Feature Extraction of a Strain-Gauge Signal Using Wavelets for Monitoring of a Tumbling Mill

Abstract Techniques that measure the force acting on a lifter bar, when it hits the charge inside a tumbling mill, have got an increased interest because of its direct physical relation to the behavior of the grinding charge. The possibility to combine it with discrete element modeling (DEM), which gives an opportunity to visualize the charge motion, opens new possibilities to understand the complex phenomena that takes place inside a grinding mill. In this work, a method for the measurement of the apparent filling level in a pilot ball mill, the Metso Continuous Charge Measurement system (CCM) has been used. The technique uses a strain gauge sensor, mounted on a flat steel spring, which is fitted into a recess underneath a lifter bar. Deflection of the lifter bar during every mill revolution will then give rise to a characteristic signal pattern depending on different operating conditions. In this work, which should be treated as an introduction, we show how the discrete wavelet transform can be used in multivariate calibration. It will be shown that by using the fast wavelet transform on individual signals as a pre-processing method in regression modeling on CCM measurements, good compression is achieved with almost no loss of information. The predictive ability and diagnostics of the data compressed regression model is almost the same as for the uncompressed.