Pros and Cons of AI for Predictive Maintenance

Artificial intelligence (AI) underpins today’s automated predictive maintenance (PdM) solutions. For instance, Senseye PdM’s algorithms alert users to deteriorating machinery in time to prevent a breakdown, delivering a wealth of associated benefits such as reduced downtime and a more targeted maintenance effort, boosting sustainability and efficiency. These tools are so powerful that it’s easy to get carried away about the capabilities of AI in the context of PdM.  

One common misconception is that AI-based systems can predict when and how machinery will fail by spotting clues invisible to a human expert. But the truth is that trained condition monitoring analysts set the gold standard for understanding machine health. The most significant difference between human touch and what automation can offer is scalability, not some magical form of data analysis.  

An expert can only examine one asset simultaneously, while an automated system can monitor thousands of assets simultaneously. This paves the way for major changes. In the past, the time, effort, and cost involved in manual condition monitoring kept it confined to only the most critical assets. In contrast, automated PdM systems now make it practical to extend the same approach to every machine throughout your entire operation.  

Decision support tools 
The benefits of scalability also touch on the myth that PdM solutions threaten jobs. These tools are about empowering operators and maintenance teams to make better decisions, not replacing them. PdM solutions allow users to direct their asset management efforts in frankly impossible ways until now, enabling existing teams to be more productive, with often reduced budgets. 

Real benefits depend on quality data 
An AI-based PdM solution is transformative to maintenance operations – provided it has access to the right kinds of machine data. The key is to ensure that users can have enough confidence in the automated PdM system to respond appropriately when the system raises an alert. The confidence level will depend mainly on the quality of the available data entering the system. You can’t get away from the old saying, “Garbage in, garbage out.” 

Basic condition monitoring data is the bare minimum required for PdM. This might include parameters such as the current drawn by a motor or the timing between two set-points, where a short-term glitch or longer-term trend can mean that the condition of an asset is deteriorating.  

Users can have limited confidence in alerts raised using primary data because the process or environmental changes can also affect many parameters. Switching a pump over to handle a denser product or changes in ambient temperature could have a significant impact.  

Condition indicators take into account such misleading information and aim to strip out changes related to processing and environmental factors rather than the behavior of the machinery itself.  

Advanced condition indicators go a step further by targeting specific failure modes. The classic example would be in vibration monitoring, which someone can tune to look for indicators of failure visible at specific frequencies, which may, for example, indicate that a motor shaft is misaligned. 

A user can identify failure modes at the start of a PdM implementation by examining the maintenance history of each asset and the general engineering information associated with it. Users can then choose the condition indicators that add value and set up sensors to focus on the failure modes likely to provide the most considerable benefits.  

Proven success 
AI-based PdM solutions are not magic wands, but the enormous benefits of shifting to a successful PdM regime are well proven. At Senseye, we use proprietary AI and ML-based algorithms to help clients monitor tens of thousands of machines worldwide to improve maintenance efficiency and provide the correct information ahead of time. Results include unplanned machine downtime dropping by 50%, maintenance costs down by 40%, and a rise in maintenance staff productivity and downtime forecasting accuracy of 55% and 85%, respectively.  

Get in touch with Senseye for a demo and learn more about how we can help you reach your machine reliability and sustainability targets.  


New call-to-action