Search

Batch processes have been playing a crucial role for the flexibility in producing low-volume and high-value-added products. Due to the fluctuations of raw materials and operation conditions, the ba...

In chemical engineering applications, computational efficient meta-models have been successfully implemented in many instants to surrogate the high-fidelity computational fluid dynamics (CFD) simulators. Nevertheless, substantial simulation efforts are still required to generate representative training data for building meta-models. To solve this problem, in this research work an efficient meta-modeling method is developed based on the concept of transfer learning. First, a base model is built which roughly mimics the CFD simulator. With the help of this model, the feasible operating region of the simulated process is estimated, within which computer experiments are designed. After that, CFD simulations are run at the designed points for data collection. A transfer learning step, which is based on the Bayesian migration technique, is then conducted to build the final meta-model by integrating the information of the base model with the simulation data. Because of the incorporation of the base model, only a small number of simulation points are needed in meta-model training.

Use of computational fluid dynamics to model chemical process system has received much attention in recent years. However, even with state-of-the-art computing, it is still difficult to perform simulations with many physical factors taken into accounts. Hence, translation of such models into computationally easy surrogate models is necessary for successful applications of such high fidelity models to process design optimization, scale-up and model predictive control. In this work, the methodology, statistical background and past applications to chemical processes of meta-model development were reviewed. The objective is to help interested researchers be familiarized with the work that has been carried out and problems that remain to be investigated.

As the electronic devices getting lighter and smaller, a coreless substrate technology, called Embedded Trace Substrate (ETS) is developed to meet the market requirement. However, this design causes severe warpage due to the large difference in CTE (coefficient of thermal expansion) of buildup material and Cu plate. Recently, finite element analysis (FEA) is a popular and effective method used for substrate warpage prediction and mechanical studies. Manufacturers apply FEA simulation for substrate design improvements and provide substrate warpage that satisfying the customer’s specification. Nevertheless, the computational resources needed for high-fidelity simulation are extremely expensive and time-consuming. Hence the simulation study becomes a long and arduous task if it has to be performed many times, e.g., sensitivity analysis and warpage optimization.In this paper, we propose a new method for FEA modeling of mechanical behaviors of the substrate and present an optimization strategy for substrate warpage control. In the first step, the Gerber files of each layer of the substrate are converted into high-resolution bitmap images, and the copper area of each image is divided and scanned by a pre-sized block window. After that, the effective material properties for each block are calculated with a volume average micromechanics approach, and then all blocks are stacked-up to build a block-based analysis model for FEA simulation. As compared with conventional trace mapping simulation, the proposed method significantly decreases the demands of the computing resource. Besides, we gained accurate warpage prediction results as validated by a real substrate experiment. Finally, we presented an optimization strategy that manipulates the thickness of each layer for substrate warpage optimization in pre-processing steps of packaging. In conclusion, the results show that the methodology for substrate simulation in this paper is practical, effective, and costless.

A novel and efficient RCGA for constrained optimization has been proposed.The proposed RCGA integrates three effective and novel evolutionary operators named RS, DBX and DRM.The proposed RCGA is proven to have a small complexity index and outperform many state-of-the-art algorithms.The proposed RCGA has been successfully applied to optimize the GaAs film-growth performance of an MOCVD process. This paper presents a simple and efficient real-coded genetic algorithm (RCGA) for constrained real-parameter optimization. Different from some conventional RCGAs that operate evolutionary operators in a series framework, the proposed RCGA implements three specially designed evolutionary operators, named the ranking selection (RS), direction-based crossover (DBX), and the dynamic random mutation (DRM), to mimic a specific evolutionary process that has a parallel-structured inner loop. A variety of benchmark constrained optimization problems (COPs) are used to evaluate the effectiveness and the applicability of the proposed RCGA. Besides, some existing state-of-the-art optimization algorithms in the same category of the proposed algorithm are considered and utilized as a rigorous base of performance evaluation. Extensive comparison results reveal that the proposed RCGA is superior to most of the comparison algorithms in providing a much faster convergence speed as well as a better solution accuracy, especially for problems subject to stringent equality constraints. Finally, as a specific application, the proposed RCGA is applied to optimize the GaAs film growth of a horizontal metal-organic chemical vapor deposition reactor. Simulation studies have confirmed the superior performance of the proposed RCGA in solving COPs.

Batch processing is a widely used method in process industry for its flexibility in manufacturing low-volume and high-value-added products. Due to inter-batch variations, the batch duration often varies, which may cause difficulties in operation scheduling and decision-making. The capability of predicting batch completion time offers valuable information to improved capacity utilisation, reduced workload, and reduced operating cost. To this end, several data-driven modelling methods have been reported. However, the uncertainty of the predicted completion time has not been well explored in previous research. In this paper, the challenges for batch-end prediction are discussed by stressing the importance of prediction uncertainty. This has been demonstrated by the application of probabilistic principal component analysis (PPCA) and quantitative sensitivity analysis to two batch processes. The prediction uncertainty tends to increase substantially, when the variable defining the completion time changes slowly towards the end of batch. Under such situations, we argue that the uncertainty should always be considered along with the mean prediction for practical use.

Biomedical imaging modalities in clinical practice have revolutionized oncology for several decades. State-of-the-art biomedical techniques allow visualizing both normal physiological and pathological architectures of the human body. The use of nanoparticles (NP) as contrast agents enabled visualization of refined contrast images with superior resolution, which assists clinicians in more accurate diagnoses and in planning appropriate therapy. These desirable features are due to the ability of NPs to carry high payloads (contrast agents or drugs), increased in vivo half-life, and disease-specific accumulation. We review the various NP-based interventions for treatments of deep-seated tumors, involving “seeing better” to precisely visualize early diagnosis and “going deeper” to activate selective therapeutics in situ.

This paper benchmarks a novel and efficient real-coded genetic algorithm (RCGA) enhanced from our previous work [1] on the noisefree BBOB 2012 testbed. The enhanced algorithm termed as direction-based RCGA (DBRCGA) uses relative fitness information to direct the crossover toward a direction that significantly improves the objective fitness. As a base of performance evaluation and comparisons, the maximum number of function evaluations (#FEs) for each test run is set to 105 times to the problem dimension. Extensive benchmarking test results reveal that all functions can be solved by DBRCGA in the low search dimensions. Although the DBRCGA shows the difficulty in getting a solution with the desired accuracy 10-8 for high conditioning and multimodal functions within the specified maximum #FEs, the DBRCGA presents good performance in separable function and functions with low or moderate conditioning.