“The data we’re collecting is just right as it is!”
Three considerations to make sure your data is suitable for reliable machine health analysis
Dr Rebecca French, Data Scientist, Senseye
The benefits of shifting to a successful Predictive Maintenance (PdM) regime are a matter of record: typical 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. With such bumper business benefits on offer, it’s no wonder that more companies are thinking about beginning their own PdM journey using an automated solution.
The ability of these solutions to make accurate predictions about future machine health can transform maintenance operations. But it’s a myth that simply feeding any old machine data into the latest condition monitoring tools will somehow enable organizations to transition to an effective PdM regime. The truth is that only the right kinds of data can deliver the results that companies are looking for…
Today, machines industrial machines are populated with more sensors than ever, providing a wealth of possible information for PdM and other analytics systems to work with. So the question is, how can you tell whether the machine data that you have available can be successfully redeployed to support a PdM project?
The precise answer will be specific to each site, but there are three considerations that can help users judge whether their data-gathering capabilities are likely to be ‘PdM ready’.
Online data, at a suitable frequency
First, data must be available at the right frequency. Sampling intervals vary, but the general rule is that the fresher the data, the more accurate your analytical options can be.
What exactly that means will depend on the machinery. For instance, a motor will generally need to report back several times an hour to enable a PdM system to spot any glitches or longer-term trends in its behavior. In contrast, a pick-and-place system will be carrying out a series of extremely rapid and precise cyclical movements and users may need to collect data several times per second to enable a PdM system to spot the earliest indications of deteriorating performance.
It’s important to strike an appropriate balance in terms of processing power and bandwidth; vibration data could be collected at tens of thousands of hertz but will it be useful to send that to the cloud (probably not) – consideration must be given as to where is most appropriate to process the data.
Paint a picture
Data is the only way the PdM analysis system can ‘see’ what’s happening, so it needs to provide an effective indication of the condition of each asset.
For example, vibration is a popular choice for condition monitoring information because it can flag up several common maintenance issues, such as worn bearings or mechanical misalignment. Alternatively, the torque or current being pulled can be a useful indicator of changing motor behavior. Meanwhile, small variations in cycle times might indicate that the performance of a robotic system is starting to waver.
Whatever the specific circumstances, the general rule is to focus on whichever parameters enable condition monitoring algorithms to build a clear picture of the near real-time functioning of each machine. As a rule of thumb, if a condition monitoring engineer can’t spot a fault using the data being captured then an automated system is unlikely to – the data captured needs to be appropriate for the failure mode being targeted.
The human factor
Third, the most successful PdM deployments harness the knowhow of the people most familiar with the machinery. Many experienced operators and maintenance engineers have been working with the same assets for years. They are ideally placed to provide the context that can turn raw data into useful information about how machines are likely to function – or malfunction - in the future. At the most basic level, it’s about understanding that the last time a pump made a knocking noise like that, it broke down a week later.
While the idea of building a detailed mathematical model for each asset – the so-called ‘digital twin’ approach – might be appealing in theory, it’s impractical when deploying condition monitoring at a large scale.
Modern PdM systems start by flagging up changes to experienced users who are best-placed to recognize when an anomaly calls for a maintenance intervention. This is especially important in the early stages of a PdM deployment when the automated solution is on a rapid learning curve. The more active the feedback loop between the human operators and the artificial intelligence-based solution, the faster the system learns.
By monitoring the reaction of operators to alerts, PdM software (such as Senseye PdM) soon learns which patterns in the data are of interest and uses its unique Attention Index feature to direct maintenance efforts towards the highest priorities set by users.
How ready are you?
The benefits of automated PdM solutions can be enormous but they are tools and not magic bullets. The most successful users will have considered carefully whether their existing systems are set up to support a PdM solution effectively. Without the right data and feedback mechanisms in place, no PdM deployment can be expected to deliver the full package.
Senseye works with potential customers to assess their current position and come up with a Readiness Index. This clarifies whether the user is at the right point on the journey to deploy the SenseyePdM solution successfully or if more preparatory work needs to be done.
If you’re not sure whether your organization is in a position to make the most of automated condition monitoring, why not get in touch and find out more about how we can help?