Bachelor and Master Theses

Background

The undercarriage of heavy machinery, such as excavators and bulldozers, consists of critical components, including track chains, rollers, idlers, and sprockets. These components are continuously exposed to mechanical stress, friction, and harsh operating environments, resulting in progressive wear and deterioration. Accurate assessment and prediction of component conditions are essential for predictive maintenance, reducing operational costs, and improving equipment reliability and safety.

Recent advances in computer vision and deep learning have enabled automated inspection and wear detection from images and videos. Generative deep learning models, particularly diffusion models, have demonstrated significant capabilities in generating realistic images and learning complex visual patterns. These models offer opportunities to simulate potential future conditions of machinery components based on their current appearance and operational wear characteristics.

Generating realistic future wear conditions presents several challenges, including limited historical image data, variations in machinery configurations, and the difficulty of maintaining consistency between generated images and actual physical wear progression. This thesis aims to investigate generative deep learning approaches for predicting and visualizing potential future wear conditions of undercarriage components, particularly sprockets, using images captured during machinery inspections.

Aim and Objectives

The thesis aims to develop and evaluate a generative deep learning framework capable of producing realistic images representing potential future wear conditions of heavy machinery undercarriage components, supporting visual condition assessment and predictive maintenance.

The specific objectives are to:

1.    Investigate generative deep learning models, particularly diffusion models, for generating images of undercarriage wear.

2.    Develop an image-generation pipeline that produces potential future wear conditions from current component images and specified wear levels.

3.    Explore conditional image generation techniques to control the progression of wear in generated images.

4.    Evaluate the visual realism, structural consistency, and wear-level accuracy of generated images.

5.    Investigate the potential application of generated images for predictive maintenance and condition monitoring.

Expected Outcomes

• Conditional Generative Pipeline for Wear Progression Synthesis: Development and implementation of a domain-adapted deep generative framework (e.g., conditional diffusion) capable of rendering high-fidelity, time-dependent visual wear trajectories of heavy machinery undercarriages conditioned on baseline images and target degradation levels.

• Empirical Validation and Industrial Feasibility Assessment: A rigorous multi-metric evaluation assessing synthesized image realism, structural consistency, and wear progression accuracy, alongside a critical analysis of diffusion-based models' viability and constraints within predictive maintenance (PHM) workflows.

Core Computer Vision & Deep Learning: Strong foundational knowledge of deep learning architectures, convolutional neural networks (CNNs), and computer vision fundamentals (image preprocessing, feature extraction, and spatial transformation).

Generative AI & Image-to-Image Pipelines: Familiarity with generative models—specifically Diffusion Models (e.g., DDPM, Latent Diffusion, Stable Diffusion), Generative Adversarial Networks (GANs), and Conditional Image Generation techniques (e.g., ControlNet, IP-Adapter).

Programming & Framework Proficiency: Advanced proficiency in Python and standard deep learning frameworks (PyTorch or TensorFlow), alongside experience with open-source generative libraries (e.g., Hugging Face diffusers, torchvision).

Volvo Construction Equipment (VCE)

Volvo Construction Equipment (VCE), a division of Volvo Group, is one of the world's largest manufacturers of construction equipment such as wheel loaders, haulers, excavators and compact machines.

At VCE, we are at the forefront of innovation in the construction equipment industry. We focus on data-driven decision-making across the entire business spectrum. With numerous ongoing exploratory projects, we provide a fertile ground for curious minds to delve into diverse domains and leave a lasting impact.

We seek individuals with innovative mindsets who can provide unexpected solutions and think outside the box. By joining us, you will contribute to research and findings that will help shape the direction of our business. Together, we will explore untapped potential, paving the way for a more efficient, intelligent, and sustainable future.