Category Archives: Technology

Smart part number example
May 29 2012

Why “Smart” part numbers should be replaced with keys and property lists

Nomenclature matters. What you call parts, products, and resources, and how you attach technical, commercial, and administrative data to their names determines how easy they are to retrieve in daily operations and to analyze for planning or control purposes. Manufacturing, worldwide, is overwhelmingly dominated by part numbering systems  that are called “smart” because they embed data about materials, dimensions, number of holes, or other characteristics into part numbers, as in CEP-134M10FB-TT1234\R4, where each group of characters has an encoded meaning, for example  CEP for Capacitor and R4 for Revision 4.

In concept, these systems are a 100-year old legacies of the paper-and-pencil age, and obsolete in the age of databases. The alternative is to have part numbers that are just unique IDs, or primary keys in database terms, with all the data attached in the form of property lists in plain text. I propose to call it the key approach. The case for the key approach is overwhelming, and the technology to use it available. But nearly all manufacturing professionals active in 2012 have been educated in the old method, and taught that it was smart. They are unaware that it should be replaced.

In addition, people who have spent years mastering a complicated, antiquated system are always strongly attached to it and oppose its abandonment, as it would make their hard-won expertise irrelevant. This is why, for example, command-driven information systems were successfully marketed long after they had been supplanted by graphic user interfaces in home computers. It is also the reason spelling reforms almost never succeed. Part numbers with encoded information are a deeply rooted legacy.

These are great hurdles to overcome, requiring missionary work. Of course, even if this post is persuasive, if leaves unanswered the question of how you migrate from “smart” part numbers to the key approach, in an existing organization with legacy information systems. It is an essential question, and needs to be addressed separately, once the desirability of this transition is established, which is our point here.

Let us begin with detailed explanations of the value of the key approach.  First, it does not in any way limit functions. In addition, it has the following advantages:

  1. It reduces training costs.
  2. It is not affected when product information changes.
  3. Part numbers are short, and therefore easier for humans to work with.
  4. Part numbers are distinctive, even for similar parts, thus reducing picking errors.
  5. Nomenclature errors are more easily detected and corrected.
  6. Property lists in plaint text on labels are easier to read.
  7. Property lists support data mining.
  8. Product catalogs from different companies can be merged without loss of information.
  9. Company-private information is not unintentionally disclosed through part numbers.

Much of the material below is based on a recent discussion in the APICS discussion group on LinkedIn, initiated by Elvi M., and including contributions from Patrick Doyle,  Jennifer Verellen, Joseph E. Harrington, Ph.D., Felipe Sanchez Ryckewaert, Martha M. Munson, PMP, Greg Pope, and T R Volpel CPSM.

Origins of “smart” numbering systems

Such systems have been introduced in libraries in the 19th century. In his latest movie, J. Edgar, Clint Eastwook highlights the role future FBI director J. Edgar Hoover played in the development of the Library of Congress cataloging system, in which “BX 378.5 .M38 1968”  breaks down as follows:

  • BX = Books on Religion
  • 378.5 .M38 = Shelf Address
  • 1968 = Year of publication

In the 1910s, it made sense because it allowed trained users to locate books quickly. You could search the card catalog by author, title or subject, and then locate the book by its call number, as shown in Figure 1:

Figure 1. Using a library card catalog

Awareness

Most manufacturing professionals see nothing wrong with “smart” part numbers, to the point that they extend the approach to every aspect of operations, resulting, for example, in expense reports with lines like “HK0010 DB-ENG-122 M-3 RK 23.50,” which translates to  ” Helmut Katz spent €23.50 for dinner at the Ratskeller while working on the DaimlerBenz Engineering project.”

“Smart” part numbers have grown such deep roots in manufacturing operations that few managers or engineers are even aware of the unnecessary costs they generate in training and routine decoding, and of the obstacles they place in the way of manufacturing data mining. Some academics do realize that there is something wrong: in his 2005 textbook on Product Lifecycle Management (PLM), Purdue University’s Micheal Grieves reports that “With the implementation of PLM, smart part numbers are generally replaced by sequential part numbers.” In their LinkedIn profiles, however, several recent Purdue graduates report implementing “smart numbering systems” during internships in manufacturing companies as late as 2009!

Awareness of the issues is perceptible in blogs on PLM and CAD , as, for example, in the following:

Advocating the replacement of a method that its promoters got away with calling “smart” is an uphill battle. Who would want “dumb” systems instead? The first step may be to come up with an attractive generic name for alternatives, which is why I am suggesting we call it the key approach.

The key approach

Figure 2 shows an example of the key approach to an item and the characteristics that are usually embedded in a “smart” part number.

Figure 2. The key approach to part numbers

The item is uniquely identified by the key “13AB5,” and the nomenclature database has a record for each property of each item, in the form of a name-value pair. There are different views of what an item is. In some cases, the same key is used for all revisions, and you tell the revisions apart by their number. In this example, each revision has its own key, and contains the key of the prior revision in its property list, so that you can trace back the history of the part through multiple revisions.

