Listening to the End Mill: Predicting CNC Tool Wear With a Regional Machine Shop

The question

Redgranite Precision Components machines steel hydraulic manifold blocks on a cell of vertical machining centers. Like most shops, it changes its carbide end mills on a fixed count: after a set number of parts, the tool comes out whether it is worn or not. The count is conservative because a tool that breaks mid-cut can scrap a $300 part and sometimes damage the spindle. The company suspected it was throwing away tools with plenty of life left and asked the Discovery Center whether the machines could tell them when a tool was really done.

Dr. Everett Fontaine, professor of manufacturing and industrial engineering, who integrated robotic cells for an automation house before coming to Porter, agreed to lead it as a sponsored project in the Advanced Manufacturing Center.

The approach

The team instrumented two machines at Redgranite and one teaching machine in Founders Hall with:

• spindle-load readings taken directly from each machine’s controller;
• a small accelerometer on each spindle housing, sampled during every cut;
• a tool-microscope station where students photograph and measure the wear on each end mill’s cutting edge at set intervals.

The core of the work is a dataset of tools run to the end of their useful life under controlled conditions, mostly on the Founders Hall machine so that no customer parts were at risk. Students then trained models to estimate remaining tool life from the load and vibration signals, working with the Applied Artificial Intelligence Laboratory on methods. The rule from the start: a model has to be explainable to a machinist on second shift, or it does not ship.

Students on the team

• Logan Petersen, senior, manufacturing engineering — student lead and the main contact at Redgranite.
• Yasmin Haddad, senior, data science — signal processing and the remaining-life models.
• Cole Brandvik, CNC Machining certificate student — runs the tool-life trials and measures every insert and end mill under the microscope.
• Erin McKellar, junior, mechanical engineering — accelerometer mounting and data acquisition.
• Samuel Oyelowo, sophomore, manufacturing engineering — cost analysis and the shop-floor display.

Two of the five are co-op students who will spend spring semester at Redgranite. The team drives to the shop every other Tuesday to pull data and, more importantly, to talk with the machinists who run the cell. Early versions of the model flagged tools as worn whenever a new batch of steel blocks arrived from a different mill; it was a second-shift operator, not the data, who pointed out that the harder material simply loads the spindle more. Material lot is now an input to the model, and the operator is thanked by name in the technical report.

What they found so far

After 212 end mills run to end of life:

• Under the fixed-count rule, the shop was retiring tools with about 35 percent of their useful life remaining, on average.
• A model using spindle load and two vibration features estimated remaining life to within about 12 percent once a tool was past the first half of its life. Early in a tool’s life, the signal is too quiet to predict much.
• Rising vibration in one frequency band gave a usable warning, roughly 20 to 40 parts ahead, for 17 of the 19 tools that chipped or broke.
• In an eight-week pilot on one cell, Redgranite ran tools 22 percent longer with no increase in scrap. The company estimates savings of about $41,000 a year on that cell, most of it in tooling.

“The machinists already knew a worn tool sounds different. What we did was put a number on the sound, so the second-shift operator and the scheduler are looking at the same thing.”

— Logan Petersen, senior, manufacturing engineering

What’s next and who’s involved

Phase two extends the system to the rest of the cell and to a second tool type, and tests a simple traffic-light display at each machine. Students are also writing up the method so that smaller shops, including members of the Northwoods Manufacturing Alliance, can try it with nothing more than the data their controllers already record.

Partner and sponsor: Redgranite Precision Components, under an 18-month, $148,000 sponsored-research contract, with first right to hire the students who worked on it. Students usually arrive through MACH 2100 CNC Programming and Operation I or MFGE 3300 Quality Engineering.