Open Projects and Research Positions

Industrial Supported BS, MS and PhD research projects

Government department of Forestry

1. Identification of Human Settlement and Other Indicators using GIS data for given point on map (Deployment as services on Kubernetes Platform)

ZeeCube

1. Machine Learning for Saffron Growth and Disease Detection in High-Altitude Regions
2. Predictive Modeling of Saffron Stigma Characteristics Using AI and Sensor Technologies
3. AI-Driven Predictive Modeling for Saffron Yield and Growth Optimization
4. AI-Powered Innovations in Saffron Cultivation: Collaborative Approaches to Sustainability

Sybrid
Title: Development of Embedding Models for Local Languages (Urdu, Punjabi, Pashto, Sindhi)

Natural Language Processing (NLP) has made significant strides with the advent of embedding models, enabling machines to process human languages with greater accuracy. However, existing embedding models are primarily designed for widely spoken languages such as English, often neglecting less represented languages like Urdu, Punjabi, Pashto, and Sindhi. This thesis addresses this gap by focusing on the development of embedding models tailored for these local languages.
The research explores the limitations of current embedding models in handling the linguistic nuances of these regional languages. The lack of large-scale, high-quality datasets and language-specific fine-tuning are identified as major challenges. This thesis proposes novel methods to overcome these challenges by enhancing embedding models specifically for Urdu, Punjabi, Pashto, and Sindhi. The study also involves the creation of a robust dataset that captures the unique characteristics of these languages.
The developed models will be evaluated in various NLP tasks, such as text classification, to demonstrate their effectiveness and reliability in processing regional languages. This research will pave the way for better NLP tools that cater to a wider linguistic audience, improving accessibility and representation for underrepresented languages in the digital domain.
More details

Optimization and Control in Energy Systems and AI

A Distributed Modular Co-simulation Platform for Local Energy Communities with Encapsulated Dynamics (Ref:2024-1)
Peer-to-Peer energy trading, energy sharing, self-optimization, or grid-supporting behavior are expected to become essential applications for energy communities and their participants. Due to the high complexity of physical and virtual energy flows, financial flows, and the optimization of controllable devices within a community, automated (digital) solutions are needed to handle these requirements.
Positions available: 2 Master thesis & 2 BS thesis.
Industrial partner: Austrian Institute of Technology.
More details

Optimization of Lithium-ion Battery Performance and Longevity using Reinforcement Learning on a Digital Twin by Optimal Charging and Thermal Management (Ref:2024-2)
Application of Reinforcement Learning to the digital twin of the Li-ion battery system that can improve the performance considering the dynamic health of the system, its usage and anticipated usage.
Positions available: 2 Master thesis & 2 BS thesis.
Industrial partner: Coventry University, UK, MIRA research center, UK
More details

Optimal control of an aggregate energy storage digital twin model of EV fleet for grid support (Ref:2024-3)
Digital twin for fleet of electric vehicles that can simulate an aggregate battery storage model & EV integrated can be supported by the grid services provided by this digital twin model. Positions available: 1 Master thesis & 1 BS thesis.
Industrial partner: Coventry University, UK, MIRA research center, UK
More details

Intelligent use of energy Storage in Energy Management System for buildings (INSEMS) (Ref:2024-4)
Positions available: 1 Master thesis & 1 BS thesis.
Industrial partner: Austrian Institute of Technology
More details

High Performance Computing

1. Integration of Open Source Billing Platform with Hypercoverged Infrastructure such as Proxmox, (Ref:2024-5)
Positions available: 1 BS thesis.
Industrial partner: Forbmax

2. Hosting web-services and CUDA Labs from Kubernetes Cluster with Aggregation of multiple GPUs, (Ref:2024-6)
Positions available: 1 BS thesis.
Industrial partner: Forbmax

3. Development of Wordpress based Online Learning Platform with the integration of Payment Gateways and Examination Platform, (Ref:2024-7)
Positions available: 1 BS thesis.
Industrial partner: Cyantech

Denim 1. Counter Matching System for Fabrics

Objective: Develop an AI-powered system to scan fabric swatches received from customers and identify the best matching fabric from Denim’s existing library. This system will help match key parameters such as color, slub, twill, fabric weight, and stretch. Additionally, the system should be capable of analyzing images sent by customers for accurate fabric matching.

