Classification of Ball Mill Acoustic for Predictive Grinding using PCA on Time and Frequency Domain Data
Sonali Sen1, Arup Kumar Bhaumik2, Jaya Sil3

1Sonali Sen*, Department of Computer Science, St. Xavier’s College, Kolkata, India.
2Arup Kumar Bhaumik, Department of Computer Science, RCCIIT, Kolkata, India.
3Jaya Sil, Department of Computer Science, IIEST, Kolkata, India.
Manuscript received on May 06, 2020. | Revised Manuscript received on May 15, 2020. | Manuscript published on June 30, 2020. | PP: 1934-1941 | Volume-9 Issue-5, June 2020. | Retrieval Number: C5551029320/2020©BEIESP | DOI: 10.35940/ijeat.C5551.029320
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (

Abstract: The process of comminution is nondeterministic in nature, so deriving out a designated size range on crushing by fixing the parameters of the mill is not possible in mining industry. Loss of materials in huge amount is an obvious phenomenon due to under sizing of materials in transit. The aim of the paper is to predict the state of grinding and the particle size distribution (psd) during any desired stage of crushing in the ball mill. The acoustic sensors have been used to capture audio signals at different running conditions of the ball mill and analyzed to develop the prediction model. In the proposed work first Genetic Algorithm (GA) based predictive procedure is applied on the fragmented signal to extract the parameters of genetic operators and store them in a table. We also apply Gaussian Mixture Model (GMM) to obtain the psd of each fragment and Fuzzy C-means (FCM) clustering algorithm is employed to classify the distributed signal. The psd of each fragment has been stored in another table. The experiment is conducted for different raw materials with different size distribution. At run time the material grinding procedure is operated and stopped automatically based on the trained controlled parameters corresponding to the desired stage of grinding. The psd of experimental output is verified with the desired psd obtained during training and stored in the table. The proposed method exhibits significant improvement in prediction performance and outcomes are verified with the experimental results.
Keywords: Acoustic Signature, Ball Mill, Crossover and Mutation, Fuzzy C means, Gaussian Mixture Model