Prominent industry players monitor the condition of their critical assets by scrutinizing data sourced from myriad sensors to pinpoint trends and anomalies. The resulting insights drive maintenance strategies grounded in simplistic rule-based algorithms.

However, the effectiveness of the algorithms hinges on the expertise and knowledge of personnel analysing the data and devising rules, thus, rendering the algorithms resource-intensive and occasionally unreliable. Furthermore, they fail to identify issues tied to asset disparities, operating environments, and customer utilization patterns.

Recent research has honed in on “stochastic dependence,” modelling interactive asset behaviour within complex systems, dispelling the notion of independent and isolated silos. Nonetheless, for PHM and CBM to thrive, two pivotal challenges must be tackled:

• Facilitating data and insight sharing among assets to foster system-wide visibility of deterioration and performance enhancements for optimal asset health.

• Empowering assets to autonomously and collaboratively make maintenance and operational decisions grounded in overall system performance, rather than individual asset performance, ushering in not only comprehensive fleet and individual asset health assessment but also resource-efficient maintenance allocation.

Embracing Resilience and a Human-Centric Approach: The Catalyst of COVID

The global COVID-19 pandemic ushered in a new era of challenges for European industrial firms, with a particular impact on small and medium-sized enterprises (SMEs) navigating a fiercely competitive manufacturing landscape, as economies worldwide reopen following a prolonged disruption. Concurrently, recent years have witnessed tumultuous shifts in the socio-political arena, alongside conspicuous signs of climate change and an ongoing energy dilemma. Bolstering competitiveness within EU industries rests on a critical imperative: industrial assets and systems must possess the capability to adapt to their intricate and costly operational demands through innovative designs and unwavering reliability throughout their lifecycles. Vigilant management of equipment and system health and the associated risks is paramount to safeguard a secure and thriving industrial sector.

Consequently, European industries should channel their efforts towards astute asset management, improving availability, maintainability, quality, and safety. Within this framework, Europe’s pursuit of global leadership hinges on establishing an internationally appealing, secure, and dynamic data-savvy economy, underpinned by a trustworthy artificial intelligence (AI) ecosystem.

Over the past decade, remarkable strides have been made in research and development, uniting novel data science methodologies in machine learning (ML) and AI to confront pivotal challenges in industrial automation and control. The technological evolution encapsulated by Industry 4.0 has advanced equipment diagnostics and prognostics from conventional physics-based models to data-driven ML techniques. Current strides in digitization, however, amplify the demand for human skill in dissecting extensive data sets. This calls for proficiency in data science, statistics, and programming, a paradigm shift that risks alienating the majority of “boots on the ground” – factory workers, maintenance engineers, and technicians – compelling them to either upskill or risk redundancy. Thus, humanizing digital technologies is a pressing imperative for this decade.

Beyond the human aspects, several fundamental obstacles are slowing the widespread industrial adoption of these emerging technologies. Firstly, AI algorithms rely on operational data, confining their applicability to scenarios with copious volumes of data. Secondly, prevailing strategies for mitigating data scarcity or imbalance hinge on aggregating data from vast asset fleets, and this approach falls short of delivering an optimal solution because the average behaviour of assets in a fleet fails to represent the intricacies of any individual asset. Thirdly, integration poses a formidable challenge within a system-of-systems framework, where an industrial ecosystem comprises diverse equipment types, often originating from various original equipment manufacturers (OEMs). These heterogeneous assets must flawlessly collaborate to attain overarching system objectives.

Achieving peak system performance necessitates pinpointing the ideal combination of actions across these assets in a dynamic and uncertain environment. Present-day AI techniques cannot really learn the intricate interrelationships between diverse system assets, impeding their utility in supporting decision-support systems reliant on equipment cohesion to achieve system-optimized outcomes. Collaborative AI emerges as a pivotal enabler, facilitating asset communication, data sharing among kindred assets, collective failure pattern learning, and behavioural optimization. Nonetheless, these techniques remain underdeveloped and unrefined for industrial equipment. For instance, clustering assets based on dynamic behavioural data similarity remains an elusive endeavour. Once akin assets are identified, seamless communication and operational status exchange, coupled with control action dissemination, become imperative. In this context, a fundamental challenge arises in machine-to-machine communications, exacerbated by a profusion of standards and protocols. This proliferation hampers communication between disparate equipment types from multiple OEMs—typical within intricate industrial systems.