Neither the property names nor the values contain any abbreviations or codes. They are all in plain text. In particular, you may notice that, where applicable, property values include units, while “smart” part numbers usually don’t. This is an important detail in multinational companies where metric and imperial measurements coexist. There is no need to store these characteristics in short fields. You can use as many characters as you need. 30 years ago, data storage space was expensive; now it is not.

Different items can have different property lists. Metal tubes have a length and a diameter; potato chips, expiration dates. The property values can also change over time without the item itself undergoing any change that would justify a new part number. For eample, if the item is a product that undergoes a drop in demand, its volume category may be downgraded from A to B, and it may then no longer have a dedicated warehouse location.

Reducing training costs

Unless the users of part numbers, in materials handling, production, engineering, production control, purchasing, or sales, are able to retrieve the information embedded in a “smart” part number, embedding it was a waste of time. But, in order for this to happen, hundreds or thousands of people would have to be trained on issues like the meaning of the field in characters 5 to 8 of a part number. In reality, it doesn’t happen. In pursuit of data mining, I often have to parse “smart” part numbers and find that, outside the department in charge of master data, very few people have a clue. When asked what a part number means, they have to painstakingly decode it field by field from a spec.

With the key approach and characteristics listed in name-value pairs in plain text, the information is accessible with no training, which is why it is used on e-commerce sites like Amazon. Each product’s unique ID is Amazon’s ASIN, which may designate anything from a book to a waffle iron. The Product Details are given in a list of name-value pairs, and this information is immediately obvious to a first-time visitor. You don’t have to take a class.

Historical continuity and traceability

Patrick Doyle used 1-015-113-029-2 as an example of a “smart” part number, meaning that it is a single part (1), used on a phone (015), is a diode (113), supplied by Diodes R Us (029), and is revision two (2).

Such a part number contains data that is subject to change as you may, for example, decide to switch suppliers. This puts you in a dilemma. If you change the part number so that it contains accurate supplier data, you break the continuity of your history with that part, or you keep the same part number and violate your own convention. If you do change the part number, you can still retrieve the complete history by maintaining a table of name changes, but it adds complexity.

With the key approach, instead of 1-015-113-029-2 and a dictionary to translate all properties into plain text, you have something like the following:

  • Part ID: EA5D4, a unique key unique, with no embedded data.
  • Property list in database, keyed on Part ID = “EA5D4″:

Where-used: iPhone
Family: Diode
Supplier: Diodes R Us
Revision: 2

All the properties can be changed without affecting the key, and the complete history of production volumes, deliveries, and quality problems can be retrieved with the key.

Short names, easy to work with

“Smart” part numbers tend to be preposterously long, and therefore practically impossible for people to remember. No matter how large a field an ERP system or MES provides for part numbers, some company “smart” numbering system will exceed it. In the key approach, the part number’s one and only job is unique identification within the scope of the system. There may be another object by the same name at the other end of the world, but not within the plant or the company.

A sequence of just 5 uppercase letters and digits is mercifully short and easy to remember. It provides (10+26)5 = 60.5 million possible IDs, and that is enough to meet the needs of even a high mix production organization for a long time. Some prefer to use digits only, but it reduces the name space to only 100,000 IDs.  Allowing lowercase letters would increase it to 916 million, but few people would reliably make the difference between “13AG5” and “13aG5.”

Distinctive part numbers

“Smart” part numbers give similar names to similar parts, which makes them more difficult to tell apart during picking and increases the risk of picking errors. Hergé’s  Tintin stories feature the two dim-witted detectives named Thompson and Thomson shown in Figure 3:

Figure 3. Thompson and Thomson

The only way to tell them apart is from the shape of their mustaches, and their nearly identical names don’t help. Now, in Figure 4, consider the following two, nearly-identical screws and their nearly identical part numbers:

Figure 4. Nearly identical parts with nearly identical part numbers

One of the screws is 1/8 in shorter than the other, which, with “smart” part numbers, results in only the last digit being different. How much more likely are they to be confused during picking than if their names were A23GT and 92WRT?

Detection and correction of nomenclature errors

The only way to be sure a part number is read and written accurately is respectively by reading it automatically from a bar code or an RFID tag, and by printing from the master database. Typing, handwriting, copying and pasting, and even selecting from a long pull-down menu with similar entries is error-prone, particularly when performed with data that has been extracted from an ERP, MES or PDM software system into an electronic spreadsheet. Besides typos, part numbers may have interchanged characters, be truncated, or be misinterpreted by the destination software, as when, for example, Excel treats a part number like a number and lops off leading zeros. And the slightest error in a part number is enough to make the consolidation of data about it from multiple sources impossible.

The above is true no matter what part numbering system you use. The risk of error, however, is much higher with “smart” part numbers than with keys, because they are two to three times longer and more similar, so that a part number with a typo may point to a different, existing part rather than nothing.

