Required data

Failure data

The most critical piece of data that is required by Prescriptive Maintenance on Cloud is the equipment failure history. Equipment failure is represented by a boolean flag that is either true or false. A true flag means the equipment failed, and a false flag means the equipment did not fail. This flag must be coded accurately. Any errors in coding the failure flag translate directly into prediction inaccuracy. For example, if your coding is only 50% accurate, the ceiling on model accuracy is 50%.

Operating data

Operating data describes anything that is known about the tasks that the equipment performs or the signals that the equipment emits, for example, temperatures, pressures, noise levels, and vibration levels. Good operating data enables the product to determine how the historical usage or loading of each piece of equipment corresponds with the signals that it produces and how load and signals are correlated with failure.

Frequency of the collection of operating data versus the analysis period

Prescriptive Maintenance on Cloud builds a history of operating data. When it analyzes this history, it aligns various pieces of operating data to a fixed interval so that it can examine the correlation between multiple variables that are collected at different times. The default time period for analysis is daily.

When performing daily analysis, Prescriptive Maintenance on Cloud resamples and aggregates all operating data to the daily level. Resampling and aggregation speeds up analysis and smooths the data to make patterns more clearly evident. Resampling to a fixed interval is particularly effective when loading on equipment is reasonably consistent throughout the day.

In some cases, loading of equipment is cyclical in nature, that is, the value of variables changes considerably depending on which stage of a manufacturing cycle the equipment is performing. In these cases, it is usually better not to resample. Instead, supply summary data that represents the values of operating variables for each cycle. When you supply data that is pre-summarized to the end of a cycle, supply all variables that describe a single cycle with a common time stamp that represents the end of the cycle.

Master data

Each asset that is monitored by IBM Prescriptive Maintenance requires an asset_id value and an asset_type value.

Next scheduled maintenance date

The next scheduled maintenance date is used to evaluate the current maintenance strategy. If a piece of equipment is predicted to fail before its next scheduled maintenance, Prescriptive Maintenance on Cloud flags it as under-maintained and recommends that the maintenance schedule be advanced.

Well-maintained days

This parameter indicates the wanted number of days between expected failure and planned maintenance. Tracking well-maintained days serves as a factor of safety for maintenance scheduling. It is not advisable to schedule maintenance on the expected failure date as allowances must be made for model inaccuracy and schedule delays.