Learning-based Modelling Research Articles

Deep Learning-Based Modelling of Complex Photonic Crystal Slow Light Waveguides

Deep Learning-Based Modelling of Complex Photonic Crystal Slow Light Waveguides

Evolving reliability assessment of systems using active learning-based surrogate modelling

Risk tracking involves monitoring and assessing evolving event risks in complex infrastructure systems. As these systems change over time, so do the associated risk profiles, requiring reliability updating principles. Traditional methods rely on repetitive simulations of complex models, while recent surrogate-based techniques offer efficiency and accuracy without relying on observed data. However, these approaches inadequately address time-dependent reliability updating and surrogate modeling integration. To overcome these limitations, this study presents the Time-Variant Reliability Updating method (TVRU), which seamlessly integrates Kriging-based time-dependent reliability with parallel learning. TVRU estimates prior failure probability by leveraging the adaptive training capabilities of a time-variant surrogate Kriging model. This model incorporates random variables and time as inputs, using the performance function's responses as outputs. By classifying design samples, the prior failure probability is accurately computed, and a meticulous Kriging model estimates the posterior probability of failure. The TVRU method is demonstrated using four numeric models of varying complexities, showcasing exceptional efficiency and accuracy. This novel approach enhances risk assessment in complex systems, facilitating informed decision-making.

Machine learning-based modelling using ANN for performance prediction of a solar air heater design with jet impingement

The prediction of heat transfer performance for solar air heater (SAH) designs is important to assess its potential without relying on extensive experimental studies. This work aimed to predict the heat transfer characteristics of a SAH design with jet impingement using an extensive artificial neural network (ANN) study and covering a complete year-long ambient and various operating parameters (commensurate and non-commensurate). Total 84 ANN models were developed using different network types, transfer functions, and training functions. They were tested on 174 representative datasets collected throughout a year. The best ANN model was found to be the feed forward backpropagation model with tangent-sigmoid transfer function and LM training algorithm, which produced accurate predictions for thermal efficiency, exergy efficiency, and thermo-hydraulic efficiency, with sample errors within ±5 %. The numerical values of R2, MAE, RMSE, and COV from the ANN model results were respectively 0.9972, 0.9053, 1.1884, and 2.0618 for thermal efficiency, 0.9947, 0.2631, 0.3363, and 1.7157 for exergy efficiency, and 0.996, 0.0139, 0.0193, and 0.8076 for thermo-hydraulic efficiency. The study concluded that ANN models can accurately predict the heat transfer characteristics of an SAH design, requiring less time and less computational resources compared to conventional experimental or computational methods.

Diagnostic Performance Evaluation of Deep Learning-Based Medical Text Modelling to Predict Pulmonary Diseases from Unstructured Radiology Free-Text Reports

Diagnostic Performance Evaluation of Deep Learning-Based Medical Text Modelling to Predict Pulmonary Diseases from Unstructured Radiology Free-Text Reports

Machine learning-based predictive modelling for the enhancement of wine quality

The certification of wine quality is essential to the wine industry. The main goal of this work is to develop a machine learning model to forecast wine quality using the dataset. We utilised samples from the red wine dataset (RWD) with eleven distinct physiochemical properties. With the initial RWD, five machine learning (ML) models were trained and put to the test. The most accurate algorithms are Random Forest (RF) and Extreme Gradient Boosting (XGBoost). Using these two ML approaches, the top three features from a total of eleven features are chosen, and ML analysis is performed on the remaining features. Several graphs are employed to demonstrate the feature importance based on the XGBoost model and RF. Wine quality was predicted using relevant characteristics, often referred to as fundamental elements, that were shown to be essential during the feature selection procedure. When trained and tested without feature selection, with feature selection (RF), and with key attributes, the XGBoost classifier displayed 100% accuracy. In the presence of essential variables, the RF classifier performed better. Finally, to assess the precision of their predictions, the authors trained an RF classifier, validated it, and changed its hyperparameters. To address collinearity and decrease the quantity of predictors without sacrificing model accuracy, we have also used cluster analysis.

