Leveraging AI for Complex Scheduling

Abstract

Scheduling in laboratory workflows, such as pipetting, plate movements, and washing, is a complex optimization problem often addressed using traditional scheduling solvers. While effective for standard tasks, these solvers struggle with intricate constraints like synchronization, incubation timing, and multi-sample workflows, often resorting to heuristic methods that can produce sub-optimal results. This thesis, proposed by Tecan Austria GmbH in Salzburg, explores leveraging artificial intelligence (AI) to address these limitations and provide faster, more adaptive, and optimized scheduling solutions for laboratory automation. The study focuses on enhancing Tecan’s Fluent Control system, which currently relies on Google’s OR-Tools and heuristic methods for schedule generation. Although effective, these methods face challenges in adapting to dynamic real-time changes and computational bottlenecks in complex scenarios. By integrating AI with traditional techniques, this research aims to develop a robust and scalable scheduling model that improves performance, adaptability, and efficiency. The proposed methodology involves generating synthetic datasets, encoding constraints from Fluent Control, and developing AI-based solutions using 1-dimensional Convolutional Neural Networks (CNNs) and Genetic Algorithms. The AI scheduler will be validated against heuristic methods using metrics such as task completion times, resource utilization, and optimization quality. A prototype application will also be implemented, allowing for real-time visualization and validation of generated schedules. Expected outcomes include improved scheduling performance, dynamic adaptability to real-time changes, and scalability to handle increasingly complex workflows. The findings of this research will contribute to advancing laboratory automation by demonstrating the potential of AI to outperform existing heuristic schedulers in terms of flexibility, reliability, and efficiency.

Date of Award	2025
Original language	English
Supervisor	Stephan Winkler (Supervisor)