An experimentation method known as Multivariable Testing analyzes statistical test results to solve problems, boost efficiency and cut costs.
Nearly 10 years ago, maintenance engineers at DuPont's Agricultural Products Division in LaPorte, TX, had a near-crisis on their hands. Pumps used in the manufacture of a key herbicide were failing at an alarming rate, shutting down the production line and resulting in losses mounting to $750 million a year.
After repeated, futile attempts to solve the problem, the company turned to a statistics-based experimentation method called Multivariable Testing (MVT). Pioneered by Knoxville, TN-based QualPro, Inc., the MVT technique allows simultaneous testing of up to 40 changes to the status quo and pinpoints which ones help, which ones hurt and which ones make no difference.
Tests of the DuPont pumps uncovered several surprising changes that helped correct the problem. These included switching from a stilt to a concrete base, stopping use of laser alignment to set up the pumps, and not increasing oil-change frequency because it did nothing to extend run life. After making the changes, pump failures dropped from 53 per month to 13 per month on more than 330 pumps operating in the plant.
What sets MVT apart from other process-improvement techniques is its use of advanced mathematics to test numerous variables all at one time in real-world situations. This runs counter to the "scientific method," which asserts that only a single variable at a time can be changed in order to observe its effect on the experiment. When most companies are confronted with a problem, they gather their staff, come up with ideas on how to solve it, then try as many as they can implement. Not only is this inefficient, but it can be counterproductive. QualPro has discovered, by analyzing the results of more than 13,000 experiments, that on average only 25% of these ideas typically help, 53% make no difference and 22% actually make the problem worse.
MVT gives companies a way to test many ideas rapidly and measure the impact of each one on every critical outcome. These outcomes can range from improving throughput, yield and product quality to reducing raw material and labor costs, to cutting liquid effluent and gaseous emissions. This enables companies to take the knowledge gained through testing and implement only the helpful solutions, avoid hurtful changes and take cost-advantage of those that make no difference.
MVT methodology has its roots in advanced statistics. QualPro founder Dr. Charles W. Holland developed the technique working in the nuclear weapons industry during the Cold War. He took the mathematical concept from academia to the factory floor when he used MVT to figure out why a certain atomic bomb part had a crippling 85% rejection rate.
Properly executed, MVT methodology is actually the final step in what QualPro calls the Eight-Step Procedure for Improving a Process. The steps focus on developing success measures, ensuring accurate data-gathering systems, implementing appropriate statistical process controls and correcting other conditions that could compromise the validity of the experiment.
Brainstorming first
Once groundwork has been laid, MVT begins with a brainstorming session to generate as many ideas as possible for improving the process. Including a broad cross-section of people in the initial brainstorming session is critical because the more ideas brought to the table, the greater the chances of discovering the one that can make a difference. A typical brainstorming group might consist of shift operators, technicians, engineers, quality improvement experts and others involved in the process, including suppliers and customers. In the first round of brainstorming, no idea is a "bad idea." Even unorthodox suggestions make the list.Companies have found the MVT process to be employee-friendly because it involves people at all levels. Unions have even negotiated it into their contracts because of the empowerment and involvement of front-line workers. The process focuses on the impact of ideas, not the source. Everyone's ideas get a fair shot.
After the initial brainstorming list is compiled, potential factors are pared to those that are easy to test, fast to implement and low in cost. The goal is to have experiments that are simple to execute and findings that can be put in place rapidly. In addition, ideas are eliminated that might compromise safety, sacrifice quality or would be cost-prohibitive. For example, someone might suggest installing a new piece of equipment to improve a process, but the idea would be discarded because only factors that cost little or nothing are considered.
Once the list is narrowed and factors are selected, specific options or levels for each factor are determined. In most cases, two levels are considered adequate for the effects of a change to be investigated. The first MVT experiments typically test 20 to 30 ideas for improving a process. Experimenting with lots of different ideas is recommended in order to get the best results.
After the results of the first MVT experiments are analyzed, a second round of experimentation is often used to pinpoint the most important factors and optimum levels of operation. These experiments also help identify interactions, which are frequently important in complex processes. Historically, QualPro has found that about 70% of the process improvement comes through making individual changes, while the remaining 30% comes through implementing pairs of changes that are interacting.
When all testing is complete and results are analyzed, two important steps remain. First, plants need to change their standard operating procedures based on the findings. Careful, clear communication of the changes is critical to making sure they happen. Second, operations must continue to measure and control the process over the long run in order to maintain the gains. It is crucial that the process doesn't slip back to the way it has always been done.
