Failure sensitivity and similarity of process signals among multiple machine tools

Abstract

For a monitoring system to provide considerable performance, it usually requires machine- and process-specific information. This includes information about which process signals are sensitive to failures and which signal behavior indicates these failures. However, this information is mostly unavailable when monitoring the manufacturing of individual parts or small series. The transfer of process-specific information among similar machine tools can provide the required information, thereby improving monitoring performance. Nevertheless, no systematic research exists on what process signals are best suited for such an information transfer. This paper investigates a) whether information about the sensitivity of a signal to failures is transferrable among multiple machine tools and b) whether the behavior of these signals, modelled as probability distributions, is similar among multiple machine tools. Initially, a measure is introduced that quantifies the capability of a signal to separate two process conditions, the signal overlap factor SOF. It is then demonstrated how the SOF can be calculated for transient process conditions. The SOF is then empirically determined for a set of process signals for three different machine tools, individually, to assess failure-sensitivity of the signals for slot milling in steel. Additionally, the SOF is calculated for the union of the data of the machine tools to assess the similarity of signals among machine tools. The set of evaluated process signals includes process forces, the torque of the main spindle, and the torque and position control deviation of the feed axes. All machine tools were operated with identical instructions, tools, and materials. Bores were machined in workpieces to simulate material anomalies. Results suggest that low-pass filtered process forces or position control deviations, if sensitive to failure in a machine tool with linear direct drives, are also sensitive to failure in other machine tools. Also, low-pass filtered process forces were the most similar signals among the investigated machines. Possible causes that impair the similarity of signals among machine tools are discussed.