Labels with plain text characteristics

Several participants in the APICS discussion, and particularly Martha M. Munson, and Greg Pope, expressed concern about the human-readability of labels. For example, if no information is embedded in a part number, how can material handlers locate it in a warehouse? In fact, the key approach is not to hide this information but to separate it from the identification itself. As shown in Figure 5, all the data in Figure 2, or a relevant selection from it, can be printed on a label.

Figure 5. Box label with characteristics in plain text

Note also that all the characteristics on this label are printed in plain text. The best, cheapest and least error prone standard is using whole words from the native language of the work force. If the bolt category is called “Bolt,” no training is needed to recognize it. If you call it “256,” you need a dictionary. Human readability is, of course, not the only concern in daily operations. You also need to use bar codes or RFID tags for machine readability.

Support for data mining

The most convenient input to manufacturing data mining is a table in the dimensional model, meaning that some columns contain reference data — used for filtering and aggregating — and facts — the measured quantities or observed attributes for which you want to identify patterns. All the characteristics embedded in a “smart” part number are part of the reference data. You want to be able to analyze quality for parts made from a given raw material, or within a dimension class, etc.

Retrieving this reference data by parsing “smart” part numbers is a major effort, that has to be repeated with each naming convention.With the key approach, on the other hand, turning the property list of Figure 1 into the table of Figure 6 is just a matter of using a crosstab query in Access or a pivot table in Excel.

Figure 6. Item properties crosstab for data mining

Merger support

The possibility that the company might merge or be acquired is usually not a consideration in the design of its nomenclature. However, considering how often such events occur and how far reaching their consequences, it should be. When companies merge, their catalogs of products and parts should too. Merging information models, and nomenclature in particular, is always difficult and time-consuming. Mergers are mostly of unequal companies, and the larger company’s system becomes the merged company’s standard, even when the smaller company has a better one. In true mergers of equals, this situation often goes unresolved for years, with both divisions’ legacy systems remaining in operation, and possibly, their real-time content meeting through integration middleware and their histories in a data warehouse. To focus on what happens we part numbers, we consider two hypothetical mergers. In the first, both companies have “smart” part numbers; in the second, both use the key approach. Real situations, of course, have many other patterns.

I have never seen two companies with identical “smart” part numbers. For them to use different naming conventions is good for uniqueness after the mergers, because it makes it highly unlikely that they will have assigned the same name to different products or parts. As the software systems themselves usually do not constrain the naming conventions — except for the total length of part numbers — you can upload another company’s part numbers, and end up with a catalog that commingles part numbers based on both conventions.  But this adds further complexity for data mining, because, to retrieve the information embedded in part numbers, you now need to parse names generated with two different conventions, and to identify which convention each name is based on. The management of the acquiring company intends to standardize the part numbers on its own system, but implementation has a way of being postponed indefinitely.

Two companies using the key approach have a higher risk of key conflict, especially if the keys are sequential auto-numbers, because both companies will have products with keys 1, 2, 3,…, 4384, needing disambiguation. This is a problem that is usually not anticipated when systems are first implemented, but can be avoided by instead selecting names at random from a large name space. For example, if 20,000 keys are chosen with equal probabilities among the 78.3 billion  possible combinations of 7 digits and uppercase letters, then 99.7% of the time, no two will be identical. You would still need to search for duplicates just in case, but you only have a 0.3% probability of finding one.

The European Union had this problem with ZIP codes, and resolved it with country code prefixes. Another option, if you have two products both named 4384 by different companies is to rename one of them 4384.1. Any change to keys is undesirable, because all references to that key then need to be updated. This can usually be managed within one system, but not within all the ad-hoc extracts and spreadsheets that users have generated.

Merging the property lists from two companies systems is work in any case but is easier to do with property lists than with “smart” part numbers, where it requires you to develop a sophisticated parser. The properties in both systems are likely to have both different names and sets of values for the same data. The name value pair (Material, Steel) in one system may by (Mat’l, STL) in the other, and the engineers need to build a dictionary to translate them. At first, you may translate the data only in the data warehouse, which is sufficient to jointly mine the data from both systems, but, eventually, you would want to standardize the property lists across the divisions.  When you do this with the key approach, you deal with issues of  meaning that have to be addressed anyway as part of a merger. With “smart” part numbers, you also have to resolve format issues that are unrelated to meaning.

Data security

The protection of proprietary information is the only legitimate reason to encode and encrypt data. During development, for example, the marketing names of products are not used for fear that they might leak to competitors. Once a product is made and sold, all units often bear serial numbers that must be encrypted lest they provide competitors with information about production volumes. As mentioned in Data Mining in Manufacturing versus the Web, the use of plain text in serial numbers on World War II German tanks allowed British analysts to estimate the numbers produced from the serial numbers of the units that were captured or destroyed. The serial number of my iPad today is DKVG805HDFHY, from which Apple competitors can deduce nothing unless they know how to translate it back to a production sequence number.