A sequential cross-product knowledge accumulation, extraction and transfer framework for machine learning-based production process modelling

ABSTRACT Machine learning is a promising method to model production processes and predict product quality. It is challenging to accurately model complex systems due to data scarcity, as mass customisation leads to various high-variety low-volume products. This study conceptualised knowledge accumulation, extraction, and transfer (KAET) to exploit the knowledge embedded in similar entities to address data scarcity. A sequential cross-product KAET (SeqTrans) is proposed to conduct KAET, integrating data preparation and preprocessing, feature selection (FS), feature learning (FL), and transfer learning (TL). The FS and FL modules conduct knowledge extraction and help address various practical challenges such as changing operating conditions and unbalanced datasets. In this paper, sequential TL is introduced to production modelling to conduct knowledge transfer among multiple entities. The first case study of auxetic material performance prediction demonstrates the effectiveness of sequential TL. Compared with conventional TL, sequential TL can achieve the same test mean square errors with 300 fewer training examples when facing data scarcity. In the second case study, balancing anomaly detection models were constructed for two gas turbines in the same series using real-world production data. With SeqTrans, the F1-score of the anomaly detection model of the data-poor engine was improved from 0.769 to 0.909.

Sustainability and Fast Fashion from the Executive Perspective – the Case of LPP S.A.

Abstract Objective This article aims to examine what the CEO of LPP S.A. discloses about sustainability and how sustainability topics vary over time in his communication with the stakeholders. Methodology We use a machine learning-based topic modelling analysis, specifically latent Dirichlet allocation and keyword analysis, to scrutinize the sustainability disclosure in the sustainability letters to the stakeholders of one of Poland’s biggest fast fashion companies, LPP S.A. Findings The findings of this study suggest that more attention should be placed on sustainability topics in the letters to the shareholders. The information disclosed is selective and mainly aimed at transmitting a positive message. It lacks a comprehensive and holistic disclosure of information concerning sustainable activities that substantially impact environmental and social matters. Comparing the variation of topics year to year, there is a clear shift from dedicating the majority of space in CEO letters to the stakeholders to social topics toward a broader discussion about the environment and responsible energy consumption in the most recent letter. Value Added The study contributes to the sustainability communication literature by identifying specific sustainability topics disclosed by the analyzed company in its direct message from the CEO to stakeholders that have been very limited and selective, thus underscoring the limited importance assigned to this topic by the CEO, thereby questioning the transparency concerning one of the most critical areas of fast fashion impacts. Recommendations To keep the stakeholders informed and maintain the transparency of actions and accountability of the disclosure, it is necessary to present the information holistically and comprehensively in the message directed by the CEO to the stakeholders.

Quantitative analysis of cost savings and occupants’ preferences in grid-interactive smart home operation

Many utility companies in the United States have introduced time-of-use (TOU) rates for homeowners with the goal of regulating electricity consumption during peak hours. The electrical appliances in homes include various thermostatically controlled devices, such as air conditioners (AC) for thermal comfort, and nonthermostatically controlled devices such as clothes washers. As a result, homeowners face the complicated challenge of economically operating multiple electrical appliances in their homes while maintaining comfort and convenience. This is usually due to the lack of an explicit understanding of the correlation between cost saving and the users’ comfort. To understand the correlation, this article is designed to construct a framework by integrating three major components: a multi-objective optimization method accommodating multiple competing goals with different weights, a learning-based system modeling approach describing the dynamics and thermal coupling effects of appliances, and a novel comfort index method differentiating preferred and acceptable thermal comfort. Our proposed framework can allow the indoor air temperature to fall into the "preferred" range with a marginal cost increase. The simulation result shows that an additional 8 h for the preferred thermal comfort can be achieved with a cost increase of only 1.77%.

Aware but not prepared: understanding situational awareness during the century flood in Germany in 2021

In July 2021, intense rainfall in parts of Western Europe was followed by unprecedented flash flooding. The flooding killed many people, mostly in Germany, which was subsequently blamed on a lack of preparation by either authorities or the broader populace. Knowledge about people's awareness of a hazard, its impacts and possible adaptation can help improve hazard management. We used social media data to assess situational awareness, sentiments and behaviour before, during and after the flood. We analysed nearly 58,000 German Twitter (now X) messages about the flood using machine learning-based unsupervised topic modelling. We showed that message frequency translated into four phases with message content suggesting sender priorities shifted in each phase of the flood. Besides messages with weather updates, correlated topics included ‘Solidarity, recovery and aid’ and ‘Grief and empathy’ (together 22% of all tweets) and a set of four topics about climate change attribution, extreme weather, long-term flood protection measures and politics (together 38% of all tweets). Many topics depended on the phase of the flood and on the distance from the affected areas. For those near the affected areas, tweets about evacuation (in peak phase) and damage assessment (in recovery phase) were particularly prominent. The high engagement on Twitter to seek weather information might indicate awareness of the extreme climatic conditions before the flood, but the severity was unexpected. Social media data provided unprecedent citizen-generated real-time information. We discuss how rapid automated analysis could contribute to disaster communication and risk mitigation.

