Inverse Identification Of Fixed Gaseous Pollutant Sources

In case of an accidental release of airborne pollutant in indoor environments, it is critical to promptly obtain the contaminant source information, such as the source location, the release rate profile and the release time. Sensors must be deployed to provide alarm of a pollutant release. However, current sensors can only respond with the local concentration instead of any source information. Identifying the source with limited sensor information forms the framework of inverse modeling. Inverse problems are ill-posed, i.e., the solution may not exist, or the solution is not unique or unstable. The existing inverse model can only identify a source released instantaneously or a source with known location but with other unknown information. So far, no inverse model is able to simultaneously identify the source location, release rate profile and the release time. In addition, no model deals with inverse identification of multiple pollutant sources released synchronously. This thesis proposes a two-step inverse model to fill the gap. The known information that must be specified to the model includes:the steady flow field and the temporal concentrations at n+1locations, where n is the number of the sources to be identified.The two-step inverse model contains two submodels:an inverse matrix submodel and a Bayesian probability submodel. The pollutant release rate profile is determined by the inverse matrix submodel, while the Bayesian probability submodel identifies the source location and the release time. The principles are as follows:at first, a database including all possible source locations and release time is established; then the inversion and the regularized operation is implemented to the matrix that describes the cause-effect relation between the release rate and the exhibited concentration based on the input at one sensor; thirdly, suppose the release rate profiles obtained in the previous step are reasonable, and an initial prior probability is assigned to each release rate profile; finally, based on the aforementioned inversely solved release rate profiles, the concentration responses at the other sensor location are predicted and compared with the monitored data, in which the concentration difference is converted into a posterior probability according to the Bayesian probability submodel that adopts the Gaussian normal distribution. A smaller concentration difference in the Bayesian submodel corresponds to a higher probability, and vice versa. The source release scenario with the highest probability is identified as the actual source. The above model can also deal with multiple synchronously-released sources, by grouping multiple sources into an equivalently single source. To test the above model, a single source released in a two-dimensional (2D) cavity with the measurement data as the input, and a source in a three-dimensional (3D) aircraft cabin with the simulated sensor response data were identified. In the2D cavity, the CO2tracer gas was released into the cavity from a gas tank at a constant rate. The temporal concentrations were monitored near the ceiling and the exhaust outlet at the floor. In the3D cabin, a passenger released airborne pollutant following a sinusoidal wave, in which the temporal concentration data at the mid ceiling and near the floor outlet were provided as the known input. To validate the model in identifying multiple pollutant sources, two sources in a2D cavity released in different rate profiles were identified.When implementing the inverse model, the steady flow field was obtained first by CFD, and then the response matrix was solved. After selecting an appropriate regularization parameter and the standard deviation of the measurement, the source information was identified. The results shows that the proposed model correctly determines the pollutant source location, temporal release rate profiles and the pollutant release time. The model performs excellently in identifying the pollutant source location but generates wiggles in the identified release rate profiles. Multiple release time may be provided unless small wiggles in the release rate profiles can be removed or the wiggles can be judged belonging to a valid release. When identifying two synchronously-released sources, the proposed model can still correctly identify the source locations and release profiles, providing that the number of the pollutant source is known. Some factors that impact the inverse solution accuracy were analyzed, such as the turbulent model used to solve the thermo-flow and the cause-effect matrix, regularization parameter, the standard deviation of the measurement, and the deployment locations of the sensors.