Don’t Automate a Dysfunctional Process 2

Last time I told you about the foundry General Manager and his edict to automate.  He had demanded that a robot needed to be installed and operating in the foundry or the Engineering Department would be fired.  If you're thinking about automating anything, you need to see this.  Here is the conclusion of the story:

The General Manager and the Robot (concluded)

as told to Norm Howe.

We installed the robot and brought in our best sprayer to program it.  He did this by simply replicating his normal motions holding on to the robot arm with the computer in ‘record’ mode.  He didn’t seem to be quite at ease with the plan, but we assured him that it was all for the best and he would now be relieved of a dirty, smelly job. 

We started the line back up and waited for our perfect engine blocks to come rolling one by one off the line.  But instead we started getting bad blocks.  Incredibly, the scrap rate was worse than it had ever been.  We took the robot off line and convened the brain trust to try to make some sense out of this situation.  We were at our wits’ end.  Nobody had any idea what the problem was. 

Finally someone had a novel idea.  “Why don’t we ask the sprayers how they know when they’re spraying the resin on right?”  We all agreed that it was a good idea.  But the answer we heard made us even more depressed. 

The sprayers told us that they didn’t have any set spraying pattern.  They said that they adjust the pattern depending on how the resin ‘glistens’ as it accumulates on and then soaks into the sand.  We were incredulous. 

The sprayers were actually adjusting the process on their own in response to some unknown variability that they could observe.  How could we possibly program a computer to emulate that?  The answer, of course, is you can’t. 

We were defeated.  Now we had to admit there were going to be no easy solutions.  We had to ask the plant manager for more resources.  There was nothing for us but the hard work of digging into all the intricacies of the mold fabrication process. 

How did sand, resin, wood patterns, and the physical manipulation of the process itself interact with each other to give rise to a good quality engine block?  We resigned ourselves doggedly to slogging through the daily lab reports from the last year of production, calling vendors and dissecting their brains, plotting control charts, and correlating quality related outcomes with process inputs. 

Gradually a picture began to emerge.  It turned out that sand wasn’t sand, and resin wasn’t resin.  Variabilities in the incoming raw materials had a dramatic impact on the molding process.  We set tighter specifications on the sand and resin.  We convinced purchasing that limiting the number of suppliers would ultimately lower total cost.

In the end when we had eliminated the process variability coming from the raw materials, the resin spraying step became easy.  The robot sprayed the resin onto the sand perfectly and the resin soaked into the sand the right way every time.  Our scrap rate coming from the molding process dropped to virtually zero. 

As it turned out we no longer needed the most skilled sprayer to keep the scrap rate low.  Now that we had control over the critical process inputs, anyone could spray on the resin.  We really didn’t need the high maintenance robot either.  But we didn’t pull the robot out until after the General Manager had come through on his next tour.  As far as I know it’s still sitting idle behind the foundry.

I was dumbfounded after listening to this story from so long ago.  I realized that it has so many applications today:

• Any technology that involves converting a powder into a product
• Any process that involves manipulating physical materials
• Any process whether it be physical or simply information

The epiphany for me was, Don’t Try To Automate A Dysfunctional Process.  You only create errors faster!  First, fix the process.  There are probably unknown variabilities lurking inside the process.  You need to identify those sources of error and control them.  Then automate.

The second take-away was, Talk to the people who are operating your process now.  They have a unique understanding of the process.  But don’t necessarily expect them to be articulate.  You may have to translate their speech into engineering knowledge in order for it to be useful. 

Recognize that human beings have formidable talents at recognizing patterns.  The operators who run your dysfunctional process are probably compensating for process variabilities already and nobody knows it.