The field of Reliability Engineering has been adopting tools from other disciplines for a long time. Statistical Tools, Culture Management, Risk Analysis, and Operations Planning are some of the tools borrowed. It is natural then, as the field grows, to continue its evolution by finding ways to adopt state-of-the-art technology: Machine Learning.
Here are some ways Reliability Engineers, working in Asset Management or Product Design, can adopt this technology to enhance their work.
1. Detect Anomaly
Statistical Process Control (SPC) is used when we need to detect abnormalities on a single defining parameter. What if there are multiple parameters and they are interdependent? Machine Learning can model this on a multi-dimensional scale and detect breaches to set threshold values.
2. Estimate Remaining Useful Life (RUL)
System’s current performance can be used to predict the remaining time (or cycles) it has till the performance degrades to an unacceptable level, giving an advantageous insight for maintenance planning. This can be applied at System, Sub-System, and Component levels.
3. Fault Diagnosis
Failure modes manifest themselves in the data collected from the System. Identifying the source of a fault in a complex system with millions of data points is difficult. Algorithms can be trained to classify the failure mode that occurred by comparing it to the previously trained dataset, resulting in a quick diagnosis of the failure and a way to plan improvement steps.
4. Fleet Analysis
Data from a fleet of fielded systems can be analyzed to detect outlier units, estimate the probability of a unit failing at a similar condition, and optimize the maintenance overhaul point. The results can be probed for individual units or for units in a defined geographical region.
5. Inspection Algorithms
Thermography, Ultrasound, and Vibration Analysis are prevalent non-invasive inspection tools used in industries. Large datasets from these can be ingested and analyzed by an algorithm to quickly identify faults from normal conditions. These models can be updated in real-time for continuous monitoring of safety-critical systems.
6. Text extraction from Maintenance Work Orders
A fringe application: extracting the failure mode from texts entered in thousands of filled maintenance work orders is a task machine learning can do. This data collection allows the user to structure the data needed for building algorithms mentioned in the above methods.
As the field progresses, there will be more avenues for machine learning methodologies to augment the work of the Reliability Engineer. What other unique ways have you found to apply ML to your work?
Related Posts
PBJ decision-making
Is your data good for Machine Learning-based-Predictive Maintenance (PdM)
