*Corresponding author’s email: ettefagh@tabrizu.ac.ir. Copyrights for this article are retained by the author(s) with publishing rights granted to Amirkabir University Press. The content of this article is subject to the terms and conditions of the Creative Commons Attribution 4.0 International (CC-BY-NC 4.0) License. For more information, please visit https://www.creativecommons.org/licenses/by-nc/4.0/legalcode. , 0 Amirkabir Journal of Mechanical Engineering Amirkabir J. Mech. Eng., DOI: Identification and Damage Detection of beam-like structure Using Vibration Signals Based on Simulated Model, Real Healthy State and Deep Convolutional Neural Network Zohreh Mousavi a 1 , Mir Mohammad Ettefagh 1 *, Morteza Homayoun Sadeghi 1 , Seyed Naser Razavi 2 1 Faculty of Mechanical Engineering, University of Tabriz, Tabriz, Iran. 2 Faculty of Electrical and Computer Engineering, University of Tabriz, Tabriz, Iran. ABSTRACT: Condition Monitoring of mechanical systems, such as structures and rotating machines is always a major challenge. This paper is presented a new method for damage detection of real mechanical systems in presence of the uncertainties such as modeling errors, measurement errors, varying loading conditions and environmental noises based on simulated model and real healthy state. In this method, data of real healthy system is used to updating the parameters of the simulated model. Some parts of the signals that are not related to the nature of the system are removed using the Complete Ensemble Empirical Mode Decomposition method. A deep convolutional network is designed to learn damage-sensitive features from raw frequency data of simulated model and real healthy state. Raw frequency data is extracted from vibration signals using the Power Spectral Density method. In order to train the proposed deep network, raw frequency data of the simulated model and real healthy state are used. Then, raw frequency data of the real model are used to test the proposed deep network. The proposed method is validated using an experimental beam structure. The results show that using the proposed algorithm for identification and damage detection of the beam-like structure has more accuracy with respect to the other comparative methods. ………………………………………………… Review History: Received: Revised: Accepted: Available Online: Keywords: Condition Monitoring Beam-Like Structure Vibration Signal Deep Neural Network. 1- Introduction Mechanical systems are widely used in the industrial sector and are key and important equipment. Condition Monitoring of these systems is always a major challenge and can extend their lifespan. The vibrational signals extracted from mechanical systems contain useful information, and by examining the physical characteristics of these signals, damages can be detected in different parts of them. The forces applied to mechanical systems are subjected to many changes; therefore, data acquisitioning from mechanical systems under different loads is difficult and expensive. Also, in mechanical systems, the extraction of damage data is not really cost-effective, and generally only data on a healthy state is available; so, using artificial damage data based on simulated model instead of real ones is a feasible approach to addressing the problem [1]. Feature extraction plays a crucial role in the damage detection of mechanical systems. Traditional feature extraction methods are not well capable of extracting damage-sensitive features [2]. In recent years, the use of deep neural networks to extract the effective features has attracted the attention of many researchers [3]. Deep neural networks have been widely and successfully used for image and signal processing in the time and frequency domain [2, 4-5]. In this paper, a new method for damage detection of mechanical systems is presented. The first purpose of this paper is to present a method for damage detection of mechanical systems in presence of the uncertainties such as modeling errors, measurement errors, varying loading conditions and environmental noises. The second purpose of this paper is to design a deep convolutional neural network in order to learn the damage-sensitive features from raw frequency data of the simulated model and real healthy state despite the various uncertainties. The third aim of this paper is to train the proposed deep network based on frequency data of the simulated model under simple loading condition and real healthy state, and then to evaluate the deep network with frequency data of real model under complex loading condition (for more realistic assumptions). In the proposed method, the simulated model parameters are updated based on the real model data. Some parts of the vibration signals that are not related to the nature of the system have been removed using the Complete Ensemble Empirical Mode Decomposition (CEEMD) method. Frequency data are obtained from the vibration signals using the Power Spectral Density (PSD) method. To evaluate the proposed method, a beam-like structure in a laboratory environment has been used as a case study. 2- Methodology In this section, at first the Finite Element (FE) and experimental models of the beam structure is explained. Then, the proposed algorithm for damage detection of the beam structure is expressed. 2-1- FE model Considering the small deformations and linear behavior of the system, a FE model of the simply supported Euler–Bernoulli beam structure is created. The vibration equation of the beam can be written as follows [6- 8]: () b b b MZ CZ KZ Ft    where Z , Z and Z are the displacement, velocity and acceleration vectors of the beam structure and , b M b K and b C displays the mass, stiffness and damping matrices of the whole structure, respectively. To ACCEPTED MANUSCRIPT