Bias identification and estimation based on data reconciliation and first-principle model - application to nuclear fuel recycling process.

Overview & Synopsis

This paper focuses on data reconciliation (DR) applied to the nuclear fuel treatment PUREX process. Data reconciliation improves the degree of confidence in available information and generates consistent data. The inventory and analysis of the plant data (position and type of sensors, flowsheet diagrams ) allow to represent the observability and the redundancy of the process using graph theory methods. The aggregation of the graph is a compulsory step due to the need in redundant measures. When many bias are present in the data, the redundancy is decreasing. This can lead to a loss of information and maybe the incapacity to apply DR. The proposed methodology combines DR and simulation to locate and estimate multiple biases and to make data consistent in the case of many disrupted hard sensors measuring connected flows.