http://dspace.dtu.ac.in:8080/jspui/handle/123456789/108 2026-07-23T07:14:44Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/22973 Title: EXPLAINABLE AI BASED PREDICTIVE MAINTENANCE AND ANOMALY DETECTION SYSTEM FOR INDUSTRY 4.0 USING MACHINE LEARNING AND DEEP LEARNING Authors: SOREN, SUNNY; Srinivas, K. (SUPERVISOR); Ansari, Naushad A.(CO- SUPERVISOR) Abstract: The advent of Industry 4.0 has revolutionized the way manufacturing companies operate by monitoring, maintaining and optimising industrial equipment. One of the most effective uses of this transformation is Predictive Maintenance (PdM), which moves maintenance from a reactive and time-based approach to a data-driven, condition-based and anticipatory approach. There is a major lack in the literature and currently in systems: most of the predictive maintenance models focus either on predictive accuracy or explainability, or on explainability without considering robustness under realistic data scenarios, like sensor noise and class imbalance. This big project proposes an Explainable AI (XAI) framework for Predictive Maintenance and Anomaly Detection System, designed and tested with the widely used UCI Machine Learning Repository AI4I 2020 Predictive Maintenance Dataset. To systematically compare the three representative machine learning models, Logistic Regression (LR), Random Forest Classifier (RF) and a Feed-Forward Neural Network (Multi-Layer Perceptron, MLP), three carefully designed experimental conditions are provided: (1) a clean baseline dataset, representing ideal data quality; (2) a synthetically noised dataset, with Gaussian noise added to the data, representing real world sensor drift, calibration errors and measurement uncertainty; and (3) a synthetically over-sampled dataset (via Synthetic Minority Over-sampling Technique, SMOTE), to address the class imbalance encountered in industrial failure prediction tasks. The project puts in place an Autoencoder-based unsupervised module for anomaly detection in addition to supervised classification. The Autoencoder is trained with operational data without any failure labels and considers a failure as being flagged when the reconstruction error exceeds the 95th percentile set using a statistically motivated threshold, which is determined by the learned data, setting the limit to be the values of the normal machine behaviour. Two complementary approaches are used to obtain model explainability: using SHAP (SHapley Additive exPlanations) with the Random Forest through TreeExplainer, and LIME (Local Interpretable Model-Agnostic Explanations) with the Neural Network. SHAP offers explanations at the individual instance level using waterfall plots and explanations at a higher level of abstraction by ranking the most important global features via a beeswarm plot, the mean absolute SHAP bar plot, and a feature interaction dependence plot. Local linear surrogate explanations are offered for individual predictions by LIME, allowing maintenance engineers to grasp the model behaviour at the instance level. Experimental results indicate that the Random Forest model gives the best F1 score (1.00) for the clean dataset, followed by SMOTE augmented data (0.99), and lowest under Gaussian noise (0.95), which is the highest noise robustness among the three models. Even on well-structured tabular PdM data, Logistic Regression shows a consistent F1-score of 0.99 for all three data conditions, making it a very well performing simple and interpretable model that remains competitive over a variety of data conditions. The highest variance is in the case of the Neural Network, which achieves F1=0.93 for clean data, F1=0.98 for noisy data (with implicit regularisation effects of the added noise), and F1=0.93 for SMOTE data. The binary failure mode flags — PWF (Power Failure), HDF (Heat Dissipation Failure), OSF (Overstrain Failure) — clearly stand out as the most important failure predictors, while Rotational Speed and Torque are the most impactful continuous feature drivers. LIME explanations from the Neural Network align with SHAP results, thereby offering cross-method validation of explainability results. The distribution of the reconstruction errors of the Autoencoder reveals the separation between the normal and anomalous instances, and the Autoencoder is able to isolate operational records that are anomalous. The results above clearly proof the synergy of the classical and deep learning models with the explainable AI, unsupervised anomaly detection, resulting in a strong, explainable, practically viable PdM solution that can be well adapted to the Industry 4.0 manufacturing environments. It is executed with open source Python libraries and is accessible on standard hardware, so it can be utilized by small and medium scale enterprises without any special computational hardware. 