Research Publications

Abstract:  The ability to classify images accurately and efficiently is dependent on having access to large labeled datasets and testing on data from the same domain that the model is trained on. Classification becomes more challenging when dealing with new data from a different domain, where collecting a large labeled dataset and training a new classifier from scratch is time-consuming, expensive, and sometimes infeasible or impossible. Cross-domain classification frameworks were developed to handle this data domain shift problem by utilizing unsupervised image-to-image (UI2I) translation models to translate an input image from the unlabeled domain to the labeled domain. The problem with these unsupervised models lies in their unsupervised nature. For lack of annotations, it is not possible to use the traditional supervised metrics to evaluate these translation models to pick the best-saved checkpoint model. In this paper, we introduce a new method called Pseudo Supervised Metrics that was designed specifically to support cross-domain classification applications contrary to other typically used metrics such as the FID which was designed to evaluate the model in terms of the quality of the generated image from a human-eye perspective. We show that our metric not only outperforms unsupervised metrics such as the FID, but is also highly correlated with the true supervised metrics, robust, and explainable. Furthermore, we demonstrate that it can be used as a standard metric for future research in this field by applying it to a critical real-world problem (the boiling crisis problem).

Keywords: Critical heat flux, Domain adaptation, Generative adversarial networks, Image-to-image translation, Pool boiling, Unsupervised machine learning

Abstract:  The detection of critical heat flux (CHF) is crucial in heat boiling applications as failure to do so can cause rapid temperature ramp leading to device failures. Many machine learning models exist to detect CHF, but their performance reduces significantly when tested on data from different domains. To deal with datasets from new domains a model needs to be trained from scratch. Moreover, the dataset needs to be annotated by a domain expert. To address this issue, we propose a new framework to support the generalizability and adaptability of trained CHF detection models in an unsupervised manner. This approach uses an unsupervised Image-to-Image (UI2I) translation model to transform images in the target dataset to look like they were obtained from the same domain the model previously trained on. Unlike other frameworks dealing with domain shift, our framework does not require retraining or fine-tuning of the trained classification model nor does it require synthesized datasets in the training process of either the classification model or the UI2I model. The framework was tested on three boiling datasets from different domains, and we show that the CHF detection model trained on one dataset was able to generalize to the other two previously unseen datasets with high accuracy. Overall, the framework enables CHF detection models to adapt to data generated from different domains without requiring additional annotation effort or retraining of the model.

Keywords: Critical heat fluxDomain adaptationGenerative adversarial networksImage-to-image translationPool boilingUnsupervised machine learning

Abstract:  Image-based deep learning (DL) models are employed to enable the detection of critical heat flux (CHF) based on pool boiling experimental images. Most machine learning approaches for pool boiling to date focus on a single dataset under a certain heater surface, working fluid, and operating conditions. For new datasets collected under different conditions, a significant effort in re-training the model or developing a new model is required under the assumption that the new dataset has a sufficient amount of labeled data. This research is to explore supervised, semi-supervised, and unsupervised machine learning strategies that are formulated to adapt to two scenarios. The first is when the new dataset has limited labeled data available. This scenario was addressed in chapter 2 of this thesis, where Convolutional Neural Networks (CNNs) and Transfer learning (TL) were used in tackling such situations. The second scenario is when the new dataset has no labeled data available at all. In such cases, this research presents a methodology in Chapter 3, where one of the state-of-the-art Generative Adversarial Networks (GANs) called Fixed-Point GAN is deployed in collaboration with a regular CNN model to tackle the problem. To the best of my knowledge, the approaches presented in chapters 2 and 3 are the first of their kind to utilize TL and GANs to solve the boiling heat transfer problem within the heat transfer community and are a step forward towards obtaining a one-for-all general model.

Keywords: Boiling crisis; Boiling crisis detection; Machine learning; Pool boiling; Supervised machine learning; Unsupervised machine learning

Abstract:  Image-based deep learning (DL) models are employed to enable the detection of critical heat flux  (CHF)  based  on  pool  boiling  experimental  images.  Most  machine  learning  approaches  for pool  boiling  to  date  focus  on  a  single  dataset  under  a  certain  heater  surface,  working  fluid,  and operating conditions. For new datasets collected under different conditions, a significant effort in re-training the model or developing a new model is required under the assumption the new dataset has  a  sufficient  amount  of  data.  This  research  is  to  explore  strategies  of  DL  adapting  to  new datasets with limited data available. The insights gained could help improve the practicality and reliability  of  DL  for  boiling  regime  studies.  Specifically,  convolutional  neural  networks  (CNN) and transfer learning (TL) are studied.

Keywords: Critical heat flux, deep learning, transfer learning, convolutional neural network, pool boiling

Abstract:  In this study, machine learning algorithms (MLA) were employed to predict and classify the tensile strength of polymeric films of different compositions as a function of processing conditions. Two film production techniques were investigated, namely compression molding and extrusion-blow molding. Multi-factor experiments were designed with corresponding parameters. A tensile test was conducted on samples and the tensile strength was recorded. Predictive and classification models from nine MLA were developed. Performance analysis demonstrated the superior predictive ability of the support vector machine (SVM) algorithm, in which a coefficient of determination and mean absolute percentage error of 96% and 4%, respectively were obtained for the extrusion-blow molded films. The classification performance of the MLA was also evaluated, with several algorithms exhibiting excellent performance.

Keywords: machine learning algorithms; polymeric films; extrusion-blow molding; cryomilling-compression molding

Abstract:  This paper explores the utility of supervised machine learning algorithms in predicting the tensile strength of high density polyethylene film produced by extrusion-blown molding process. Three algorithms were used: Artificial Neural Networks, Decision Tree, and k-Nearest Neighbors. Eleven input parameters, five materials related and six process related; were modeled in the algorithms. The application of algorithms demonstrated their capability in predicting the intended property of the extrusion-blown process products.

Keywords:  Machine learning algorithms, extrusion-blown molding, tensile strength, HDPE film