So I’m collecting data on a production line that I’ve been graphing as a simple time series, but I don’t think a time series is an accurate representation of the data, nor does it make it easy to derive the information I intended the data for. I’m looking for some advice on visualizing the data.
The Data:
I have a production line that consists of say 5 operations. I’m timing individual elements of that operation. Operator Load Time, Weld time, Part Transfer Time, Total Time, etc… for every operation every single cycle.
I’m currently graphing this in two ways.
First as as a time series for each step, x axis is time data was logged, and y axis is the step time. I don’t feel this is entirely accurate though because because the time between cycles logged is irrelevant. It does allow me to quickly see if I was welding for 30 seconds yesterday, but today it’s taking 37 seconds.
Second, I do a stacked bar graph for each operator. X-axis is operation, Y-axis is the stacked step times. This allows me to quickly see bottlenecks in the line.
Does anyone have any better ideas on how to represent this data? Key areas I want production leaders and managers to be able to figure out is: Is the line running slower today than usual and why?
What about using the Moving Analog Indicator?
What I’m thinking is you can set what a desired range for each time is. I.E. welding should take 25 to 35 seconds. Then set the overall range something like 0-2 minutes. The KPI would visually show if an executed operation is within that range or outside of it.
1 Like
I don’t know about better, but we use a table (I don’t necessarily care for it, but it’s what they wanted because it shows all stations at once) showing takt times for each station in a histogram form, Also have average and standard deviation columns. The standard deviation lets them know how consistent an operation is. Anyway, a histogram may be a good option for you.
I later added a takt-over-timebase graph to display slowdowns and speed ups.
Hope this sparks some ideas.
That’s a nice idea.
In PLC I specify a target time for each step. So I could do like ±10% is warning level, and ±20% is alarm level. Then having a moving indicator for each step to quickly see if a particular process in a operation is out of spec.
How do you handle those gaps in your times? When I graph it out, I end up with LONG straight lines where a line hasn’t been running or been on break.
Here is what I have now.
and
You mean to show the gaps? That’s a pen setting:
BTW, I really like the stack chart-- or Yamazumi chart, if you work for a Japanese-owned company like I do. 
Don’t show my bosses, though, or I’ll be up late-night coding again. Heh.
Thanks.
Yeah I did all this for fun on 1 production line, not very pretty code and not as “accurate” as I’d like it to be. Also uses a crap ton of PLC resources because I actually store a 25 part history for every single step and sort it and grab the median which is what is logged in ignition as a historical tag.
Boss asked me one day, “They are telling me OP 330 is running about 7 seconds slower, can you investigate?”. I was like oh yeah I’m cool and got this, pushed him aside on computer, pulled up one of those charts and pointed that the weld time increased 3 days ago, probably because someone added a double pass.
His response, “Hey that’s pretty neat…can you set that up on every line in plant?”
…1000+ robot facility…FML
So now I’m going back and rewriting it all for this line, pulling individual cycle data into database instead of a calculated median. Then I’ll do all the calculations on ignition side this time. Also standardizing the tags so I can create the graphs easier by just aliasing the Line and Operation.
I’m kind of worried though that these tables are going to grow out of control and break. I was using tag historian and just taking a snapshot of median every time it changed. Works fine I suppose, but produces long straight lines in chart when line doesn’t run for a while. Now I’m doing my own transaction group stuff.
You’ll probably need to ask how much would actually be useful.
I have my groups set to purge at 14 days. I had started at six months, but trimmed it back when I found out no one was looking further than a week into the past.
Sorry to resurrect an old thread, but I have just started looking in to doing something similar for our biggest assembly line to help with cycle time improvements and identifying bottle necks. I’m curious how yours ended up turning out!
