The availability of the power unit for electricity production is one of the most important issues of the power plant operator. Considering the power unit, boiler is the main source of its malfunction. Depending of the boiler construction, source of problem can be different. In case of usage of the circulating fluidized bed boiler the potential issue that can lead to interruption of the energy production is its failure caused by leakages of heating surfaces. In order to mitigate this risk different treatments or procedures are introduced. Nowadays, thanks to continuous development of mathematical tools its is possible to introduce a new solution to reduce the risk of heating surface erosion cause by the friction of solid material used in fluidization. One of possible option, that can help to resolve such a problem is application of machine learning technique. Based on real observation of the boiler operation and data analysis, it is believed that the uniform temperature distribution at the lower part of the combustion chamber should has positive impact on erosion reduction at the kick-out level where tapered walls changed to vertical one. This can be attain by careful manipulation of selected boiler operating parameters. Due to the reason that in order to find requires setup dozen of input data configuration need to be considered appropriate toll need to be developed. That is the main reason way the machine learning technique need to be applied for such purpose. Indeed, of this work is to develop artificial models that can help in adjustment boiler setup. In order to check models functionality, they were on-site tested by boiler operator. Developed model shows tremendous potential and confirm that it is worth to investigated this topic farther.
With technology development, there is a growing need for an accurate simulation tools, allowing the best possible representation of the reality. Developed model found not only application in prototyping process but can provide significant knowledge into the artificial intelligent model extending their comprehension. Computational performance and accuracy of the numerical model, dedicated for granular flows are mostly a function of mathematical model that track the mutual interaction between particles. Currently the kinetic theory of granular flow or soft- and hard-sphere collision models are used for modeling particle interactions. Each of them suffers from some imperfection that often limit the problem sizes. The purpose of this work is building new mathematical approach, not only fast, but also accurate regarding predicting collisions and determining particle trajectories by application of machine learning technique.
This study focuses on developing a simplified particle collision model in terms of the machine learning technique to a simplified standard hybrid Euler-Lagrange (HEL) technique. The developed surrogate collision model was implemented in the solution procedure of the HEL technique to replace the kinetic theory of granular flow collision model. The Hybrid Euler-Lagrange Surrogate Collision model (HELSCM) results were juxtaposed with the numerical results obtained by applying standard HEL and discrete element method approaches using solidphase velocity profiles. The experimental data were collected in-house using a two-stream particle collision test rig to verify the accuracy of the reduced model. A qualitative check was conducted based on particle distribution. Keywords: multiphase flow, particle tracking, machine learning, particle collision, circulating fluidized bed, CFD
