TRC-2019 37:26
by AJ Alexander, ITG Technologies
This presentation will address the business value of adopting machine learning technologies and how to properly measure the operational and economic impact these technologies may have on your line of business. All industries, no what the vertical, work hard to create machine, asset, and process reliability. The history of reliability centered maintenance has an interesting way of repeating itself and these typical breakdowns will continue to happen leading to catastrophic results. The human element, more often than not, the fault lies with the way we approach reliability and not with the machine. Emerging machine learning technologies brings a fresh approach to reliability centered maintenance with “evidence-based” hard data that replaces customary policy and subjective opinions regarding reliability best practices. In hindsight, reliability is simple, keep machines running and prevent breakdowns. Reliability initiatives have evolved over time to accommodate this, progressing from run-to-failure, to calendar/ condition-based, to prescriptive/descriptive maintenance. Unfortunately, these best practices are resource intensive and can’t always deliver the business value impact that was originally promised. I like using the below quotes to heighten the importance of addressing unplanned downtime and poor reliability.
“63% of all maintenance is unnecessary and causes more problems than it fixes.” – Emerson
“85% of all equipment failures happen on a time-random basis regardless of inspection and service.” – Boeing
Contemporary reliability applications use techniques to “trap” anomalies or changes in operational behavior of a machine that might indicate a problem. Such methods are complex, limited to certain equipment, prone to error, and ALWAYS require further expert investigation and validation; producing high levels of false positives. Machine learning agents learn operational behavioral patterns using actual data from sensors on and around a machine. Recognizing industrial diverse patterns in the sensor signals that indicate degradation, failure, and root cause.