2026-05-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/22969 Title: FREE INFLATION OF AN ANISOTROPIC HYPERELASTIC CIRCULAR MEMBRANE Authors: BISHWAKARMA, SUNIL; Sahu, Satyajit (SUPERVISOR) Abstract: Inflation of a thin elastic membrane is a well-studied classical problem in nonlinear continuum mechanics; however, the impact of material anisotropy on the inflation response of hyperelastic materials has yet to be thoroughly researched. Therefore, this research examines the axisymmetric axial deformation behaviour of a circular membrane composed of an incompressible anisotropic Mooney-Rivlin material being blown up against a uniform transverse pressure load, without any initial radial pre strain. The derivation of governing equations comes from the principle of stationary potential energy, with the strain energy density function being expressed using the two principal stretches from the plane of the membrane and an anisotropic invariant 𝐼4 = 𝜆1 2 for the thickness of the membrane. The invariant thus gives rise to one stretch in the thickness direction, since there can be no volume change under the assumption of incompressibility. Each of these principal stretches is identified by introducing the appropriate field variable so that the governing equations are reduced from two second-order ordinary differential equations into a system of three first-order ordinary differential equations. The converting or changing of the governing equations into a different format allows for easier numerical integration using standard initial value problem solvers. The two-point boundary value problem resulting from this research has been solved using the combination of the shooting method and the optimization procedure using fminsearch. The primary advancement in the numerical solution is the established verification of the scaling invariance property inherent to the equilibrium equations applies to the anisotropic equations, thus allowing for the elimination of the arc-length continuation methods in all cases where the pressure-stretch relationship has limit point instability. Presented here are numerical results for two different pairs of Mooney-Rivlin material parameters, 𝛼 = 𝐶2 𝐶1 =0.01,0.03,0.1 and four different levels of anisotropy, 𝜁 = 0,0.01,0.03,0.05, with no pre-stretch (𝛽 = 1). The results show that anisotropy causes a stiffening effect in the inflation response, through the pressure-deflection curves. The pressure-deflection curves for 𝛼 = 0.01 (close to Neo-Hookean behaviour) show that the limit point is always there regardless of connectivity level 𝜁. Critical pressures remain about the same for each value of 𝜁. When working with 𝛼 = 0.1 (which has a large component for 𝐶2), the response is monotonic and stable; but as the value of 𝜁 increases, maximum deflections will decrease by as much as 56 at a pressure of 10. The stress resultants show that inflation creates anisotropic behaviour in the meridional and circumferential components of the material properties (highest stresses at the clamped edges). The peak stress at the clamp reduces and circumferential stress increases at the interior of the structure when the anisotropy parameter is increased. The 3D surface reconstruction and membrane profiles have shown that hysteresis in the anisotropic membrane is less than that of the isotropic membrane at similar elevations (pressure). This study will benefit the designer of any inflatable structure requiring uniform strains and controlled directional stiffness. 2026-06-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/22945 Title: CRITICAL FACTORS FOR GENERATIVE AI- DRIVEN GREEN VALUE CREATION IN SUPPLY CHAINS: A HIERARCHICAL FUZZY BEST WORST METHOD APPROACH Authors: SISODIYA, ABHISHEK KUMAR; Kumar, Pravin (SUPERVISOR) Abstract: Green value creation has become a central objective for modern supply chains as organizations increasingly adopt circular and sustainable practices. While prior research has identified multiple environmental and sustainability-related factors across supply chain activities, the role of advanced digital technologies—particularly Generative Artificial Intelligence (GenAI)—in shaping and prioritizing these factors remains insufficiently explored. Moreover, existing studies often rely on conventional analytical approaches and lack structured decision-making frameworks capable of addressing uncertainty and expert subjectivity. This study aims to identify and prioritize the critical factors influencing GenAI-driven green value creation in supply chains using a Hierarchical Fuzzy Best–Worst Method (HFBWM) approach. Based on an extensive review of the literature and expert consultation, five key supply chain dimensions—Supplier, Product, Packaging, Logistics, and Consumption—along with eighteen associated sub-factors are identified and validated. The HFBWM is employed to systematically capture expert judgments under uncertainty and to derive local and global priority weights. The results reveal that product-related factors, particularly design