CSI Transactions on ICT最新文献

Irrespective of its potential, mining sector in India is not developed to the extent it is developed in the countries with similar geological history. There are multiple reasons for it: less investment in mineral exploration, lack of a common repository of data used for mineral exploration, less advanced mineral and geological mapping, lack of digitization in end to end mining operations are a few important ones. This article addresses how some of them, especially digitization of various operation in mining industry, could be achieved using advanced remote sensing technique such as a hyperspectral imaging. The discussion makes a case for integrated sensing and analysis in different stages of mining value chain using a common knowledge repository which is accessible to all the connected sensing devices through a common protocol. It attempts to address how integrated sensing platform can be leveraged especially for challenging multi-scale mining and exploration operations.

Manufacturing of steel involves conversion of raw iron ores into different steel products through a complex network of unit operations. Optimizing manufacturing operations and ensuring high availability of associated equipment are the key challenges faced by plant engineers. Artificial intelligence and machine learning technologies can play an important role in this. Development and deployment of digital twins for some of the unit operations in the ironmaking process are described in this article. The generic architecture of a digital twin system is presented and its adaptation for sintering, pelletization, cokemaking and blast furnace ironmaking is explained with relevant details of their industrial scale implementation and realization of tangible business benefits. The importance of developing hybrid digital twins combining physics-based models, machine learning algorithms and domain knowledge is emphasized. Potential future directions for applying physics-informed neural networks and large language models in the development and deployment of digital twins are indicated.

Global metal industry is progressively relying on various digitalization tools i.e. information and communication technology (ICT) for improved process control and optimization. Hindalco, major metal producer, leverages high-fidelity ICT tools for smooth and optimized operation of refineries and smelters. In the present study, the application of ICT at Hindalco alumina refinery is detailed, wherein alumina is extracted from bauxite ore, which is further processed to get aluminium metal. Evaporation and calcination are key stages in Bayer process defining the quality of alumina as well as the carbon footprint. In the present work, a framework of modelling tools, which include predictive models based on the machine learning algorithm as well as physics-based models are developed for these key processes in alumina refinery. For evaporation circuit, first-principle based model using Aspen is developed to get better insights into the operation as well as to provide essential guidelines to develop ML model. Then, Random forest ML model is employed using historian data to predict steam economy. Validation using real-time DCS data on a minute-wise basis is performed. The developed model is capable of real time prediction of the steam economy within acceptable deviation of ± 5%. The model is now integrated with a control system at Hindalco alumina refinery for online monitoring as well as providing necessary predictive and corrective actions to plant personnel for stable and energy efficient operation. For calcination stage, physics-based model using CFD is developed for calciner and holding vessel to get necessary understandings into flow, temperature, and concentration profiles to predict alpha alumina generated. Additionally, extreme gradient boosting type ML model is developed for predicting alpha alumina and LOI using plant historian data. The validation showed that 77% of the predictions are falling in the acceptable range of 0–10% deviation. The predictive model as well as suggestion is now connected through graphical user interface/dashboard (GUI) in Hindalco refinery control panel for taking corrective action.

Process modelling in steelmaking started from mid-sixties and witnessed rapid growth and wide spread applications during the last fifty years or so. In the early years, key roles in steelmaking process modelling were played, mainly by researchers from Imperial College, spearheaded subsequently by three North American Professors, Brimacombe, Guthrie and Szekely, as well as their associates. By the nineties, process modelling in steelmaking became popular and a reasonably matured research approach. Significant R&D efforts continued in the academia and industry in the intervening period and these resulted in many applications of mathematical modelling in steelmaking process analysis, design, optimization and control.

