You can’t control your way out of poor mechanical design

Leland Teschler, Executive Editor
On Twitter @DW_LeeTeschler

With all the hoopla surrounding software startups, you might think the only innovative technology these days involves C++ programs and SIM cards. Worse, a lot of software-types seem to have the same opinion: A product that doesn’t depend on some kind of esoteric algorithm in software can’t be any good.

You aren’t alone if you sense this sort of arrogance coming from software engineers. “I have a LTeschlerTHhealthy respect for software people, but they don’t have the same respect for mechanical engineers,” says Dr. Nevan Hanumara, a research scientist at the Massachusetts Institute of Technology. “Software types think they can control their way out of any problem. I will tell them bluntly that a sloppy machine is going to be a complete pain to control. You can’t control your way out of poor mechanical design. An elegant machine will be beautiful to control from the perspective of software.”

Hanumara and fellow MIT professor Alexander Slocum know a lot about elegant machines. They teach, among other things, a professional development course in engineering design called Innovative Precision Product Design that conveys some of the same ideas made famous in MIT’s famed annual student competition of remotely operated robots, held annually as the culmination of an undergrad “Introduction to Design and Manufacturing” course.

It looks as though even manufacturers with a reputation for innovative software ought to take heed of the design principles that Slocum and Hanumara teach. Consider Tesla, for example. The car maker gets kudos for its software-control features and its automatic remote software updates. But Tesla is experiencing mounting mechanical problems and manufacturing defects, including brake troubles that led to a voluntary recall of 53,000 cars, another issue of exposed adhesive on roof liners, out-of-alignment trunk latches, and sunroof leaks. All these difficulties, of course, have roots in the car’s mechanical design.

If Tesla engineers sat in on the MIT design course, they might be told that the solution to such woes lies in adopting what Slocum and Hanumara call a deterministic design process. “The magic is how to get high performance and precision without paying buckets for it,” Hanumara says. “You can machine your way to any tolerance you want, but that is the expensive way to do it.”

The approach taken to tolerancing in the MIT class sounds enlightened. “”Life is a box of springs and everything is flexible,” says Hanumara. “You need to visualize how and where the structure will bend and then analyze around those points. Do it this way and you can make significant progress. You have to figure out where to put the tolerance stack-up so it doesn’t matter and where it might even be helpful. A key message is that not everything has to be difficult. Sometimes you need to step back and see if there is a better way to approach what you’re doing.”

The MIT precision product design course where Slocum and Hanumara teach these ideas lasts just five days. Though it is structured for practicing engineers and engineering managers, it sounds as though almost anybody can talk their way into it. Hanumara says there’s a lot of hands-on work involved as well as a sharing of ideas and case-history-style story telling.

Perhaps best of all, the textbook for the course is online and is free for anybody to see:


  1. Thomas Tachovsky says:

    Good to see that the more things change the more they stay the same. BTW,I now have a Tesla and have not experienced any problem. Maybe they heard you.I trust all is well. My best to Alex.

  2. I see things a lot differently. The MIT professors are talking mostly about quality control issues with Teslas. I/we see machines with real design flaws. We are often asked if our controller can meet some extreme specification. What the person asking the question doesn’t realize is that I am thinking ‘can he design a machine that can meet this same specification’. It seems like the mechanical guys are what to have a guarantee that our motion controller can fix mechanical design errors. Motion controllers are fast and have floating point that can provide much more precision than can be achieved with mechanics.

    My challenge to the MIT professors is to teach their mechanical engineering students to model their designs before sticking an unknowing PLC programmer with it. In 35+ years of control I have yet to see a mechanical system with a transfer function in the documentation. If these existed then tuning and control the system would be simple. From my point of view a ‘designed’ system should have a transfer function. In reality machines are evolved from previous designs or worse, kludged. Aircraft and space engineers design their machines but I have yet to see a ‘designed’ industrial machine with a transfer function..

    Also, I have yet to see a software ( control theory ) engineer that thinks he can control his way out of any problem. What usually happens is the a PLC programmer/system integrator is told he must control a machine or system. He is stuck with a poor design to control. I/we get a tech support call where the PLC programmer describes his problem. We tell him about how the machine must be fixed to work as intended. Then I hear the phrase I hate. “It is what it is”. Now he just wants help, a ‘get out of jail card’ so he can go home.

Speak Your Mind