Minimizing radiation exposure in construction projects

A multinational engineering firm responsible for designing, constructing and managing large-scale energy infrastructure projects. Our client’s projects include major international hydro and nuclear plant construction, public transportation systems ranging from subway to airport infrastructure.

Today’s construction and energy firms are seeking to improve efficiencies on multi-billion dollar projects, while simultaneously improving the safety and quality of work performed by tradespeople. Such projects are prone to cost overruns due to the complexity of scheduling activities dependant on thousands of people, equipment, materials and related tasks. Radioactive sites present an even bigger challenge as radiation exposure also needs to be taken into account.

Annually, millions in cost overruns are incurred as project plans fail to account for the numerous factors that need to be considered when scheduling trades on a real-time basis.

Our client was looking to drive cost-savings on behalf of their customers by more efficiently scheduling tradespeople while also ensuring their workers’ safety by accounting for radiation exposure in near-real-time.

Our analysis revealed that every week nearly 2.5% of trades scheduled to be onsite could not perform their tasks either due to materials or equipment being unavailable, or workers being at their maximum radiation exposure for a given time period, adding millions of dollars to the project’s costs.

To address the cost overruns, we developed a cutting edge irregular time series deep learning model that leverages security access logs for workers, equipment and materials to predict future radiation exposure and the probability of work completion.

This model eliminates the scheduling of workers who near their radiation exposure limit for a given time period, and flags potential work completion issues, saving millions annually.

This model eliminates the scheduling of workers that are near their radiation exposure limit for a given time period, and flags potential work completion issues, saving millions annually.

Crater Labs developed a long-short-term-memory (LSTM) neural model that leverages state-of-the-art technology called Neural Ordinary Differential Equations to account for irregular distributions in time-series data. The model predicts future radiation exposure with 99.1% accuracy and flags work completion issues with 92% accuracy.

The benefits of this project are two-fold. First, our model eliminates scheduling conflicts where workers would exceed their radiation exposure limit for a given shift period, ensuring that workers scheduled for a shift can complete the work. In addition, our model flags in near real-time possible situations where equipment or materials are unavailable or late, allowing project managers to look at rescheduling workers. These two insights have generated millions of dollars in savings to date.