Industry 4.0: Without Stable Processes, Nothing Works | Wiegand’s Watch

Michel Baudin‘s comments:

Wiegand’s main point is that it’s not worth networking all your machines just to find out how many of them are down on the average. In fact, for a manager to see this information on a screen in the office is not an improvement over seeing it as red Andon lights on the shop floor. This is especially true if it is expressed as OEE (Overall Equipment Effectiveness), an aggregate measure of performance that is not easy to interpret. In practice, to take any action, you end up having to break it down into its constituent factors of availability, performance, and quality.

Where Wiegand’s argument is less compelling is when he suggests that investing in advanced information systems is never a good idea when your challenge is stabilizing processes. While presented in generic terms, his example is about machines that get oily and dirty and produce “chips,” by which he means the bits of metal that come off when you turn, mill, drill or grind a piece of metal, not the electronic chips that power your smart phone. So he is really talking about machine shops and machine tools.

If you actually make electronic chips in a semiconductor wafer fabrication facility, you are using machines performing deposition, patterning, and doping processes based on solid state physics, loaded with sensors and driven by embedded controllers that are themselves powerful computers, networked with higher-level supervisory controllers and Manufacturing Execution Systems (MES). Together with the analytical tools attached to testers for finished chips, these systems are key to the stabilization of the process, known in the industry as “yield enhancement.”

The key reason why, contrary to machining, semiconductor wafer processes require these systems is that they never get a chance to mature. High-volume production starts at yields of 10% and, by the time you reach 60% or 70%, the chips you make with this process are obsolete and you need to start on the next generation. This is an example of a strategically important sector where the issues of process stability differ from machine shops. We must be wary of blanket statements on manufacturing in general.

Wiegand then evokes Lean Maintenance as an approach to both stabilizing processes and avoiding unplanned downtime. He describes this as focusing on identifying the root cause of instability on the bottleneck machines. Bottleneck machines, by definition, throttle production, but their problems may not be the easiest ones to solve for the available maintenance team, and focusing on them may lead to neglect other machines until they deteriorate enough to become bottlenecks themselves, which is not the desired outcome.

To eliminate unplanned downtime, Wiegand then suggests applying Failure-Mode-Effect-Analysis (FMEA) on the bottleneck machines to target modules or subsystems for redundancy, so that the stand-by unit kicks in when the current one fails, and allows the machine to keep operating until the next planned maintenance shutdown. This sounds very much like Reliability-Centered Maintenance (RCM), an approach developed at United Airlines in the late 1960s when it took delivery of the first Boeing 747s and applied since then to nuclear power and chemical plants as well as aircraft maintenance.

In the same spirit, in a machining center, putting the same tool in two different pockets lets you cut with it for twice the tool life before replacing it. In general, however, FMEA is the only tool from RCM that is not widely known in manufacturing, if not extensively used.

What I would understand under “Lean Maintenance” would instead start with the Autonomous Maintenance component of TPM, which involves delegating the simplest maintenance tasks, like switching lightbulbs, to production operators, thereby getting the small problems solved faster while freeing up technicians to address bigger challenges. Then you organize a group of maintenance first-responders with responsibility for all the equipment in a section of the plant, cross-trained to solve the most common mechanical, electrical, plumbing or control problems. The first responders are then backed up by specialists in one of these trades to address the more difficult problems.

In principle, information systems should substantially help make maintenance effective. However, the focus of the commercially available software, called Computerized Maintenance Management Systems (CMMS) or Enterprise Asset Management (EAM), to this day, appears to be on work order management, as opposed to the technical content of the maintenance work. In health care, your clinic’s information system holds your history of symptoms, diagnoses, treatments and outcomes, in a way that (1) helps your physician understand what is ailing you on your next visit, and (2) by aggregating patient data, identify which bugs are running around. It is focused on medical content, with administrative information on visit scheduling and billing as a by-product.

Maintenance technicians, by contrast, do not turn to their CMMS or EAM system to figure out what’s wrong with a machine. These systems are in the category of business, not engineering software. They put out performance metrics for managers, not diagnostic help for technicians. They may locate spare parts once a technician has identified a need, but are no help in relating error messages, gauge readings and direct observation to failure causes.

As a consequence, the administrative data is not automatically generated as a by-product of the work  itself, but instead requires transactions that add labor and are usually entered by a clerk at the end of the shift rather than in real time. This hurts the quality of the data, particularly regarding the use of technicians’ time. The maintenance department must account for the use of its technicians, and it is more a matter of allocating their time to various jobs than recording how much each job actually consumed.

I believe the reason software suppliers focus on administration is that these functions are generic to maintenance, and needed regardless of the nature of the object being maintained. This is why you see customer lists that include food processors, steel mills, research labs, and municipalities. Technical content, on the other hand, is specific. A system addressing the diagnosis of rolling mill failures would have a much smaller niche market. A generic system to diagnose equipment would have to be populated with specific equipment models for each application. It is more difficult to develop and more work to implement.

In maintenance, I agree with Wiegand that most companies “don’t reap what they have sown,” but I don’t think that lack of stability in the process is the reason. Maintenance information systems could be help stabilize processes, and the reason they don’t is their focus on work order management.