OR WAIT 15 SECS
Volume 44, Issue 1
Data collected through the Industrial Internet of Things enable predictive maintenance.
Performing equipment maintenance to prevent breakdowns or unplanned process stops is an obvious best practice; how to know when to do that maintenance is not as simple. Preventive maintenance is the conventional pharmaceutical industry practice that involves setting a time-based maintenance schedule for a piece of equipment, typically using models based on experience and original equipment manufacturer (OEM) recommendations. Preventive maintenance schedules are usually set conservatively, to be short enough to have a low risk of failure.
Digital tools are helpful in managing this routine and scheduled maintenance. “Many companies still rely on paper records to manage their manufacturing floors, including equipment maintenance activities. But in a high-paced and high-volume production environment, you simply can’t be proactive when you operate on paper,” suggests Matt Lowe, MasterControl’s president of laboratories. “Digital preventive maintenance systems are the bare minimum in today’s competitive manufacturing marketplace. The goal is to avoid missed or delayed maintenance tasks and keep equipment in good working condition,” says Lowe.
Some pharmaceutical manufacturing facilities run 24 hours a day, seven days a week, with two-week shutdowns twice a year for preventive checks, notes Jon Biagiotti, product marketing manager at Augury. This standard approach, however, may not be cost-effective or an optimal use of resources. “Preventive maintenance may be done too late, not addressing potential issues until the next scheduled check, so that it degrades to a more expensive fix. Or it may be done too early, when it isn’t needed yet,” he says.
“The pharmaceutical community is showing great interest in predictive maintenance because the conservative nature of our applications results in frequent preventive maintenance. Preventive maintenance is not always needed and results in costly downtime,” adds Pamela Docherty, industry manager at Siemens.
Continuous monitoring and condition-based predictive maintenance offer the potential to improve efficiency and quality compared to time-based preventive maintenance.
Predictive maintenance takes asset health analysis to the next level, by collecting data from equipment using sensors connected through the Industrial Internet of Things (IIoT) and analyzing that data to predict how an asset will perform in the future. Decisions are tailored for a specific situation, rather than following a general expectation.
“In the past, predictive analytics on a set of many assets was too time consuming to be practical, but advanced analytics enables faster, cost-effective insights,” explains Michael Risse, vice-president and chief marketing officer at Seeq. Using the IIoT and predictive analytics, “the assets that need attention provide advanced warning on what they will need in terms of spare parts and maintenance in enough time to take action at the best price and timing for the organization.”
In many cases, data needed for predictive analytics are already collected and available in historians or other databases, says Risse. If more data are needed, sensors and wireless networks are easily added. The barrier to predictive maintenance is, thus, not the availability of data, but the ability of subject matter experts to leverage the data. “It’s the ability to create actionable insights and deliver it through an easy-to-use interface that creates value,” notes Risse.
Enabling process engineers to analyze the data themselves, with self-service analytics, gives these experts the knowledge they need to optimize maintenance, says Edwin van Dijk, vice-president of marketing at TrendMiner. Another key to self-service analytics is contextualizing the data coming from the equipment using process-related data. “The goal of predictive maintenance is to be able to perform maintenance at a time when it is not only the most cost-effective, but also when it will have the least impact on operations,” says van Dijk.
“Human intervention is critical to determine the best course of action based on the available information,” adds Lowe. For example, with insight into an upcoming problem, “manufacturers can proactively reassign equipment and divert upcoming batches to other production lines.”
Advances in computing power and in artificial intelligence (AI)-particularly machine learning-have enabled predictive maintenance. For example, digital twin libraries (i.e., collections of models) were originally developed by OEMs for specific equipment, and now general models that can be tuned to specific pieces of equipment are increasingly available, says Elinor Price, senior product manager at Honeywell Process Solutions. She says that the role of the asset digital twin is to alert the maintenance team to be proactive rather than reactive. For example, advanced pattern recognition analytics (i.e., machine learning) can identify potential equipment problems by spotting changes in flow or temperature before they are large enough to trigger an alarm on the control system.
“Machine learning algorithms build a model of the machine to learn how it operates,” explains Biagiotti. “By comparing current performance to past performance, anomalies can be detected. Full fault diagnostics can be conducted by looking at the frequency spectrum, applying pattern recognition, and comparing signals to similar machines. Based on [these diagnostics], specific, actionable recommendations are made to improve the health of a machine.”Machine learning is often based on vibration analysis, but it goes beyond a conventional rules-based system. “The system learns how the machine operates so that you don’t receive false alarms,” says Biagiotti. “By comparing one machine’s data to similar machines, the accuracy improves exponentially as we collect more data. Because the IIoT is being leveraged, manufacturers can benchmark equipment and production lines at a global level, comparing plants around the world.”
Biagiotti says that one of the main uses of machine learning algorithms is monitoring cleanroom utility equipment, which are especially critical because shutdowns result in the time-consuming and expensive process of reconditioning the cleanroom. Air-handling units, for example, are usually enclosed and difficult to access, but wireless sensors can be placed in the enclosure to send data through the IIoT. In one case, machine learning algorithms identified bearing wear on two air handling units, and correcting the problem prevented an unexpected shutdown.
