Data management and the maintenance of clean, usable data for asset performance metrics pose great challenges for manufacturers today. The abundance of irregular and inaccurate data limits the analysis reliability engineers can conduct on data sets. The resulting cycle of failure and maintenance plagues manufacturers with inconsistent data processes and limited resources available to correct historical data.
In the age of big data, manufacturers must resolve data entry issues to achieve high-quality asset performance analytics that drive cost savings for organizations. One of the most traumatic events for any industrial organization is asset failure. From stopped production and costly physical damage to safety concerns for humans and the environment, the impacts of asset failure remain the top risk for industrial companies.
Fortunately, manufacturers now have more advanced software and computing capabilities at their fingertips, and accurately calculating failure rates can be achieved with machine learning technology known as cognitive analytics. This methodology uses natural language processing on work order descriptions to more intelligently identify bad actors, determine asset failures and improve reliability.
The Society for Maintenance and Reliability Professionals (SMRP) defines a failure as a situation when “an asset is unable to perform its required function.” Despite this clear definition, many engineers and operators hesitate to mark an asset as failing when an incident occurs, which results in many true failures mischaracterized in maintenance data sets. If manufacturers can’t properly track failures and determine an accurate mean time between failure (MTBF) for an asset, asset performance and the organization’s bottom line suffer.
One reliability team tried to determine how often their assets failed and prioritize the list, but they ran into a situation where they realized that the breakdown indicator field in their computerized maintenance management systems was rarely populated and it was difficult to determine whether the asset actually failed. As a result, the MTBF calculated using the raw data was 16,000 months.
After investigating the work orders, the team found several that should have been marked as failures according to the SMRP definition, but were marked incorrectly. The work order description clearly outlined the problem and why the asset failed, including “the pump needed to be rebuilt because of a seal failure.” This work order should have been marked properly to indicate asset failure and get counted in the MTBF metric.
Cognitive analytics technology builds models trained on input data and then applies them to new data to make predictions. The company with an MTBF of 16,000 months applied cognitive analytics to its data and the MTBF was recalculated to be just 14 months. Several work orders incorrectly marked were corrected and this classified data was verified for accuracy by engineers.
Cognitive analytics includes algorithms for identifying maintainable items and failure mechanisms from the text fields. To determine whether an asset failed, the machine learning classification model uses text information about an event as input data and returns a prediction and score on whether that event was a failure or not. Users go from having missing or “unknown” fields to performing analyses on detailed data for insights that help organizations improve asset performance.
Incomplete and inaccurate failure data can be effectively solved using cognitive analytics technology and reliability practices. Clean data can provide critical insight into machine health and help predict mechanical failures. With cognitive analytics, organizations can more effectively reduce failure risks and unplanned maintenance costs and improve operational excellence. More than ever, progressive data science is helping industrial operating facilities adopt proactive maintenance techniques that ensure compliance, improve safety and ultimately save a lot of money.