“Smart” part numbers do not provide much security, as anyone who cares to can reverse-engineer them from actual products and labels. For example, on the Lands’ End catalog, once you find out that Men’s Regular Short Sleeve Jacquard Polo shirts are item#40600-5A63 in Regular sizes, and item# 40600-6A69 in large sizes, it doesn’t take Sherlock Holmes to guess that the numbers preceding the dash identify the polo, and the alphanumerics following it, a size class. A “smart” part number is like a house key that would carry information about which door is opens; a thief cannot infer your address just from your house key simply because this information is not on it. Likewise, in the key approach, a product ID by itself will not reveal anything to your competitors.

Raku-Raku seat
Mar 13 2012

How to eliminate “Muri,” or overburdening

“Muri, Muda, Mura” is often mentioned in the Japanese manufacturing literature as a trio of evils to avoid. Of the three, Muda gets the most attention. Usually translated as waste, it designates everything we do in a factory that is unnecessary.  For a change, let us focus on Muri.

Muri, in everyday Japanese, means impossible, with the nuance of unreasonable or unsustainable. A person working exceptionally hard is said to be doing Muri. Other words are used to say that something would violate the laws of physics, or that it is socially improper or inopportune. When there is Muri in your process, it means that you are asking people to work too hard, which results in  defects, burnout, repetitive stress injuries, or even accidents. Conversely, removing Muri means making your process humanly sustainable, so that is can be executed as well at the end of a shift as at the beginning, by a 50-year-old  or a 20-year-old, a man or a woman, 5 or 7 feet tall.

It cannot be repeated often enough that Lean is not about making people work harder but instead, in the tradition of Frank and Lillian Gilbreth, in making the work easier to do. When you observe a truly Lean plant, you do not see operators hurrying. Instead, you see them working steadily, at a sustainable pace, at jobs that are carefully choreographed for effectiveness and efficiency. A key example of Muri elimination is the raku-raku seat shown above. It is a device introduced at Toyota in the 1990s and now adopted by many car makers to remove the need for operators to crawl into car bodies in order perform assembly tasks inside.

There are many tools to remove Muri. You can easily notice that an operator is overburdened by direct observation in the shop. A more systematic approach is to use Toyota’s TVAL to rate jobs based on the weight operators have to carry and how long they have to carry it. TVAL establishes an equivalence between combinations, so that, for example, carrying 4 lbs for 200 seconds is equivalent in terms of fatigue impact to carrying 10 lbs for 4 seconds. You then focus on the jobs with highest TVAL ratings and improve these jobs to reduce it.

Once you know which job to focus on, you record it on video and review it with the operator to identify ways to make it easier or to offload parts of it to others with lighter burdens. If the job involves interactions between operators and machines, you analyze with with a work combination chart to improve task sequencing and identify tasks within the job that need better tooling or a better work station layout.

The-Most-Complicated-Formula
Feb 16 2012

Safety Stocks: Beware of Formulas

A formula you find in a book or learn in school is always tempting. It is a “standard.” If you follow it, others are less likely to challenge your results. These results, however, may be worthless unless you take a few precautions. Following are a few guidelines:

  1. Don’t use a formula you know nothing about. Its validity depends on assumptions that may or may not be satisfied. You don’t need to know how to prove the formula, but you need to know its range of applicability.
  2. Examine your data. Don’t just assume they meet the requirements. Examine their summary stats, check for the presence of outliers, generate histograms, scatter plots, time series, etc.
  3. Don’t make up missing data. If you are missing the data you need to estimate a parameter, find what you can infer about the situation from other parameters, by other methods. Do not plug in arbitrary values.
  4. Make your Excel formulas less prone to error by using named ranges rather than cell coordinates. If a formula is even slightly complicated, referring to variables by names like “mean” or “sigma” makes formulas easier to proof-read than with names like “AJ” or “AK.”

The safety stock formula for the reorder point method

Safety stock is a case in point. The literature gives you a formula that is supposed to allow you to set up reorder point loops with just the minimum amount of safety needed to prevent shortages under certain conditions of variability in both your consumption rate and your replenishment lead time. It is a beautiful application of 19th century mathematics but I have never seen it successfully used in manufacturing.

Let us look more closely at what it is so you can judge whether you would want to rely on it. Figure 1 shows you a model of the stock over time when you use the Reorder Point method and both consumption and replenishment lead time vary according to a normal distribution. The amount in stock when the reorder point is crossed should be just sufficient to cover your needs until the replenishment arrives. But since both replenishment lead time and demand vary, you need some safety stock to protect against shortages.

Figure 1. The reorder point inventory model

If your demand is the sum of small quantities from a large number of agents, such as sugar purchases by retail customers in a supermarket, then the demand model makes sense. In  a manufacturing context, there are many situations in which it doesn’t. If you produce in batches, then the demand for a component item will be lumpy: it will be either the quantity required for a batch or nothing. If you use heijunka, it will be so close to constant that you don’t need to worry about its variations.

