Operating condition recognition in ball mill based on discriminant PLS

Operating condition recognition of a ball mill is an important part in grinding process. The load status can only be determined according to expert´s experiences. If common operating conditions (such as under-load, best-load, and over-load) are not monitored and handled promptly and accurately, the quality of the grinding product may deteriorate or even the grinding production may come to a stop. This paper estimates the power spectral density of the input vibration and acoustic signals using Welch´s averaged modified periodogram method of spectral estimation. Unsupervised Fuzzy C-Means classification and Partial Least Squares for Discrimination (DPLS) are used to build operating status model for the ball mill. Experimental results shown recognition capability is enhanced, the false alarming rate is decreased.