Who would have imagined how dramatically the industrial internet of things (IIoT) would elevate reliability and maintenance practices? Today, we have sophisticated sensors monitoring multiple variables, closing information gaps, eliminating data silos, and populating Big Data repositories in the cloud, where artificial intelligence (AI), advanced pattern recognition (APR), machine learning (ML), and advanced analytics work their magic on common industrial challenges.
Predictive maintenance (PdM) gave us our first taste of the power of monitoring individual machine conditions. With prescriptive maintenance (RxM), data is assimilated from diverse process and performance variables and woven into actionable recommendations (or “prescriptions”) on what to do, when to do it, and how.
The benefits are readily evident – better-quality data, earlier problem detection, more timely and accurate response, and perhaps of the most importance, less reliance on manual knowledge capture. Following are some companies that are on the cusp of this new level of maintenance maturity called RxM.
Network preparation at Penn State
Maintenance strategies such as PdM and RxM are possible only in connected environments. Tempered Networks recently helped Penn State’s Office of Physical Plant (OPP) instantaneously connect, segment, secure, and manage all of its network devices cohesively despite unique building and campus challenges. As a result, OPP is now making real-time control adjustments based on conditions, entering the predictive stage of maintenance and preparing for a future in which recommendations will be prescribed
Previously, each building was a separate entity. A lot of the systems in use were standalone, and there was a server for every application. “It causes headaches for maintenance when buildings are disjointed like that,” says Tom Walker, environmental systems design specialist at Penn State.
Now, about 300–350 buildings are connected at University Park, with all or most servers housed at the data center. Everything is on a virtualized server; hardware is shared among multiple systems; and authorized personnel have instant access to the systems. “This increased our resiliency, reliability, and overall uptime,” Walker says. “It also gave us the path to start sharing data with other systems and stakeholders.”
For instance, OPP is now working to enable fault detection and diagnostics within the building automation systems, which is expected to help reduce energy use and maintain optimum facility operation. OPP’s new energy dashboard visualizes when an energy problem emerges in a building so the issue can be addressed proactively. In the future, OPP would like it to prescribe what to do based on ML and data analytics from the connected systems.
Efforts are also underway to automate work orders in IBM’s Maximo based on certain fault conditions and eventually prescribe corrective actions. “Right now the work orders are only telling that there’s an issue that needs to be investigated,” Walker explains. “We’re working with our Maximo group on being able to feed more data on the assets.”
Walker’s biggest lesson learned so far is that the use of analytics packages that read directly from the server is a better option than pulling data directly from the controllers, which does not scale. There are also issues with legacy control systems. “With Tempered Networks, we’re putting a shell around all of our legacy systems by locking them out and using microsegmentation to say only this device can talk to this server,” says Walker. “It’s really solved a lot of problems.”
Segmentation and isolation has become a best practice, but it is fragile using traditional technologies. “You can set it up once, but as time goes on, it becomes impossible to maintain, so it’s important to keep it simple,” observes Erik Giesa, vice president of products at Tempered Networks. Instead of using a traditional enterprise IT solution to force-fit connections, Tempered Networks technology was borne in an ICS and OT data environment and bridges legacy systems in a simplified manner, Giesa says.
Prescriptive services for Refining NZ
Industry has come to expect maintenance service providers to employ state-of-the-art technologies and practices. The outcome-based maintenance service for industrial control systems from Honeywell Process Solutions is relied upon by companies such as Refining NZ, New Zealand’s only oil refinery.
Peter Smit, head of process control at Refining NZ, says: “The Honeywell Assurance 360 program we have in place provides us with the confidence that we have our Honeywell distributed control systems and Honeywell Advanced Solution applications at an agreed level of availability. We are very clear what outcomes we expect, and this allows Honeywell to leverage their knowledge and resources to meet the agreed outcomes in a structured and planned way.”
Steve Linton, director of programs and contracts at Honeywell Process Solutions, explains the underlying goal. “We are trying to facilitate achievement of our customers’ business drivers and provide the outcomes they expect,” he says, “whether it’s control system performance, control system availability, or reduced incidences on the control system.”
Tools such as planned, preventive, predictive, prognostic, and prescriptive analytics and maintenance aid in driving toward those outcomes. Prescriptive approaches are being beta-tested at some customer sites.
With RxM, Honeywell’s goal is to amalgamate data across multiple control systems to provide insights that say, “There is X probability in X time frame that X is going to happen, so go look at these things to prevent an undesirable outcome.” To do this, information from multiple customer systems is put into a data lake in the Honeywell Sentience IoT platform, which is appropriately controlled, cordoned off, and anonymized. Self-learning algorithms use and analyze the data and provide information that the customer can use to better maintain its control systems.
Prescriptive reliability analytics for MOL
Corrosion, fouling, opportunity crudes, and resulting process fluctuations are the most common operative challenges faced daily at MOL, an integrated oil, gas, and petrochemicals company based in Hungary. It is a member of MOL Group, one of the largest companies in Central and Eastern Europe.
MOL Group’s 2030–Enter Tomorrow program and recent strategic initiatives require a dynamic enterprise-operations-focused data and information infrastructure to improve productivity and increase process safety performance, says Gábor Bereznai, maintenance engineering manager at MOL. “Crude analysis, process simulations, continuous data monitoring, and early failure detection are the only possible answers to keeping our processes safe and under control,” Bereznai says.
