ARUP saves 790 engineering hours using machine learning to detect utility clash points planning a light rail system for Auckland

"As with any urban infrastructure project, underground utility interfaces posed a huge risk to the cost, programme and on-site safety of Auckland Light Rail. As a result, it was critical to understand how the proposed route alignment impacted these utilities. Traditionally, the only acceptable process to identify utility clashes has been for teams of engineers to compare the alignment with horizontal and vertical vertices of individual utilities (clipping). Use of this method of alignment requires large amounts of re-work and in some cases complete reassessment of the utilities. This manual assessment was extremely tedious and costly."

"When Auckland Transport investigated the introduction of light rail to address congestion and growth, Arup and Jacobs Joint Venture (AJJV) were commissioned to create the reference design for the proposed 29km Auckland Light Rail (ALR) route. This included 24 stations, overhead wire (OHW) pole installations, depot and related infrastructure, and road realignment. This major construction project was located in a heavily congested corridor that contained multiple major utilities including gas, water and electricity, which provided essential services to the operation of Auckland City. After extensive research and tests it was decided to utilise a multi-classification Neural Network to determine the location of utility clashes. This algorithm provided the best accuracy of the tested algorithms available. The use of algorithms for both RAG [traffic light risk values] and HLWP [High Level Works Plan] classifications on the Auckland Light Rail enabled a further amount of reduced assets that required manual assessment. An entirely new automatic system was created to detect clashes and the existing utilities’ asset was consolidated into a federated asset information model. A Machine Learning algorithm was then applied to further reduce any manual assessments."

"For clash detection, the algorithm can be retrained to understand the risk and treatment requirements specific to individual Network Owners. Combined functional analysis allowed for further reduction of clashes required for manual project assessment."

Details undisclosed