Predictive Maintenance

Structured frameworks such as RCM (Reliability Centered Maintenance) and FMEA (Failure Mode and Effect Analysis) are invaluable for defining the optimal maintenance strategy for each equipment piece. Yet while they can significantly enhance reliability, implementing these approaches the traditional way is very time-consuming and expensive.

Predictive Maintenance (PdM) is vital to any proactive maintenance strategy, as it focuses on identifying failures before they occur. Common PdM techniques only look at the condition of equipment. This is problematic, since 70% of all downtime is caused by process-induced failures – only 30% is caused by equipment-induced failures.

Operators, planners and maintenance technicians should be able to intervene on both asset-induced and process-induced failures, so as to reduce downtime and all its ramifications.

AI makes it possible to move beyond univariate analysis and identify more complex patterns across thousands of variables, to get a true indication of asset health by combining equipment data with process, operational and context data.

We know which data to collect and store to feed our algorithms, using best practices derived from RCM and FMEA methodologies in combination with our own knowledge of failure modes.

Unsupervised learning techniques are used to alert operators and technicians to abnormal operating behavior and provide them with actionable insights, so that they can validate and label anomaly scores.

Supervised machine learning techniques afford more accurate failure predictions, Remaining Useful Life (RUL) estimates for machines, and insights into root causes, enabling planners and operators to optimize maintenance work orders, order replacement parts and minimize downtime.

Simulations allow for assessing how downtime and running assets in a sub-optimal state impact overall productivity.