Hackathon 2019 solutions – Team Purple
Our proposal was divided in two parts, first part focused on analyzing the data on a large scale, second part was to construct a scenario where results from the data would create value for the construction. The approach was to let data be a central factor that drives the whole process.
The objective of our proposal was to investigate anomalies in the data and use them to compare different datasets. The approach was to attack the data holistically where comparing different types of data would give us information on the bridge dynamic behavior. The results from the data analyses would then be collected in a central module that shows connection between dataset and illustrates the dynamic behavior. By identifying anomalies in the data, and by collecting the comparison of the data, the central model would then serve as a platform for prediction of the dynamic behavior.
The scope we had set us was far to large to be able to construct all the results during the hackathon. Therefor our focus was to prove the concept by analyzing limited range of data for two, and then three datasets. Interesting objectives such as anomaly analyzing approach and machine learning would be used for the final product. For the prove of concept we had to settle on a simple analysis of the data.
We had great challenges in retrieving the data in a constructed way due to complexity and inconsistency of the datasets. In the end we did manage come up with interesting results that showed a direct connection between anomalies of the different datasets. Below is an example of the illustrated data compares looks like.
To ensure that these results would give value to the construction we investigated how a data driven approach could benefit the construction. Our vision is that the data collected would drive a prediction model constructed around anomalies. The results would then be used to calibrate the dynamics of the bridge and with that update the FE model. The data model should then extend with additional accelerometer sensors to get results suited for dynamic analyses.
Calibrating the FE model omits some of the assumptions bound to uncertainty. The expected outcome of the calibrated model is extended life time of the construction and more accurate maintenance estimate.
See the ppt presentation here.
Read more about the other solutions here: