Process data collected by blast-furnace sensors

Blast-furnace production, which concentrates enormous energy and technological flows of high-temperature and highly toxic materials, is classified as explosion- and fire-hazardous; accidents can cause severe consequences, resulting in significant material damage and multiple casualties. In hot-metal production, toxic combustible gases are both consumed and generated, making the provision of industrial safety a critical task. Timely identification of the furnace operating condition based on informative indicators obtained from real-time measurements of technological parameters, along with corresponding management decisions by process engineers, helps reduce the number of accidents, decrease the specific coke rate, and lower the probability of producing off-spec hot metal. The datasets include the following parameters: total and specific gas pressure differences, peripheral gas temperature, top gas temperature, CO2 content of top gas, H2 content of top gas, top gas pressure, ore-to-coke ratio, blast flowrate, blast oxygen flowrate, hot blast pressure, hot blast temperature, cold blast pressure, cold blast temperature, and Si content in hot metal. The silicon content in hot metal at the furnace taphole characterizes the energy efficiency of the process; therefore, forecasting this indicator is essential. The data are collected from 26 different sensors installed at various locations around the blast furnace. All the attributes, except one, are numeric, and the collected examples form a time series. This data can be applied to a wide range of analytical tasks, including regression, forecasting (prediction), filtering, smoothing, classification, and both time-series and stream-data analysis. The data are presented in two datasets: the first contains raw data, while the second includes data that have been validated for reliability and labelled into classes.

Related research article
V.B. Trofimov, Automated expert systems in blast-furnace process control. Metallurgist, 64 (2020) 3-12.