Deep Learning-based structure modelling illuminates structure and function in uncharted regions of β-solenoid fold space

Repeat proteins are common in all domains of life and exhibit a wide range of functions. One class of repeat protein contains solenoid folds where the repeating unit consists of β-strands separated by tight turns. β-solenoids have distinguishing structural features such as handedness, twist, oligomerisation state, coil shape and size which give rise to their diversity. Characterised β-solenoid repeat proteins are known to form regions in bacterial and viral virulence factors, antifreeze proteins and functional amyloids. For many of these proteins, the experimental structure has not been solved, as they are difficult to crystallise or model. Here we use various deep learning-based structure-modelling methods to discover novel predicted β-solenoids, perform structural database searches to mine further structural neighbours and relate their predicted structure to possible functions. We find both eukaryotic and prokaryotic adhesins, confirming a known functional linkage between adhesin function and the β-solenoid fold. We further identify exceptionally long, flat β-solenoid folds as possible structures of mucin tandem repeat regions and unprecedentedly small β-solenoid structures. Additionally, we characterise a novel β-solenoid coil shape, the FapC Greek key β-solenoid as well as plausible complexes between it and other proteins involved in Pseudomonas functional amyloid fibres.

Deep learning-based modelling of pyrolysis

Deep learning-based modelling of pyrolysis

Investigations on machine learning-based control-oriented modeling using historical thermal data of buildings

Machine learning-based control-oriented modelling is a promising solution to model the complex system dynamics of buildings. It can benefit learning-based controllers, e.g., Model Predictive Control and Reinforcement Learning. However, obtaining reliable models is non-trivial, especially when only limited historical data generated by the baseline controller is available. This paper analyses the modelling tasks and identifies the sources of prediction error to obtain suitable machine learning models, and then proposes a training framework to provide a well-designed training procedure, mitigating the reducible errors. Six typical machine learning methods are evaluated using simulated and real-world datasets using in- and out-of-distribution data. Comprehensive evaluation metrics are also proposed. The results of absolute error and prediction interval coverage percentage indicate that purely point estimation is not enough to provide reliable predictions, and uncertainties must be quantified. Further results from analysis of the out-of-distribution data show that although some models have satisfactory predictions on in-distribution data, their capability of interpolation and extrapolation is poor. Some predictions may deviate from the targets. Point estimation may provide over-confident results. It is, therefore, essential to quantify the uncertainty caused by the lack of data to ensure that the true value falls into the prediction interval with high probability. The decomposition of the total uncertainty provides more information and detects out-of-distribution data. These investigations demonstrated the importance of accurate point estimation and uncertainty quantification for models' reliability.

Machine Learning-based Time Series Modelling for Large-Scale Regional Wind Power Forecasting: a Case Study in Ontario, Canada

Recently, time series forecasting has acquired considerable academic and industrial interests in various areas for different applications. Machine learning (ML) algorithms are known for their ability to capture the chaotic temporal non-linear relations in time series data. In this paper, various ML-based algorithms are employed and analyzed for time series forecasting of “regional wind power” in Ontario, Canada. To this end, the meteorological and spatial parameters with seasonal and temporal features are filtered and selected by a proposed deep feature selection approach. Then, multiple ML algorithms, including artificial neural network (ANN), deep neural network (DNN), long short-term memory (LSTM), bagging tree (BT), and support vector machine/regression (SVM/SVR), are used for training one-step ahead forecasting models. Finally, a comprehensive assessment of the constructed models is conducted based on different error criteria metrics. By evaluating and analyzing the performance of the models using testing data, it is observed that SVR/SVM is one of the most promising robust ML-based forecasting models. This technique results in reliable generic models that perform well with new data, where the testing MAPE % reaches a value of 13 %. Almost a similar MAPE is obtained from the ensemble modeling approach, which means combining process of the generated ML-based models does not significantly improve the predictions, in comparison with the developed SVR/SVM model. On the other hand, when constructing the multi-step ahead forecasting models, the predictions obtained from the multi-input multi-output (MIMO) LSTM approach are more reliable with higher accuracies. In other words, it is shown that the performance of the MIMO multi-step strategy is superior to the direct multi-step forecasting method, while employing algorithms with recursive properties.

