discrete variables. Detecting and not being misled by unusual results. Presenting the results of an MC© effectively to management and operators.

(Learning Efficiently and Effectively)

Haller & Company’s experimental design package (EDO©) is very efficient on a variety of difficult problems and can be tailored to fit any unique situation. Here are two examples:

Problem: 504 experiments were run to study 6 "inner" controlled variables and 6 "outer" noise variables in a powder metallurgy study to find optimum process conditions. Using EDO© only 18 experiments were run yielding a superior set of conditions.

Problem: Optimize a quality characteristic from 3 mixture variables with constraints and 3 process variables. Classical design called for 72 combinations. EDO© called for only 12 experiments, determined the same effects and found the same optimum.

Topics to be covered include:
Evaluating and comparing experimental plans. Designing an efficient experimental plan. Designing plans which allow for curvilinear and interacting effects. Restricting combinations from the design to make sure the design fits the process. Building on existing information to further reduce costs. Setting up very efficient screening designs.

MULTIPLE PROPERTY OPTIMIZATION (MPO©) (Finding the best of all worlds)

The optimization program is easy to use and very effective on all problems with many complicated effects, stringent demands on quality, costs and productivity and responses that are in conflict with one another but no obvious solution is indicated.

Problem: Reduce 7 raw material costs while continuing to meet 6 quality specifications on strength, color, viscosity, density, specific gravity and smoothness of surface. Using MPO© a recipe was determined which met all quality specs and reduced raw material costs $5000/day on one product line.

(Detecting Shifts and Non-Normality)

Statistical Process Control (SPC) is often equated with the concept of control charting. SPC encompasses much more than control charting. In fact all the previously described components of HITS™ are essential for effective process improvement and control. But the proper use of control charts plays an important role in any quality improvement effort.

Topics to be covered include:
Analyzing X-MR, C, and P type Control Charts. Separating the contribution to total variation from testing, process, and systematic shifts.
Detecting skewed and bimodal distributions and quantifying the appropriate parameters
for each case.
Developing control limits for skewed distributions.