What about replenishment lead times? If in-plant transportation is by forklifts dispatched like taxis, replenishment lead times cannot be  consistent. On the other hand, if it takes the form of periodic milk runs, then replenishment lead times are fixed at the milk run period or small multiples of it. With external suppliers, the replenishment lead times are much longer, and cannot be controlled as tightly as within the plant, and a safety stock is usually needed.

Let us assume that all the conditions shown in Figure 1 are met. Then there is a formula for calculating safety stock that you can find on Wikipedia or in David Simchi-Levy’s Designing and Managing the Supply Chain (pp. 53-54).  Remember that it is only valid for the Reorder Point method and that it is based on standard deviations of demand and lead time that are not accessible for future operations and rarely easy to estimate on past operations. The formula is as follows:

Where:

  • S is the safety stock you need.
  • C  is a coefficient set to guarantee that the probability of a stockout is small enough. You can think of it a number of sigmas above the mean item demand needed to protect you against shortages, while the other factor is the corresponding sigma. In terms of Excel built-in functions, it is given by:

C = NORMSINV(Service level)

Service level    C
90.0% 1.28
95.0% 1.64
99.0% 2.33
99.9% 3.09
  • μL and σL are the mean and standard deviations of the lead time.
  • μD and σD are the mean and standard deviation of the demand per unit time, so that the demand for a period of length T has a mean of μD xT and a standard deviation of σDx √T

Case study: Misapplication of the safety stock formula

This formula is occasionally discussed in Manufacturing or Supply Chain Management discussion groups, but I have only ever seen one attempt to use it,  and it was a failure. It was for the supply of components to a factory, and 14 monthly values were available for demand, but only an average for lead times.

The first problem was the distribution of the demand, for which 14 monthly values were available. This is too few for a histogram, but you could plot their cumulative distribution and compare it with that of a normal distribution with the same mean and standard deviation, as in Figure 2. You can tell visually that the actual distribution is much more concentrated in the center than the normal model, which is anything but an obvious fit.  Ignoring such objections, the analyst proceeded to generate a spreadsheet.

Figure 2. Actual versus normal cumulative distribution

The second problem is that he entered the formula incorrectly, which was not easy to see, because of the way it was written in Excel.  The formula in the spreadsheet was as follows:

C*SQRT((AJ4*AL4^2)+(AI4^2*AM4^2))

then, looking at the spreadsheet columns, you found that they were used as follows:

  • AJ  for Standard Deviation of Daily Demand, and
  • AL for Average Replenishment time.

And therefore the first term under the square root sign was σDxμL2 instead of μLxσD2.

The third problem was that the formula requires estimates of standard deviations for both consumption and replenishment lead times, but not data was available on the latter. To make the formula produce numbers, the standard deviations of replenishment lead times was arbitrarily assumed to be 20% of the average.

For all of these reasons, the calculated safety stock values made no sense, but nobody noticed. They caused no shortage, and the “scientific” formula proved that they were the minimum prudent level to maintain.

Sizing safety stocks in practice

There is no universal formula to determine an optimal size of safety stocks. What can often be done is to simulate the operation of a particular replenishment loop under specified rules. For a simulation of a Kanban loop using Excel, see Lean Logistics, pp. 208-213.

No calculation or simulation, however, is a substitute for keeping an eye on what actually happens on the shop floor during production. One approach is to separate the safety stock physically from the regular, operational stock and monitor how often you have to dig into it. If, say, six months go by without you ever needing it, you are probably keeping too much and you cut it in half. With a Kanban loop, you tentatively remove a card from circulation. If no shortage results, then the card was unnecessary. If a shortage occurs, you return the card and look for an opportunity to improve the process so that the card can be removed.

TaktTimesHomepageLogo
Feb 7 2012

Takt times and falling sales: How to Respond?

Question from Jean-Baptiste Bouthillon on The Lean Edge:

We have all learned that overproduction is muda, and that production must follow the takt of customer demand.
Is there a lean way of dealing with falling sales ? Should we just adjust production to customer takt time or stabilize sales by giving rebates ?
Is it important to level sales and give some stability to production or should we just adjust the production takt time ?

My response:

You question implies that takt time is only a function of customer demand. It is not. When you calculate it, you divide your production time by the demand, which means that it is as much a function of how long you decide to work as of how much you have to produce. Without any change in customer demand, you double the takt time by working two shifts instead of one.

The takt time of a production line is the time that elapses between two consecutive unit completions when the line runs. It is not the rate at which customer orders arrive.

So how do you respond to falling sales?

You have to distinguish between fluctuations in sales, for which you should not change the pace of production, and major changes, for which you should.

Once you have set up a large assembly line to work at a takt time of 57 seconds, changing it to 60 seconds is a major effort, involving the balancing of tasks among stations and adjustments in part supplies. In car assembly, unless you are hit by something like the Fukushima earthquake, you don’t do it more than once in four months, even if you are Toyota. During this period, you use heijunka to respond to fluctuations in mix, and adjust overtime for fluctuations in total volume.

