Part 2 of 3 Parts (Please read Part 1 first)
     The Outage AI currently employs eight years of past outage data in order to make these predictions. It also brings the schedules of upcoming outages into the preparation for future preventative maintenance outages. This is a powerful solution that actually learns over time as more data is delivered and processed by its algorithms.
     The Outage AI solution contains three important components that are applied to a variety of specific functional and technical requirements which are provided by the outage team during the solution design. First, dummy tasks are created that serve to identify tasks that were in previous historical schedules but are missing from the current schedule. Second, as-yet-unscheduled tasks are automatically scheduled. Third, logic ties that would cause loops in the schedule are eliminated.
    The first two of these components focus on a NLP based text-match algorithm which analyzes the text of work orders and tasks scheduled for the current outage and compares it to analyses of the text of previous outage work orders. The third component employs an algorithm which detects loops. It identifies logic ties that could generate loops and removes them.
      In order to create dummy tasks, the text-matching algorithm compiles a list of all historical tasks that are related to a specific work order. Then it performs the same compilation of tasks for the current work order. Finally, it compares the two lists. When the algorithm finds a historical task that is not present in the current work order, it brings the historical task into the current work order as a dummy task and place holder for a task that needs to be incorporated into the current schedule.
      These dummy tasks provide a quality assurance check for situations where the current list of tasks may be incomplete. This tells the work schedulers that they should double check the integrity of the current schedule. This dummy task insertion process is carried out for all the work order descriptions in the schedule for the current nuclear unit outage. All historical outage schedules and task lists are checked during this stage of the process.
      Following the completion of the dummy task stage, the text-matching algorithm is employed to assist the creation of logic ties between the tasks in the current schedule. This process proceeds by finding the best matching historical task for a given task in the current schedule. Then, the algorithm infers the proper predecessor and successor logic ties for the current task based on the way in which the historical task was scheduled. The creation of the logic ties is repeated for all unscheduled tasks in the current schedule that have been tagged to be scheduled by the AI.
    The final stage of the AI operation carries out a quality control check in order to be certain that any logic ties created in the second stage of the process do not lead to loops in the schedule. This stage of the process is straightforward. The loop detection algorithm checks all predecessors of the current task to see whether or not it is a predecessor of itself. If it turns out that it is a predecessor of itself, then the upstream logic tie is eliminated and the loop is prevented.
Please read Part 3