In another case, vibration analysis was used to detect misalignment and bearing failure on a chilled water system pump. The pump was required to keep a constant temperature for experimental product. “The system detected a failure 120 days in advance, saving batch experiments that, if lost, could have wasted months of time,” says Dennis Belanger, director of Operational Certainty Consulting at Emerson.
He reports on another use, “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.”
Heat-exchanger performance is crucial for process control and offers an opportunity for maintenance optimization, says van Dijk. “Fouling of heat exchangers increases the cooling time, but scheduling maintenance too early leads to unwarranted downtime. Scheduling too late leads to degraded performance, increased energy consumption, and potential risks,” he explains. “In a reactor with subsequent heating and cooling phases, the controlled cooling phase is the most time-consuming, and it is almost impossible to monitor fouling when the reactor is used for different product grades and when a different recipe is required for each grade. In one instance, a monitor was set up to look at the cooling times of a company’s most highly produced products. If the duration of the cooling phase started to increase, a warning was sent to the engineers who could then schedule timely maintenance, sometimes two to three weeks in advance. The gained benefits are extended asset availability, predictive maintenance leading to operational and maintenance cost reduction, and reduction of safety risk.”
Predictive algorithms can also prevent the breakthrough of a filter in a suspension tank, which is used for removing impurities in a product before it is fed into the batch. “Sometimes one of the valves can leak, and gas can enter the system. But sometimes the valve can really be stuck due to solids, and the pressure keeps on building up until the filter eventually breaks,” notes van Dijk. “Using self-service analytics, process engineers set up the monitors to identify when the valves were leaking, which could be an early indicator of a filter breakthrough that could contaminate an entire batch. With the predictive monitors, the equipment can be replaced sooner, or the process can be controlled differently.”
Belanger concludes, “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.”
Prescriptive maintenance describes a method for automatically scheduling required maintenance based on predictive algorithms. “This type of maintenance requires even more data from many more sources than the ‘few’ sensors at the equipment. Operational contextual information is required to artificially assess all circumstances to generate the adequate prescription for the maintenance required,” explains van Dijk.
“Prescriptive maintenance is being adopted by best-in-class pharma manufacturers to drive better production through more informed decision making,” adds Belanger. He explains that this method uses analytics tools 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.” A corrective action is then “prescribed” to minimize or prevent failure.
The pharmaceutical industry is not ready for a “fully artificial intelligence-led prescriptive analytics system for running an autonomous factory,” says van Dijk. “A human-interacted artificial intelligence system is currently a much safer bet.” In this system, the process engineers and operators use all the available information to create “process monitors.” These automated monitors send “prescriptions” for future maintenance action to the appropriate people or systems in the plant.
For both predictive and prescriptive maintenance, understanding the process, building data models, and analyzing the data are key, says van Dijk. Subject matter experts-the process engineers-can use “self-service analytics” tools to search and filter data, perform root cause analysis, test hypotheses, and build monitors to predict process and equipment performance. Van Dijk explained three ways to analyze data using this self-service analytics approach.
“The first is event-based. If a certain signature behavior is detected that can affect another part in the process that typically occurs later, a notification can be generated. This notification can include instructions for the required preventive actions or required maintenance.
“The second is probabilistic. The current behavior is interpreted, and a likeliness of future behavior is calculated, optionally resulting in automatically scheduled maintenance work orders with the needed instructions.
“The third type is regressive. The prediction is based on certain conditions that must be met and verified, and in case of deviations, the instructions can be given to the control room, or maintenance can be scheduled for the near future.
“For all three situations, the events can be captured in case they occur, providing more information for improving future predictive and even prescriptive maintenance work.”
When getting started, companies should first analyze which data are already available and whether existing networks are adequate for data collection. If so, they should move forward with analyzing data, “find the low hanging fruit,” and use it to optimize maintenance activities, suggests Donald Mack, industry manager at Siemens.
Quality teams must be educated on the reliability of predictive maintenance, adds Docherty. “It is likely that companies will ‘watch’ the predictive maintenance data and get an understanding, while slowly pushing the time interval between each predictive maintenance,” she says.
Data integrity is crucial for IIoT-connected equipment. “Digitalization and cybersecurity go hand in hand. What were once isolated, nearly impossible to access devices are now being brought on to the information superhighway,” says Mack.
“A strong IIoT solution requires a detailed, security-driven system architecture that can effectively represent multi-layered security within the solution,” adds Brycen Spencer, IoT consultant at Siemens. “Companies should seek a solution designed to be scalable, resilient, and efficient. Features such as strict access management, encryption, network security, tenant and environment separation, and filtered communication channels are fundamental to good IIoT architecture."
Vol. 44, No. 1
When referring to this article, please cite it as J. Markarian, “Optimizing Machine Health," Pharmaceutical Technology 44 (1) 2020.