Machine Learning Framework for Causal Modeling for Process Fault Diagnosis and Mechanistic Explanation Generation - Doctor of Philosophy Thesis

Aug 5, 2026

Upload by Jiven

Machine learning models often trade explainability for performance, but mechanistically grounded explanations require first-principles, at least mechanistic, rooting. This dissertation develops a causal-modeling machine learning framework spanning different levels of process knowledge for process fault diagnosis and mechanistic explanation generation. It presents data-centric causal model construction, then shows deep learning-based classification and regression limitations in aligning hidden representations with system mechanisms. It further integrates process knowledge to produce explainable models via automated variable transformations, extracting system mechanisms from provided model knowledge.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages155

Reading Time6 h 31 min

AccessFree to download

Acknowledgments

Chapter 1: Introduction and Background

Chapter 2: From data to causal models

Transfer Entropy as a Measure of Causality

Tennessee Eastman Benchmark Process

Hierarchical framework for developing causal maps

Chapter 3: Neural Networks for Classification

Peeking Under the Hood of a Deep Neural Network

Chapter 4: Neural Networks for Regression

Peeking Under the Hood of a Neural Network

Chapter 5: Mechanistic Explanation Generation (MEG)

AI for Mechanistic Explanation Generation – XAI-MEG

Results and Discussion

Machine Learning Framework for Causal Modeling for Process Fault Diagnosis and Mechanistic Explanation Generation Abhishek Sivaram Submitted in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy under the Executive Committee of the Graduate School of Arts and Sciences COLUMBIA UNIVERSITY © 2022 Abhishek Sivaram All Rights Reserved Abstract Machine Learning Framework for Causal Modeling for Process Fault Diagnosis and Mechanistic Explanation Generation Abhishek Sivaram Machine learning models, typically deep learning models, often come at the cost of explainability. To generate explanations of such systems, models need to be rooted in ﬁrst-principles, at least mechanistically. In this work we look at a gamete of machine learning models based on diﬀerent levels of process knowledge for process fault diagnosis and generating mechanistic explanations of processes. In chapter 1, we introduce the thesis using a range of problems from causality, explainability, aiming towards the goal of generating mechanistic explanations of process systems. Chapter 2 looks at an approach for generating causal models purely through data-centric approach, with minimal process knowledge with respect to equipment connectivity and identifying causality in the domains. These causal models generated can be utilized for process fault diagnosis. Chapter 3 and chapter 4 show how deep learning models can be used for both classiﬁcation for process fault diagnosis and regression. We see that depending on the hyperparameters, i.e., purely the breadth and depth of a neural network, the learned hidden representations vary from a simple set of features, to more complex sets of features. While these hidden representations may be exploited to aid in classiﬁcation and regression problems, the true explanations of these representations do not correlate with mechanisms in the system of interest. There is thus a requirement to add more mechanistic information about the features generated to aid in explainability. Chapter 5 shows how incorporating process knowledge can aid in generating such mechanistic explanations based on automated variable transformations. In this chapter we show how process knowledge can be used to generate features, or model forms to generate explainable models. These models have the ability of extracting the true models of the system from the model knowledge provided. Table of Contents Acknowledgments .................................... ix Chapter 1: Introduction and Background ........................ 1 Chapter 2: From data to causal models ......................... 5 2.1 Transfer Entropy as a Measure of Causality .................. 8 2.1.1 Generating digraph based on transfer entropy ............. 11 2.2 Tennessee Eastman Benchmark Process ..................... 14 2.3 Hierarchical framework for developing causal maps .............. 17 2.3.1 Tier 1: Plant-level DAG ......................... 19 2.3.2 Tier 2: Subsystem-level graph with possible cycles ........... 30 2.4 Major Results ................................... 33 Chapter 3: Neural Networks for Classiﬁcation ..................... 35 3.1 Mathematical Background ............................ 38 3.1.1 Problem Formulation ........................... 38 3.1.2 Classiﬁcation with Neural Networks ................... 40 3.2 Peeking Under the Hood of a Deep Neural Network .............. 42 3.2.1 Feature Extraction: Node-speciﬁc Selective Activation of the Input Space ................................... 43 3.2.2 Wider vs Deeper Networks: Complexity of Features .......... 45 3.2.3 From Features to Feature Spaces .................... 49 3.2.4 The Final Layer: Separating Hyperplanes for Classiﬁcation ...... 50 3.2.5 Degeneracy of parameters using softmax activation .......... 52 3.3 How does a Neural Network Learn the Mapping?: From Parts to Whole ... 56 3.4 Major Results ..........................

FAQs

Why does the thesis focus on mechanistic explanations for machine learning models?

Deep learning models often lack explainability, so the work emphasizes generating explanations rooted in first-principles and mechanistic understanding of process systems.

How are causal models generated with minimal process knowledge?

Chapter 2 proposes a data-centric approach to generate causal models using measurements and causality identification, including a transfer-entropy-based method to build causal structure.

What role does process knowledge play in improving explainability?

Later chapters show that hidden representations from standard deep models may not reflect real mechanisms, so process knowledge is incorporated to generate explainable mechanistic models through automated variable transformations.

Aug 5, 2026

Upload by Jiven