A nuclear utility was experiencing difficulty with analytics models that were placed in service five years prior. The software provider created the models using suboptimal practices. Due to inaccuracies in model predictions, analysts had to manage over 100 nuisance alerts per week.
Processing these alerts was non-value added work, forcing analysts to spend valuable time on model maintenance instead of focusing on diagnostics associated with the small amount of actual alerts that were actionable.
Integral Analytics LLC assessed the utility’s equipment models and developed a customized remediation plan. This evaluation included checking each model for configuration, accuracy, and its ability to detect failure modes and associated degradation mechanisms.
Following a detailed assessment, Integral Analytics, LLC identified and implemented improvements, including:
Restructured large models into smaller, more manageable models.
Added operational filters, refocusing models on expected patterns.
Redefined the process for training data selection that is more robust than standard practices.
Customized alert settings based on each model rather than rule-of-thumb based guidance.
Developed an issue management flowchart that results in more consistency in actions taken between analysts.
The client realized immediate results. Nuisance alerts were reduced by over 95%, recovering 500-plus person-hours on average, annually.