Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using artificial neural network
Thesis (PhD. (Civil Engineering))
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Universiti Teknologi Malaysia
2023
|
| Subjects: | |
| Online Access: | http://openscience.utm.my/handle/123456789/653 |
| _version_ | 1825623453953163264 |
|---|---|
| author | Padil, Khairul Hazman |
| author_facet | Padil, Khairul Hazman |
| author_sort | Padil, Khairul Hazman |
| collection | OpenScience |
| description | Thesis (PhD. (Civil Engineering)) |
| first_indexed | 2024-03-05T17:32:55Z |
| format | Thesis |
| id | oai:openscience.utm.my:123456789/653 |
| institution | Universiti Teknologi Malaysia - OpenScience |
| language | English |
| last_indexed | 2024-03-05T17:32:55Z |
| publishDate | 2023 |
| publisher | Universiti Teknologi Malaysia |
| record_format | dspace |
| spelling | oai:openscience.utm.my:123456789/6532023-08-28T13:00:16Z Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using artificial neural network Padil, Khairul Hazman Neural networks (Computer science) Frequency response (Dynamics) Feedback control systems Structural analysis (Engineering)—Data processing Thesis (PhD. (Civil Engineering)) The artificial neural network (ANN) has become a popular computational approach in the field of vibration-based damage detection, owing to its ability to relate the nonlinear relationship between structural vibration characteristics and damage information. Hence, many researchers have employed modal domain data as ANN input variables. However, modal data suffer the drawbacks of providing insufficient information needed for damage detection and being highly dependent on the quality of measured time history data obtained from sensors. In this regard, the frequency response function (FRF) is often chosen as the input variable for ANN to detect structural damage due to its ability to prevent information leakage. The main concern surrounding damage detection using FRF with an ANN is the amount of FRF data. The use of a full FRF data set will result in a wide composition range for the ANN input layer, thus affecting the iteration divergence in the network training process and, in turn, resulting in computational inefficiency. In most applications, principal component analysis (PCA) has been used to reduce the amount of FRF data before it is fed to an ANN model. However, as the structures become more complex, the amount of FRF data increases. Large amounts of FRF data may diminish the effectiveness of PCA to select important information from the FRF data, thus leading to false damage detection. Moreover, the existence of uncertainties from modelling error and measurement error may amplify errors in damage detection. Hence, the present study proposes the use of a combination of a non-probabilistic method with PCA to investigate the problems of existing uncertainties and the inefficiency of using FRF data for ANN-based damage detection processes. In this study, an ANN is used to relate FRF data to a damage feature. The input data for the network are compressed real FRFs, and the outputs are the elemental stiffness parameters (ESPs). The compressed FRF data obtained from the PCA provide a new damage index (DI) that is used as the input layer of the ANN. Based on the interval analysis method, the uncertainties in the new DI are considered to bind together to obtain the interval bounds (lower and upper bounds) of the DI changes. The possibility of damage existence (PoDE) is designed to ascertain the relationships between the input and output parameters in the form of undamaged and damaged conditions. Numerical models of a two-tier steel bookshelf and a steel truss bridge with a laboratory tested steel truss bridge model are tested to demonstrate the proposed method. Based on the results obtained, the proposed method can detect structural damage with a relatively low computational time, even when uncertainties are present. Moreover, when the damage severity is low at 5%, the proposed non-probabilistic ANN predictions become less accurate. The highest PoDE and the most reliable damage measure index (DMI) values are obtained when the uncertainty levels in the training data and testing data are the same, which either 0%, 2%, 5%, 7% or 10% uncertainties level. In overall, the proposed method is efficient in dealing with FRF data and uncertainty while consuming less computational effort when using FRF data as the ANN input for damage detection. Faculty of Engineering - School of Civil Engineering 2023-08-28T04:27:44Z 2023-08-28T04:27:44Z 2020 Thesis Dataset NA NA http://openscience.utm.my/handle/123456789/653 en NA; NA application/pdf Universiti Teknologi Malaysia |
| spellingShingle | Neural networks (Computer science) Frequency response (Dynamics) Feedback control systems Structural analysis (Engineering)—Data processing Padil, Khairul Hazman Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using artificial neural network |
| title | Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using artificial neural network |
| title_full | Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using artificial neural network |
| title_fullStr | Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using artificial neural network |
| title_full_unstemmed | Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using artificial neural network |
| title_short | Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using artificial neural network |
| title_sort | non probabilistic method to consider uncertainties in frequency response function for vibration based damage detection using artificial neural network |
| topic | Neural networks (Computer science) Frequency response (Dynamics) Feedback control systems Structural analysis (Engineering)—Data processing |
| url | http://openscience.utm.my/handle/123456789/653 |
| work_keys_str_mv | AT padilkhairulhazman nonprobabilisticmethodtoconsideruncertaintiesinfrequencyresponsefunctionforvibrationbaseddamagedetectionusingartificialneuralnetwork |