Wear of high-speed cutting tools can be monitored with high accuracy using a new acoustic sensor-based system
Manufacturers of precisely engineered products such as engine components are increasingly turning to online monitoring of high-speed fabrication tools to ensure product quality and reliability. Jun-Hong Zhou at A*STAR’s Singapore Institute of Manufacturing Technology and co-workers have now developed an analytical method that gives online systems new predictive abilities through the use of acoustic emission sensors—small, inexpensive devices that can assess tool quality by processing sound waves.
Acoustic sensors offer rapid and unobtrusive monitoring capabilities. Interpreting the complex signals they generate, however, has proved difficult. Features ranging from the burst-like pulses of strain energy that occur when a tool chips to the amplitude fluctuations of a continuous sound wave have been proposed as key indicators of tool wear. A software decision program is typically used to scan a mix of these signal patterns to determine if the machine needs to be shut down for maintenance. However, the use of too many parameters slows down the computation and compromises its accuracy, making online monitoring difficult to implement.
Zhou and her co-workers developed a new dominant-feature identification (DFI) algorithm to resolve these issues. In this approach, machine tool data are collected using embedded acoustic sensors and converted into a low-dimensional mathematical matrix. A procedure called ‘singular value decomposition’ is then applied to factor out the matrix into a series of linear approximations that reveal which elements dominate the acoustic signal. By eliminating the need for processing of the full data set, DFI can perform signal analysis 80% faster than typical computational methods.
Zhou notes that because most algorithms for predicting tool wear use static models based on one type of cutting environment, they fail to account for many real-world situations. To rectify this situation, the researchers’ approach instead involves the use of an ‘autoregressive moving average with exogenous input’ (ARAMX) model to update the decision software dynamically with current and past dominant features, and with previously predicted tool-wear values.
The experiments showed that the DFI-ARAMX method needed only four dominant features to predict the cutting tool lifetime for a ball nose cutters in a milling machine with a stunning 92.8% accuracy rate—significantly better than other processing systems that use up to 16 different elements. “DFI is very efficient for identifying key input parameters, and combining it with the ARAMX model provides accurate predictions for online machine condition monitoring,” says Zhou.
The research team plans to apply their acoustic-based system in the fabrication of function-critical devices such as aircraft gearboxes and wind turbine generators in the near future.
The A*STAR-affiliated researchers contributing to this research are from the Singapore Institute of Manufacturing Technology.