If you have a major downturn, you have to reduce production, and the challenge then is to do it without going bankrupt while retaining the work force you spent so much time and effort developing.
It is in such times that having your money tied up in inventory can bankrupt you. When the recession hit in 2008, management in manufacturing companies suddenly took an interest in working capital, but it was too late. Downturns come brutally, and it is when they occur that you must be ready.

Keeping your work force intact and prepared for the next upturn is just as essential. So you stop using temps, cut all overtime, go on four-day weeks, or three-day weeks, and use the available time to solve nagging engineering problems, experiment with new technology, etc. I remember an auto parts plant in Japan, in which recession time had been used by a team to build in-house a pick-to-light system with their own AGV out of Creform. Even though they did not explain it, you could tell that they would know exactly what to require from vendors and how to deploy this technology when the upturn came.

spc1
Dec 27 2011

Is SPC obsolete?

In the broadest sense, Statistical Process Control (SPC) is the application of statistical tools to characteristics of materials in order to achieve and maintain process capability. In this broad sense, you couldn’t say that it is obsolete, but common usage is more restrictive. The semiconductor process engineers who apply statistical design of experiments (DOE) to the same goals don’t describe what they do as SPC. When manufacturing professionals talk about SPC, they usually mean Control Charts, Histograms, Scatter Plots, and other techniques dating back from the 1920s to World War II, and this body of knowledge in the 21st century is definitely obsolete.

Tools like Control Charts or Binomial Probability Paper have impressive theoretical foundations and are designed to work around the information technology of the 1920s. Data was recorded on paper spreadsheets, you looked up statistical parameters in books of tables, and computed with slide rules, adding machines or, in some parts of Asia, abacuses (See Figure 1). In Control Charts, for example, using ranges instead of standard deviations was a way to simplify calculations. These clever tricks addressed issues we no longer have.

Figure 1. Information technology in the 1920s

Another consideration is the manufacturing technology for which process capability needs to be achieved. Shewhart developed control charts at Western Electric, AT&T’s manufacturing arm and the high technology of the 1920s. The number of critical parameters and the tolerance requirements of their products have no common measure with those of their descendants in 21st century electronics. For integrated circuits in particular, the key parameters cannot be measured until testing at the end of a process that takes weeks and hundreds of operations, and the root causes of problems are often more complex interactions between features built at multiple operations than can be understood with the tools of SPC. In addition, the quantity of data generated is much larger than anything the SPC techniques were meant to handle. If you capture 140 parameters per chip, on 400 chips/wafer and 500 wafers/day, that is 28,000,000 measurements per day. SPC dealt with a trickle of data; in current electronics manufacturing, it comes out of a fire hose, and this is still nothing compared to the daily terabytes generated in e-commerce or internet search  (See Figure 2).

Figure 2. Data, from trickle to flood, 1920 to 2011

What about mature industries? SPC is a form of supervisory control. It is not about telling machines what to do and making sure they do it, but about checking that the output is as expected, detecting deviations or drifts, and triggering human intervention before these anomalies have a chance to damage products. Since the 1920s, however, lower-level controls embedded in the machines have improved enough to make control charts redundant. The SPC literature recommends measurements over go/no-go checking, because measurements provide richer information, but the tables are turned once process capability is no longer the issue. The quality problems in machining or fabrication today are generated by discrete events like tool breakage or human error, including picking wrong parts, mistyping machine settings or selecting the wrong process program. The challenge is to detect these incidents and react promptly, and, for this purpose, go/no-go checking with special-purpose gauges is faster and better than taking measurements.

In a nutshell, SPC is yesterday’s statistical technology to solve the problems of yesterday’s manufacturing. It doesn’t have the power to address the problems of today’s high technlogy, and it is unnecessary in mature industries. The reason it is not completely dead is that it has found its way into standards that customers impose on their suppliers, even when they don’t comply themselves. This is why you still see Control Charts posted on hallway walls in so many plants.

But SPC has left a legacy. In many ways,  Six Sigma is SPC 2.0. It has the same goals, with more modern tools and a different implementation approach to address the challenge of bringing statistical thinking to the shop floor. That TV journalists describe all changes as “significant” reveals how far the vocabulary of statistics has spread; that they use it without qualifiers shows that they don’t know what it means. They might argue that levels of significance would take too long to explain in a newscast, but, if that were the concern, they could save air time by just saying “change.” In fact, they are just using the word to add weight to make the change sound more, well, significant.

