(446e) Optimal Sensor Placement for Fault Diagnosis in an Acid Gas Removal (AGR) Unit for an Integrated Gasification Combined Cycle (IGCC) Plant with CO2 Capture

Future integrated gasification combined cycle (IGCC) power plants with CO₂ capture will face stricter environmental emission targets. The acid gas removal (AGR) unit for an IGCC plant plays a key role in satisfying the environmental constraints. However, a number of faults can occur in an AGR unit that can result in violation of the environmental constraints. For examples, fly ash present in the syngas can deposit on the trays of the absorber(s) and stripper(s). Heat exchangers may leak causing internal mixing of shell and tube side materials.  An early detection of these faults can help to satisfy the emission targets, and to ensure safe operation of the unit in the presence of faults. It can also help to plan for maintenance in advance and thereby reduce the overall down-time.

In this work, a comprehensive model of a selective, dual-stage, Selexol-based acid gas removal (AGR) unit is developed in Aspen Engineering Suite (AES). Using a chilled solvent, the AGR unit captures about 90% CO₂ and more than 99.9% H2S present in the syngas. The captured CO₂ is recovered at three pressure levels and is sent for compression. The H₂S that is thermally stripped off from the loaded solvent is sent to the Claus unit.

The effect of faults gets manifested in process variables. However, it is not obvious which variables to measure for diagnosing a fault. It is neither economically viable nor practically possible to measure every process variable.  A sensor placement (SP) algorithm can help to identify the number, type, and location of the sensors required for fault diagnosis.

In this work, first, a sensor placement (SP) algorithm is developed for the desired observability of the faults. For this, a set cover problem is formulated based on the Directed Graph (DG) representation of the fault propagation. The algorithm is further extended for a desired fault resolution by using a Signed Directed Graph (SDG) representation. The candidate sensors considered in this algorithm are those that can be measured with current state-of-the-art technology. Sensors with different levels of precision are considered in this algorithm. In this presentation, we will also discuss the impact of the budget for sensors on the overall fault resolution and the trade-off between precision and fault resolution. The proposed approach shows great promise for system level sensor placement for fault diagnosis in plants with a large number of possible sensor locations.