General Electric adaptive weld control feedback timing failure

Same engagement as above, but specifically the technical finding — the adaptive controller's response lag (six cycles) exceeded the weld duration (four cycles). Tom presents this finding in front of GE Lynn's boss and the GE Research engineer who admits "we fixed that last week."

I go off to Japan, and Gordon and Carl, my students, are working on this, and I get back from Japan thirteen months later. It turns out that Gordon and Carl had figured out — they had sent me notes while I was over in Japan — that General Electric Research would not tell us anything about the system. We found out very quickly that General Electric Lynn had cut the million dollars a year of funding from General Electric Research for having spent seven million dollars and not having a product yet. They hired us as the whipping boys to punish General Electric Research. So Gordon and Carl had to go into this seven-million-dollar system and figure out the software on their own. They actually didn't even have access to it. All they could do was run experiments. They ran some experiments and did some tests, and they concluded that — even though this is 60-cycle current and you're supposed to control it in four cycles of current, one fifteenth of a second — the adaptive controller wasn't even kicking in until like six cycles after the weld was over. It was trying to adapt the weld.

GE Research Schenectady spent $7M over 5-6 years developing adaptive feedback control for spot welding combustor cans. Tom (with postdoc and grad student, while Tom was in Japan as ONR liaison) demonstrated the algorithm didn't start measuring until 0.2 sec into a process that completes in 0.1 sec — controlling after the fact. Tom was funded as a "whipping boy" to discipline GE Research; he received ~$40-50K against their >$1M/year.

I didn't know it at the time, but I was the whipping boy for General Electric Research in Schenectady, New York. They had been working for five or six years on SWAC — S-W-A-C, spot welding adaptive control. General Electric Research had spent seven million dollars developing this technology. While I was off in Japan, I had a postdoc and a graduate student working on this project, and they showed that this process was not working at all. GE Research in Schenectady had blown seven million dollars and had nothing to show for it. They were supposed to spend their money with GE Research, rather than going off and giving it to some stupid university. But the managers up at Lynn decided to cut out GE Research and give a portion of the money to me. They were giving over a million dollars a year to them, and they gave me like forty thousand dollars to do some work, fifty thousand — a lot less. But we're cheaper.