Manufacturing and Factories – Predictive Maintenance

Within manufacturing an ‘unplanned shutdown’, due to asset failure, is unacceptable. And two potential approaches to deal with this situation are: -

Reactive Maintenance: Not Scheduled – targeted replacement of parts in response only to their failure. Occurs during the subsequent unplanned shutdown.

Inefficient (need stock of replacement parts, costly line stoppages).

Preventative Maintenance: Scheduled – bulk replacement of parts, which are within a margin of their end of useful life. Performed at set intervals during planned shutdown.

Inefficient (all service parts replaced, some still useable).

 

“Users require timely notification of the potential failure of an asset. Allowing maintenance to be performed, at a time convenient to operations, before the failure occurs.”

Avoiding unplanned shutdowns in Manufacturing using Machine Learning to predict failure states in equipment.

Predictive Maintenance: sensors, applicable to condition monitoring (e.g., temp, pressure & vibration) are mounted on the asset. An ‘Internet of Things’ type set up provides for highest levels of functionality. The DataStream from said sensors allows capture & storage of an asset’s state, in real time.

Predictive Analytics, i.e., Machine/Deep Learning Algorithms applied to the data detects the minutiae present during the transition from the Normal State, into a Failure State and allows predictions identifying the failing subsystem and ‘Time till Failure’. Before o deployment, these algorithms require training with a mixture of data, between the Normal & Failure states. A technique known as ‘anomaly detection’; in which novel data, not being outliers, allows classification between Good/No-Good.

By detecting the assets degradation into a failure condition, before it occurs, users can plan for targeted part replacement at an appropriately scheduled downtime.

 

Common challenges: Issues associated with appropriate sensor availability, employee knowledge/training and data quality.