http://repository.iiitd.edu.in/xmlui/handle/123456789/874 Sat, 11 Apr 2026 11:49:02 GMT 2026-04-11T11:49:02Z http://repository.iiitd.edu.in/xmlui/handle/123456789/948 Data-driven detection and isolation of wear-induced faults in industrial robot V, Sathish; Butail, Sachit (Advisor); Orkisz, Michal (Advisor) Industrial robots are complex systems that require technical expertise for condition monitoring and diagnostics. In small and medium scale industries, adequate skilled resources may not be available to monitor the robots consistently. In such cases, domain experts can provide maintenance advisory based on data collected through remote monitoring. The goal of this thesis is to develop and validate data-driven approaches to detect and identify the source of mechanical degradation in industrial robots using data typically collected through remote monitoring. Performance of data-driven methods is influenced by both data and algorithms. The factors that influence the data include robot application, tasks, and environment. Algorithms differ in terms of the extent to which they rely on an underlying model, and how they weight bias versus variance to isolate anomalous situations. With a remote monitoring solution, knowledge about all the factors may not be available and thus create uncertainty about the data generation process. Therefore, we adopted a two-pronged approach for the study: a) evaluate data-driven methods on simulated data, b) apply the algorithms that worked well with simulated data on real data enhanced with preprocessing methods. For evaluating data-driven methods, we identified wear through significant changes in torque values from normal operating conditions using principal component analysis and studied the effect of source and type of training data on detecting failures in industrial robots. Towards application of these results on real data, we next formulated strategies to detect the occurrence of wear-induced fault using supervised learning algorithm in a systematic hypothesis-driven study. That study sought to identify effective combinations of four preprocessing techniques on data collected from twenty-six industrial robots. Our results show that preprocessing techniques improved the fault detection performance. Finally, we investigated the problem of isolating the axis of fault by inferring pair wise directional relationships between all axes using an information-theoretic approach called Transfer Entropy (TE). The approach was validated on simulated data generated with an in-house robotic simulation tool. The axis responsible for wear was always detected when the wear was 10% above nominal value. The results of these two studies form the basis for informed data-driven strategies for fault detection and isolation in industrial robots and sets the stage for advanced adaptive detection approaches. Mon, 01 Feb 2021 00:00:00 GMT http://repository.iiitd.edu.in/xmlui/handle/123456789/948 2021-02-01T00:00:00Z http://repository.iiitd.edu.in/xmlui/handle/123456789/947 Facial analysis under low resolution and disguise variations Singh, Maneet; Singh, Richa (Advisor); Vatsa, Mayank (Advisor) Automated facial analysis has widespread applicability in scenarios related to image tagging, access control, and surveillance. Initial research focused primarily on face recognition in constrained settings, where the captured face image had variations due to pose, illumination, or expression. With the increased applicability of facial analysis models in real world scenarios, dedicated research was required for data captured in unconstrained settings including resolution variations. When subjects are captured at a large stand-off distance from the acquisition device, the resulting region of interest (ROI) capturing the face is often small (less than 32 ⇥ 32), requiring recognition of low resolution or very low resolution facial regions. Data captured in such unconstrained scenarios also often contain people using different disguise accessories or occluded faces, resulting in the obfuscation of the face region, rendering automated face recognition challenging. To this effect, this dissertation focuses on facial analysis under low resolution and disguise variations. Two facial recognition algorithms have been presented for data captured in low and very resolution settings: Dual Directed Capsule Network and DeriveNet model, followed by two novel datasets (Disguised Faces in the Wild 2018 and 2019) for facilitating research on disguised faces in the wild along with a Disguise-Resilient face verification framework. This is followed by designing facial analysis models for attribute prediction in low and very low resolution settings. We begin with developing deep learning algorithms for low or very low resolution face recognition, which suffers from the challenge of limited interpretable information in the face images, thus resulting in ineffective feature extraction and classification. In order to address this challenge, we propose two novel algorithms: Dual Directed Capsule Network (DirectCapsNet) and DeriveNet model. Since low resolution face images contain limited meaningful information, we propose utilizing a small set of high resolution samples for directing the classification model towards learning richer features. The DirectCapsNet is built using a combination of convolutional and capsule layers, and is trained via three loss functions: HR-Anchor loss, Reconstruction loss, and Margin loss. DeriveNet thus learns rich feature representations for very low resolution samples by utilizing the auxiliary high resolution samples during training. While capsule layers encode rich features, they are computationally expensive and contain a larger number of trainable parameters. In order to address the above limitation, a novel DeriveNet model has been proposed for low and very low resolution face recognition. The proposed model utilizes a set of high resolution images for learning an effective recognition model via combination of two loss functions: Derived-Margin softmax loss and Reconstruction-Center loss. The proposed Derived-Margin softmax loss estimates the inter-class variations between low resolution samples and models it as a margin for learning improved classification boundaries. Experimental analysis is performed on different challenging real-world datasets including the Unconstrained College Students (UCCS) dataset for facial regions having less than 24 ⇥ 24 resolution. Comparison with recent techniques demonstrates state-of-the-art results by the proposed algorithm. The next contribution of this dissertation lies in the area of disguised face recognition, where individuals attempt to obfuscate the face region, either intentionally in order to fool the automated system, or unintentionally by the use of day-to-day facial accessories. To the best of our knowledge, most of the research focused on disguised face recognition in constrained scenarios, with limited disguise accessories