Neural Parameter Calibration of Digital Shadow Models for Hysteresis in Bolted Joint Assemblies

• Grant number: 25/24724-5
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: February 15, 2026
• End date: February 14, 2027
• Field of knowledge: Engineering - Mechanical Engineering - Mechanics of Solids
• Principal Investigator: Samuel da Silva
• Grantee: Estevão Fuzaro de Almeida
• Supervisor: Chevallier
• Host Institution: Faculdade de Engenharia (FEIS). Universidade Estadual Paulista (UNESP). Campus de Ilha Solteira. Ilha Solteira , SP, Brazil
• Institution abroad: Centre De Recherche De L'École De L'Air, France
• Associated to the scholarship: 22/16156-9 - Digital Twin for Structural Health Monitoring in Bolted Joints using Physics-Informed Machine Learning, BP.DR

Abstract

Calibrating parameters and selecting suitable models to describe hysteresis effects in bolted joints is a highly challenging task. First, it requires defining parameters that govern hysteretic and memory behavior in dynamic systems based on controlled-vibration data, while acknowledging that several existing models can describe these effects. Moreover, performing this identification within a Bayesian framework incurs significant computational cost for classical Bayesian calibration algorithms, making real-time implementation unfeasible, despite this being one of the main advantages of using such models for assessing the state and tightening condition of bolted joints. In this context, the present project proposes an original approach that implements a physics-informed neural network (PINN) with neural differential equations (NDEs), in which the loss functions incorporate a joint model selection process that can describe hysteresis in bolted joints under different excitation amplitudes and tightening torques. The proposed framework also enables simultaneous model selection and hyperparameter tuning. An additional key requirement is the quantification of uncertainty in the estimated parameters and the assessment of model reliability, while ensuring that the procedure remains computationally efficient enough for real-time implementation. This allows its integration into a digital shadow model for use in conjunction with a structural health monitoring (SHM) system for bolted joints. This plan presents the motivation for this study, the Ph.D. candidate's ongoing research activities, including preliminary results, and the next steps for the BEPE internship in France. (AU)