Monday, January 12, 2009

How to Regain American Manufacturing Jobs (Part III)

This blog is broken into three parts. Part I describes costly manufacturing sector problems. Part II explains the primary causes of the problems. Part III offers solutions.

The Solution
The fields of applied mathematics and statistics are well-armed with quantitative methods available for predicting product performance and for predicting the output from manufacturing processes. If a manufacturer can predict performance, it can also prevent poor performance.

Unfortunately, manufacturers suffering from failed new product launches, high internal scrap, and premature failures in the marketplace are NOT predicting or preventing. As a result, they add even more cost by purchasing inspection equipment to try to detect defective product (so it doesn’t get to customers).

Many companies argue that they do have predictive models for predicting product performance. Unfortunately, most are computer-based models (with only theoretical relationships included), and clearly, many are inadequate—otherwise, there wouldn’t be such devastating dollar losses due to internal scrap, warranty costs, recall costs, litigation costs, losses in customer loyalty, high inspection costs, excessive equipment down-time costs, etc.

Physical models are required for predicting product performance. Physical models are equations developed from actual data (collected under a variety of strategically determined conditions/scenarios). .

Collecting physical data from testing can be expensive, but there are many optimization methods for minimizing the amount of data required while maximizing the amount of information produced.

Physical models can predict product performance under infinite sets of conditions, so engineers understand how their products will perform in a wide variety of environments (i.e. identify robustness). Furthermore, the predictive models can uncover design flaws quickly, so design adjustments can readily be made.

Process performance and tool wear can be best predicted by Statistical Process Control (SPC). A plant without proper SPC is a plant that reacts to problems—rather than preventing them. Unfortunately, I have rarely seen SPC applied properly during the past 20 years (while working in numerous plants world-wide).

Many quantitative methods can be used to make U.S. manufacturing plants competitive and highly profitable. Unfortunately, we generally do not see these methods being used—at least not properly. These quantitative methods are mastered by applied mathematicians and industrial statisticians, but rarely do we see highly credentialed mathematicians or statisticians working in manufacturing environments.

If our manufacturers continue to ignore these predictive modeling and optimization techniques, U.S. manufacturing will continue its precipitous decline.

1 comment:

Greg M said...

I enjoyed reading your very nice analysis.

This needs to be brought up in the 2010 election year now more than ever.