A Data Cleaning Approach for a Structural Health Monitoring System in a 75 MW Electric Arc Ferronickel Furnace

Abstract

Within a model of scientific and technical cooperation between the smelting company Cerro Matoso S.A. (CMSA) and the Universidad Nacional de Colombia (UNAL), a project was developed in order to take advantage of the data that were obtained from a sensor network in a ferronickel electric arc furnace at CMSA to improve the structural health monitoring process. Through this sensor network, online data are obtained on the temperature measurement along the refractory lining of the electric furnace, as well as heat fluxes and chemical characterization of the minerals on each stage of the process. These data are stored in a local database, which stores several years of historical data with valuable information for control and analysis purposes. These data reflect the behavior of the industrial process and can be used in the development of machine learning models to predict some of the electric arc furnace operation parameters, and thus improve the decision-making process. Currently, most of the data are analyzed by the experts of the structural control department, but, due to the large amount of data, the development of analytical tools is necessary to support their work. This paper proposes a data cleaning approach for improving data quality by creating a set of rules and filters based on both expert judgment and best practices in data quality. A statistical analysis was also carried out in order to detect variables with anomalies and outliers, which do not reflect real operation parameters and belong to anomalous data that should not be considered for modelling. With the proposed process, the quality of the data was improved and abnormal data were eliminated in order to consolidate a clean data set for later use in the development of machine learning models. This work contributes on understanding data cleansing rules that must be considered in order to reflect the real behavior of the electric furnace operation for further analysis and modeling tasks.

Attention-Based Deep Recurrent Neural Network to Forecast the Temperature Behavior of an Electric Arc Furnace Side-Wall

“…Several data preprocessing steps were performed to detect abnormal behavior in the used variables. These data preprocessing steps are listed below [32] After verifying the data preprocessing, it was concluded that the 49 variables used to train and test the models did not present abnormal behaviors.…”

Monitoring of the refractory lining in a shielded electric arc furnace: An online multitarget regression trees approach

“…Recently, different studies related to signal processing and data acquisition systems to monitor the state of furnaces and their lining inside CMSA have been developed. A work related to data preprocessing to handle a large amount of operation data in the CMSA furnaces presents a set of rules and filters in order to detect variables with anomalies and outliers 26 . Another work related to the development of a predictive temperature model based on deep learning time series reached low mean square error errors, predicting accurately temperatures in different sectors of the furnace 27 .…”

“…A work related to data preprocessing to handle a large amount of operation data in the CMSA furnaces presents a set of rules and filters in order to detect variables with anomalies and outliers. 26 Another work related to the development of a predictive temperature model based on deep learning time series reached low mean square error errors, predicting accurately temperatures in different sectors of the furnace. 27 In the process of shut down and initialize again the furnace, a contraction and expansion process of the hearth lining occurs; to measure the possible gap formation, an ultrasound-based method was deploy in the work of Tibaduiza et al in 2020.…”