Problem Statement

All of the significant production theories and programs (Deming, TQM, Lean, Goldratt, automation, etc.) agree on one thing. To do a job well in a repetitive business, you have to minimize process variation. Many of the systems and tools that companies use are designed to either keep process variation from happening or to tease it out of the system if it occurs. Yet, the struggle to make production consistent continues without victory. Some process variation remains and it raises the cost of production. In our view, one of the major reasons for the stubborn persistence of process variation is the fact that most front-line workers don’t really grasp what process variation is, or what it does. I would wager that, for as many SPC and Six-Sigma green belt courses as they may take, many don’t really get the “aha” moment.


The factory has installed a continuous, comprehensive video system to cover its operations. Some have wide area views, but others are HD cameras (using anything less is now silly) that are aimed at key parts of the process, including inside machines and on critical workstations. You can find and extract  video clips from any or all of these cameras, for any point in time in the recent past.


There are sections of the process where highly repetitive operations happen and these are intentionally covered by suitable cameras. With a little bit of video processing magic, you can fairly easily generate video clips like the following. In this clip, four different videos have been overlaid on one another, with each being mostly transparent. The timing of the video clips were adjusted so the background machine trays line up in the same place in every clip … so they seem in focus. Play the clip and watch the margins of the dough balls. Some are almost  in focus (very low variation), while others exhibit visible differences in placement.

This gives a crystal clear image of the variation in the dough depositing process. You don’t need to understand statistics or be an engineer to grasp the fact that this process could be a lot tighter. In fact, if there were a reasonably quick way to generate this type of video, you could whip them up every so often and see if variation is increasing, decreasing or moving around. Finally, note there is an overlaid scale. With that, you can even measure the amount and shape of the variation and generate objective data for analysis.

As it turns out, this type of analysis is pretty easy to do. You can do the overlays with most video editors, and you can add the scales and extract the analytical data if you apply a video performance analysis tool like Dartfish. After you have done this once or twice, it is pretty simple. The hard part is getting the relevant raw video to work on. If you have a video system for operations with continuous recording, that is already done.