for reuse, modular design, and circular product design, are the most influential drivers of green value creation, followed by sustainable packaging and consumption-oriented factors. Scenario-based analysis further demonstrates that GenAI capabilities—through iv interactive and non-interactive knowledge search—enhance decision-making quality, reduce dependence asymmetry, and strengthen inter-organizational collaboration, thereby reshaping the prioritization of green value creation factors. The study contributes to the literature by integrating fuzzy multi-criteria decision making with GenAI-enabled supply chain capabilities and offers a practical decision support framework for managers seeking to prioritize high-impact sustainability initiatives. The proposed approach provides actionable insights for leveraging GenAI to support strategic green value creation in complex and uncertain supply chain environments. 2026-06-01T00:00:00Z http://dspace.dtu.ac.in:8080/jspui/handle/repository/22926 Title: NUMERICAL ANALSIS AND MULTI-OBJECTIVE OPTIMISATION OF THE NACA 2415 AIRFOIL USING A TAGUCHI-FUZZY FRAMEWORK Authors: SONI, PULKIT; Zunaid, Mohammad (SUPERVISOR) Abstract: Small fixed-wing unmanned aerial vehicles operating at chord Reynolds numbers of Re = 10⁵–5×10⁶ face a fundamental aerodynamic challenge: the gap between two dimensional section performance predicted by computational fluid dynamics and the drastically degraded efficiency of real, aspect-ratio-constrained three-dimensional wings. This study addresses that gap through a three-phase hierarchical investigation of the NACA 2415 aerofoil, integrating Taguchi design of experiments, Mamdani fuzzy multi-objective optimisation, steady-state Reynolds-Averaged Navier–Stokes simulation, and surrogate-assisted optimisation into a single systematic framework validated against NACA Technical Report 824 experimental data. Phase 1 deploys a Taguchi L25(5⁵) orthogonal array — reducing a 3,125-run full factorial to 25 balanced simulations — to simultaneously screen five RANS turbulence closures (Spalart–Allmaras, k-ε Realizable, k-ω SST, SST γ–Reθ, and Reynolds Stress Model), five Reynolds numbers (1–12×10⁶), five angles of attack (−4° to 16°), five turbulence intensities (0.05%–5.00%), and four surrogate optimisation strategies (RSM-Kriging, NSGA-II, Sparse RSM, and Neural Network Screening). Three conflicting aerodynamic responses — lift coefficient, drag coefficient, and lift-to-drag ratio — are unified into a scalar Multi-Performance Characteristic Index via a 27-rule Mamdani fuzzy inference system with corrected strict-inequality boundary membership evaluation, a previously unreported defect whose correction changes the turbulence model ANOVA contribution from a spurious 12.30% to the physically correct 1.83%. One-way ANOVA identifies angle of attack as the dominant factor (ρ = 80.99–85.70%), with Reynolds number second (ρ ≈ 8–10%). The k-ω SST model achieves the highest multi-objective η(MPCI) level mean (−9.202 dB) due to its Bradshaw adverse-pressure-gradient limiter and structural turbulence-intensity insensitivity via cross-diffusion. Sparse RSM achieves the highest Weighted Composite Score of 9.13/10, uniquely detecting the NACA 2415 drag-bucket interior minimum at α ≈ −0.75°, independently confirmed by Neural Network Screening at α ≈ −0.77°. Phase 2 deploys a Taguchi L9(3³) array exclusively with k-ω SST across a refined design space (Re = 6–12×10⁶, α = 4°–8°, TI = 0.05%–0.50%). The confirmed optimal configuration — Re = 12×10⁶, α = 8°, TI = 0.10° — yields CL = 1.038, CD = 0.015711, and |CL/CD| = 66.08, with a Taguchi additive model prediction error of only 0.26%, validating negligible factor interactions. Turbulence intensity contributes ρ ≈ 0.00% (F = 0.04) within the tested range, providing a practically significant result that eliminates TI as a source of CFD modelling uncertainty for this application. Phase 3 extends the Phase 2 optimum to a three-dimensional finite-wing RANS simulation at AR = 0.25 (b = 0.5 m, c = 2.0 m, A_ref = 1.0 m²). The aerodynamic outputs — CL = 0.12921, CD = 0.014899, |CL/CD| = 8.67, Lift = 607.908 N, Drag = 70.096 N — reveal an 86.9% efficiency collapse from the two-dimensional optimum, driven by a tip-vortex-induced downwash of ε ≈ 9.33° that reduces the effective angle of attack from +8° to approximately −1.43°. Three independent CFD visualisations — velocity pathlines, static pressure vectors, and velocity magnitude vectors — provide mutually corroborating topological, thermodynamic, and kinematic evidence confirming that the entire span lies within the tip-vortex induction zone and no two dimensional flow region exists. The study conclusively establishes that the binding aerodynamic limitation of the platform is planform geometry rather than section performance, motivating a redesign to AR = 6–8 to recover 75%–86% of the two dimensional efficiency ceiling. 2026-05-01T00:00:00Z