Victor Clark

Bringing online a million dollar system

Note: some information has been scrubbed from these charts, but the story told by the data is still intact.

After joining a startup specializing in hardware, some of my main responsibilities were to bring online a million dollar manufacturing system, to maintain system reliability, ensure system uptime, and keep executive management updated on the progress. Pulling upon my engineering background and R&D experience,  I relied heavily on data analysis and visualization to complete this task. By the project's completion, my efforts increased quarterly yield by 10% and quarterly throughput by 3.5x. Afterwards, I developed capacity planning for high-volume production based on scaling several modules with various throughputs. The result was increased daily throughput by 6x and mitigated system downtime by monitoring performance.

The system required a high purity environment and was equipped with the appropriate sensors - so my first order of business was to get a handle on the environment. The result, shown in Figure 1, was establishing a system log to determine maintenance intervals.

This system was also capable of high-volume production, so I partnered with a vendor capable of supplying the volume of parts in order to run at full capacity. Figure 2 shows the distribution of a critical parameter, D, of parts received from the vendor. With specification limits determined by senior members of the research team, a more in-depth analysis was required in order to steer the vendor's process within the spec limits.

Figure 3 displays the results of a multivariate analysis of including parameters. The analysis, combined with geometric knowledge of the parts suggested parameter B had a slightly larger effect than C. Parameter A is not related to parameter D; correlation does not always imply causation.

Figure 4 shows a more in-depth analysis of the relation between B and D. Parameter B was within specification limits, determined by a process upstream of the vendor. With this in mind, I requested that vendor slightly increase parameter C -- the result was a more robust product.

With the correct material in flow, the next task was to monitor yield. Figure 5 is the yield dashboard I established, maintained and would present quarterly. At a single glance, stakeholders can see trends for

• individual yield
• number of parts delivered
• cumulative yield
• average yield to date

This was instrumental in communicating status to executive management, as well as to provide a sense of accomplishment and visibility to operators.

From the beginning of the project, I established certain failure modes which were  captured throughout the project cycle. Twelve months into the project there was an event that resulted an abnormally high number of failures for a single process, captured by failure mode A. This event accounted for 16% of annual loss. However, dropping this event from tracked failure mode provides a more accurate picture of the system's performance over the course of the project, as shown by Figure 6, allowing for a more effective execution of the 80/20 rule.

Finally, to get a sense of the progress of the system performance over the course of the project, Paretos were generated by quarter, Figure 7.  Q4's performance compared against Q1: 

• 10% more yield
• 3.5x throughput
• Just two failure modes accounting for 80% of failures, instead of four
• Not shown here, but daily throughput increased by 6x

As the program lead, I know the answer may not always be immediate, and complete data may not always be conveniently available. For this project's success, it was critical to have a clear concept of the question that demanded an answer and to present data that delivered actionable items. Success arrived after crafting a meaningful story around the data that  wove its message to all stakeholders.