Dissipative Latent Residual Physics-Informed Neural Networks for Modeling and Identification of Electromechanical Systems

2026-04-20

AI Summary

This work addresses the challenge that traditional mechanistic models of electromechanical systems often fail to accurately capture dissipative effects such as joint friction and stray losses, while existing residual learning approaches may violate physical energy conservation principles. To overcome these limitations, the authors propose DiLaR-PINN, which uniquely embeds a strict skew-dissipativity constraint into a latent-variable residual network, guaranteeing non-increasing system energy under arbitrary parameters. Furthermore, they introduce a curriculum-based training strategy featuring incrementally increasing sequence lengths via recurrent unrolling, tailored for scenarios with partially observable states. Experimental validation on a real helicopter platform demonstrates that DiLaR-PINN significantly outperforms purely mechanistic models, unconstrained MLP residuals, soft-constrained variants, and black-box LSTMs in both dissipative modeling accuracy and long-term extrapolation capability.

Technology Category

• Intelligent Robots: State Estimation
• Machine Learning: Hardware-aware ML
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Accurate dynamical modeling is essential for simulation and control of embodied systems, yet first-principles models of electromechanical systems often fail to capture complex dissipative effects such as joint friction, stray losses, and structural damping. While residual-learning physics-informed neural networks (PINNs) can effectively augment imperfect first-principles models with data-driven components, the residual terms are typically implemented as unconstrained multilayer perceptrons (MLPs), which may inadvertently inject artificial energy into the system. To more faithfully model the dissipative dynamics, we propose DiLaR-PINN, a dissipative latent residual PINN designed to learn unmodeled dissipative effects in a physically consistent manner. Structurally, the residual network operates only on unmeasurable (latent) state components and is parameterized in a skew-dissipative form that guarantees non-increasing energy for any choice of network parameters. To enable stable and data-efficient training under partial measurability of the state, we further develop a recurrent rollout scheme with a curriculum-based sequence length extension strategy. We validate DiLaR-PINN on a real-world helicopter system and compare it against four baselines: a pure physical model (without a residual network), an unstructured residual MLP, a DiLaR variant with a soft dissipativity constraint, and a black-box LSTM. The results demonstrate that DiLaR-PINN more accurately captures dissipative effects and achieves superior long-horizon extrapolation performance.

Problem

• dissipative dynamics
• electromechanical systems
• physics-informed neural networks
• residual learning
• energy consistency

Innovation

• dissipative modeling
• latent residual
• physics-informed neural networks
• skew-dissipative parameterization
• recurrent rollout