This Small Business Innovation Research Phase I project proposes to develop the Data-fusion Predictive Control for the Flaws in the Bulk of the Continuously Cast Products ("DPC") in which (a) various sensors are used to acquire surface conditions of the cast products in a steel mill, (b) a diagnostic module predicts whether the cast products meets quality requirements in both internal and surface conditions, and (c) a software application suggests corrective actions to enable reduction or elimination of defects. The DPC will be a product that is commercially viable and have high impact in the continuous casting, resulting in a new energy efficient control paradigm in the operations through improved yield, reduced material removal and enhanced direct charge. The current practice by continuous casters, which is the primary steel making process in the U.S., has room to improve for better efficiency and energy savings.
The boarder/commercial impact of this project will be in-line sensors; the DPC has the potential of over $10 million per annum per installation in yield improvement or energy savings, along with the savings of 130 million KWh of energy and 1.5 billion gallons of water reduction, as well as the reduction of 37,500 tons of CO2 emission. This project represents a unique multi-model data fusion (soft as well as hard sensors, hydrogenous data, in-line/off-line information) approach to controlling a highly stochastic and non-linear process. This predictive system approach will have wide applications to other processes that are difficult to monitor and control by conventional statistical methods.
("DPC") in which (a) an in-line imaging system and other sensors are used to acquire surface conditions of the cast products, (b) a diagnostic module predicts the internal state of the continuously cast products and the presence of internal defects based on the surface conditions, and (c) an inference engine suggests corrective actions to enable the reduction or elimination of defects. The Phase I objective was accomplished, in that the R&D team has demonstrated the ability to estimate of the likelihood of having internal defects based on the data of the slab external state and conditions, collected using various sensors such as optical, thermal and forces. The accomplishments of the Phase I work provide a solid ground for several perspectives of the proposed DPC system, summarized as follows. First, the Phase I work provided the essential explorative research and convincing validation on the proposed innovation in establishing a model that links the slab external state to its internal state in a probabilistic manner. Second, the success of the Phase I work demonstrated the R&D capability of the proposer and its team formed for this project. Third, the close interactions with the target industry during the Phase I helped in the planning for the Phase II work such that the DPC technology can be materialized as a commercial product. Fourth, the interactions with the target industry affirmed the identified commercial applications and the commercial potential of the DPC product. The objective of the Phase I project was to provide the feasibility verification of the proposed DPC innovation, developing a data-fusion predictive control for continuous casting. The current practice in the continuous casters, the primary steel making process in the US, has room to improve for better efficiency and energy savings. A study funded by the US Department of Energy and performed by Dr. Stubbles explicitly points out that "The ultimate solution is the use of sensors on the casters to ensure surface and internal quality as the product exits the caster." The Steel Technology Roadmap, published by the American Iron and Steel Institute, documents the technological R&D needs of Advanced Process Control Strategies, Advanced Vision Systems for Defect Detection and Identification, and Advanced Computer Diagnostic Controls for the casting operations. The proposed DPC directly responds to the needs that have not yet been addressed. The R&D team carried out the Phase I research plan and successfully accomplished the objectives with the following answers. Verified the feasibility of acquiring the product surface data and the process data with in-line sensors. Conducted the fusion of surface image information and real time data from sensors on the caster for the study of process variations and instability associated with the surface conditions. Demonstrated a probabilistic causal model of the internal state and internal defects from surface conditions based on various analytical approaches, resulted in a prediction accuracy of greater than 90% with 27% or less in false prediction. This distinguishing capability of this model was also tested statistically.
