Robust Period Estimation of Automated Cutting Systems by Improved Autocorrelation & Linear Regression Techniques

Abstract

Condition monitoring is an important asset in the industry to improve the safety and efficiency of the production chain. However, in heavy machinery - such as edge trimmers in steel mills - it is often impractical and unsafe to install intrusive sensors to get the data needed for condition monitoring. Non-intrusive monitoring techniques based, e.g., on acoustic data captured by microphones placed in the vicinity of the assembly being monitored are attractive options. Our application deals with the acoustic monitoring of rotational blades cutting steel strips at high speeds. Knowing the correct period of the cutting process is important for quality evaluation purposes. We propose two novel robust methods to estimate the periodicity based on the audio captured by a microphone near the blades. One is an improved autocorrelation function and the other is based on linear regression, both using incorporating an novel test for the correctness of the estimated period. We compare our methods against the standard autocorrelation-based periodicity measurement techniques on real data recordings. The proposed method estimates the correct period about 87% of the time, compared to an accuracy of only 51% using standard periodicity measurement approaches.