Notable efforts by various groups of researchers in the arena are briefly reviewed in this work to trace the growth path of process modelling research in steelmaking. It is shown that early applications, while were primarily restricted to grossly idealized situations, presently, however, significantly more advanced and complex process models are increasingly applied to investigate steelmaking. These have accordingly led to improved understanding, providing useful insights of the underlying process dynamics. Case studies from the arena of converter steelmaking, AOD (argon oxygen decarburization), ladle metallurgy, vacuum degassing, tundish metallurgy, continuous and ingot casting are included and discussed briefly to present the state of the art. The review confirms substantial progress and suggests that future endeavors need considerably more emphasis towards integrating coupled chemical reactions, mass transfer and multiple species transport, nucleation, and growth phenomena etc., in the ambit of the currently available process models, such that fully predictive models are developed to carry out comprehensive industrial scale simulations. In such context, integration of actual steelmaking process features and concurrent industrial scale validation of mathematical models are also highlighted. A few, practically relevant but unaddressed problems, needing attention are also mentioned in the text.

Greening of coal through sustainable mining in India requires extensive deployment of technology. This article briefly highlights the role of technology in the mining value chain—exploration, mine planning, operation, monitoring, and mine closure—and suggests the way forward to make mining sustainable and cost effective.

In this paper, we propose an AI based approach to Trailer Generation in the form of short videos for online educational courses. Trailers give an overview of the course to the learners and help them make an informed choice about the courses they want to learn. It also helps to generate curiosity and interest among the learners and encourages them to pursue a course. While it is possible to manually generate the trailers, it requires extensive human effort and skills over a broad spectrum of design, span selection, video editing, domain knowledge, etc., thus making it time-consuming and expensive, especially in an academic setting. The framework we propose in this work is a template-based method for video trailer generation, where most of the textual content of the trailer is auto-generated and the trailer video is automatically generated, by leveraging Machine Learning and Natural Language Processing techniques. The proposed trailer is in the form of a timeline consisting of various fragments created by selecting, para-phrasing or generating content using various proposed techniques. The fragments are further enhanced by adding voice-over text, subtitles, animations, etc., to create a holistic experience. Finally, we perform user evaluation with 63 human evaluators for evaluating the trailers generated by our system and the results obtained were encouraging.

Students Performance (SP) analysis is regarded as one of the most important steps in the educational system for supporting students' academic success and the institutions' overall outcomes. Nevertheless, it is tremendously challenging due to the numerous details that many students have. Data Mining (DM) is the most widely used approach for SP prediction that extracts imperative information from a bigger raw data set. Even though there are various DM-centered performance prediction approaches, they all have low accuracy and high training time and don't produce the desired output. This paper proposes a hybrid deep learning framework using Deer Hunting Optimization based Deep Learning Neural Networks (DH-DLNN). A self-structured questionnaire covers all aspects of using information and communication technology, including increased access, knowledge building, learning, performance, motivation, classroom management and interaction, collaborative learning, and satisfaction. Data Cleaning and data conversion preprocess the dataset. The prediction of the student's level is then performed by extracting imperative features from the preprocessed data, followed by feature ranking using entropy calculations. The obtained entropy values are inputted into the DH-DLNN, which predicts the students' academic performance. Finally, the accuracy of the proposed system is evaluated using K-fold cross-validation. The experiment results revealed that DH-DLNN outperforms the other classification approaches with an accuracy of 96.33%.

Sleep affects our bodily functions and is critical in promoting every individual’s well-being. To that end, sleep health monitoring research has gained interest recently, including coupling data-driven AI techniques with mHealth adaptations of wearable, smartphone, and contactless-sensing modalities. Regardless, prior works, by and large, require gathering sufficient ground truth data to develop personalized and highly accurate sleep prediction models. This requirement inherently presents a challenge of such models underperforming when inferring sleep on new users without labeled data. In this paper, we propose SleepLess, which uses a semi-supervised learning pipeline over unlabeled data sensed from the user’s smartphone network activity to develop personalized models and detect their sleep duration for the night. Specifically, it uses a pre-trained model on an existing set of users to produce pseudo labels for unlabeled data of a new user and achieves personalization by fine-tuning over selectively picking the pseudo labels. Our IRB-approved user study found SleepLess model yielding around 96% accuracy, between 12–27 min of sleep time error and 18–25 min of wake time error. Comparison against other approaches that sought to predict with fewer labeled data found SleepLess, similarly yielding best performance. Our study demonstrates the feasibility of achieving personalized sleep prediction models by utilizing unlabeled data extracted from network activity of users’ smartphones, using a semi-supervised approach.