A path to efficiency
Over the years, MVT has proven it can solve problems that may have previously been deemed insurmountable. In chemical manufacturing, for example, this has involved finding a way to exceed theoretical yield estimates. In a paper mill, it has been used to coax higher throughput from a 20-year-old machine so the factory can survive price wars. And in another DuPont plant, it meant finding the source of a perplexing defect in paint uniformity.A good example is the work Olin Chlor Alkali Products has been doing at its plant in Charleston, TN, during the past year. The facility produces chlorine and caustic soda for pulp, paper, chemistry, textile, clay and other industries. Its largest variable cost is power consumption with more than 750 million kw/hours of power consumed by the plant each year and an annual power bill of about $20 million. After numerous attempts to reduce power usage had failed, and with the price of power skyrocketing, the company turned to MVT to test ideas for reducing consumption. A series of tests revealed that anode leveling caused the voltage coefficient to drop and could contribute more than $300,000 in power savings per year. Tests also zeroed in on an interaction between the voltage coefficient and the flow of mercury used in the process, pushing annual power savings to more than $435,000. Changes implemented as a result of the tests also reduced anode damage, increasing average lifespan from 32 to 48 months, and reduced pump maintenance.
MVT has also helped plants achieve objectives that appear to be at odds with each other, such as increasing quality while cutting costs or boosting efficiency while reducing waste. At its Hexacomb paperboard facility in Trenton, IL, for example, Pactiv Corp. recently completed an MVT with a goal to reduce set-up time on two different production lines without affecting product quality. The test identified various changes that lead to a 35% reduction in setup time, or roughly 125 hours saved per year per line. That not only translated into an overall increase in uptime of as much as 4.6%, but reduced waste on each line by nearly 14%.
Oil-Dri Corp., headquartered in Chicago, experienced a similar result in a plant that had recently switched from plastic jug packaging to paperboard cartons for its Cat's Pride scoopable litter. After learning that the cartons were leaking product on store shelves, the plant tested and tweaked equipment and materials, but failed to solve the sealing issue. The MVT process, however, determined that the flap sealer had been running too slowly, which caused glue to drip down the flaps and lose its tackiness. Speeding up the machine not only solved the sealing dilemma, it also allowed Oil-Dri to increase line speed by 35% and save up to $125,000 per year in glue costs.
MVT can be used to help avoid unnecessary capital investment by identifying the majority of all ideas tested in a given situation that have a negative or no impact on the process. For example, several years ago, the DuPont Agricultural Products plant in LaPorte, TX, used MVT methods to increase the capacity of its wastewater treatment facility by 50%, eliminating the need for a $500,000 dredging of its polishing lagoon and a $20 million expansion of the facility. An experiment conducted over a six-month period showed that solids-removal capacity could be increased dramatically simply by adding more ammonia nutrient and increasing aeration.
The same team also improved incinerator capacity by 41% using MVT, removing the need to ship wastes off-site to be burned at five times the cost of incinerating them at the plant. An experiment tested factors such as the recycling water purge flow, scrubber delta flow, the number of blowers used, and breech temperature. The experiment was completed in 33 hours with virtually no experiment costs or disruptions to the process.
A third project discovered "unfound" capacity in the plant's toxicity characteristic stripper, which uses steam to remove methyl chloride from the waste stream. The stripper was at capacity keeping pace with production demands, and the plant was facing a significant tightening of permits by the Environmental Protection Agency. The first two of three experiments shocked those familiar with the process by zeroing in on the waste-stream feed rate and overhead temperature as key factors limiting capacity. Even more surprising was the finding of the final experiment: After the stripper reached a 50-gal.-per-minute feed rate, it was learned that the only critical control necessary was to maintain the overhead temperature at a minimum of 210 degrees F. As a result, the stripper's capacity soared 300%, from 40 gpm to as high as 120 gpm. The level of methyl chloride in the wastewater meanwhile plummeted from 78 ppm to 0.2 ppm, a level well below the EPA limit. The modifications also yielded a cost savings of $400,000 per year and avoided a massive investment in a new stripper.
Organizational focus
A side benefit of MVT is that it forces organizations to focus. Some manufacturing plants, for example, keep control charts on hundreds of things when only a handful truly matter. Others have never instituted statistical process control, have let it lapse, or are not rigorous in analyzing the data. The MVT process enables companies to define what is really important, then control it.Elo TouchSystems, a Raychem Corp. subsidiary based in Oak Ridge, TN, for example, had tried hard to change a devastating 25% reject rate on its touch-sensitive computer screens. The company was losing $3 million a year producing flat and curved touch screens plagued by "chicken skin" and was on the brink of shutting down. Within months of running an MVT that focused on 18 factors, the manufacturer had instituted a series of changes that lowered its defect rate to less than 1%. This led to an $8 million per year improvement and the ability of the company to capitalize on an expanding market.
Similarly, a manufacturer of machine-cast manifolds for V-8 engines also credits MVT for helping the company remain viable. An astounding 32% of the aluminum parts produced by its casting operation failed inspection and had to be remelted. After identifying important flaws in a key measurement system, QualPro also discovered a human factor: the order in which one operator did things during the casting process made a huge difference in the scrap rate. When the company instituted the prescribed changes, its scrap rate dropped to 9%. That translated into an $11.5 million annual profit improvement —- and survival of the company.
QualPro Vice President of Operations David Cochran and Account Director Kerry Stone have worked with hundreds of manufacturing plants to improve processes. For more information on Multivariable Testing, contact QualPro at [email protected] or QualPro, PO Box 51984, Pellissippi Parkway, Knoxville, TN 37950; (800) 500-1722; www.qualproinc.com.