INTELLIGENT SYSTEM BASED ON MULTIAGENT SYSTEM FOR AUTONOMOUS BRIDGE CONDITION ASSESSMENT USING WIRELESS SENSOR NETWORK

• Main Author: Adi Putra, Seno
• Format: Dissertations
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/47381
• Institution: Institut Teknologi Bandung

Wireless Sensor Network (WSN) is a small embedded device installed on a large scale network with the capability of sensing, computing, and communication. WSN combines modern sensors, microelectronics, computing, communication, and distributed processing technology. WSN has potential contribution to the bridge condition assessment system. This work proposes the development of the autonomous bridge condition assessment system based on its dynamic response with a multiagent system approach. The bridge condition assessment includes natural frequency identification, mode shape assembly, and bridge capacity determination. Due to the computational capability in wireless sensor nodes, it is very interesting to implement in-network processing in WSN which means that data processing is done in the network to prevent data flooding in WSN and servers. One of the in-network processing approaches that needs to be explored is multiagent-based processing. It is proposed by employing agents in accelerometer sensor nodes, including the mobile agent, that collaborate each other for data processing and aggregation. The proposed in-network processing consists of identifying global information of WSN condition, mobile agent migration plan, data pre-processing at each sensor node, and mobile agent dispatching hop by hop from one sensor node to other nodes. Uncontrolled in-network processing causes a problem such as waste of battery energy. It is not a good solution if all sensors are always set to wake up or asleep deterministically. What the system needs is measuring the dynamic response of the bridge when the bridge is traversed by a single heavy truck, which the arrival is random. Therefore, in this work the two-player game and reinforcement learning algorithm are designed to control in-network processing. Here, the control decision is based on the selection of probabilistic actions conducted by the agent of the accelerometer-based weight-in-motion. Simulation and experimental results show that the proposed multiagent system for in-network processing has been successfully carried out and the collaboration between weight-in-motion sensors for its controlling shows satisfactory effectiveness. The construction of a laboratory-based test-bed bridge was also carried out for system validation and accuracy test. The measurement results produced by the proposed system are compared with the results produced from finite element analysis, which show the close value between finite element analysis and proposed measurement system.test-