For many years, the process of analyzing lubricants, coolants and fuels to improve the reliability and maintenance of machines has remained relatively unchanged. Fluids typically are sampled from an asset on a regular cadence and sent to a lab to be processed. The results are interpreted by experts or simple rules-based analysis, and a report is returned.
The asset owner must then use the report to decide whether to take action or continue normal operation. Although laboratory analysis produces extremely valuable data, the process can become somewhat cumbersome and may not provide a consistent spectrum of deep insights that asset owners need to increase performance and reliability.
Today, the rise of artificial intelligence (AI) and machine learning is allowing the aggregation of all lab data into a single platform as well as the ability to look at assets in a deeper and more granular way. This is resulting in greater insight that can deliver better precision, consistency and lead times than previous methods.
Fluid analysis was first pioneered by the rail industry in the 1940s. Railroads quickly realized the analysis of fluids could proactively identify potential issues with diesel engines, air compressors and other rail equipment. By the 1960s, the analysis of industrial lubricants, coolants, fuels and other fluids by commercial laboratories had become common.
The data delivered to industry by these laboratories facilitated the transition away from time-based maintenance toward condition-based maintenance for critical components.
Oil analysis is now being utilized for a number of benefits, including root cause analysis, preventive maintenance, condition-based oil changes and proactive component change-outs. For new equipment, it can be used as a supplement to the data generated by sensors and onboard computers.
On legacy equipment with no telematics, it serves as one of the only insights operators have for asset health. The practice of oil analysis has become widely utilized and recognized as valuable across all industrial sectors, including agriculture, aviation, energy, transportation, mining, manufacturing, and oil and gas.
Laboratories and industrial companies currently utilize oil analysis data in a number of different ways. While each of the following methods has benefits, there are also some important drawbacks.
Manual analysis is the process of employing domain experts to review hundreds of lab samples per day and produce recommendations from those results. For large industrial businesses, this method requires a significant number of trained employees to continually monitor laboratory results. As a quarter of the industrial workforce becomes eligible to retire in the next decade, retaining existing experts while training new, younger technicians will become a growing challenge.
Analysts tasked with reviewing many samples on a daily basis across several asset types can experience fatigue. At worst, this can yield missed opportunities for alerting on issues that may result in catastrophic machine failures. In addition, human-centered analysis can be inconsistent due to differing views from analyst to analyst. All of these factors can lead to varying levels of value obtained from laboratory analysis.
Manufacturers of industrial equipment and lubricants often provide acceptable levels of wear metals, contaminants and fluid quality. Because of a lack of context concerning the usage of each asset, the acceptable ranges from original equipment manufacturers (OEMs) often lean toward the conservative end of the spectrum to protect the asset. In practice, conservative ranges can produce a high number of alerts where no defect is found on the asset or with the lubrication.
For example, a 2017 case study involving a Class I railroad found no defects were actually present in 86 percent of the alerts generated. This lack of precision by the OEM alerts has created distrust in oil analysis at many organizations. Moreover, these false alerts ultimately result in thousands of dollars in unnecessary labor and material costs every year.
Statistical analysis is used to build acceptable ranges for lab results based on a representative repository of historical sample data. These ranges are then converted to alerts for each customer. In this method, industrial assets must be paired with a dataset with similar assets so expected normal ranges can be created. Statistical analysis can produce better alerting than previous methods but is difficult to manage and is dependent on the historical dataset used.
Over time, the asset’s normal acceptable ranges may change based on operating conditions, weather, age and other factors. Operators also may switch lubricant types or top off the fluids in their equipment. With each change, the accuracy of statistical analysis will inevitably deteriorate over the life of the equipment, and the rate of false alerts or missed recommendations will increase. In the end, statistical analysis proves capable in a never-changing application but often falls short in today’s highly variable industrial spaces.
Fortunately, AI is able to resolve many of the drawbacks of these other methods. AI and machine learning techniques utilize both lab analysis data and asset failure data to recognize the difference between normal and alarming lab results. These techniques can be significantly more precise, as they take into account an asset’s full dataset over its lifespan as opposed to relying on acceptable high and low set points to produce recommendations.
By utilizing multiple signals at once, AI can provide very specific recommendations, such as alerting on a bearing failure based on tin, lead and copper content changes in the oil. AI and machine learning can also distinguish the difference between the slow, acceptable rise of soot in a normal operating engine and the fast rise of soot in an engine with an injector issue. Through feedback from the end user, AI and machine learning can even adapt with machinery to ensure false alerts or missed alerts don’t occur due to changes in the lubricant manufacturer, machinery age or a new operation.
Some organizations have begun using AI on lab data to obtain better insights from the analysis they already perform. In a recent case study conducted for a Class I railroad, 7,683 assets were tracked using conventional laboratory oil analysis as well as AI and machine learning. Over the course of the study, the AI and machine learning analytics proactively identified twice as many failures as compared to the conventional lab alerts.
The AI and machine learning alerts also saw an increase in precision by 3.9 times as compared to the conventional alerts. Additionally, the predictive ability of AI and machine learning increased the number of critical alerts with at least 30 or more days of forewarning by 4.5 times.
This increase in alerts, accuracy and lead time offered by AI and machine learning is causing many organizations to take notice. As the reliability and uptime improvements continue across multiple industries, more and more companies are likely to jump on the AI bandwagon. Will yours be one of them?