and other variations. Therefore, as part of this dissertation, we propose the Disguised Faces in the Wild (DFW) 2018 and DFW2019 datasets containing face images with unconstrained disguise variations, captured across different resolutions, acquisition devices, lighting, pose, and expression. The datasets were released as part of two international workshops for facilitating research in this direction. We also present the Disguise-Resilient framework using a novel Disguise Encoder-Decoder network, with application to face verification. The efficacy of the proposed framework has been demonstrated on the challenging DFW2018 and DFW2019 datasets, where it achieves state-of-the-art performance. Further, the arduous task of disguised face recognition in low resolution settings has also been explored and presented to the research community. Baseline results and performance of the proposed framework for face images with resolutions varying from 32⇥32 to 16⇥16 demand dedicated research focus from the community. The final contribution of this dissertation focuses on developing deep learning algorithms for learning discriminative features, with application to attribute classification in low resolution face images. Automated prediction of attributes such as gender (male/female) or adulthood (adult/child) can be useful as ancillary information for person identification, enhanced human computer interaction, or for restricting age-based access. As part of this contribution, two supervised variations of the deep learning based Autoencoder model are proposed for learning class-specific features: Class Specific Mean Autoencoder and Class Representative Autoencoder. Both models utilize the concept that the mean feature of a given class contains class-specific information which can be incorporated for learning discriminative rich features. To the best of our knowledge, this is one of the initial research focused on analyzing attributes in low resolution facial regions. The proposed autoencoder models are able to extract meaningful information by modeling the inter-class and intra-class variations, resulting in improved performance for low resolution attribute classification from face images. Experimental evaluation on different datasets for facial images of 24 ⇥ 24 and 16 ⇥ 16 resolution demonstrate the effectiveness of the techniques. Wed, 01 Dec 2021 00:00:00 GMT http://repository.iiitd.edu.in/xmlui/handle/123456789/947 2021-12-01T00:00:00Z http://repository.iiitd.edu.in/xmlui/handle/123456789/944 Statistical and machine learning-based approaches to precise characterization of cellular phenotypes Gupta, Krishan; Sengupta, Debarka (Advisor); Ghosh, Abhik (Advisor); Ahuja, Gaurav (Advisor) Delineation of the complex layers of biological system requires a cumulative effort from multiple disciplines of science. The present thesis work utilizes some of the interdisciplinary approaches by combining the automation and accuracy of computation to the in-depth concepts of Biology. In my thesis, I have addressed three fundamental biological problems. In one of my initial projects, I developed a computational framework by utilizing Machine Learning-based approach to build a classification model for the detection of Circulating Tumor Cells (CTCs). Moreover, I validated the authenticity of our model on a large number of publicly available scRNA-seq datasets and a newly generated CTC dataset of breast tumour cells, captured using a newly developed microfluidic system for label-free enrichment of CTCs. In my second project, I utilized single cell genomics approach coupled with stringent statistical and structural biology frameworks to dissect the cellular basis of the loss of smell in COVID-19 infected patients. Of note, one of the prevalent, but largely ignored symptoms during the early COVID-19 pandemic was the loss of smell and taste. Our work utilized the known information about the viral entry proteins, and viral-human protein-protein interaction map. Our integrative analysis clearly suggests that the non-sensory (sustentacular, Globolar Basal Cells and Bow-man’s gland) cell-types are vulnerable to SARS-CoV-2 infection. In my third project, I explored the potential of modelling expression-ranks, as robust surrogates for transcript abundance. Here I examined the Discrete Generalized Beta Distribution (DGBD) performance on real data and devised a Wald type test to compare gene expression between two phenotypically divergent groups of single cells. We carried out a comprehensive assessment of the proposed method, to understand its advantages as compared to some of the current best practice approaches. In addition to striking a reasonable balance between Type 1 and Type 2 errors, we concluded that with increasing sample size, Rank Order- Sequencing (ROSeq), the proposed differential expression test, is remarkably robust for expression noise and scales rapidly. Fri, 01 Oct 2021 00:00:00 GMT http://repository.iiitd.edu.in/xmlui/handle/123456789/944 2021-10-01T00:00:00Z http://repository.iiitd.edu.in/xmlui/handle/123456789/941 Detecting and reasoning collusive activities in online media Dutta, Hridoy Sankar; Chakraborty, Tanmoy (Advisor) Online media platforms have enabled users to connect with individuals, organizations and share their thoughts. Other than connectivity, these platforms also serve multiple purposes - education, promotion, updates, awareness, etc. Increasing the reputation of individuals in online media (aka social growth) is thus essential these days, particularly for business owners and event managers who are looking to improve their sales and reputation. The natural way of gaining social growth is a tedious task, which leads to the creation of unfair ways to boost the reputation of individuals artificially. We refer to such unfair ways of bolstering social reputation in online media as collusion. This thesis covers various aspects of collusion: a large-scale analysis of collusive entities and designing state-of-the-art models for detection of collusive entities in multiple online media platforms. First, we design approaches using user’s metadata properties to identify collusive users on Twitter who request for artificial retweets from the blackmarket services. Here, we also explore the differences between the working of various types of blackmarket services. Second, we extend our previous approaches to identify collusive Twitter users using user’s network properties. Third, we consider another type of collusive Twitter appraisal (followers) and study the collusive entities present in another online media platform (YouTube). Fourth, we propose an approach to detect core users of the blackmarket services and show the differences in the working of core and non-core users. Finally, we release a multi-platform data repository of collusive entities collected from two blackmarket services. Wed, 01 Sep 2021 00:00:00 GMT http://repository.iiitd.edu.in/xmlui/handle/123456789/941 2021-09-01T00:00:00Z