Nov 12 2017
Mark DeLuzio started a discussion on LinkedIn with the following question:
“My Sensei Mr. Nakao once told me: ‘The hardest thing to do in TPS is to create flow.’ What do you think about that?”
It started a spirited debate, with the following participants, in alphabetical order: Bruce Andersen, Rob Beesley , Vincent Bozzone, Mark DeLuzio, Michael Dunne, Okan Gurbuz, Shahrukh Irani, Kerry McPherson,, Gregoire Nleme, Okan Gurbuz, Egidijus Karitonis, Sunil Malagi, Paul Van Metre, Jerry O’Dwyer, John Peck, Luis Saenz, Ravi Vaidiswaran, Prasad Velaga, Raka Rao, Sandur Subramanyam, Mark Warren
Sourced through LinkedIn
The following is a digest of my own answers, collated before they vanish in the replies-of-replies bowels of LinkedIn.
Flow in assembly and fabrication
Shahrukh Irani asserted that flow, as understood at Toyota, is a concept that only applies to assembly lines. Flow and assembly, however, are distinct topics. There are assembly plants that don’t have flow, because they do fixed-station assembly or batch-and-queue, and there are fabrication processes that are organized to flow.
In the late 80s, I had the opportunity to work in a roller bearing plant operated by an American car company. It was laid out like a job-shop, with rows of identical machines so long that, standing by by the first one, you couldn’t see the last in the oil mist. It made 8 products, all in steady, high volumes.
The company was not considering any change in layout. Instead, they thought that Eli Goldratt’s OPT scheduling software was the solution. It didn’t pan out. Layouts are not a given. Often, they must be changed to achieve flow.
The job-shop layout is not intrinsically wrong. It’s fine if you have a demand for which it fits. Even then, however, you don’t have to lay out your machines in neat rows.
Machining Job Shops Within Manufacturing
Shahrukh Irani, Prasad Velaga, and Vincent Bozzone steered this discussion towards the needs of high-mix/low-volume machine shops with unstable demand, the clientele they are serving as consultants. It is, however, a tiny segment of the manufacturing business and shouldn’t hog the discussion, as it does in the pages of academic journals and in university curricula, where cell design and operation is considered too trivial a subject. Designing, implementing, staffing, and running competitive cells is not a trivial endeavor. It is worthy of attention and should be taught to students of industrial and manufacturing engineering.
Looking at the most recent Economic Census data, I don’t see any evidence that this job-shop niche is a large chunk of US manufacturing, either in terms of value added — its contribution to GDP — or employment. Dunn & Bradstreet defines Machine Shops as using machine tools to modify metal, plastic, and composite materials to produce finished products. According to them, the US machine shop industry includes about 20,000 companies with combined annual revenue of about $37 billion. The US industry is highly fragmented: the 50 largest companies generate about 10% of revenue.
According to the Bureau of Labor Statistics (BLS) 2016 data, US machine shops (NAICS 332710) employ 276,240. It works out just under 14 employees per shop and $1.85M/year in sales on the average. I don’t believe these numbers include the in-house machine shops of plants that make more complex products, like car engines. These can employ hundreds. I have seen one with >1,000 employees and >100 machines.
The Economic Census of 2015 showed that machine shops (NAICS 332710) accounted for about 1% of Manufacturing’s contribution to the US GDP, and 2% of US manufacturing employment. The machine shop is an icon of manufacturing with a tool cutting metal as the featured image of choice. The reality, however, is that machine shops are a small niche.
It certainly deserves support, but it shouldn’t be the center of attention in a discussion of manufacturing. What they do may be to other manufacturing activities as running a marathon is to 100m. It’s harder to run a marathon than 100m, but it’s no easier to win a 100m race than a marathon.
The top two markets for machine shops are automotive and aerospace. In automotive, you remove small amounts of metal from many pieces, and in aerospace, large amounts from few pieces but in neither case does the demand vary daily. Two other markets, construction and medical, also have repetitive demands. In the list, only general machining and die/mold qualify as producing a high-mix in low-volume with unstable demand.
In the absence of market survey data, I would bet on automotive and aerospace accounting for 50% to 60% of US sales, with another 20% to 30% for construction and medical, and the rest for job-shops. The remaining 99% of manufacturing plants compete at managing flow, and it’s no easier than scheduling a job shop.
