Americanmachinist 1961 87409am0602camp00000060046

Smart Tools for Machine Set Up

June 2, 2011
A new patent brings neural network technology to the shop floor
Cessna Aircraft has filed a patent for a programming process that uses a solid or mathematical model of a part and the machine tool functions to conduct component part machining in a virtual plane — and then analyzes the results to deliver optimized machine programming commands.

I love the elegance of math. I also realize not everyone shares my eccentricities. Indeed I am hoping readers got this far without being repulsed by the very word. The fact of the matter, however, is that even the most basic machining processes require a bit of math, at least until one becomes a skilled machinist when it largely becomes an art form. However, even the skilled machinist needs time to set up and adjust a machine to optimal performance, and time as they say is money. Companies are looking for and finding more and more ways to trim time and costs from every process.

One of the most recent successes in that search for time and cost savings has been patented by Cessna Aircraft Co., the designer and manufacturer of business jets in Wichita, Kan. Interestingly enough the patent isn't for a new thing, but rather for a new process that uses existing technologies. The patent bears the rather uninteresting title, “Method for Analyzing and Optimizing a Machining Process,” when in fact it is largely about the application of sophisticated mathematics and cutting-edge computing.

Cessna's new process calls for vibration analysis, finite element analysis (FEA) of the cutting tool and work material interaction, and then mechanistic modeling of the cutting process, using the results of the finite element analysis. Now here's the rub: None of this is done on a machine. At least not a machining machine. It's all done on a computer, and the results are exported to a machine tool, where the actual metal cutting happens.

The process uses a solid model, or mathematical model, of the part and the machine tools, and the cutting process happens in the virtual world. Specialized software for vibration analysis processes the effect of the machining, as does another FEA software package. Then, the outputs from these two pieces of software are passed to a mechanistic modeling software package, and this is where the math gets weird — as if vibration analysis and FEA weren't already mathematically challenging enough.

Mechanistic modeling uses the entire data set collected by the other two sets of software and built-in information about the physical and chemical properties of the materials under consideration and returns recommendations on how to increase the efficiency of the operation. "In contrast to FEM, [Finite Element Method] mechanistic modeling provides [that] a complete tool path can be analyzed for a series of operations by a cutter or set of cutters in sequential order." The recommendations are fed back into the virtual machining program in a loop until the process is optimized. Once the optimal parameters are determined they are stored in a database and eventually exported to a machine as the speeds, feeds, depths, temperatures and other information that the machine will use to complete the operation.

As an option, this entire set of software packages can have their outputs processed in a two-stage artificial neural network. An artificial neural network is yet another type of mathematical model that in theory mimics a real brain. It receives inputs, processes them, and returns results, or answers if you will, as outputs. But, an artificial neural network is capable of learning. In other words they are used in Artificial Intelligence (AI). This means that an artificial neural network can start the development of the process loop with a reasonably close set of inputs already, based on its previous experience with similar sets of parameters.

Interestingly, the patent actually calls out potential candidates for some of these software packages by name and manufacturer, such as the mechanistic modeling software. It doesn't, however, call out a specific artificial neural network. Commercially available artificial neural networks are available for Windows, Mac, and Unix. (No, I'm not making that up, you really can install an artificial neural network on your home computer. The problem is that artificial neural networks are designed to receive very specific kinds of data as inputs, so it is possible that a suitable one was not readily available and Cessna was compelled to develop their own.)

The result of all of this is that once a new part hits a machine the program setting for every machining operation is already determined and set in the machine. Program time on the machine settings drops to zero, and given the hourly cost of most modern machining centers the resulting cost savings, especially over time, is significant. It is reasonable to extrapolate that for the remainder of the 21st Century machinists will receive more sophisticated education, sometimes resembling or even overlapping that of their counterparts in engineering.

Paul DQ Campbell is an author and technology freelance writer. For more information or to contact Paul, visit