Most people in maintenance do not really understand advanced analytics like artificial intelligence and machine learning (AI/ML). Personally, I think advanced analytics have a lot to offer maintenance professionals, but I am an analytics geek. Eventually analytics will be used across all maintainable assets and perhaps even on run-to-failure assets. Advanced analytics and low-cost sensors will allow the vast majority of repairs to be conducted well before an asset has a functional failure. I would like to share a few thoughts that I hope will allow more people to adopt AI/ML as practical maintenance tools.

Advanced analytics are software applications that take numeric data from one or more sensors and identify and memorize patterns in individual and groups of sensors. The most important capability they have is to identify normal data patterns and detect when data patterns deviate from normal. Their capacity to analyze groups of data allows maintainers to quickly evaluate whether individual assets or complex asset systems are behaving normally. Once the data indicate a deviation from normal, the designed functions of the asset (first principles) are the key to interpreting what is occurring and most importantly what should be done about it. Some advanced analytics applications have embedded first principles information and automate the translation of abnormal conditions and identify specific failures.

The math for defining data patterns is not new. In the late 1980’s, I ran statistical software on a 386-desktop computer. It was a laborious process involving spreadsheets and floppy disks but the software greatly simplified finding patterns in the data and helped me overcome my fear of matrix algebra. During the 1990’s, I used similar techniques to build econometric models that forecasted energy demand for more than 50 countries. We now have much better computing, data management and analytic software capabilities allowing faster and continuous processing of more data. We can simply apply more analytics to more problems than ever before.

Applying advanced analytic techniques to asset health is not new either. More than a decade ago I was working on a partnership involving a powerful algorithm for finding early indications of functional problems. It worked really well. (It was far more accurate than my economic forecasts!) At the time, these techniques were mostly applied to very large, expensive and critical assets where data consisting of pressures, temperatures, flows and vibration readings were readily available. Nowadays pattern recognition can be applied to assets more broadly. A host of supporting factors like lower cost sensors, lower sensor installation costs, and greater access to operational data help make advanced analytics available to more industries and able to cover more assets than ever.

While advanced analytics offers a lot to be excited about, be wary of claims that artificial intelligence and machine learning can overcome poor data. Garbage in – garbage out is as immutable a law as the physical laws of thermodynamics. If you have not maintained good data practices in your CMMS, from your asset master data to your maintenance history, you will need to invest a lot of time and energy for your legacy data to add value to your new analytics.

What you do with the output of analytic algorithm matters most. No one should be rewarded for finding a potential failure. Rewards should only be handed out for finding and repairing the problem without a functional failure! I once heard a client tell how they had abandoned their predictive maintenance program because it failed to deliver value. Their rationale was not what you might think. They were actually very successful at finding potential failures but they garnered no value because they could not move the required actions through their work processes before the assets actually failed! If your work processes are broken or your people do not follow your processes, you will not get value from better early warning systems such as advanced analytics.

Top performers on the other hand can easily adopt these systems to augment their existing predictive maintenance programs and get even better performance. If you are not a top performer, you should start by covering your most critical systems because you will be able to manage the findings on a few assets much more easily in an ad hoc process than a system covering dozens or hundreds of assets.

I expect that most of our clients will add advanced analytic capabilities to their maintenance and reliability programs. Some may take years and some may start tomorrow. Advanced analytics will change how they identify maintenance work. The software may prescribe a specific repair and time requirement or it may simply issue a slate of diagnostic tasks that are designed to identify the failure more precisely. Algorithms may also require user inputs to continue learning, such as classification of events. Having another system involved in maintenance means that processes for identifying and closing work will need to change. Importantly, people’s behaviors will have to change too. PCA is good at solving the old problems of process design and culture change. Once you have decided on an advanced analytics solution, we can help you realize success with it!