This page contains a Flash digital edition of a book.
statistics in manufacturing industry


numerous interacting variables continually alter their relative contribution to marginal return per item over thousands or millions of occurrences, though, multifactorial data analytic methods are increasingly central. Tat balancing act, in case you


Sampling of normalised tool setting drift data, gridded and used to plot a surface in SigmaPlot, reveals a double sinusoidal deterioration model


the purpose. Tere are others which use in- house statisticians running desktop soſtware with MiniTAB, SigmaPlot or Xlstat being popular choices. Such considerations, which have their


analogues in every area of manufacture, have effects up and down the supply chain and are also part of a much wider mesh of factors. In what a recent Quality Digest seminar succinctly described[2]


were wondering, is the reason for my unpredictable mince pie hunt. I asked a large supermarket chain why it was that I could, at the same time in one year, eat their own- brand mince pies in packets of 12 while an identically presented packet of six contained an unacceptable ingredient which makes no material difference to the pie. Te answer took me up through a supply


chain in which the central decision-making discriminator was a margin per pie of a small fraction of one cent. One manufacturer leſt that ingredient out for the cost saving, but couldn’t pack sixes without losing profit margin. Another could pack sixes, and included the ingredient for another market, and thereby gained an economy of scale which reduced costs. Supermarkets and suppliers both have


as ‘the complex supply


chain ecosystem’, customer perceptions of disparate product factors are only part of a larger, multifactorial dance; an ever shiſting global structure of ‘manufacturers, strategic partners and multiple tiers of small- and medium-sized enterprises’. At each stage in those multiple tiers, an entity is both supplier and client. Each entity is, in external system modelling and data analytic terms, a black box. Internally, each is a dynamic decision-making system which has to juggle its own clients’ (various) expectations against its own supplier capabilities and its own economic balancing act. Manoeuvring and competing within such


an ecosystem has a surprising amount of ground in common with financial markets analysis. Te currents and flows are not as fast moving, but the structures are the same and so is the basic principle of planning for many occurrences of very small margins. I didn’t use the word multifactorial lightly, in the last paragraph. Specific analyses like the relation between a product and its putative real world effects are oſten based on classical significance testing. Te MVD/ Parkinsonism studies mentioned elsewhere (see box: In sickness or in health), for instance, use Fisher’s exact test, Pearson’s chi-square statistic and the Wilcoxon rank-sum test because the investigation space can be tightly defined to isolate pairs of variables. In making decisions where


www.scientific-computing.com


one eye on the sensitivity of consumers to ingredients, but my own requirement is not common enough to outweigh other factors, so what I am actually seeking out is a combination of other factors which deliver it as a by-product. Both of these mince pie manufacturers


were fine tuning their production strategies using multivariate data analysis soſtware products also common in the chemical,


microprocessor, bioengineering and aerospace industries. StatSoſt’s Statistica, Kovak’s MVSP and Camo’s Unscrambler all raised their heads in different parts of the supply and demand web stretching out behind these pies, with discriminant and principle components analyses the most oſten named approaches. One university department with strong manufacturing research relationships was evaluating a proposed move to mathStatica (which works as a specialist add-on for mathematical


CHOICES ARE


HEAVILY INFLUENCED BY INTUITION AND EXPERIENCE, BUT MODELLING ON THE BASIS OF DATA ANALYSIS PLAYS A LARGE PART, TOO


statistics to Wolfram’s Mathematica) as its multivariate tool of choice. Similar soſtware was also used by the supermarket’s own planners, producing a feedback environment which tends towards stability but is vulnerable to unexpected ripple effects as changes propagate through the structure. Data analysis within the food and


beverage region of the manufacturing ecosystem doesn’t stop with supply lines.


Keeping the wheels turning


Statistics in manufacture brings to mind issues concerned with the manufactured product, but data analytic attention to the means of production is equally important. Georgia Tech, some of whose product analyses have already been mentioned in the main text, also focuses its attention on how the manufacturing environment functions. Shabbir Ahmed, George Nemhauser and Joel Sokol, members of the School of Industrial and Systems Engineering, are collaborating on a data-driven project to minimise congestion in the physical movement of semiconductors from one processing station to another throughout a factory. Routing of semiconductors between processing machines differs from item to item, so there is no assembly-line procedure in the usual sense: items are instead conveyed from one processing point to the next by hundreds of automated vehicles. Any congestion as these vehicles move around the facility delays production, so the team is developing means to


optimise their routing and scheduling. The same trio also devised algorithms to maximise the efficiency (reportedly to 99 per cent) of machines which handle glass plates as they are added to and removed from a production line. By again analysing operational data to inform scheduling of production sequences they were able to reduce production delays, increase capacity and cut waste.


Nagi Gebraeel, an associate professor in the same faculty, conducts research on ways in which gathered data can be linked back to maintenance decision making. The aim is to monitor deterioration in engineering systems over time, detecting likelihood of failure and triggering prevention at the optimum time. By turning multivariate metrics into accurate estimates of remaining life for individual components, the most favourable balance for the overall manufacturing system can be struck between the costs of downtime and unnecessary maintenance.


FEBRUARY/MARCH 2012 19


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48