These multifaceted challenges collectively relegate system-level optimization to a distant aspiration. Yet system-level optimization constitutes the very essence of efficient and effective 21st-century enterprises. How can data, information, and insights be seamlessly shared among asset fleets and human stakeholders, culminating in system-level optimization?
In the realm of consumer technology, data sharing through Internet of Things (IoT)-enabled devices via social networks is on the rise, offering avenues for benchmarking and performance optimization. Notably, companies like Garmin and Nike have pioneered platforms enabling consumers to share and compare data on their exercise routines. These data, collected through GPS and IoT-enabled wristbands, can provide an inherent health boost, as the exchange of health data, habits, and insights among peers promotes collective well-being. This social application of data holds immense promise in the realm of asset health management.

Presently, the bulk of research and development attempts to leverage IoT and social media to target end-consumers. These endeavours range from smart home appliances to data mining to harness consumer data from social networks and drive more refined and targeted marketing strategies. Our overarching objective, however, is to channel the potential of these groundbreaking technologies into the domain of manufacturing and industrial systems.

Elevating Digital Twins into Social Entities

The advent of the Digital Twin (DT) concept has ushered in the era of digitally replicating physical assets. It endows assets with true intelligence by incorporating software agents, paving the way for machines to communicate, collaborate, and cooperate indirectly, through their digital doppelgängers within what is known as the metaverse. This innovation holds the promise of surmounting the challenge posed by incompatible data standards and protocols, while bestowing assets with collaborative learning and decision-making capabilities.

Although a gamut of DT models with varying functionalities has emerged in the era of Industry 4.0, integrating these DTs for deployment in complex systems and fleets remains a formidable task. The detailed exploration of architectures that amalgamate collections of DTs is largely uncharted territory. The focus has been on individual assets, necessitating more concerted efforts to materialize an interconnected federation of DTs. In a complex and dynamic system, such as infrastructure networks or expansive industrial plants, we envision a hierarchical structure for DTs. DTs representing virtual collections (e.g., subsystems or sub-fleets) of assets will reside in the upper levels of the hierarchy. These collections may dynamically form based on the “friendships” cultivated among social assets. Within this hierarchy, a “supervisory” DT that encapsulates the entire collection becomes indispensable. The demand for cognitive DTs equipped to design and deploy other DTs dynamically and autonomously, coupled with seamless interaction with humans in close cooperation – integrating avatars as part of the process – is a commendable aspiration within this context.

However, the technology for facilitating communication between DTs is still in its nascent stages; standardization is wanting, and industry-wide best practices are notably absent. Multiple disparate working groups have already developed a medley of standards describing heterogeneous assets at various levels. These standards offer generalized blueprints for DTs and have yet to gain substantial traction within the industry. Indeed, existing standards often suffer from overgeneralization or fall short of accommodating the swift evolution of DTs. Consequently, data shared across contemporary cyber-physical systems seldom adhere to a format readily comprehensible by human stakeholders beyond data specialists. Addressing this imperative entails devising solutions for sharing learning data and information. Messages exchanged between DTs and the social platform must incorporate ample context to ensure transparency, safety, and robustness, while enabling human agents to seamlessly participate in message processing. Augmenting transparency necessitates crafting a schema vocabulary for a standardizable information model, extending its capabilities to match the communication requisites between DTs and between DTs and human stakeholders. Safeguarding safety and robustness hinges on the implementation of explicit indicators for device health and data integrity.

Cultivating a Collaborative Digital Ecosystem: The Vision of Social Networks for Industrial Assets

Picture a world where individual machines in factories across the globe and infrastructure assets within vast networks compile, upload, and disseminate condition and operational performance data, alongside human-comprehensible “status updates,” via a purpose-built social network platform. The potential for learning and optimization within this landscape is immense. Assets spanning a system or network can engage in collaborative learning, pattern recognition, and problem diagnosis, harnessing collective wisdom to adapt their behaviour, alleviate the burden on ailing equipment, and bolster long-term system performance. Operators, maintenance engineers, and managers gain the ability to peruse these status updates, pinpointing opportunities for efficiency enhancements and orchestrating measures to optimize their system’s performance.

