Scope 

Use of Qualitest’s machine learning for predictive maintenance solution to predict which transformer is likely to break one week ahead of time.  

By predicting breakdowns more accurately, the company increased its ability to provide better SLA and operational efficiency.   

  • The first phase was to gather breakdown cases with time-related explanatory variables.  
  • The second phase was applying predictive algorithms to the data to identify transformers that are at risk of failure.   
  • The third phase was implementing the solution into the company’s production environment.  

In a short project time of 14 weeks, a proactive maintenance prediction solution was ready for rollout for different locations. 

Benefits 

  • Predicted device & infrastructure components’ failure proactively with high accuracy and achieved stability over time 
  • Reduced time and effort by technical teams to repair ‘likely to break’ devices.   
  • Achieved high customer satisfaction, effectiveness, and efficiency by providing higher SLA nationwide 
  • Increased commercial advantage by providing better pricing based on advanced prediction capabilities   

Qualitest Solution 

The company partnered with Qualitest to use its Machine Learning (ML) solution. The final model included three main components  

(a) Device failure prediction per infrastructure category  

(b) Root cause analytics  

(c) Optimize predictive maintenance for each territory and infrastructure specifically for each electricity distribution facility.  

Project Stages 

Qualitest used complex data streams from multiple devices, transformers, and geo-location information.  The main model of demand prediction was built using a machine learning model ensemble. The next phase included the use of a prediction modeling component for each territory. An optimized time-to-maintain for each device type was applied. The model components are now part of the company’s production environment. 

  1. Planning and data reprocessing by the company data analysts, guided by the Qualitest team. Exploration of data and creation of a unified panel of explanatory variables. 
  1. Running on-premises the Qualitest solution that provides 40 Machine Learning algorithms with automatic finetuning to find the best explanatory variables combination. 
  1. Delivering a stable and transparent prediction model, ready for deployment in production. 
  1. Qualitest solution provided a variety of implementation methods and processes for constantly changing data streams. 

Endnotes 

Qualitest’s ‘AI-powered Machine Learning Predictive Analytics Solution’ proved to be fast, easy to use, and accurate when applied to complex data.  The solution will enable the company to correctly predict transformer failure, increase SLA, and increase operational efficiency.