Research Focus: a) AI and image processing techniques for fabric identification. b) Data-driven matching algorithms for textile parameters. c) Integration of color and texture recognition. d) Development of a user-friendly interface for fabric comparison.

Potential Impact:
This system will streamline the process of fabric selection, saving time and reducing errors in customer orders, leading to higher customer satisfaction and optimized inventory usage.

2. Counter Development System for Fabric Construction

Objective: Create an AI-based system that can suggest the optimal base fabric construction by analyzing the desired outcomes entered by the user, such as stretch, texture, weight, and durability.
Research Focus: a) Machine learning models to predict fabric construction. b) Predictive modeling based on textile properties. c) Automated suggestions for fabric composition and weave patterns. d) Collaboration with textile experts to develop a comprehensive fabric library.

Potential Impact:
The system will reduce the manual effort of fabric construction design, increasing efficiency in fabric development and allowing for faster response to customer demands for new fabric types.

3. Slub Pattern Analysis System

Objective: Build a system that can scan a fabric sample and provide detailed analysis of its slub pattern, including slub length, pause length, and slub thickness, crucial for understanding fabric aesthetics and performance.
Research Focus:
a) High-resolution image processing and pattern recognition. b) Algorithms for analyzing irregularities in fabric textures. c) Development of robust slub metrics. d) Integration with existing textile quality control systems.

Potential Impact:
This project will significantly improve Denim’s quality control processes by automating slub pattern detection, ensuring consistency in fabric production and offering better customization options to customers.

4. Auto Bleaching System for Denim Garments

Objective: Design an AI-based auto-bleaching system that can scan the shade of a denim garment and automatically adjust the bleaching process to achieve the desired shade. This system aims to replace the current manual bleaching practices, which are time-consuming and variable.

Research Focus: a) AI-driven image analysis for accurate shade detection. b) Development of automated bleach application techniques. c) Feedback loops for real-time shade adjustment. d)Integration with existing garment production lines.

Potential Impact:
The auto bleaching system will lead to greater precision in achieving desired shades, reducing waste, and making the bleaching process faster and more environmentally friendly, giving Denim a competitive advantage in sustainable fashion.

PhD Positions

Dr. Khan is HEC recognized PhD Supervisor, accepting self/organization funded proposals for the following PhD positions at Pak-Austria Fachhochschule: Institute of Applied Sciences and Technology.

On the Use of Reinforcement Learning Techniques of Advanced Distribution Management Systems with Policy Gradients, (Ref:2024-PhD-1)
This research will investigate how policy gradient-based RL algorithms can be leveraged to improve decision-making processes within ADMS, with a particular focus on real-time optimization tasks such as load balancing, voltage regulation, and fault management. By directly learning optimal policies from data through interaction with the environment, policy gradient methods offer a promising approach to address the dynamic and stochastic nature of distribution networks. The study will involve the development and simulation of novel RL-based models, followed by validation using real-world data and scenarios. The outcome of this research has the potential to significantly advance the state-of-the-art in smart grid management, offering more adaptive, scalable, and efficient solutions for distribution network operators.
Jointly supervised by: University of South Wales, UK

Improvement of Hosting Capacity in Unbalanced Distribution Grids by Topology Control Activated Using Reinforcement Learning Agents, (Ref:2024-PhD-2)
The proposed study aims to address these challenges by leveraging RL techniques to dynamically manage grid topology, optimizing the configuration of network switches and reclosers in real-time. RL agents will learn to identify optimal topology changes that enhance grid stability and hosting capacity, particularly in unbalanced networks where phase imbalances can lead to significant operational issues. The research will develop and test RL-based control strategies in simulated environments, with validation against real-world grid data and scenarios. The anticipated outcomes of this research include the development of intelligent, adaptive control strategies that significantly improve the hosting capacity of unbalanced distribution grids, thereby supporting the integration of higher levels of renewable energy and improving overall grid resilience and efficiency.
Jointly supervised by: University of South Wales, UK

Note: Interested students may send their interest by email to: sohaildotkhan@spcai.paf-iast.edu.pk (remove “dot” from email) with your CV/CVs attached and a reference number of position mentioned as e.g., Ref:2024-1 in the email subject. Looking forward!