There is a new breed of software which is a subset of artificial intelligence (AI), called machine learning. And if you feed them the proper data, these “intelligent” software systems can find relationships, provide advice, and make even decisions for your plant and take action.
Weather-Based Operational Decisions
As a very basic example, it may be determined that on hot, humid days the speed of operation should be reduced to reduce the number of equipment failures.
While operators may develop anecdotal observations that lead a person to conclude that “this equipment always seems to fail when it’s hot and humid,” if the data proved the relationship, then operational decisions could be made based on weather conditions to avoid failure and downtime: reduce production rate by 5% under certain circumstances, reduce by 10% under more extreme circumstances, and so on.
Industry 4.0
All of this ties in with Industry 4.0, where production lines control themselves based on production demands, weather, equipment health, and other information.
Rather than humans trying to make these decisions on the fly, smart systems can optimize production if they have access to live data and a rich history.
Predictive Maintenance
In the modern era, predictive maintenance is driven by artificial intelligence. Predictive maintenance systems utilize vibration and other data to determine the health of an asset and provide automated diagnoses and, optionally, corrective maintenance recommendations. In this case, the machine learning system must be taught how to interpret the data in relation to the assets being monitored.
However, some systems are being developed to teach themselves. To learn about the relationship between condition monitoring parameters, and optionally operational parameters (speed, temperature, etc.), a system might “watch” those parameters and also watch the maintenance management system. When it sees that a work order is generated to replace the bearing because the bearing has failed, it can correlate that to what was observed in the condition monitoring and operational data.
Unfortunately, unless there is already a history of condition monitoring data, operational data, and maintenance history, it must learn that relationship over multiple failures.
The field of machine learning is powerful, and it will shape our future. We must prepare for it by accurately storing analyzable data at every opportunity.