Virtualization, Cloud, Analytics, Edge Computing
Next-gen asset performance management software shows the big picture
APM 2.0 integrates information from other control systems to deliver a more comprehensive view and analysis of production and performance.
Manufacturing is at the brink of the next generation industrial automation revolution, driven by advances in robotics, artificial intelligence, and machine learning. To remain competitive, complex, capital-intensive industries need to deploy industrial automation more than ever, as global competition requires companies to increase efficiency through reduced operating costs, increased production, higher quality, and lower inventories. The highest priority is increasing asset utilization and eliminating production losses caused by unplanned downtime–a $20 billion a year process industry problem.
Current maintenance practices
For the past 50 years, maintenance practices have evolved in terms of equipment reliability and availability. Maintenance strategy has progressed through run-to-failure; calendar-based; usage-based; condition-based and reliability-centered maintenance (RCM). Outcomes are improved but the equipment continues to fail. Why?
Despite successive techniques becoming more sophisticated, they do not address the main issue: failures caused by the operation of equipment outside its stipulated design and safety limits. Current reliability practices that only focus on maintenance issues like wear and age have failed to detect seemingly “random” equipment failures, particularly those caused by excursions in operational conditions that cause the most damage to equipment.
With the emergence of the Industrial Internet of Things (IIoT), Big Data, machine learning and other analytics, there is a huge market opportunity to better address reliability and availability.
A new generation of analytical capabilities that can provide deeper insights is needed, however. Operators need accurate, predictive solutions to provide much earlier warnings of impending trouble. They also require prescriptive guidance to avoid or mitigate the forecasted outcomes. However, data science alone cannot solve the problem. Solution providers need to offer deep domain and process expertise, as well as experience and knowledge about design, operations, and maintenance systems.
Here comes APM 2.0
The next generation of APM–APM 2.0–gets its power from integrating information from other control systems, such as manufacturing execution systems (MES), to deliver a more comprehensive view and analysis of production processes and asset performance. By incorporating data from all systems, it can better assess patterns of normal and abnormal behavior, which helps to predict future conditions and their inherent causes. Those who were early to adopt this new mindset have made significant improvements in reliability and overall operational excellence initiatives. One leading pulp and paper manufacturer, for example, modernized its APM strategy, instituting real-time monitoring that correlated archived data in the plant historian with posted failure incidents in its asset management system. Shortly after the monitoring began, it uncovered a recurring overheating solution in a kiln, predicting an imminent breakdown with nine days’ warning. The company used the predictive analytics to change operating conditions and prevent costly shutdowns
These sorts of data-driven insights help operations and maintenance staff collaborate on decision making and promote shared goals. The amalgam of data supports shared and improved understanding of risk, which provides the ability to balance operational constraints and efficiency opportunities to improve return on assets (ROA).
The development of a new collaboration between maintenance and operations suggests that the availability and longevity of equipment is a shared responsibility, requiring a common comprehensive viewpoint. Both maintenance and operations need data-driven mechanisms to assess degradation and fine-tune activities, as well as provide early intervention to change process operations and avoid damage respectively. With superior predictive diagnostics–such as early detection of cavitation, dirty feed, liquid carryover, etc.–and prescriptive guidance, operators can respond faster, earlier, and in a more effective manner, adjusting process set-points to guide the process away from conditions with the potential to degrade and damage assets. Thus, process-induced maintenance is eliminated.
Key technologies providing breakthroughs such as these are centered upon fundamental analytics and data science strategy, especially machine learning. Proper application of machine learning to manufacturing assets requires domain-specific knowledge of the chemical processes, mechanical assets, maintenance practices, etc. It requires nuance and inside knowledge to interpret complex, problematic sensor and maintenance event data. Chemical company LyondellBasell has witnessed this firsthand, saying, “new Asset Analytics contains a unique set of modeling and data science-based technologies. Utilizing the additional process insight available from this promising new software solution brings with it the potential to operate closer to the true flooding limit on this tower. For a world scale olefins unit, this would be worth millions of dollars per year.”
Machine learning: a driving force
To increase availability, failure prevention must be based on data-driven truths powered by machine learning, a computer science sub-field that has evolved from the study of pattern recognition using algorithms that learn and predict from data without programming or rules. Applied within context, it can cast a “wider net” around machines to capture process-induced degradation that causes most failures.
Advanced machine learning software picks up behavioral patterns from streams of digital data produced by sensors, which reside on and around machines and processes, combined with other data events. Autonomous by nature with little need for human intervention, this advanced technology constantly learns and adapts to new signal patterns when operating conditions change. Failure signatures learned on one machine “inoculate” it, ensuring that the same condition does not happen. Learned signatures readily transfer to similar machines, preventing the same degradation conditions from affecting them.
With this insight, organizations can decipher patterns of looming degradation to secure sufficient warning. In doing so, failures are prevented and outcomes changed. The experience of one oil and gas producer and refiner demonstrates this well. For over a decade, they had experienced a compressor failure that went undetected by RCM methodologies and state-of-the-art vibration systems. Because modern APM makes it possible to separate the behavior of machine and process, this company was able to get to the root cause of the compressor failure.
Another North American energy company was losing up to a million dollars in repairs and lost revenue from repeated breakdowns of their electric submersible pumps. An advance machine learning software application learned the operational behavior of 18 pumps from archived historical values and maintenance events. During the learning period, on one pump it detected the explicit pattern leading to a casing leak that caused an environmental incident. By applying this failure signature to all 18 pumps, the application provided an early warning on another pump, which was about to suffer the same failure. Early action to pull and repair the pump avoided a repeat incident and major losses.
A new frontier of performance
Technology advancements like sensorization, the IIoT, machine learning and cloud technologies are not only making new manufacturing analytics methods possible, but also more scalable and affordable. Paired with process industry company’s dire need to find new ways to enhance existing operational excellence initiatives, it’s catapulted us into a new era of manufacturing: Industrie 4.0.
This new phase overcomes a key reliability challenge: the sheer number of asset types to cover. It uses cyber physical systems to monitor physical processes and make decentralized decisions. Asset and process analytics jointly are responsible for creating a multi-faceted view to enable fact-based decision making that considers a broader set of tradeoffs. While machine learning is a new technology, conventional methods to predict performance emerged four decades ago.
The difference between then and now lies in the degree of human involvement, as well as the accuracy level of prediction. Modeling techniques continue to be successful, where first principles, specific behavior needs to be understood. Real-time, dynamic models offer prediction of forecast behavior at (any) points in time, offering in-depth understanding of expected performance. A combination of models and machine learning is the best path forward. This can detect and avoid risky process operating conditions. It can explain explicit conditions anytime, calibrate and tune the model automatically via machine learning to achieve timely, accurate process status with simpler calibration.
However, only companies with a keen sense of urgency and commitment should embark on this journey. To successfully use analytics to improve safety and reliability, several elements need to be present, many of which are based on human (company) behavior and culture. Companies must focus on specific business problems before finding appropriate solutions (not technology) that align with business goals. But once a company identifies those challenges and evolves its strategic thinking about APM, it will strengthen existing operational excellence initiatives and gain a competitive advantage that will soon be in practice far beyond early adopters.
Michael Brooks is senior director for the asset performance management business consulting at AspenTech.