Demand analysis versus Group Technology
In his first comment, Okan Gurbuz went straight from postulating “100,000 different part numbers…” to grouping them by feature similarity, which, as Mark Warren pointed out, is group technology. The problem in starting with group technology is that it ignores the distribution of the demand. I have never seen a plant were none of the products had a higher demand than others.
Before deploying group technology, you need to break your product mix into runners, repeaters, and strangers. Runners are individual products with sufficient demand to dedicate a production line to each; Repeaters are products that you can group into families to which you can dedicate a flexible line; the rest are Strangers. Group technology is useful with Repeaters.
It’s Not All About Scheduling
Based on Prasad Velaga‘s own words, his focus is on scheduling. When you consider a factory, it may be the elephant’s trunk but it is not the whole elephant. A complete approach must encompass (1) the engineering of the facilities, the equipment, the processes and the interactions between people and machines, (2) organization and people, (3) performance management, and (4) production control, including scheduling, and logistics.
While not a panacea, TPS/The Toyota Way is a complete approach. Six Sigma is not. Neither is TOC. I would consider an FMS of 5 identical machining center with a shuttle and its own supervisory control system making 200 different items at rates on the order of 30/month to be a High-Mix/Low-Volume system.
If, however, these items are grouped into sets each going into the same assembled unit for 7 different products with takt times ranging from 3 to 10 days, then it makes sense to organize production in the FMS to put out these sets at the corresponding regular intervals. My point is that it’s not the mix and volume that determines the relevance of takt time but the stability of the demand. The above example is a real one, from the commercial aircraft industry.
Repetitiveness as the essence of manufacturing
Repetitiveness, flow, using lines to produce a catalog of standard products, etc. is the essence of manufacturing, as opposed to handicraft. The village cobbler made custom shoes for everyone in the village, from start-to-finish one pair at a time; the shoe factory uses production lines to make models and sizes of shoes in a fixed set large enough to meet almost everybody’s needs. And yes, occasionally, someone needs custom shoes and you need the capability to make them, but you don’t organize the entire factory around this need.
The toys that will be sold this Christmas are already made and on their way from China, including a few hits that will be in short supply and many duds that will be discounted. Hits and duds, however, won’t be known until Thanksgiving, and wouldn’t be great for a factory in the US to turn out more of the hits between Thanksgiving and Christmas? Except that no one has found a way to build one. The demand exists for 3.5 weeks every year, for products unknown ahead of time and ranging from dolls to game consoles. However you deal with this level of uncertainty, it’s not manufacturing.
Mix And Volume Versus Stability
Whether you can organize production to flow not a question of size or mix but of stability. I don’t ever recall seeing a plant run as a flow shop when, based on its demand pattern, it should have been a job-shop. On the other hand, most traditional plants are laid out and run like job shops to meet a flow shop demand. Identical machines are in work centers because the A&E firms drawing the layouts have always done it this way.
They focus on facilities issues, not flows of materials or movements of people. And the machines are in neat rows because, 100 years ago, they used to receive power through belts from overhead shafts. These mistakes create the kind of problem Prasad was showing, and failing to take advantage of flow extends lead times, and tanks both productivity and quality. In working with machines, correcting these mistakes is the bulk of the work.
There is always a need for a job-shop work to make strangers.Spare parts for obsolete products are a major group of strangers. The other groups include R&D prototypes, pilot production runs, and sample quantities for OEM customers. Today’s stranger may be next year’s runner. While strangers should not hog management and engineering attention, they shouldn’t be ignored either.
Flow And The Sourcing Of Materials
The sourcing of materials has nothing to do with the use of takt time and heijunka in production. If you have to buy Titanium two years in advance to make sure you have it when you need it, it’s just a cost of doing business. It’s the job of Materials Management and their call. Whatever you have to do to obtain this material, Production happens once you have it. Takt time is a driver for the design and operation of production lines; heijunka, a short-term sequencing principle. They are both about what happens inside the plant, not supply chain management.T
Historically, the first plants set up to flow were not in assembly but in process industries, flour mills for example. If flow is so natural for assembly, why was the first large-scale assembly line built in 1913? Before then, products like Singer sewing machines were assembled by the millions without assembly lines.