Identification of the tension force in cables with insulators

• Published in: Meccanica (Milan) 2019-01, Vol.54 (1-2), p.33-46
• Main Authors: Rango, Bruno J., Serralunga, Fernando J., Piovan, Marcelo T., Ballaben, Jorge S., Rosales, Marta B.
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Automotive Engineering; Bayesian analysis; Cables; Civil Engineering; Classical Mechanics; Computer simulation; Configurations; Finite element method; Frequency response; Heuristic methods; Insulators; Laboratories; Markov chains; Mechanical Engineering; Mechanical systems; Physics; Physics and Astronomy; Random variables; Resonant frequencies; Stress concentration
• ISSN: 0025-6455; 1572-9648
• DOI: 10.1007/s11012-018-00941-w

The present paper explores two approaches which, based on the measurement of the two first natural frequencies, allow the identification of the tension force in cables with insulators. For this purpose, the nonlinear mathematical model of the mechanical system and its Finite Element discretization are firstly stated. Besides, free-vibrations experiments on both a laboratory and a real-scale simulated configuration of cables with insulators are performed in order to derive their frequency response. During the laboratory experiments, a vision-based methodology is implemented for the register of the time series displacements of the cable. On this basis, a Bayesian approach is first addressed. In this framework, the cable tension is regarded as a random variable and the Bayes rule is applied to combine the experimental natural frequencies with the prior information about the random variable to derive the posterior distribution of the tension force. The Markov Chain Monte Carlo-Metropolis Hastings algorithm is implemented for the evaluation of the posterior distribution. On the other hand, a heuristic approach is proposed through the implementation of an Artificial Neural Network (ANN) as an inverse model between the parameters of the cable—including the natural frequencies—and its tension force. The training patterns are obtained from computational simulations of different cable configurations. The experimental natural frequencies are then applied to the trained ANNs to infer the tension force of the laboratory and real-scale configurations. Both approaches provide estimates of the tension force within admissible error margins.