Physics-Informed Machine Learning for Transformer Condition Monitoring -- Part I: Basic Concepts, Neural Networks, and Variants

TL;DR

The paper addresses transformer health monitoring under uncertainty and data scarcity, proposing neural-network–based solutions for diagnostics, prognostics, and adaptive control. It provides a structured overview of neural networks, covering fully connected MLPs, CNNs for time–frequency data, and reinforcement learning for control, with concrete case studies on OLTC acoustic monitoring and inrush current minimization. Loss functions such as $L_{MSE}$ and $L_{CE}$ guide supervised learning, while discussions point to data-efficient and physics-consistent modeling strategies. Part II will introduce physics-informed neural networks (PINNs) and Bayesian uncertainty quantification to yield robust, uncertainty-aware PHM for transformers and grid operations.

Abstract

Power transformers are critical assets in power networks, whose reliability directly impacts grid resilience and stability. Traditional condition monitoring approaches, often rule-based or purely physics-based, struggle with uncertainty, limited data availability, and the complexity of modern operating conditions. Recent advances in machine learning (ML) provide powerful tools to complement and extend these methods, enabling more accurate diagnostics, prognostics, and control. In this two-part series, we examine the role of Neural Networks (NNs) and their extensions in transformer condition monitoring and health management tasks. This first paper introduces the basic concepts of NNs, explores Convolutional Neural Networks (CNNs) for condition monitoring using diverse data modalities, and discusses the integration of NN concepts within the Reinforcement Learning (RL) paradigm for decision-making and control. Finally, perspectives on emerging research directions are also provided.

Figures (7)

• Figure 1: Multilayer Perceptron (MLP) with activation functions in the hidden layer and a loss function at the output.
• Figure 2: Schematics of the CNN architecture. The numbers $a \times b \times c$ inside the Conv2D boxes indicate the kernel size and the number of filters Secic25.
• Figure 3: (a) ELIN test environment (b) control scheme with acoustic sources; (c) segmented audio signal Secic25.
• Figure 4: Agent–environment interaction in reinforcement learning. The agent learns an optimal policy through repeated interaction with the environment.
• Figure 5: Transformer simulation environment. Inputs: remanent fluxes and closing angle. Output: peak inrush current RL_Jone.