Joint Master Thesis - Analyzing the Supply Network Process via Multi-Object Causal Models @ Celonis

 

Description

Functional Causal Models are a great tool to perform counterfactual reasoning on a system with features that are observed for a single isolated object. Based on causal relationships a graph is constructed and each feature is defined by a functional (machine learning) model dependent on its causal parents. For supply-chains, the object might be a material with causal relationships between its inventory characteristics, order lead-times and faultiness. In this example, a causal model could help identify the effect a new supplier or inventory plan has on the materials availability.

However, in the real world, we are often dealing with multiple objects interacting with each other and introducing direct or indirect causal relationships with different cardinality. In supply-chain networks, we are not dealing with a single material but many materials that depend on each other in their manufacturing pipeline.

In this thesis, the student performs a case-study in the supply chain process where we design a Functional Causal Model that leverages causal relationships between multiple objects, i.e. materials, and their behavior with regards to production and consumption. The goal is to analyze lead times and on-time delivery based on interventions on stock and supplier behavior.
As the main contribution, the student will design and evaluate graph structures and machine learning mechanisms that support the challenges in this multi-object scenario. As a secondary deliverable, they will transfer their approach to a generic multi-object process model.

The approach will be implemented as part of Celonis’ in-house causal reasoning framework, developed in Python. The student will be employed as a working student and supervised in collaboration with the CeloAI and Supply-Chain department.
The student presents the achievements in the context of their thesis paper as well as an intermediate and final presentation.

Prerequisites

• Basic understanding of causal concepts, including confounding and the Simpson Paradox.
• Good knowledge of process mining and machine learning techniques.
  • Expertise in Multi-Object Process Representation.
• Proficiency in Python programming.

Pointers

Causal Reasoning

• Causality - Blog Post Series
• Causal Inference in Statistics: A Primer
• Lecture Notes, Jonas Peters (ETH Zurich)

Graphical Causal Models

• DoWhy-GCM: An extension of DoWhy for causal inference in graphical causal models
• Probabilistic Graphical Models - Principles and Techniques
• Causality

Multi-Object Process Mining

• Defining Cases and Variants for Object-Centric Event Data
• A Framework for Extracting and Encoding Features from Object-Centric Event Data

Supervisor

Prof.dr.ir. Wil van der Aalst

Advisor

Mahsa Pourbafrani

For more information

Send an e-mail to mahsa.bafrani@pads.rwth-aachen.de. Make sure to include detailed information about your background and scores for completed courses.