ARTIFICIAL INTELLIGENCE BASED ACTIVE MASS DAMPER

• Main Author: Felix Sinjaya, Michael
• Format: Theses
• Language: Indonesia
• Subjects: Teknik sipil
• Online Access: https://digilib.itb.ac.id/gdl/view/79630
• Institution: Institut Teknologi Bandung

In recent years, the development and application of artificial intelligence (AI) has attracted enormous attentions. Active control systems for structures to reduce dynamic loads, such as earthquakes, can be improved using AI technology as the controller. Non-AI controllers such as linear quadratic regulator (LQR) control require full-state variable measurements of structures that are rarely possible or use state observers which increase computing time and delays. To overcome this problem, two AI models, namely artificial neural networks (ANN) and fuzzy logic (FL), have been tried as AI-based controllers in various studies. However, existing research tends to focus on one model only. In this research, both AI models were investigated to see their practicality and effectiveness so that they can support the development of AI applications in control structures. The AI model was implemented to control the active mass damper (AMD) on top of a 3-story prototype scale building. The structure was modelled using the finite element method in MATLAB. System identification and model updating were carried out to obtain a finite element model that can represent the original structure. Identification was carried out by converting the original structural response data into the frequency domain with fast Fourier transform (FFT). The element model is updated with a direct updating method based on the dynamic properties of the identified structure. The results of structural simulation with LQR were used as a benchmark and used as training data for the AI model. The ANN model used was an autoregressive network with exogenous (ARX) with an architecture consisting of 7 inputs, 3 hidden layers and each consisting of 20 neurons, and 1 output (7×20×20×20×1). The input to the ANN model was acceleration at time t and t-1 for the entire floor and control force at time t-1. The fuzzy logic controller model used was a Sugeno fuzzy inference system whose parameters were optimized using the symbiotic organism search (SOS) algorithm. The inputs to the fuzzy model were the acceleration on the third floor for time t and t-1 and the control force at time t-1.. Comparisons were made between a structure controlled by ANN and structure controlled with fuzzy logic controller control against 8 earthquakes. The variables compared included performance values, changes in dynamic properties of the structure, and controller stability. The performance measured included maximum deflection, drift-ratio, maximum absolute acceleration, average absolute acceleration, and maximum control force. Changes in the dynamic properties of the structure were obtained through analysis in the frequency domain and system identification of the structure with AI-based control. Controller stability was seen from the ability of AI-based control to return the structure to a rest state when given an initial condition. Then a study of variations in structural stiffness and mass was also carried out to see the ability of AI-based control of structures with different dynamic properties. Stiffness and mass variations were carried out for each floor while maintaining the stiffness and mass ratio between floors. The research results show that both AI models could reduce structural responses and were stable. Both AI models could also reduce the response for structures with differences in stiffness and mass of up to 50%. Nevertheless, ANN is a more practical and effective AI model than FL as an AI-based controller for the structure in this study. The accuracy of control based on fuzzy logic in this study could be increased, but it would take much longer than using ANN. Then, the effectiveness of ANN-based AMD was also successfully demonstrated by experimental results