Listening for tired machinery

Cheap sensors and sophisticated software keep expensive machines running smoothly

Software is making its way into places where it hasn’t usually been before, like the cutting surfaces of very fast, ultra-precise machine tools.

A high-speed milling machine can run at 42,000 RPM as it fabricates high-quality machine components within tolerances of a few microns. Excessive wear in that environment can lead to a failure that ruins an expensive part, but it’s difficult to use physical means to detect wear on cutting surfaces: human operators can’t see it and detailed microscopic inspections are costly. The result is that many operators simply replace parts on a pre-determined schedule — every two months, perhaps — that ends up being overly conservative.

The researchers’ milling machine, shown with sensors near the cutting device. (Source: X. Li, M.J. Er, H. Ge, O. P. Gan, S. Huang, L.Y. Zhai, S. Linn, Amin J. Torabi, “Adaptive Network Fuzzy Inference System and Support Vector Machine Learning for Tool Wear Estimation in High Speed Milling Processes,” Proceedings of the 38th Annual Conference of the IEEE Industrial Electronics Society, pp. 2809-2814, 2012.)

Enter software: in a paper delivered to the IEEE’s Industrial Electronics Society in Montreal last Thursday*, a group of researchers from Singapore propose a way to use low-cost sensors along with machine learning algorithms to accurately predict wear on machine parts — a technique that could cut costs for manufacturers by lengthening the lifespan of machine parts while avoiding failures.

The group’s demonstration is a promising illustration of the industrial Internet, which promises to bring more intelligence to machines by linking them to networks and integrating them with sophisticated software. Techniques from areas like machine learning, which can be computationally intensive, can thus be available in monitoring parts as small and common as cutting surfaces in milling machines.

“This is a simple optimization problem,” says Meng Joo Er, a professor at the Nanyang Technological University and an author of the paper. “But you’re talking about a very expensive piece of equipment working on a very expensive product. We have to be very careful.”

The cutting tool, enlarged under a microscope, after four cuts (Source: ibid.)

Er and his colleagues outfitted a high-speed computer-controlled milling machine with a handful of common sensors: accelerometers to measure vibrations, an acoustic emission sensor to measure stress waves and a dynamometer to measure cutting forces. The researchers then used data from these sensors along with measurements of wear taken through a microscope as a training set for two machine-learning algorithms (one ANFIS model and one SVM model). The researchers managed to predict wear at above 90% by the ANFIS method and 85% by the SVM method.

The same surface showing wear (Source: ibid.)

That’s accurate enough to find its way into application although, Er notes, industrial users will likely combine this sort of method with conventional fallbacks, and it will take longer to make its way into very-high-stakes fields like aviation. Computation time for running the trained model — now on the order of 25 seconds for the more accurate model and half a second for the less accurate — will also need to come down, and engineers will need to find ways to integrate these types of models into the industrial control systems that are ubiquitous in automated manufacturing.

Nevertheless, this is a good lens for peering into the future of connected machines: generate lots of data, let software swallow it up, and optimize away.

*Available for a fee from IEEE: X. Li, M.J. Er, H. Ge, O. P. Gan, S. Huang, L.Y. Zhai, S. Linn, Amin J. Torabi, “Adaptive Network Fuzzy Inference System and Support Vector Machine Learning for Tool Wear Estimation in High Speed Milling Processes,” Proceedings of the 38th Annual Conference of the IEEE Industrial Electronics Society, pp. 2809-2814, 2012.


This is a post in our industrial Internet series, an ongoing exploration of big machines and big data. The series is produced as part of a collaboration between O’Reilly and GE.

tags: , , , ,