Energy infrastructure is essential to maximising production performance, any cracks or defects can at best reduce performance or at worst lead to significant danger.
The customer uses eddy current based non-destructive testing (NDT) methods to monitor pipeline surfaces. The analysis of measurements to classify sound and unsound pipe sections takes up a lot of expert time that could be better spent on the analysis of actual defects/cracks.
DiagFit was used to learn NDT measurements on a healthy part of the infrastructure (unsupervised learning) to build a predictive model, then was able to run the model to classify thousands of other measurements as healthy/defective
A 100% true positive rate was achieved on the historical data with only 17% false positive rate. The predictive model proved even more useful in correcting the initial labels of the experts who finally agreed with the diagnosis of the DiagFit model after re-analysing some conflicting assessments.