In factories, the promoters of SPC, over decades, have not succeeded in getting basic statistical concepts understood in factories. Even in plants that claimed to practice “standard SPC,” I have seen technicians arbitrarily picking parts here and there in a bin and describing it as “random sampling.” When asking why Shewhart used averages rather than individual measurements on X-bar charts, I have yet to hear anyone answer that averages follow a Bell-shaped distribution even when individual measurements don’t. I have also seen software “solutions” that checked individual measurements against control limits set for averages…

I believe the Black Belt concept in Six Sigma was intended as a solution to this problem. The idea was to give solid statistical training to 1% of the work force and let them be a resource for the remaining 99%. The Black Belts were not expected to be statisticians at the level of academic specialists, but process engineers with enough knowledge of modern statistics to be effective in achieving process capability where it is a challenge.

I'm six sigma - I'm Lean
Dec 9 2011

Six Sigma R.I.P.

If you google Six Sigma, you get the impression that it is a going concern, with all sorts of organizations offering training and consulting on how to implement it. If you dig just a bit deeper, you run into a Business Week article from June 11, 2007 entitled Six Sigma: So Yesterday? It explained how the best known Six Sigma icons, like GE, 3M, Home Depot, or Motorola were “dialing it back.” Whatever this may mean, it is difficult to imagine ambitious employees in a company showing enthusiasm for a program that is being “dialed back.”

The same article attributes the following statement to GE’s former CEO Jack Welch about Six Sigma: “Even if the concept is applied in areas where perhaps it shouldn’t be, it’ll be worth it in the long run.” It makes you wonder how he would have liked to work in such an area, with management knowingly pressuring him to implement an irrelevant method.

Now that the Six Sigma craze is over, there is no much merit in criticizing it. Ever since I was first exposed to it in the 1990s, I have perceived it as a welcome update of the now 90-year-old tools of Statistical Process Control (SPC), useful in industries where, if your process is mature, your product is obsolete. This applies in semiconductors and other high-technology manufacturing sectors, but not in mature sectors like automotive.

It never struck me as having the potential to be a revolution in business or comparable in scope and impact to Lean. Saying so 10 years ago made many people angry but I did worse: I put in writing, in an article entitled Six Sigma and Lean Manufacturing that was published by the SME in a Six Sigma newsletter in July, 2002.

If you google Motorola +Six-Sigma, you learn that Motorola no longer teaches Six Sigma business improvement. Given that Motorola is where Six Sigma was invented, the equivalent would be for Toyota to dump Lean. Maybe it is time to dial down the Six Sigma training programs.

kanbancardfromliker
Nov 29 2011

More about Kanbans, and What it Takes to Use Them.

Ever since the world outside of Toyota started noticing its production system in the late 1970s, the Kanban system has received a disproportionate amount of attention compared to other features. It does not mean, however, that it has been accurately implemented in many of the factories that claim to have done it.  To anyone who cared to study it, details have been available in English at least since Robert Hall’s Zero Inventories (1983), the JMA’s Kanban, Just-In-Time at Toyota (1985), Yasuhiro Monden’s Toyota Production System (1993), or an updated treatment in Lean Logistics (2005).

Pressured to implement Kanbans by executives to whom it was little more than a buzzword, many manufacturing professionals found it more expedient to take old, familiar approaches like the two-bin system or reorder-point and call them Kanban. One such system implementing reorder-point through cards placed on a board has become so popular in France that I suggested calling it “French Kanban.” As can be seen in Figure 1, each column on the board is a mirror of the inventory level for an item. Each pocket filled by a card corresponds to an empty slot in stores, so that the remaining amount is visually indicated by the empty pockets on top. The reorder point is crossed when the cards reach the red zone.

Figure 1. French Kanban: Reorder-Point with Cards

Meanwhile, a few academics like J.T. Black at Auburn University or Robert Hall at Indiana University took the trouble to thoroughly investigate the Toyota system as a whole and the Kanban system in particular, but most of their colleagues didn’t, preferring a simplistic rendition of the Kanban system that made their own ideas stand out by contrast. In this context, insisting on the genuine Kanban system is perceived as nitpicking, because the differences are not in the big idea but in the details. You can easily dismiss these details as insignificant until you consider their cumulative effect on thousands of shop floor transactions every day.

Here are two examples, found today in a blog post:

  1. A common misconception is that you pull a Kanban from a bin when it is empty. If this were true, you would just be using a card to implement the Two-Bin system. The Kanban is not pulled when the bin is empty but when you withdraw the first part from the bin, to allow the bin to cover consumption during the replenishment lead time.
  2. Another in the same post was that the eKanban system did not involve physical cards. It is conceivable that, in the future, goods in transit will only be identified by RFID tags, but it is not the state of the art. They still need some form of human-readable identification and routing, for which a purely electronic system would require some kind of screen on each container. In fact, the electronic signal is used only in the return part of the loop, to eliminate the labor-intensive, slow and error-prone handling of unattached cards. On the supplier side, you print single-use cards that are attached to bins for transfer to the customer. When you detach the car on the customer side, you scan its barcode, and this triggers the electronic replenishment signal.

