Data Analytics Technologies Help Reduce Equipment Failure

Data analytics software and machine learning capabilities give pharmaceutical manufacturers new tools to move beyond preventive maintenance into predictive and prescriptive maintenance, which promises to be more effective and save time and money. Pharmaceutical Technology spoke with Dennis Belanger, director of Operational Certainty Consulting at Emerson, about these new methods. 

Maintenance technology trends

PharmTech: What is the difference between preventive, predictive, and prescriptive maintenance? What do you see as the trend in maintenance of equipment in pharma manufacturing?

Belanger (Emerson): Today, prescriptive maintenance is being adopted by best-in-class pharma manufacturers to drive better production through more informed decision making. Prescriptive maintenance uses Big Data analytics to find patterns or anomalies in large amounts of seemingly unrelated data-understanding and evaluating the performance of a process or system rather than measuring the condition of a single piece of equipment. Finding these indicators can help organizations identify areas of good and bad performance that might be impossible to identify through standard monitoring procedures. The prescriptive element comes in when organizations can associate a corrective action or adjustment to the anomalies and provide specific actions to take to minimize or stop the failure.

Many organizations still use preventive and predictive maintenance, and there is an important distinction between them. Preventive maintenance tasks are performed on a fixed schedule or a metric such as run hours or cycles. Preventive tasks inspect assets or bring them back to a certain condition and typically require a scheduled system shutdown. Preventive maintenance tasks should only be used when it is not possible to detect failure modes in a timely manner with predictive monitoring solutions  because they provide only few weeks’ notice of potential failure. Filter changes, leak inspections, and alignment checks are good candidates for preventive maintenance.

Predictive maintenance uses technology to monitor and trend an asset to detect change in conditions that might indicate a future problem. Maintenance data are collected regularly while equipment is running, through manual rounds or through devices installed at the collection site. Predictive maintenance eliminates the need to schedule a shutdown and provides a better indication of the actual condition of equipment. Predictive technologies have become quite popular because they can provide early indication of potential asset failure-typically allowing many months of advance warning, which supports organizations’ planning of turnarounds and scheduled outages to minimize overall downtime. Vibration analysis, infrared testing, and condition monitoring are all common examples of predictive maintenance technologies widely used today.

Best practices

PharmTech: What are some best practices for using predictive maintenance?

Belanger (Emerson): One of the best practices a pharma organization can apply is to perform a maintenance strategy analysis using a failure modes and effects analysis (FMEA). This analysis would typically incorporate critically ranking information and will ultimately show an organization where best to apply the resources of a predictive maintenance program to have the strongest impacts. Often, organizations view predictive maintenance as a costly expense and use it only sparingly on their most critical assets, but applying it more widely can result in much higher savings if the application is properly targeted. Top quartile performers typically use vibration analysis on 80–90% of their rotating equipment. These organizations know that predictive maintenance is approximately 50% less costly than preventive maintenance, while still providing a much better indication of potential failure. Moreover, online predictive maintenance can also provide earlier detection of issues, making it easier to plan repairs, which typically cuts repair costs by 40–50%.

PharmTech: How are models for maintenance developed, and how can they be best used?

Belanger (Emerson): Common assets and systems typically have common problems and impacts on a business. Emerson uses these commonalities as a starting point to develop models that can be used hundreds or thousands of times across many clients with minimal customization. Organizations can use these models to effectively and efficiently simulate common assets such as compressors, distillation columns, heat exchangers, pumps, valves, generators, and turbines, making advanced analytics more achievable and sustainable. 

Many models for high-impact, known problems can be applied across industries or organizations because the models are built for common equipment. As Emerson expands solutions to address business problems for complex systems and processes, more specific models are possible and will be useful to target a company or plant’s unique environment. When individual systems begin to vary greatly in configuration, engineering design, age, and monitoring techniques, there are more complex custom solutions that can be built from scratch and can deliver end user-specific insights. 

PharmTech: What is the importance of detecting and addressing issues in real time?

Belanger (Emerson): The importance of detecting issues in real time is linked to the duration between the detection of a failure mode and the onset of a problem. For example, some electrical failures occur in a matter of minutes. Such a failure would require real-time or near-real-time detection to provide enough time for maintenance to intervene. On the other hand, failure modes such as bearing deterioration may take months to go from detection to failure. When the time to failure is that long, more intermittent monitoring may be appropriate. 

While monitoring every piece of equipment in real time would be nice, an organization would need to be able to handle the volumes of data such monitoring would generate. Understanding that time to failure dictates real-time detection in only a handful of cases, most organizations opt to monitor many assets in near real-time or even using periodic readings, which still allows them plenty of time to react to problems.

If you are going to spend the effort and resources setting up and maintaining infrastructure to detect issues in real time, then you also need to be able to act in real time. Detection is only a part of the picture. Organizations also need to examine the established work management process so that significant lead time translates to timely and precise action. 

Examples of addressing equipment failures

PharmTech: Can you give some examples of detecting and addressing equipment failure modes in pharma manufacturing?

Belanger (Emerson): There are many examples of varying size and scope, but three examples in particular stand out. Emerson worked with one organization to develop a machine learning system that could detect sensor drift on a temperature sensor for a heat-treat skid. That implementation detected an aberration 60 days in advance, which allowed the organization to save a batch worth over $1 million.

Another organization used vibration analysis to detect misalignment and bearing failure on a chilled water system pump for a system performing batch experiments that require constant temperature. That system detected a failure 120 days in advance, saving batch experiments that, if lost, could have wasted months of time.

In one innovative example, a pharmaceutical organization used machine learning to capture knowledge and monitor systems with the objective of minimizing the impact of a potential loss of 25% of the organization’s workforce due to expected retirements. 

What all successful examples have in common is that decision makers closely examined critical points of failure in the organization and developed solutions that gave the organization the time it needed to react efficiently but thoughtfully, to drive more positive outcomes overall.