In our previous blog, we wrote how a development process called the experimental approach can supercharge your prototyping productivity. The next step after successfully developing a high-quality prototype: moving it to industrial scale production. But it’s hard to meet the specifications on thousands of copies, so you will need to work towards operational excellence to maintain top quality. By significantly optimizing our production process, we significantly improved yields and decreased lead times. In this blog, we give insights on how to achieve just that. Hint: it has something to do with process mining.
Operational excellence in practice
What better way to make a term like operational excellence tangible than to provide a real example. A couple of years ago, at Veco, we introduced a quality and process control system called IRIS. This system enables us to determine where insufficiencies in the production process occur. All this qualitative data generated from IRIS is stored in a data lake, along with the corresponding timestamps.
I’ll go into timestamps — and how we use process mining to make sense of this data — later.
Here’s a visual representation of the process described above:
In IRIS, we register:
Which part is being processed at which production station
The rejection rate — including rejection reasons — at each production station
As you can see, this type of data generation is all about production quality: which parts of our process can we optimise to improve yield?Here’s an example of how the process works for the production of encoder discs:
Creating top quality encoder discs
When we create encoder discs through electroforming, we do it on a large scale. Up to hundreds of encoder discs are produced on one metal sheet — which is massive and highly productive. But not all of these 200 copies will be produced exactly according to the specifications of the design.
Even though a lot of the copies will be perfect, some of them won’t. And this is nothing to worry about; it's due to the chemical process.
But what is worrisome, is if you can't determine the production stage and root cause of the product failure. Without that knowledge, after all, how you can determine what part of your process to improve?
So that’s why we built IRIS. Employees use this database and reporting tool make a data entry during every phase of the manufacturing process.
The operator reports that during a certain phase, only 5 of the 6 plates (each with up to hundreds of discs) met the specifications.
In the next process phase, a quality control employee reported that out of the 1.000 remaining discs, 30 copies were rejected because of tolerances that deviated too far from the tolerance baseline, and 50 more were rejected because of an incorrect thickness of the disc.
All this data is then combined — together with corresponding timestamps — into the data lake, to see how it’s all related.
This type of qualitative data mining enables us to determine at what process phase an unwanted number of rejects is produced. It helps us discover trends and find the root cause of the problem, so that we can improve the process and create batches with an even higher success rate.
Now that we have described the generation of qualitative data, we want to go into the quantitative side of the process and how process mining comes into play.
How process mining comes into play
Every time an employee finishes a process step and logs it, a corresponding timestamp is created. The timestamp reveals when the process step has been taken, and how long the step took to complete.All these timestamps combined form an event log — a database of transactions that details how (fast) your process works.
Actually, “fast” shouldn’t be in parentheses. In some production processes, speed is the major driver for operational excellence. Why? Because in some processes, the final quality of the part is not known until inspection. In these cases, production lead times are absolutely essential because you want to get to quality control as quickly as possible.
Back to that database of transactions — the event log. Your event log is most likely big: you could easily have 10 or 100 thousands of timestamps in there. And it’s almost impossible to make sense of that data and distill any actionable insights from it with Excel. So that’s where process mining comes into play.
What is process mining?
For those who are not familiar with process mining, let’s first talk about the concept.Process mining means extracting process data from your event logs and visualize it in order to improve your process — for the purpose of improving lead times. Process mining takes the event log as input and provides a visual representation of your process (workflows) as output.
Here’s an example of such a visual, from Veco’s production site:
This visual shows what the process workflow at Veco’s production site looked like at 04-08-2016, 16:10.
As you can see, process mining provides you with a snapshot of your process at a given time and date. All of a sudden, you can pinpoint where your production process becomes clogged. Identifying and fixing these bottlenecks is what you want when you strive for operational excellence.
Process mining and quality control for process optimization
Process mining hasn’t been around for that long, especially in the manufacturing industry. But the more advanced process mining techniques available today enable manufacturers to applyadvanced analytics to their revenue-generating assets and processes in order to improve it.
Process mining — combined with a quality and process control system like IRIS — can make all the difference in the world when it comes to process optimization and improving yield and lead times. It’s a powerful way for manufacturers to achieve operational excellence and create more and better products as a result.
Curious how you can build even better and smaller products? Then looking into co-development services might be a good idea. Why? Because the biggest design challenge for a next-generation product or component often involves the smallest and most dimensionally precise part — and as an engineer, you shouldn't have to be on your own.