Analyzing the data points from human inspections, SCADA or sensor collections we could look for patterns in temperature,time and date of routine cleaning or maintenance such as lubrication or calibration and by whom, weather, ego-location,pressure, RPM, age of belts, age of hoses, fluid levels, age of filters, age and type of senors,diagnostic codes, voltage,time running continuously with out stopping per day, time idle per day,incidents such as exposure to unusual temperatures or other environmental factors such as exposure to wind, hurricane, sun,rain,snow,extremely dry air, salt water, corrosive chemicals, sand,floods, presence of corrosion is visible, presence of damage to the asset, cracks or weak points in installation hardware or welds, vibration, unusual smells, type of damage, date damage was discovered,unusual noises, introduction of a foreign object or animal into the assets, and correlate these over time, against, weather, service history,Geo-location, age of the asset, total hours in service, expected service work, expected life and then compare the same to similar assets and rank these assets in groups or clusters.
This way we could see what these values are normally over time and see if anomalies occur before a break-down occurs, therefore alerting the company to take action to avoid dangerous or costly asset failures. In order to properly develop this I think we need sample data and we could use an algorithm of multiple variants to pin the anomaly which occurs with some lead time prior to the break down. Then through machine learning the system could learn to predict when to issue a particular type of work order to intervene with procedures to mitigate the predicted failure.
The exact algorithms for correlation the variables and the algorithm for machine learning I havent decided on yet but will update this blog with these soon.