Midhun Xavier

Midhun Xavier

AI Research Engineer @ Aerobase | Neural Combinatorial Optimization & Reinforcement Learning

Sitemap

A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.

Pages

Posts

publications

Process mining in industrial control systems

Published in 2022 IEEE 20th International Conference on Industrial Informatics (INDIN), 2022

Recommended citation: M. Xavier, V. Dubinin, S. Patil, V. Vyatkin. "Process mining in industrial control systems." 2022 IEEE 20th International Conference on Industrial Informatics (INDIN), 2022. /files/paper2.pdf

An interactive learning approach on digital twin for deriving the controller logic in IEC 61499 standard

Published in 2022 IEEE 27th International Conference on Emerging Technologies and Factory Automation (ETFA), 2022

Recommended citation: M. Xavier, V. Dubinin, S. Patil, V. Vyatkin. "An interactive learning approach on digital twin for deriving the controller logic in IEC 61499 standard." 2022 IEEE 27th International Conference on Emerging Technologies and Factory Automation (ETFA), 2022. /files/paper5.pdf

A Comprehensive Approach for Gearbox Fault Detection and Diagnosis Using Sequential Neural Networks

Published in 2023 IEEE International Conference on Prognostics and Health Management (ICPHM), Montreal, QC, Canada, 2023

Recommended citation: P. Sobha, M. Xavier, P. Chandran. "A Comprehensive Approach for Gearbox Fault Detection and Diagnosis Using Sequential Neural Networks." 2023 IEEE International Conference on Prognostics and Health Management (ICPHM), Montreal, QC, Canada, 2023. /files/paper8.pdf

Formal verification of observers supervising a cyber-physical system implemented using IEC 61499

Published in 2023 IEEE 32nd International Symposium on Industrial Electronics (ISIE), Helsinki, Finland, 2023

Recommended citation: P. Ovsiannikova, E. Le Priol, V. Perret, P. Jhunjhunwala, M. Xavier, V. Vyatkin. "Formal verification of observers supervising a cyber-physical system implemented using IEC 61499." 2023 IEEE 32nd International Symposium on Industrial Electronics (ISIE), Helsinki, Finland, 2023. /files/paper11.pdf

Formal modelling, analysis, and synthesis of modular industrial systems inspired by Net Condition/Event Systems

Published in 44th International Conference on Application and Theory of Petri Nets and Concurrency (PETRI NETS 2023), Lisbon, Portugal, 2023

Recommended citation: M. Xavier, S. Patil, V. Dubinin, V. Vyatkin. "Formal modelling, analysis, and synthesis of modular industrial systems inspired by Net Condition/Event Systems." 44th International Conference on Application and Theory of Petri Nets and Concurrency (PETRI NETS 2023), Lisbon, Portugal, 2023. /files/paper6.pdf

Generating Portable Test Cases for IEC 61499 FBs from Interface Behaviour Specifications

Published in 2023 IEEE 28th International Conference on Emerging Technologies and Factory Automation (ETFA), Sinaia, Romania, 2023

Recommended citation: B. Wiesmayr, M. Xavier, S. Patil, A. Zoitl, V. Vyatkin. "Generating Portable Test Cases for IEC 61499 FBs from Interface Behaviour Specifications." 2023 IEEE 28th International Conference on Emerging Technologies and Factory Automation (ETFA), Sinaia, Romania, 2023. /files/paper10.pdf

Probabilistic Model Checking for IEC 61499: A Manufacturing Application

Published in 2024 IEEE International Conference on Industrial Technology (ICIT), 2024

This paper applies probabilistic model checking to IEC 61499 control applications, using a manufacturing case study to quantitatively analyse the reliability and stochastic behaviour of the control software.

Recommended citation: I. Faqrizal, T. Liakh, M. Xavier, G. Salán, V. Vyatkin. "Probabilistic Model Checking for IEC 61499: A Manufacturing Application." 2024 IEEE International Conference on Industrial Technology (ICIT), 2024.

A framework for the generation of monitor and plant model from event logs using process mining for formal verification of event-driven systems

Published in IEEE Open Journal of the Industrial Electronics Society, 2024

Recommended citation: M. Xavier, V. Dubinin, S. Patil, V. Vyatkin. "A framework for the generation of monitor and plant model from event logs using process mining for formal verification of event-driven systems." IEEE Open Journal of the Industrial Electronics Society, 2024. /files/paper14.pdf

LLM-Powered Multi-Actor System for Intelligent Analysis and Visualization of IEC 61499 Control Systems

Published in IECON 2024 – 50th Annual Conference of the IEEE Industrial Electronics Society, Chicago, USA, 2024

Recommended citation: M. Xavier, T. Liakh, S. Patil, V. Vyatkin. "LLM-Powered Multi-Actor System for Intelligent Analysis and Visualization of IEC 61499 Control Systems." IECON 2024 – 50th Annual Conference of the IEEE Industrial Electronics Society, Chicago, USA, 2024. /files/paper17.pdf

Enabling dependable flexibility in industrial automation with formal methods integrated to development toolchains

Published in PhD Dissertation, Luleå University of Technology, 2025

This thesis develops methods and toolchains that integrate formal verification into the engineering of flexible, reconfigurable industrial automation systems, enabling dependable flexibility across the development lifecycle of IEC 61499-based control software.

