Optimized Planning for the Construction of Heavy Assets
The construction of heavy assets such as oil platforms, subsea gas compressors, and offshore wind farms requires careful planning. These assets are split into individual construction assemblies that can consist of hundreds of thousands of objects that have to be carefully constructed and assembled in the correct order, with little or no room for errors.
This is a highly complex process that is usually spread over several years at hundreds of different construction sites, often in different countries. The process is supervised by dedicated planners that schedule activities and allocate resources to minimize cost, risk, and delays. This is not an easy task since activities can have complicated dependency structures and there is often a limited amount of resources available to execute them. Currently, planners mostly rely on their domain knowledge and experience in the planning process. At Aize, we have been investigating how data-driven approaches can assist planners in their daily work.
Planning consists of setting a start and finish date for activities and allocating an appropriate amount of resources (construction workers) to execute them. A central part of this process is to estimate the duration and resource requirements for all the different activities while also taking the dependency structure between them into account. If activity B depends on activity A being finished, then the planner needs to know when activity A is finished before setting a start date for activity B. In addition, if there is another activity C that no other activity depends on, then it is better to prioritize allocating resources to activity A since any delay in this activity will lead to a delay in activity B, while a delay in activity C will lead to no further delays. In this way, the duration and resource requirements are closely related; if the planner allocates more resources to the activity, then we would expect that the activity is finished in a shorter amount of time. The key is to schedule the activities in the correct order and allocate the correct number of resources to them such that the overall duration of the entire project is minimized.
Accurately estimating the duration and resource requirements for construction activities can be a daunting task since there exists a multitude of factors that can influence how and when an activity is executed. At Aize we have explored how we can leverage the vast amounts of data collected from previously completed projects to optimize the planning process. The main idea behind our approach is to first train a supervised-learning algorithm on historical data to learn what aspects of construction activities affect the duration and resource requirements and then use the framework of mathematical optimization to create an optimal schedule for the project. The optimization algorithm is designed to minimize the total duration of the project while satisfying not only the dependencies between the activities but also the predicted duration and resource requirements from the trained supervised-learning algorithm.
However, there are some complications that are important to consider. First of all, the historical data does not always give the full picture of what has happened since many factors affect the execution of construction activities that are simply not captured in the data. Furthermore, all projects are unique and it is naive to assume that the factors that affect the duration and resource requirements are the same across many wildly different projects. Taking this into account, we opted to tweak our prediction model such that it gradually shifts focus from previously completed projects to the current project as more data is collected from the current project. Moreover, after solving the optimization problem we give the optimal schedule to the planners for inspection. The planners can then make adjustments to the plan if they believe there are other external factors affecting the construction activities that the algorithms have not taken into account.
An optimal planning tool for capital-intensive projects allows for increased efficiency and can drastically reduce costs. At Aize we are dedicated to utilizing data in the best way possible to give planners the tools they need to succeed in a challenging and increasingly competitive landscape. Stay tuned for more of our adventures!
