Development of a semi-automated impact device based on human behaviour recognition for in service modal analysis of structures / Fahad Zahid

Fahad, Zahid (2023) Development of a semi-automated impact device based on human behaviour recognition for in service modal analysis of structures / Fahad Zahid. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (198Kb)
	PDF (Thesis PhD) Restricted to Repository staff only until 31 December 2024. Download (3096Kb)

Abstract

Current Impact Synchronous Modal Analysis (ISMA), an operational modal analysis technique incorporated with either manual impact hammer or Automated Phase Controlled Impact Device (APCID) faces challenges on its effectiveness and practicality in real industrial applications. Studies show that the manual operation of this technique is laborious and time intensive due to the lack of control and knowledge of the impact with respect to phase angle of disturbance. APCID solves this problem as it provides knowledge and control of impact with respect to phase angle of disturbance. However, its large size and heavy weight makes it unsuitable for real world applications. As automated impact application makes APCID bulky, automated impacts can be replaced with manual impacts while still using APCID control. However, the randomness in human behaviour significantly reduces the effectiveness of APCID control due to the lack of control of impact with respect to phase angle of disturbance. Studies show that human behaviours can be recognized using devices like Inertial Measurement Unit (IMU) and Electroencephalogram (EEG) in conjunction with machine learning. In this study, machine learning models are developed using IMU and EEG to recognize physical and cognitive human behaviours respectively, to replace APCID with semi-automated impact device (IMU-ISMA/EEG-ISMA) while still using APCID control. For physical human behaviour recognition, impact classification model was developed to classify 13 different impact types using orientation data from IMU and then used with reaction time and impact speed (measured through IMU), to predict impact time and adjust APCID control accordingly. Impact time was predicted with 5.2% mean prediction error. For improved practicality, EEG was used for cognitive human behaviour recognition. Machine learning model was developed for predicting impact time before the impact and make adjustment in APCID control accordingly. Impact time was predicted with 8.3% mean prediction error. The IMU and EEG based time prediction models were integrated with APCID control to perform ISMA. Using both IMU-ISMA and EEG-ISMA, the cyclic load components at 20 Hz and 30 Hz running frequencies were reduced by over 80%. For both the devices, the extracted modal parameters were in very good correlation with the benchmark, Experimental Modal Analysis (EMA) data and all the modes were identified with less than 3% and 6% difference in natural frequencies and damping respectively, and Modal Assurance Criterion (MAC) values greater than 0.9 for all modes. Using IMU-ISMA and EEG-ISMA, on average, it took 24-26 impacts and 18-19 impacts, respectively, to complete one test. Manual ISMA, (i.e., Random ISMA) was also performed at 20 Hz and 30 Hz and results show that IMU-ISMA and EEG-ISMA were able to extract more modes with better accuracy compared to Random ISMA. Additionally, IMU-ISMA took 12-20% less number of impacts and EEG-ISMA took 35-40% less number of impacts to get better results. Both variations of semi-automated impact device present a portable and practical solution for in-service modal analysis with comparable accuracy however, EEG-ISMA is a less labour intensive and user-friendly solution as it takes around 24-27% smaller number of impacts to complete a test.

Actions (For repository staff only : Login required)