Millaray Curilem

Neural Networks and Support Vector Machine Models Applied to Energy Consumption Optimization in Semiautogeneous Grinding

Semiautogenous (SAG) mills for ore grinding are large energy consumption equipments. The SAG energy consumption is strongly related to the fill level of the mill. Hence, on-line information of the mill fill level is a relevant state variable to monitor and drive in SAG operations. Unfortunately, due to the prevailing conditions in a SAG mill, it is difficult to measure and represent from first principle model the state of the mill fill level. Alternative approaches to tackle this problem consist in designing appropriate datadriven models, such as Neural Networks (NN) and Support Vector Machine (SVM). In this paper, NN and a SVM (specifically a Least Square-SVM) are used as Nonlinear autoregressive with exogenous inputs (NARX) and Nonlinear autoregressive moving average with exogenous inputs (NARMAX) models for on-line estimation of the filling level of a SAG mill. Good performances of the developed models could allow implementation in SAG operation/control hence optimizing its energy consumption.

Comparing NARX and NARMAX models using ANN and SVM for cash demand forecasting for ATM

A comparative study between NARMAX and NARX models developed with ANN and SVM when used to forecast cash demand for ATMs is conducted. A simple methodology for developing SVM-NARMAX models is proposed. The best results were obtained with NARX-ANN models. In addition no significant differences were found between NARX and NARMAX for both ANN and SVM. Hence it seems advisable to choose simpler models, such as NARX and a user-friendly tool like ANN at least for this particular application.

Feature Analysis for the Classification of Volcanic Seismic Events Using Support Vector Machines

This paper shows a preliminary study to perform a pattern recognition process for seismic events of the Llaima volcano, one of the most active volcanoes in South America. 1622 classified events registered from the Llaima volcano were considered in this study, taken from 2009 to 2011. The events were divided in four classes: TREMOR (TR), LONG-PERIOD (LP), VOLCANO-TECTONICS (VT) and OTHERS (OT). All of them correspond to specific activities. TR and LP events, are related to magmatic fluid through the ducts: continuous flux correspond to TR and discrete flux to LP. VT events occurs when excess of the magmatic pressure provides enough energy for rock failure. The group of OT contains events not related to the three first volcanic classes. Many features extracted from de amplitude, the frequency and the phase of the events were used to characterize the different classes. A classifier step based on Support Vector Machines was implemented to evaluate the contribution of each feature to the classification. The paper shows the results of this process and gives insights for future works.

Non-invasive Intracranial Pressure estimation using Support Vector Machine

Intracranial Pressure (ICP) measurements are of great importance for the diagnosis, monitoring and treatment of many vascular brain disturbances. The standard measurement of the ICP is performed invasively by the perforation of the cranial scalp in the presence of traumatic brain injury (TBI). Measuring the ICP in a noninvasive way is relevant for a great number of pathologies where the invasive measurement represents a high risk. The method proposed in this paper uses the Arterial Blood Pressure (ABP) and the Cerebral Blood Flow Velocity (CBFV) - which may be obtained by means of non-invasive methods - to estimate the ICP. A non-linear Support Vector Machine was used and reached a low error between the real ICP signal and the estimated one, allowing an on-line implementation of the ICP estimation, with an adequate temporal resolution.

Comparison of Neural Networks and Support Vector Machine Dynamic Models for State Estimation in Semiautogenous Mills

Development of performant state estimators for industrial processes like copper extraction is a hard and relevant task because of the difficulties to directly measure those variables on-line. In this paper a comparison between a dynamic NARX-type neural network model and a support vector machine (SVM) model with external recurrences for estimating the filling level of the mill for a semiautogenous ore grinding process is performed. The results show the advantages of SVM modeling, especially concerning Model Predictive Output estimations of the state variable (MSE < 1.0), which would favor its application to industrial scale processes.

Improving the Classification of Volcanic Seismic Events Extracting New Seismic and Speech Features

This paper presents a study on features extracted from the seismic and speech domains that were used to classify four groups of seismic events of the Llaima volcano, located in the Araucania Region of Chile. 63 features were extracted from 769 events that were labeled and segmented by experts. A feature selection process based on a genetic algorithm was implemented to select the best descriptors for the classifying structure formed by one SVM for each class. The process identified a few features for each class, and a performance that overcame the results of previous similar works, reaching over that 95% of exactitude and showing the importance of the feature selection process to improve classification. These are the newest results obtained from a technology transfer project in which advanced signal processing tools are being applied, in collaboration with the Southern Andes Volcano Observatory (OVDAS), to develop a support system for the monitoring of the Llaima volcano.

Evaluation of Llaima volcano activities for localization and classification of LP, VT and TR events

Abstract Evaluation of seismic signals is one of the most important research topics on Volcanology. Volcanoes have daily activity; therefore, high speed evaluation of recorded signals is a challenge for improving the study of the natural phenomena occurring inside these natural formations. The aim of this paper is the evaluation (denoising, localization and classification) and analysis of Llaima volcano activities, one of the most actives volcanoes in South America. Different already proposed methods, such as, Butterworth, Spectral Subtraction (SS) and Wiener Filter (WF) are compared to the proposed Modified Spectral Subtraction (MSS) and Modified Wiener Filter (MWF) to find the best method for denoising the volcano signals. Then, event localization based on received signals of volcano is performed. In this step, Time Delay Estimation (TDE)-based method is used on data acquired from 3 mechanical sensors located in the volcano area. The proposed method is used to estimate the area for event location. The proposed denoising methods make the starting point for the event more evident to increase the localization accuracy for events where the starting point is difficult to find. In the last step, a method based on the novel DNN technique is proposed to classify the three main events occurring in the Llaima volcano (TR (Tremor), LP (Long Period) and VT (Volcano Tectonic)).

NARX neural network model for predicting availability of a heavy duty mining equipment

In this work a neural network NARX model has been developed in order to predict availability of a heavy duty equipment of an important copper mining site in Chile. Four exogenous inputs have been considered (Number of Detentions, Mean Time to Repair, Mean Time between Failures and Use of Physical Availability) while Availability is the autoregressive variable. A 30 days moving average has been performed over the data. Results confirm that availability can be adequately multiple-step-ahead predicted using this arranged data and a NARX model including the 4 above mentioned variables as exogenous inputs.

Prediction of the criticality of a heavy duty mining equipment

In this paper, we are concerned by the improvement of the maintenance procedures in heavy mining equipment. Criticality is an important concept in the maintenance task because it points out which piece of equipment has a higher probability to fail, and how this failure may impact the general productive process. The paper proposes a definition of criticality of equipment, based on an experts experience and on historical data of seven asset management variables of a mining process. We propose a criticality semaphore and a method to classify the criticality of the next month and the subsequent month, based on the present values of the asset management variables. This is a preliminary work, however the results show a good prediction performance of criticality.

Predictive Models Applied to Heavy Duty Equipment Management

In this work we present the development of nonlinear autoregressive with exogenous inputs models to predict some relevant variables for asset management of heavy mining equipment, like Mean Time between Failures (MTBF), Mean Time to Repair (MTTR) and Availability is presented. The models were developed using support vector machine with historical data obtained on a daily basis during 2013 from one heavy mining equipment of an important copper mine site in Chile. One-step-ahead predictions of the predicted variables confirmed good performance of the dynamic models.