Data analytics and machine learning are revolutionizing the way to monitor the health of critical machine components.
The goal of preventive maintenance is to avoid the costly consequences of unplanned downtime when machine parts fail. In many cases, the replaced components are still within spec, but we replace them anyway instead of taking the risk – especially when the equipment is critical to the output.
One accusation that is regularly made about this approach is that it can lead to excessive maintenance. Operators are much more likely to know when a machine part fails, so they can avoid work until it’s absolutely necessary. Ideally, this would be the exact remaining life of the part, but predictive maintenance technology is a long way from achieving that level of precision.
What we often see happening today, when it comes to critical machines, is a condition-based approach to maintenance that uses condition monitoring solutions that rely on sensors to detect signs that a part could be deteriorating. Once a certain threshold is reached – be it pressure, temperature, vibration, or whatever – an alarm is triggered to inform the technicians to replace the machine part.
While this is better than periodically changing components, it still uses pretty rough limits to determine when a part needs to be replaced.
However, the rise of anomaly detection is changing the game by enabling next-generation predictive maintenance. This technology uses multidimensional data and machine learning to evaluate various factors that can affect the condition of a part and how they are likely to affect each other before determining whether the data indicates a real problem with a machine.
The advantage of this multidimensional approach is that it adds context to the situation. For example, it understands whether variances are due to deteriorating component conditions or normal seasonal changes due to production demand.
Process parameters sampled from a conveyor electric motor may exhibit abnormal activity characteristics that – under a traditional condition-based maintenance approach – could trigger an intervention alarm. However, when other contextual factors such as seasonal workloads are taken into account, the algorithm can conclude that these levels are normal and that no intervention is required.
Not only does this save operators time for unnecessary maintenance, but it can also help identify more accurately why components fail in the first place.
If it turns out that a machine part only needs to be changed for three years in a design life of, say, ten years, anomaly detection can reveal which factors are responsible for this. Perhaps the logic that drives a motor is misconfigured, causing it to overload. By identifying such a problem, the operator then prevents the same problem from occurring again and again with subsequent replacement components.
This level of understanding also allows companies to perform service eligibility checks on their legacy machines, or to perform due diligence on another organization’s assets before an acquisition. The insights gained from anomaly detection can also help Original Equipment Manufacturers (OEMs) improve their specifications and designs. This will help make future equipment more responsive to operators’ needs.
In many cases, without detecting anomalies, organizations would have to purchase substantive expertise, at significant cost, to determine the health status of an asset.
This technology isn’t just changing the game for businesses looking to reduce critical equipment downtime. Anomaly detection is a cost-effective solution that can be deployed without disrupting the output, so it can also be deployed on less critical machines. As long as the machine is sufficiently data-rich and has good connectivity, anomaly detection can be trained using data sets collected from healthy components.
While anomaly detection cannot yet infer how much time apart has before it fails, it is one step closer to that predictive maintenance goal. By enabling us to better understand the potential for failures, it helps to reduce planned maintenance and operational downtime, improve operational safety, provide sustainable maintenance practices and save businesses huge costs.