A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Bridging the Gap Between Data and Outliers: Using Machine Learning to Automate Outlier Sensor Behaviour for Bridge Structural Health Monitoring
https://orcid.org/http://orcid.org/0000-0002-4693-4357
https://orcid.org/http://orcid.org/0009-0001-2508-1842
Structural health monitoring (SHM) is a process where a system is monitored over time through periodically recorded system response measurements related to changes in material and geometric properties. SHM aims to provide accurate and just-in-time information concerning structural conditions and performance through the recording of representative parameters over the short or long term. A proper SHM system can give an owner better knowledge of the structural conditions in real time, facilitating a better-educated approach to planning maintenance activities which may prolong the structure’s service life beyond the design life, and decrease potential future major repair costs. Advances in structural monitoring enable the collection of large data sets from strain gauges for usable information. However, making sense of these data remains a challenge. In addition to complexities with working with time series data and considering seasonality effects, the sensors are exposed to harsh environmental conditions, which can impact the accuracy and validity of recorded results. In this research, data collected from a steel truss cantilever bridge in Canada, ON, were used as a case study to propose an automated workflow for sensor data outlier detection. Thirty-five (35) weldable strain gauges, having a nickel–chromium alloy grid encapsulated in fibreglass-reinforced epoxy phenolic, were placed along the bridge, where data were collected every one (1) minute for a two (2)-year period. Using Python3, input data were transformed into features for a fitted classification machine learning model to identify outlier periods. The results were cross-validated using a time series CV fold strategy to assess the model’s predictive power whilst also considering environmental effects. To perform SHM accurately, it is important to implement processes which can capture when there are anomalies in the sensor data quality. Ontario has over 15,500 bridges as of 2020, the most out of any province, with another ~ 500 bridge projects currently in construction. Statistical methods and machine learning algorithms can be applied to existing bridge sensor data to ensure that information is being delivered accurately, which can inform important decisions on cost and repair in an automated manner for engineers.
Bridging the Gap Between Data and Outliers: Using Machine Learning to Automate Outlier Sensor Behaviour for Bridge Structural Health Monitoring
https://orcid.org/http://orcid.org/0000-0002-4693-4357
https://orcid.org/http://orcid.org/0009-0001-2508-1842
Structural health monitoring (SHM) is a process where a system is monitored over time through periodically recorded system response measurements related to changes in material and geometric properties. SHM aims to provide accurate and just-in-time information concerning structural conditions and performance through the recording of representative parameters over the short or long term. A proper SHM system can give an owner better knowledge of the structural conditions in real time, facilitating a better-educated approach to planning maintenance activities which may prolong the structure’s service life beyond the design life, and decrease potential future major repair costs. Advances in structural monitoring enable the collection of large data sets from strain gauges for usable information. However, making sense of these data remains a challenge. In addition to complexities with working with time series data and considering seasonality effects, the sensors are exposed to harsh environmental conditions, which can impact the accuracy and validity of recorded results. In this research, data collected from a steel truss cantilever bridge in Canada, ON, were used as a case study to propose an automated workflow for sensor data outlier detection. Thirty-five (35) weldable strain gauges, having a nickel–chromium alloy grid encapsulated in fibreglass-reinforced epoxy phenolic, were placed along the bridge, where data were collected every one (1) minute for a two (2)-year period. Using Python3, input data were transformed into features for a fitted classification machine learning model to identify outlier periods. The results were cross-validated using a time series CV fold strategy to assess the model’s predictive power whilst also considering environmental effects. To perform SHM accurately, it is important to implement processes which can capture when there are anomalies in the sensor data quality. Ontario has over 15,500 bridges as of 2020, the most out of any province, with another ~ 500 bridge projects currently in construction. Statistical methods and machine learning algorithms can be applied to existing bridge sensor data to ensure that information is being delivered accurately, which can inform important decisions on cost and repair in an automated manner for engineers.
Bridging the Gap Between Data and Outliers: Using Machine Learning to Automate Outlier Sensor Behaviour for Bridge Structural Health Monitoring
https://orcid.org/http://orcid.org/0000-0002-4693-4357
https://orcid.org/http://orcid.org/0009-0001-2508-1842
Structural health monitoring (SHM) is a process where a system is monitored over time through periodically recorded system response measurements related to changes in material and geometric properties. SHM aims to provide accurate and just-in-time information concerning structural conditions and performance through the recording of representative parameters over the short or long term. A proper SHM system can give an owner better knowledge of the structural conditions in real time, facilitating a better-educated approach to planning maintenance activities which may prolong the structure’s service life beyond the design life, and decrease potential future major repair costs. Advances in structural monitoring enable the collection of large data sets from strain gauges for usable information. However, making sense of these data remains a challenge. In addition to complexities with working with time series data and considering seasonality effects, the sensors are exposed to harsh environmental conditions, which can impact the accuracy and validity of recorded results. In this research, data collected from a steel truss cantilever bridge in Canada, ON, were used as a case study to propose an automated workflow for sensor data outlier detection. Thirty-five (35) weldable strain gauges, having a nickel–chromium alloy grid encapsulated in fibreglass-reinforced epoxy phenolic, were placed along the bridge, where data were collected every one (1) minute for a two (2)-year period. Using Python3, input data were transformed into features for a fitted classification machine learning model to identify outlier periods. The results were cross-validated using a time series CV fold strategy to assess the model’s predictive power whilst also considering environmental effects. To perform SHM accurately, it is important to implement processes which can capture when there are anomalies in the sensor data quality. Ontario has over 15,500 bridges as of 2020, the most out of any province, with another ~ 500 bridge projects currently in construction. Statistical methods and machine learning algorithms can be applied to existing bridge sensor data to ensure that information is being delivered accurately, which can inform important decisions on cost and repair in an automated manner for engineers.
Bridging the Gap Between Data and Outliers: Using Machine Learning to Automate Outlier Sensor Behaviour for Bridge Structural Health Monitoring
Lecture Notes in Civil Engineering
Desjardins, Serge (editor) / Poitras, Gérard J. (editor) / El Damatty, Ashraf (editor) / Elshaer, Ahmed (editor) / Ossetchkina, Ekaterina (author) / Mylonas, Paraskevas (author) / Sabamehr, Ardalan (author)
Canadian Society of Civil Engineering Annual Conference ; 2023 ; Moncton, NB, Canada
Proceedings of the Canadian Society for Civil Engineering Annual Conference 2023, Volume 11 ; Chapter: 13 ; 161-175
2024-09-26
15 pages
Article/Chapter (Book)
Electronic Resource
English
A New Outlier Detection Method Considering Outliers As Model Errors
| British Library Online Contents | 2015
Proposed Data Specifications for Bridge Structural Health Monitoring Sensor Data
| British Library Conference Proceedings | 2011
Filtering Outliers in GNSS Time Series Data in Real-Time Bridge Monitoring
| Springer Verlag | 2023
WM-4-3 Impedance-based structural health monitoring using an outlier analysis
| British Library Conference Proceedings | 2007
Monitoring Structural Health of a Bridge Using Wireless Sensor Network
| British Library Conference Proceedings | 2010