A major services company has higher equipment downtime due to slow turn-around in routine maintenance & servicing functions. Delays in the triaging process to identify the right level of service required the creation of a backlog, thereby increasing maintenance turn-around time. The company wanted to deploy a recommendation platform to suggest the type of service required by the operator, therefore reducing investigation effort.
Data from multiple sources, including the equipment’s physical properties, lifetime history, and sensor operating parameters, are captured and processed using data engineering tools. Text mining methods are used on operator comments and observations to enrich the system data. Once data has been processed and transformed, appropriate features that influence maintenance decisions are selected using feature engineering methods. Selected features are then used to build a supervised machine learning model (random forest classification) that suggests the right level of maintenance required for the Equipment based on current conditions.
The output is presented as a suggestive dashboard to the triage operator who makes the final decision and directs the equipment to the appropriate job center for the right level of maintenance.
With 80% accuracy in suggesting the right level of maintenance required, the predictive solution not only improves the turn-around time of triage function – It also reduces the scrap re-work cost.