Large Area High-Resolution 3D Mapping of the Von Kármán Crater: Landing Site for the Chang’E-4 Lander and Yutu-2 Rover

We demonstrate the creation of a large area of high-resolution (260 × 209 km2 at 1 m/pixel) DTM mosaic from the Lunar Reconnaissance Orbiter Camera (LROC) Narrow Angle Camera (NAC) images over the Chang’E-4 landing site at Von Kármán crater using an in-house deep learning-based 3D modelling system developed at University College London, called MADNet, trained with lunar orthorectified images and digital terrain models (DTMs). The resultant 1 m DTM mosaic is co-aligned with the Chang’E-2 (CE-2) and the Lunar Orbiter Laser Altimeter (LOLA)—SELenological and Engineering Explorer (SELENE) blended DTM product (SLDEM), providing high spatial and vertical congruence. In this paper, technical details are briefly discussed, along with visual and quantitative assessments of the resultant DTM mosaic product. The LROC NAC MADNet DTM mosaic was compared with three independent DTM datasets, and the mean differences and standard deviations are as follows: PDS photogrammetric DTM at 5 m grid-spacing had a mean difference of −0.019 ± 1.09 m, CE-2 DTM at 20 m had a mean difference of −0.048 ± 1.791 m, and SLDEM at 69 m had a mean difference of 0.577 ± 94.940 m. The resultant LROC NAC MADNet DTM mosaic, alongside a blended LROC NAC and CE-2 MADNet DTM mosaic and a separate LROC NAC, orthorectified image mosaic, are made publicly available via the ESA planetary science archive’s guest storage facility.

Identification of Potential p38γ Inhibitors via In Silico Screening, In Vitro Bioassay and Molecular Dynamics Simulation Studies.

Protein kinase p38γ is an attractive target against cancer because it plays a pivotal role in cancer cell proliferation by phosphorylating the retinoblastoma tumour suppressor protein. Therefore, inhibition of p38γ with active small molecules represents an attractive alternative for developing anti-cancer drugs. In this work, we present a rigorous and systematic virtual screening framework to identify potential p38γ inhibitors against cancer. We combined the use of machine learning-based quantitative structure activity relationship modelling with conventional computer-aided drug discovery techniques, namely molecular docking and ligand-based methods, to identify potential p38γ inhibitors. The hit compounds were filtered using negative design techniques and then assessed for their binding stability with p38γ through molecular dynamics simulations. To this end, we identified a promising compound that inhibits p38γ activity at nanomolar concentrations and hepatocellular carcinoma cell growth in vitro in the low micromolar range. This hit compound could serve as a potential scaffold for further development of a potent p38γ inhibitor against cancer.

Integration of multi-physics and machine learning-based surrogate modelling approaches for multi-objective optimization of deformed GDL of PEM fuel cells

The development of artificial intelligence (AI) greatly boosts scientific and engineering innovation. As one of the promising candidates for transiting the carbon intensive economy to zero emission future, proton exchange membrane (PEM) fuel cells has aroused extensive attentions. The gas diffusion layer (GDL) strongly affects the water and heat management during PEM fuel cells operation, therefore multi-variable optimization, including thickness, porosity, conductivity, channel/rib widths and compression ratio, is essential for the improved cell performance. However, traditional experiment-based optimization is time consuming and economically unaffordable. To break down the obstacles to rapidly optimize GDLs, physics-based simulation and machine-learning-based surrogate modelling are integrated to build a sophisticated M5 model, in which multi-physics and multi-phase flow simulation, machine-learning-based surrogate modelling, multi-variable and multi-objects optimization are included. Two machine learning methodologies, namely response surface methodology (RSM) and artificial neural network (ANN) are compared. The M5 model is proved to be effective and efficient for GDL optimization. After optimization, the current density and standard deviation of oxygen distribution at 0.4 V are improved by 20.8% and 74.6%, respectively. Pareto front is obtained to trade off the cell performance and homogeneity of oxygen distribution, e.g., 20.5% higher current density is achieved when sacrificing the standard deviation of oxygen distribution by 26.0%.

