Ore ball-milling process load identification method based on milling sound signal

Abstract

The invention discloses an ore ball-milling process load identification method based on a milling sound signal, and the method comprises the steps: firstly carrying out the preprocessing of a collected original signal, including the elimination of a DC component and the filtering, then proposing a geometric spectral subtraction inhibition noise signal based on the autoregression (AR) spectrum estimation, and improving the signal to noise ratio; secondly, decomposing the grinding sound signal into K intrinsic mode components IMF based on an ensemble empirical mode decomposition (EEMD) method, and selecting a mode component with high reliability by utilizing permutation entropy to reconstruct the grinding sound signal; calculating the box fractal dimension of the reconstructed signal under each load type, and taking the box fractal dimension as a final load classification basis; and finally, establishing a Bagging and extreme learning machine load identification model based on fuzzy C-means clustering to realize load identification. According to the method, the actual production load of a certain domestic multi-metal separation plant is recognized, and the result shows that the recognition model can accurately recognize the load states of different mills and has good generalization performance and recognition precision.

Description

technical field [0001] The invention relates to the technical field of ore grinding process detection, in particular to a load identification method for ore ball milling process based on grinding sound signals. Background technique [0002] Mill load refers to the instantaneous total load in the mill, including new ore feed, cycle load, water volume and medium load, etc. The mechanism of the grinding process of the mill is relatively complex, with the characteristics of large inertia and long time lag, which makes the optimal working point of the mill load time-varying. However, the internal state of the mill is a black box, and the mechanism is complex, so it is difficult to establish an accurate model for this process. There are often three states in the internal load state of the mill: underload, normal, and overload. When the mill is in the underload state, the efficiency of the mill cannot be fully utilized, and when the mill is in the overload state, there will be "sw...

AI Technical Summary

Problems solved by technology

There are also certain grinding sound analysis products on the market, but due to the different application scenarios of each dressing plant, the results are not satisfactory. Foreign products have disadvantages such as expensive prices and inconvenient maintenance, while it is difficult to provide effective and accurate monitoring products in China.

At this stage, most manufacturers still rely on experienced field experts to judge the load status of the mill by "listening". This method has strong real-time performance and can detect the load of the mill without damage, but artificial listening is highly subjective

Images