Project & Thesis Market at ITU 2024 - Master thesis & projects

Topic 1: Shop test digitalization
Keywords: Image processing, document AI, software engineering, human-in-the-loop, diverse data

The first of our three projects will try to accomplish a dream which has existed for many years in the company. We receive performance measurements of vessels when they were originally built (shop tests), and our goal is to detect performance deficiencies by comparing operational conditions with these initial performance measurement (shop test). The challenge is, that there are multiple conductors of the performance measurement process, and results are often supplied with PDF documents.

The task is produce structured information from the unstructured nature of PDFs. We know that every shop test must include the information, but the way that information is presented differs from each supplier. We have developed an old-school proof-of-concept solution centering around inferring the underlying row- and columnar format of the underlying Excel table, and creating a "signature" from this. This signature can then be compared to a memory bank of template signatures, and if we find a match, we can "overlay" the template and extract the text fields using e.g. OCR technology or using PDF metadata. One goal is to complete this "signature" method as a possible baseline, which may involve a human-in-the-loop with a LabelStudio interface.

Once a baseline is developed, you could try to improve upon it using cutting-edge technologies such as PaddleOCR table detection and using deep-learning to produce a more flexible method of extracting the information. Once you have built a solution, you can validate your method against our thousands of human annotated measurements to quantify the quantitative improvements of your method, but maybe more importantly qualitative benefits such as flexibility or the omission of a human-in-the-loop.

Succeeding in this task will improve the data quality of existing products helping to guide vessel owners towards more efficient usage of their engines.

Topic 2: Unsupervised imputation
Keywords: Generative models, timeseries data, prediction, timeseries embeddings, auto-encoder, uncertainty estimation

The real world is messy, and at MAN-ES we experience that daily. Sensors may break and stop responding, or never have been mounted at all.

We would like to be able to fallback to an estimation of a sensor. We have years of data for hundreds of vessels at a minute resolution. Can you develop a solution which can impute missing values based off other values? Training such a model should be no problem, as you can generate testing data simply by masking out a column and trying to predict the masked out values. But how do you deal with uncertainty? What if a vessel only has a hundred out of the thousands of possible signals, can we still trust your model? How (if at all) will you utilize the temporal dimension of the sensor data?

An accurate imputation model will help bring our solutions to more vessels, increasing the global impact of our services. To support you with the task, you’ll have access to subject matter experts who understand the engine behavior, placement and structure of sensors.

So, if this sounds interesting, and you have a friend or two who would like to join you in developing a solution with a real-world impact, we'd love to meet you!

Topic 3: Enhance an AI-powered chatbot with service ticket data
This project aims to enhance an existing chatbot, which currently processes manuals and engine performance data, by integrating AI capabilities to manage service tickets. The aim is to generate insightful summaries, perform detailed analysis and provide auto-generated responses within our messaging system. This enhancement will improve service operations, increase efficiency, reduce response times and increase user satisfaction.

Objectives:
Service Ticket Integration: Integrate service ticket data into the chatbot processing framework, enabling sophisticated analysis and automated responses.

AI-driven summarization: Develop advanced algorithms using state-of-the-art techniques to create concise, informative summaries of service tickets.

In-depth data analysis: Implement analytical tools to uncover patterns, trends and common issues in service tickets, facilitating proactive support and service improvements.

Automated response generation: Enhance the chatbot to deliver accurate, auto-generated responses to frequently asked questions and common service issues, informed by analyzed ticket data.

About the product:
PrimeServ Assist is an engine performance monitoring platform with 24/7 technical support. The cloud-based platform is accessible via web application or mobile app. Even when customers are not logged in, their 2-stroke engines are actively monitored. Experienced marine engineers contact customers when anomalies occur, and actions need to be taken to improve engine performance. They then create a service ticket to record the communication with the customer.

The constant flow and analysis of data provides customers with an accurate status of their engines and advice on how to keep them running at peak efficiency for longer.

About the team:
Join Our Dynamic Team at Group Digital!

Are you ready to be part of an exceptional team of highly motivated professionals? Look no further! At Group Digital, we’re not just busy—we’re driven by a shared purpose: creating tangible value through our projects. Here’s why you’ll thrive with us:

Collaboration at Its Best:
Our full-stack development teams bring together diverse skill sets under one roof. Data scientists, back-end developers, front-end experts, and subject matter specialists collaborate seamlessly.

While we work closely within our team, we also foster cooperation with other departments. It’s all about synergy!

Creating Real Value:
We don’t just write code; we build solutions that matter. Our projects directly impact our customers and organization.

Whether it’s enhancing user experiences or streamlining processes, your work will leave a lasting mark.

Global Connections:
Our team spans the globe—from Copenhagen to the USA, India, Germany, Spain, and Switzerland. You’ll interact with colleagues from diverse backgrounds.

Highlight your cross-cultural collaboration experience on your CV—it’s a valuable asset!

Support When You Need It:
The team recognizes the value of your project. Rest assured, you’ll receive the support you need, whether it’s answering questions or overcoming challenges. Each project will have dedicated person to coordinate but you will be interacting with the whole team to ensure best outcomes.

Ready to make an impact? Join us today!

Lokation: Teglholmsgade 41, 2450 København SV

Application deadline: 30 Sep 2024