Recommended citation: M. Xavier. "Enabling dependable flexibility in industrial automation with formal methods integrated to development toolchains." PhD Dissertation, Luleå University of Technology, 2025.

ReACT-Gen AI Agents for Reasoning, Planning, and Testing in IEC 61499-Based Control Systems

Published in IECON 2025 – 51st Annual Conference of the IEEE Industrial Electronics Society, 2025

This paper presents ReACT-Gen AI agents that combine reasoning, planning, and automated testing to support the design and validation of IEC 61499-based control systems.

Recommended citation: M. Xavier, S. Patil, C.-W. Yang, V. Vyatkin. "ReACT-Gen AI Agents for Reasoning, Planning, and Testing in IEC 61499-Based Control Systems." IECON 2025 – 51st Annual Conference of the IEEE Industrial Electronics Society, 2025.

Agentproof: Static Verification of Agent Workflow Graphs

Published in arXiv preprint, 2026

This work presents Agentproof, an approach for the static verification of agent workflow graphs, enabling correctness properties of LLM-based agent workflows to be checked before execution.

Recommended citation: Melwin Xavier, Vaisakh M A, Melveena Jolly, Midhun Xavier. "Agentproof: Static Verification of Agent Workflow Graphs." arXiv preprint, 2026.

IndustriConnect: MCP Adapters and Mock-First Evaluation for AI-Assisted Industrial Operations

Published in arXiv preprint, 2026

This work introduces IndustriConnect, a set of Model Context Protocol (MCP) adapters together with a mock-first evaluation methodology for developing and assessing AI-assisted industrial operations.

Recommended citation: Melwin Xavier, Melveena Jolly, Vaisakh M A, Midhun Xavier. "IndustriConnect: MCP Adapters and Mock-First Evaluation for AI-Assisted Industrial Operations." arXiv preprint, 2026.

teaching

Generalization and Scaling of Neural Solvers

Generalization & Scaling, LTU, 2025

A key challenge in Neural Combinatorial Optimization is generalizing to instances that are larger or drawn from different distributions than those seen during training. I investigate curriculum learning, data augmentation, and inference-time search strategies to improve out-of-distribution robustness, and I benchmark learned solvers against exact and metaheuristic baselines (OR-Tools, LKH, Gurobi) on solution quality, optimality gap, and runtime.

Learning Improvement Heuristics and Neural Local Search

Learned Heuristics, LTU, 2025

Beyond constructing solutions in a single pass, I research learned improvement operators that iteratively refine an existing solution. This includes neural local search, guided and large-neighborhood search, and hybrid schemes that combine learned policies with classical operations-research techniques to reach lower optimality gaps within a fixed compute budget.

Best-of-K Objectives: Optimizing for Pass@K and Max@K

Objectives & Evaluation, LTU, 2025

At inference time, neural solvers typically sample multiple candidate solutions and keep the best one. I am interested in training directly for these best-of-K settings by optimizing Pass@K and Max@K objectives, so that the training signal matches how models are actually deployed. This work explores objective design, variance reduction, and the trade-off between solution diversity and per-sample quality.

Reinforcement Learning and Policy Optimization for Combinatorial Problems

Policy Optimization, LTU, 2025

I study how modern policy-optimization algorithms can train neural solvers end-to-end without labelled optimal solutions. This includes policy-gradient and actor–critic methods, REINFORCE with greedy and rollout baselines, and recent algorithms such as GRPO, PKPO, RSPO, and PPO. My goal is to make training more stable, sample-efficient, and better aligned with the true optimization objective of the underlying combinatorial problem.

Neural Combinatorial Optimization with Attention Models

Neural Combinatorial Optimization, LTU, 2025

My core research interest is Neural Combinatorial Optimization (NCO): training deep neural networks to solve classic combinatorial optimization problems such as the Travelling Salesman Problem (TSP), Capacitated Vehicle Routing (CVRP), scheduling, and assignment. I focus on attention-based encoder–decoder architectures, Pointer Networks, and graph neural networks that learn to construct high-quality solutions directly from data, complementing or replacing hand-crafted heuristics.

Physics-Informed Reinforcement Learning for Defect Minimization in Additive Manufacturing

Applied RL (Aerobase), Aerobase, 2025

I develop reinforcement-learning methods that minimize defects — such as porosity, cracking, residual stress, and distortion — in parts produced by Directed Energy Deposition (DED), Laser Powder Bed Fusion (LPBF), and Wire Arc Additive Manufacturing (WAAM). The RL agents learn from multiphysics simulation feedback, including thermal and mechanical finite element (FEM) analysis, microstructure analysis, and melt-pool computational fluid dynamics (CFD) simulations, closing the loop between process parameters, deposition-path planning, and predicted part quality.

Combinatorial Optimization for Additive Manufacturing Toolpath and Robot Path Generation

Applied NCO (Aerobase), Aerobase, 2025

At Aerobase, I apply neural combinatorial optimization to robot path generation and toolpath planning across metal additive-manufacturing processes — Directed Energy Deposition (DED), Laser Powder Bed Fusion (LPBF), and Wire Arc Additive Manufacturing (WAAM). I formulate deposition- and scan-path sequencing as a combinatorial optimization problem and train attention-based neural solvers to generate paths that reduce travel, balance thermal load, and improve build quality, throughput, and material efficiency.