Water Detection Framework for Industrial Electric Arc Furnaces
MetadataShow full item record
This thesis develops a framework for water detection in an industrial electric arc furnace. The objective of the framework is to prevent water leak furnace explosions. This framework consists of a hybrid algorithm and a fault detection method. The hybrid algorithm consists of a mechanistic model and an empirical model. The hybrid algorithm and the fault detection method developed in this work are implemented on two industrial AC electric arc furnaces. The names of the plants and details of the operations were withheld for confidentiality reasons. The first problem treated in this work was collecting the required data. The data required for this work included EAF operational data and off-gas composition. Both melt-shops did not have off-gas analysis systems and hence an off-gas analyzer with an HMI/SCADA data collection system was installed for each furnace. EAF operational data was sent to the data HMI/SCADA collection system installed at each melt-shop. The off-gas compositions measured in both melt-shops were CO, CO2, O2, H2, N2, and H2O. Once all required data was collected then the framework to detect water was developed. In order to test the water detection framework developed in this work, industrial trials were completed where water was intentionally added into the furnace by increasing the electrode spray water flow rate. The mechanistic model is completed by performing a mass balance on the furnace. The model provides a boundary with upper and lower limits in real-time of the expected EAF off-gas water vapor leaving the furnace. The mechanistic model of the hybrid algorithm has shown in both industrial EAFs that it provides a valuable on-line monitoring tool to the operator on what boundary to expect for the off-gas water vapor. There are many input variables and historical heats in an EAF operation; hence before building the empirical predictive component of the hybrid algorithm, heats selection model and input variables selection model are constructed based on latent variable methods. The outcome of the heats selection model is heats with normal operation. The outcome of the input variables selection model is variables that are highly correlated with the off-gas water vapor. Once the heats and the input variables are selected, then the empirical predictive models are developed. Empirical predictive models investigated in this work are: statistical fingerprinting, artificial neural network, and multiway projection to latent structures. Robustness issues with each method are discussed and a performance comparison between the methods is presented. The last section of this thesis proposes a novel approach to detecting water leaks in the furnace.
Cite this version of the work
Hamzah Alshawarghi (2016). Water Detection Framework for Industrial Electric Arc Furnaces. UWSpace. http://hdl.handle.net/10012/11140