back to list

Project: Explainable AI for Deep Learning-Based Industrial Time-Series Models

Description

Internship Title
Explainable AI for Deep Learning-Based Industrial Time-Series Models

Introduction
ASML is developing deep learning models that learn from thousands of time-series signals collected across its worldwide machine fleet. While these models can achieve strong predictive performance, understanding why they make specific predictions remains a major challenge. This project focuses on developing explainability techniques that provide insights into the behavior of sequence models used for industrial analytics.

Objective
Develop and evaluate attribution methodologies that explain predictions made by deep learning sequence models, enabling engineers and data scientists to better understand model behavior, validate learned patterns, and increase trust in AI-driven decision-making.

Background and Scope
ASML uses sequence models such as GRUs and Temporal Fusion Transformers (TFTs) to analyze high-dimensional industrial time-series data for applications such as KPI prediction, degradation monitoring, and root-cause analysis. The student will investigate state-of-the-art explainability and attribution techniques, including gradient-based, perturbation-based, and model-agnostic approaches. The project will assess how effectively these methods identify the signals and temporal patterns that drive model predictions and will develop visual analytics techniques to communicate these insights to end users.

Deliverables

·         Literature review of explainable AI methods for sequence models and time-series analysis.

·         Evaluation of multiple attribution techniques on industrial datasets and models.

·         Framework for generating and comparing model explanations across architectures.

·         Visual analytics prototype for interactive exploration of explanations and prediction drivers.

·         Assessment of explanation quality, robustness, and usefulness for domain experts.

·         Final report, thesis, and presentation summarizing methodology and results.

Details

Student

GS

Gauri Sharma

Supervisor

Stef van den Elzen

Secondary supervisor

Nicola Pezzotti

External location

ASML