Pattern recognition, classification and diagnosis of acoustic emission signals in applications for mining

In this work, Acoustic Emission Technique (AET) and Pattern Recognition Technique (PRT) are used in combination for classification of AE signals obtained from two experimental applications in the mining sector. AE signals are collected and used to form features or parameters in order to identify AE events of interest. Signal processing in time domain and in frequency domain are applied to extract these features from the AE signals.The aim of this work is to identify and classify similar AE signals from experimental mining tests. It is achieved by means of using unsupervised and supervised methods of PRT. The main reason for performing these tests at a laboratory scale is to obtain a correlation of the AE signals with each mining process as well as to use the results as a guide for other specific mining applications.Beginning with the formulation of the problem and data acquisition, each step of the pattern recognition process is carried out for classification of the AE signals, namely pre-processing of the data, feature selection and extraction, unsupervised or supervised classification, assessment and interpretation of the results.In the pre-processing of the data, Standard Score Normalization is applied to the input data for further comparisons of AE features. For feature extraction, three algorithms are used, namely Principal Components Analysis (PCA), Linear Discrimination analysis (LDA), and Multidimensional Scaling (MDS). For supervised classification three methods are employed, these are linear Support Vector Machine, non-linear Support Vector Machine and Back-propagation Neural Network. These supervised classifiers are evaluated using classification accuracy. For unsupervised classification, also three classification algorithms are used, namely K-means Clustering, Fuzzy C-Means Clustering and Vector Quantization Clustering. These unsupervised classifiers are evaluated using similarities between clusters by Rand Index (RI).Two experimental applications are studied using the algorithms and the selected AE features. First, AET in a laboratory column flotation cell is used to monitor the bubble activity and bubble size, as a means of improving the efficiency of the column flotation process. Next, in rock cutting, AET and PRT are employed to identify the rocks being cut as a means of automation in the operation of underground mining and tunneling.These experiments were carried out in a laboratory scale, using AE wave measurements to analyze the processes. Pattern Recognition Technique is used in combination with classic and advanced signal processing techniques to characterize the collected AE signals.