Soft Sensor Design for Product Gas Composition Monitoring Including Fault Isolation in a Dual Fluidized Bed Biomass Gasifier

Dual fluidized bed gasification represents a promising way to produce product gas from woody biomass and residues. While the main dry product gas components concentrations can be easily measured during plant operation with standard instrumentation, other quantities such as the water content of the product gas and the amounts of char and tar cannot be measured online. The soft sensor presented in this work aims to provide a complete picture of the product streams leaving the gasification reactor. Particular emphasis is laid on the estimation of the product gas water content as it influences key performance indicators like cold gas efficiency and water conversion. The soft sensor is based on a Hammerstein model which separates the static nonlinearities of the system and its dynamics. This model provides the prediction for an extended Kalman filter. A second extended Kalman filter is used to improve future model predictions. In addition, a simple heuristic fault detection mechanism is implemented which allows the soft sensor to simulate the system state when measurements are not available. An implementation test at a 100kW pilot plant demonstrated robustness against sensor faults and showed reliable results with a maximum relative error in the predicted product gas water content of 5.2%.