The industrial systems of the future will comprise genuinely intelligent collaborating assets, seamlessly leveraging AI in conjunction with human expertise, fostering heightened efficiency and judicious resource allocation. The underpinning models driving these systems will be fortified by explainable AI (XAI), instilling trust in autonomous behaviour among human managers.
Realizing a vision of collaborative industrial assets through a social network within the realm of AI is a tough challenge. This interconnected and intelligent “social network” of assets draws inspiration from the social networks observed in biological entities. In its pursuit, the construction of a multi-agent ecosystem for a collaborative asset social network via a dedicated social network platform must duly acknowledge the indispensable presence of human stakeholders. This necessitates a robust foundation encompassing techniques for efficacious social network formation, algorithms empowering agents to cultivate contextual awareness, federated algorithms facilitating collaborative learning, and multi-agent reinforcement learning strategies for cooperative decision-making. The ultimate objective is to cultivate a network of diverse assets adept at collectively optimizing operations while mitigating climate impact and data vulnerabilities. The framework prominently involves humans, serving as essential contributors who both instruct and learn from software agents through cutting-edge data mining algorithms adept at deciphering intricate data and distilling actionable insights.

Professors Diego Galar, Ramin Karim and Uday Kumar from the Luleå University of Technology, Sweden

Unlike motors, pumps are rated by head and flow, not by power. There’s no such thing as a 50 hp pump or a 100 kW pump. A pump can operate over a range of heads and flows, and the power required is determined by those and by the pump’s efficiency at the particular head-flow operating point. It’s helpful to know that “head” correlates to a measure of pressure. For water, it’s a simple conversion: 2.31 ft head = 1 psi (1 m head = 9.8 kPa). Here’s a simple formula that describes the relationship between head, flow, pump efficiency and pump power: (where k depends on chosen units)

While this formula is helpful for quickly estimating the power required for a rotodynamic pumping application with known head and flow values, you can only get accurate power values from the manufacturer’s pump curve. How to read pump curves is beyond the scope of this article. What is important here is that the power requirements vary with flow rate, so knowing the range of flow rates for the pump is essential to sizing a motor to the pump.

Sizing the pump

The process of sizing a pump and motor starts with sizing the pump for the application’s range of head and flow requirement. The following basic concepts are evident on the pump curve.

Flow requirement. A pump may operate across a wide range of flow rates, known as the Allowable Operating Range. Ideally, the pump should be designed to operate as close as possible to the Best Efficiency Point (BEP) and within the Preferred Operating Range. Pump efficiency will drop dramatically as flow rates move away from the BEP, and turbulent flow will reduce the reliability of the pump.

Head requirement. The head that a pump can deliver must match the application. If the maximum pump head is below the system demand, the pump will not produce flow (bad!). If the maximum pump head is much greater than the system demand (more than double), the operating point will not be near the BEP, and both efficiency and pump reliability will suffer.
Cavitation. Another important concern when selecting a pump for a specific application is the possibility that cavitation may occur. If the pump is to operate across a range of flow rates (rather than always operating near a single flow rate), cavitation will be more likely at the higher flow rates. Pumps have Net Positive Suction Head Required (NPSHR) ratings, which allow evaluation of the likelihood of cavitation at any flow rate using NPSHR values from the pump curve. Generally, lower-speed pumps are less susceptible to cavitation than higher-speed pumps. If the application has low suction head demands, a lower operating speed will be an advantage. At lower operating speeds, a larger pump impeller diameter will be required, and thus a physically larger and more expensive pump may be needed.

Sizing the motor

Once a pump of the proper size is selected for the application’s range of head and flow, the motor can be sized and selected to match the pump’s requirements.

Minimum power requirement. For most pumps, the power requirement varies with flow rates. Power requirements may increase or decrease with increased flow. The pump curve will provide that information. Obviously, the motor must have adequate power to meet the pump demand at the application flow rate with the highest power requirement. That’s the minimum power requirement for the motor. But it is likely the pump will have an Allowable Operating Range wider than the application demands.
Maximum power requirement. If application demands were to change at some future time, the pump might be expected to operate at a point where the power requirements are greater than the minimum power requirement. Therefore, it’s wise to consider the maximum power the pump could require under any operating conditions. This value is provided on the pump curve as the No Overload Power (NOL) rating. In some cases, the difference between the minimum power requirement for the application and the NOL rating may be absorbed by the motor service factor. In other instances, sizing for NOL power may require a higher power motor.

Conclusion

Sizing a pump and a pump motor for an application is not a trivial endeavor. The application head and flow requirements must be known. The pump power formula provided above, with the “k” to match the selected units, will provide a good estimate of the size of the machine. Pump vendors have pump selection charts which are generalized versions of the pump curve that will help with pump selections. Those charts and related reference data will provide NOL power ratings. The person responsible for selecting a pump and motor should have the appropriate pump curves and motor data and know how to read them.

Eugene Vogel, Pump and vibration specialist at at the Electrical Apparatus Service Association (EASA, Inc.)