When evaluating or learning a tool like the Kanban system, you have to consider the following:

  1. The objects. They may be cards carrying specific data, bins of particular sizes and configurations, electronic messages of a given structure, … This is what we have to play with. Their physical nature makes a difference, not in a philosophical way but in basic, practical ways. For example, cards can be shuffled and posted on boards but bins cannot. When you send a card, you no longer have it, but when you send an electronic message, you still do. With the former, you have to make sure it doesn’t lose its way; with the latter, that it isn’t accidentally sent multiple times.
  2. The rules. These are protocols for users to follow. They specify who is allowed or required to do what to which objects when. In the Kanban system, the rules say who can issue new Kanbans or remove them from circulation, who attaches Kanbans to bins and detaches them, and what events trigger these actions. The rules give the objects meaning, as the rules of poker do to a deck of cards.
  3. The mapping to reality. This is what happens to materials and goods  in production and logistics when people follow the rules. When applied rigorously in the right context, the Kanban system tells production operators and materials handlers exactly what they should work on. Unlike the traditional dispatch lists, instructions in the form of Kanbans leave no ambiguity and require no judgement call by the leader or supervisor.

Within its range of applicability, the Kanban system is both simple enough for people to apply and sophisticated enough to get the job done. This is a tall order, and we should not underestimate what it takes.

Even in Japanese, the word Kanban has many different meanings, the most common being a sign advertising a store on the street, as you can see by searching Google images for “看板” (Kanban). Figure 2 shows, on a sidewalk,  the Kanban of a beauty salon located on the 2nd floor of the building.

Figure 2. A Kanban in everyday Japanese

Shape sorter
Nov 22 2011

Key details on Poka-Yoke/Mistake-Proofing

The following are  my inputs to a discussion on AME’s LinkedIn group initiated last August by Xola Qhogwana, which also  included contributions from Steve Bathe, Richard Foster, Karen Wilhelm, Steven Wade, Wesley Bushby, Ron Turkett, and Trevor Krawchyk.

When to use Poka-Yoke

Poka-Yokes prevent human error, and are therefore relevant when and only when human error is the main cause of your quality problems.

If you have process capability issues, focus on resolving them, not on preventing human error. What you need is deep understanding of your technology combined with statistical tools to enable your process to consistently hold tolerances.

If your process is capable but you are still producing in batches, focus on converting to flow to prevent your defectives being buried in WIP. Your problem is that it takes you too long to detect problems, not human error.

If your process is capable and you practice one-piece flow, then the defects you still produce are due to human error. At this point, and not before, Poka-Yoke is the relevant technique.

For details, see When to use statistics, one-piece flow, or mistake-proofing to improve quality.

Poka-Yokes do not require extensive a-priori analysis

Poka-Yokes are usually small devices, such as a permanent magnet to suck up a panel already containing a metal bracket, or a hole in a container to prevent overfill.

Doing an FMEA do decide whether to design and implement a Poka-Yoke is more expensive than just doing it. If you sort of think a process might need a Poka-Yoke and you have an idea of what it might be, just go ahead, try it, and document it in your company-specific Poka-Yoke library to inspire others. Don’t over-analyze it upfront. On the other hand, if you are building a spacecraft, you should definitely do an FMEA.

If it adds labor, it’s not a Poka-Yoke

By definition, also a Poka-Yoke device adds no labor. Manually scanned barcodes on parts to validate picks, for example, do not qualify as a Poka-Yokes because they add labor. A barcode that is automatically read or an RFID tag , on the other hand, would qualify. A Poka-Yoke has to become part of the process in the long run. If you look at the old big red book of Poka-Yoke from Productivity Press, you will notice that none of the examples adds labor, and there is a reason: any device that adds labor is likely to be bypassed under pressure.

This even happens with safety. Take, for example, the traditional approach of requiring the pressing of two buttons to start a press. How many times to you see plants where one button is taped down so that you can start the press with just the other one? By contrast, safety light curtains add no labor, and are not bypassed.

Using bar codes reading for data acquisition effectively eliminates the errors due to keyboarding because it is faster. If it weren’t, operators would revert to keyboarding and typos would creep back in. This is exactly what you see happening after two or three failed attempts at scanning a code. A barcode on a workpiece that is automatically read can be a Poka-Yoke. The workpiece passes under a reader in the proper orientation and under good lighting conditions and the barcode is reliably read. Under these conditions, it can even drive the lighting of the proper bins in a pick-to-light system. It does not work as a Poka-Yoke s if an operator has to wave a bar code gun in front of a part for pick validation.

Just because you use a device with the intent of preventing mistakes doesn’t mean it works. You have to make sure it does, and not just at the time you implement it. If you don’t pay attention, Poka-Yokes tend to deteriorate and to be set aside, for example when new operators are assigned to the station.

Nov 21 2011

What is chaku chaku? definition and meaning

Via Scoop.it – Cellular manufacturing
The definition of chaku chaku in the online business dictionary is missing the concept of machines with automatic unloading and incorrectly states that the line must encompass the entire production process, which is not a requirement.

Via www.businessdictionary.com