A machine learning-based diagnosis modelling of type 2 diabetes mellitus with environmental metal exposure

Background and Objective:Increasing and compelling evidence has been proved that urinary and dietary metal exposure are underappreciated but potentially modifiable biomarkers for type 2 diabetes mellitus (T2DM). The aims of this study were (1) to identify the key potential biomarkers which contributed to T2DM with effective and parsimonious features and (2) to assess the utility of baseline variables and metal exposure in the diagnosis of T2DM.Methods:Based on the National Health and Nutrition Examination Survey (NHANES), we selected 9822 screening records with 82 significant variables covering demographics, lifestyle, anthropometric measures, diet and metal exposure for this study. Combining extreme gradient boosting (XGBoost), random forest and light gradient boosting machine (lightGBM), a soft voting ensemble model was proposed to measure the importance of 82 features. With this soft voting ensemble model and variance inflation factor (VIF), strong multicollinear features with low importance scores were further removed from candidate biomarkers. Then, a soft voting ensemble classifier was adopted to demonstrate the efficiency of the proposed feature selection method.Results:With the novel feature selection method, 12 baseline variables and 3 metal variables were selected to detect patients at risk for T2DM in our study. For metal variables, the dietary copper (Cu), urinary cadmium (Cd) and urinary mercury (Hg) metals were selected as the most remarkable metal exposure and the corresponding P-values were all less than 0.05. In a classification model of T2DM with 12 baseline biomarkers, the addition of 3 metal exposure improved the classification accuracy of T2DM from a traditional area under the curve (AUC) 0.792 of the receiver operating characteristic (ROC) to an AUC 0.847.Conclusions:This was the first demonstration of T2DM classification with machine learning under urinary and dietary metal exposure. Improved prediction precision illustrated the effectiveness of the proposed machine learning-based diagnosis model facilitated lifestyle/dietary intervention for T2DM prevention.

Quality assessment of VHH models

Heavy Chain Only Antibodies are specific to Camelid species. Despite the lack of the light chain variable domain, their heavy chain variable domain (VH) domain, named VHH or nanobody, has promising potential applications in research and therapeutic fields. The structural study of VHH is therefore of great interest. Unfortunately, considering the huge amount of sequences that might be produced, only about one thousand of VHH experimental structures are publicly available in the Protein Data Bank, implying that structural model prediction of VHH is a necessary alternative to obtaining 3D information besides its sequence. The present study aims to assess and compare the quality of predictions from different modelling methodologies. Established comparative & homology modelling approaches to recent Deep Learning-based modelling strategies were applied, i.e. Modeller using single or multiple structural templates, ModWeb, SwissModel (with two evaluation schema), RoseTTAfold, AlphaFold 2 and NanoNet. The prediction accuracy was evaluated using RMSD, TM-score, GDT-TS, GDT-HA and Protein Blocks distance metrics. Besides the global structure assessment, we performed specific analyses of Frameworks and CDRs structures. We observed that AlphaFold 2 and especially NanoNet performed better than the other evaluated softwares. Importantly, we performed molecular dynamics simulations of an experimental structure and a NanoNet predicted model of a VHH in order to compare the global structural flexibility and local conformations using Protein Blocks. Despite rather similar structures, substantial differences in dynamical properties were observed, which underlies the complexity of the task of model evaluation. Communicated by Ramaswamy H. Sarma

A novel approach of tool condition monitoring in sustainable machining of Ni alloy with transfer learning models

Cutting tool condition is crucial in metal cutting. In-process tool failures significantly influences the surface roughness, power consumption, and process endurance. Industries are interested in supervisory systems that anticipate the health of the tool. A methodology that utilizes the information to predict problems and to avoid failures must be embraced. In recent years, several machine learning-based predictive modelling strategies for estimating tool wear have been emerged. However, due to intricate tool wear mechanisms, doing so with limited datasets confronts difficulties under varying operating conditions. This article proposes the use of transfer learning technology to detect tool wear, especially flank wear under distinct cutting environments (dry, flood, MQL and cryogenic). In this study, the state of the cutting tool was determined using the pre-trained networks like AlexNet, VGG-16, ResNet, MobileNet, and Inception-V3. The best-performing network was recommended for tool condition monitoring, considering the effects of hyperparameters such as batch size, learning rate, solver, and train-test split ratio. In light of this, the recommended methodology may prove to be highly helpful for classifying and suggesting the suitable cutting conditions, especially under limited data situation. The transfer learning model with Inception-V3 is extremely useful for intelligent machining applications.

A Machine Learning-Based Diagnosis Modelling of Type 2 Diabetes with Environmental Metal Exposure

A Machine Learning-Based Diagnosis Modelling of Type 2 Diabetes with Environmental Metal Exposure