Digitalization’s Influence on Maintenance Strategies

Informing condition-based maintenance

Condition-based maintenance (CBM) was the idyllic maintenance strategy of the Lean Six Sigma age. The premise of CBM is that an event, trigger, or exceedance drives maintenance, rather than a schedule. The challenge with CBM was never in defining the condition that prompted a maintenance activity, because these conditions are often predefined by ancillary equipment constraints, like a maximum allowable temperature or pressure, or a minimum flow requirement. Rather, the challenge was in determining which limit would be exceeded, and when.

CBM is much more valuable when combined with model construction, marrying monitoring and forecasting techniques that together make up predictive analytics. For example, using near-real-time vibration data to do CBM on a fleet of pumps requires setting conservative triggers to effectively avoid running to failure in the time it takes to schedule and perform maintenance activities following detection. The downside of a conservative trigger is the inevitable presence of occasional false positives.

By contrast, when past and current vibration data is leveraged using a model, the model can construct a vibration forecast and compare it with historical runs in relation to other process signals. This better-informed prediction can approximate not only when a trigger will be hit, but when the failure is likely to occur, providing the greatest possible lead time for maintenance. In addition, these predictions can actively adjust over time, informing and updating the urgency of service based on how operation continues.

Digitalization-enhanced predictive analytics examples

Many manufacturers in the chemical and other process industries have transformed their operations by developing maintenance strategies built on the foundation of predictive analytics. As a result, these companies are minimizing downtime, unnecessary maintenance and operational uncertainty, saving millions of dollars every year. The most successful predictive analytics applications often combine first principles models with statistical techniques to develop forecasts based on theory and behavior.

Filtration membrane predictive maintenance

When producing certain biopharmaceutical compounds, the desired molecules are separated from other species using membrane filtration systems. During each batch, particles build up on the membrane, and a clean-in-place (CIP) procedure must remove the accumulation between batches. Over time, these filtration membranes can degrade, causing CIP procedures to become less effective. 

Engineers at a major bioprocessor suspected this was the case with one of their operations. Fearing unplanned downtime, they sought a way to identify long-term particle buildup on the permeate filter to predict when maintenance was required.

Using an advanced analytics application, the manufacturer calculated the filter membrane resistance based on pressure and flow sensor data, and on known values of surface and fluid viscosity, by applying Darcy’s Law. This reduced the variables of interest, providing clear visual indicators of degrading membrane performance, confirming the engineers’ suspicions (Figure 2).