A POD-based Inverse Method To Promptly Design The Enclosed Environment

People in modern society spend more than 80%of their time in enclosed environments,such as buildings and vehicles.Evidence shows that most illnesses are related to the enclosed environment.So it is important to promptly design a desired enclosed environment.Current designs require a lengthy iterative trial-and-error procedure by enumerating every possible air-supply parameter to find an optimal solution.Such methods suffer from high computing cost,but may still not be able to find a desired environment.An inverse design is able to fulfill the task by starting with the design objective and using a method that can search for the desired design parameters.However,the computing efficiency of current inverse design methods can still be improved due to the large number of forward CFD(Computational fluid dynamics)simulations involved.This study developed a CFD-based POD(proper orthogonal decomposition)method for a typical enclosed environment,which accelerated the inverse design.The proposed POD method first sampled representative thermo-flows using full CFD simulations,and then the POD modes and their coefficients were extracted.To construct a much larger amount of thermo-flow data,an appropriate interpolation was adopted to interpolate the coefficients of the POD modes.Once the cause-effect maps from all the design variables to the resulting environmental performance were established,the design parameters that led to satisfactory performance were achieved,and the specific solution corresponding to the best performance of each design target could also be obtained.To extract the POD modes for an inverse design,a sufficient number of original data samples must be provided by full CFD simulations as snapshots.Because the cost of a full simulation 1s extremely high,a trial and data sample increase method was proposed to keep a minimum load of such simulations while still ensuring high accuracy.The method determined the key samples by evaluating gaps between the interpolated fields and the fields by full CFD simulation in the middle of two neighboring data samples.The proposed POD method was validated using two different cases,in which the design targets were from the expected velocity and temperature values at specific locations,and from the demand of inside occupants,respectively.The validated POD model was then used in cases with single and multiple design targets.Finally,an integrated method was developed to further improve the accuracy and efficiency of the proposed POD method.The integration first adopted genetic algorithn for initial design to circumscribe ranges of the air-supply parameters.Then POD was applied for basic design to further narrow the ranges and estimate the optimal air-supply parameters for each design criterion.Finally,the estimated optimal design from POD was supplied to the adjoint method for fine tuning.The aforementioned investigations showed that the extracted POD modes and the interpolated coefficients of the POD modes could be used to construct thermo-flow fields that were in good agreement with experimental data.Moreover,the POD?based design method was effective in improving the design of an indoor environment,whether a single-objective design or a multiple-objective design was used.The model required a relatively small amount of computing resources m achieving reasonably good accuracy.Most of the computing time was devoted to preparation of the snapshot data by CFD;the POD analysis,interpolation of the POD mode coefficients,and determination of the design solutions were much more efficient.The proposed trial-and-data-sample-increase method required a number of samples in the regions where the nonlinearity prevailed,while in the approximately linear regions,only several samples were required.In this way,the method improved the efficiency of the sample determination.However,comparison of results using the proposed POD method and those using CFD simulation suggested the accuracy of current POD method was limited,likely due to the fields were obtained by the POD modes.So the initial ranges provided to POD may not be too large to ensure the accuracy.By combining the genetic algorithm and the adjoint method with POD,the method could not only expend the initial ranges for optimization,but also improve the computing accuracy in providing the best comfort status for each target.In addition,the integrated method showed higher efficiency than the pure genetic algorithm,and was able to provide better initial designs for the adjoint method,which showed the proposed integration could stand a high ground.