MOL began its journey to refinery maintenance excellence with reliability-centered maintenance (RCM) almost two decades ago. At that time, a race to acquire software led to implementation islands and a lack of deliberate business process re-engineering.
In the next era, the focus was on software integration and connecting the systems with the corporate SAP ERP solution. MOL’s daily operations have come to rely on the company’s successful integration of asset management software, including Emerson AMS with SAP EAM and OSIsoft’s PI System with SAP PM.
The PI System provides the real-time operational data infrastructure and configurable, streaming analytical platform for MOL’s refining division. Predictive and condition-based maintenance, data aggregation, and health scoring is done in the PI Asset Framework (PI AF) and sent to SAP PM, which generates the work orders.
MOL is using a “layers of analytics approach,” with human analytics and real-time/streaming analytics providing a foundation for higher-level, operationally focused ML/AI, explains Craig Harclerode, global industry principal for O&G/Petrochem at OSIsoft. MOL built momentum and awareness of the power of analytics by asking the operations managers what problems needed to be solved and then quickly solving them.
“Once they had an analytical foundation, they moved to identifying areas where more-advanced prescriptive and predictive analytics would have value and began developing ML applications accordingly,” Harclerode says, noting that MOL currently has more than 25 ML-based applications in production.
This approach works because, as Bereznai explains, IT/OT transformation is a long journey that involves not only architectural and analytical method changes but also multilevel synergies among people and processes.
“This is a really long journey, especially in terms of mindset change and cultural development,” Bereznai says. “The technology and software side is much easier to change than the mindset, and the impact of this is underestimated.”
The efforts are paying off. MOL’s digital and downstream business transformation has delivered $1 billion in its first four years, and the goal for the next two-year period (2017-2018) is an additional $500M in EBITDA.
Prescriptive performance analytics for Tata Power
Software companies such as AVEVA are working quickly to answer the call for RxM. “We are building prescriptive maintenance and analytic capabilities into all of our asset performance management solutions to help our customers optimize the entire asset lifecycle and to ensure they have access to the most advanced technology available,” says Sean Gregerson, global director of asset performance management sales at AVEVA.
Tata Power, one of the largest integrated power companies in India, has rolled out AVEVA’s Predictive Asset Analytics software to 10 units at three plants to enhance the reliability of its critical-power plant equipment. The rollout is putting Tata Power in a position to quickly incorporate RxM capabilities.
The utility set its sights on remote, fleetwide continuous monitoring and diagnostics of critical asset health and performance in 2014 with the goal of improving efficiency, enabling proactive maintenance, and avoiding unplanned downtime. It built a new Advanced center for Diagnostics and Reliability Enhancement (ADoRE) powered by Predictive Asset Analytics.
The software learns an asset’s unique operating profile during all loading, ambient, and operational process conditions. When existing machinery sensor data is compared with real-time operating data, subtle deviations are revealed. Alerts and fault diagnostics are generated and plant personnel are dispatched quickly to take corrective action.
One recent catch yielded an estimated $270,000 (U.S.) in cost savings. Analytics revealed that the top thrust and guide bearing temperatures of some circulation water pumps were exceeding expected levels. During a brief planned outage, clogging in the bearing-cooling water line was identified and cleared, thus normalizing subsequent operation.
“Tata Power demonstrates the power of using analytics to move away from a reactive maintenance strategy,” AVEVA’s Gregerson says. “By catching problems early using APR and ML, the company was able to reduce maintenance costs, minimize unscheduled downtime, and prevent equipment failures.”
One recent catch yielded an estimated $270,000 in cost savings for Tata Power.
Prescriptive scheduling for Devon Energy
Prescriptive approaches can be simple to introduce incrementally. Devon Energy has thousands of batteries of tanks that collect water and oil during the course of operations, and how that liquid is scheduled for haul-off has recently become prescriptive. Real-time data engineer Don Morrison described the transition in a presentation at the ARC Industry Forum in Orlando in February.
Previously, scheduling liquid tank haul-offs for the Oklahoma City-based independent oil and gas company involved collecting data from multiple parties in an Excel spreadsheet and then using that file to create schedules. A centralized, more-accurate, on-demand process was needed to prescribe when, where, and how haul-offs would be needed.
Morrison explained: “We already had SCADA systems monitoring oil and water tank levels, so why not use them to detect when haul-off trucks are on site and how many; whether water or oil is removed from the tanks and how much – we only want full loads – and the fill rate?”
Two specific answers were sought: Could the engineers predict when the next load needed to occur so they could schedule the right number of trucks 3–4 days out? Could they gain enough data to “grade” their service providers?
Devon Energy chose Seeq analytics software to quickly detect haul-off events based on real-time OSIsoft PI data. With the push of a “get loads” button, all of the data from PI are pulled; forecasts up to three days out are generated; and the spreadsheet gets filled automatically. The results are reported in Microsoft Power BI, where they can be sliced and diced as needed.
Excel was retained in the first stage because “we didn’t want to change everything the users were doing and they were comfortable using it,” Morrison explained. Other future goals for Devon Energy include auditing and grading haul-off vendor performance and potentially incorporating opportunities such as RxM, smart contracts, and blockchain.
As more companies like these advance to prescriptive analytics and RxM, prescriptive maintenance has the potential to further heighten visibility and respect for the maintenance profession and its positive impact on the bottom line.