تجزیه و تحلیل حساسیت از تجهیزات تکنولوژیکی مبتنی برروش رحم جایگزین

The computational cost associated with the use of high-fidelity computational fluid dynamics (CFD) models poses a serious impediment to the successful application of formal sensitivity analysis in engineering design. Even though advances in computing hardware and parallel processing have reduced costs by orders of magnitude over the last few decades, the fidelity with which engineers desire to model engineering systems has also increased considerably. Evaluation of such high-fidelity models may take significant computational time for complex geometries. In many engineering design problems, thousands of function evaluations may be required to undertake a sensitivity analysis. As a result, CFD models are often impractical to use for design sensitivity analyses. In contrast, surrogate models are compact and cheap to evaluate (order of seconds or less) and can therefore be easily used for such tasks. This paper discusses and demonstrates the application of several common surrogate modelling techniques to a CFD model of flocculant adsorption in an industrial thickener. Results from conducting sensitivity analyses on the surrogates are also presented.

For many industrial fluid dynamics problems, it is impractical to perform experiments on the physical world directly. Instead, complex, physics-based simulation codes are used to run experiments on computer hardware. Accurate, high-fidelity computational fluid dynamics (CFD) models are typically time consuming and computationally expensive, this poses a serious impediment to the successful application of formal sensitivity analysis in engineering design. While advances in High Performance Computing and multi-core architectures have helped, routine tasks such as visualisation, design space exploration, sensitivity analysis and optimisation quickly become impractical [1] and [2]. As a result, researchers have turned to various methods to mimic the behaviour of the simulation model as closely as possible, while being computationally cheaper to evaluate. This work concentrates on the use of data-driven, global approximations using compact surrogate models in the context of computer experiments. The objective is to construct a surrogate model that is as accurate as possible over the complete design space of interest using as few simulation points as possible. Once constructed, the global surrogate model is reused in other stages of the computational engineering pipeline, such as sensitivity analysis. Sensitivity analysis of model output aims to quantify how a model depends on its input factors. Global sensitivity determines the effect on model output of all the input parameters acting simultaneously over their ranges. Most global sensitivity techniques are variance-based methods and determine the fractional contribution of each input factor to the variance of a model output. The main difficulty with global sensitivity analysis is that the number of model evaluations required is often large. As a result, CFD models are often impractical to use for design sensitivity analyses. The objective of this paper is to discuss and demonstrate the application of several common surrogate modelling techniques to a case study of a CFD model of flocculant adsorption in an industrial thickener. A sensitivity analysis is then conducted on the produced surrogate models.

Three types of surrogate models (RBF, ANN and LS-SVM) have been described along with the approach to optimising their parameters. The output from a CFD model of thickener feedwells that incorporates flocculant adsorption has been used to produce surrogate models of the adsorption process within thickener feedwells. A space filling method was used for three model types and an adaptive sampling method was used with an ANN model to allow comparison between the two sampling approaches. For this particular case study it was found that the adaptive sampling offers no real benefit over the space filling sampling due to the uniformly non-linear behaviour of the response. Both ANN models (space filling and sequential design) are similar from an error perspective. The resulting network structure of the sequential design ANN is too complex because not enough pressure is applied to the modelling process to keep the model complexity down. The produced surrogate models have been used to investigate the sensitivity of the model output (flocculant loss) to the various models inputs (A, B and C). The calculated sensitivity indices for each of the model types were very similar in value for both the main and total effect for each of the parameters. This indicates that the global response for each of the models is very similar despite their underlying architecture being very different. The sum of the main effects indicated the models are non-additive with significant interactions between variables. The most important parameter is C (distance from feedwell wall to flocculant sparge), the next most important parameter is B (vertical location of flocculant sparge), with the effect of parameter A (radial angle between flocculant sparge and feed pipe) being negligible. The total sensitivity indices showed strong interaction effect for C followed by B and again minor interaction for A. Therefore most of the model output can be explained by C and B with strong interaction between these two parameters.