Optimising Wind Turbine Performance and Availability with Augmented Intelligence
Recently the term “Predictive Maintenance” has received much attention across many industries. Perhaps this is due in part to the synergies with other rapidly developing areas of technology such as Artificial Intelligence, Cloud Computing, Internet of Things and Big Data Analytics. It is certainly not all hype; in most areas of industry where operational costs represent a significant proportion of the overall lifecycle cost of a product, a strong business case can be made for the introduction of predictive maintenance.
Wind energy can certainly benefit from such innovations, and perhaps the potential reduction in the overall cost of energy would be the final push that is needed to bring the industry through grid parity and beyond. Although the use of condition monitoring in wind turbines is nothing new, recent technical innovations provide the framework that is needed for a more complete implementation of predictive maintenance.
Following early momentum with the introduction of drivetrain vibration monitoring systems in the wind industry more than a decade ago, further progress towards the development of more holistic condition monitoring solutions has been relatively slow. With the aid of the latest software technologies, a revolution is now on the horizon.
High Potential Limited by Challenge of Change Management
Essentially the operator of a fleet of wind turbines aims to maximise total energy production, and to do so at minimal cost. Analysis of operational data and service documentation from large numbers of turbines has demonstrated significant potential to increase the power performance of individual turbines, reduce overall downtime and reduce O&M costs through the selective introduction of a range of analytical and predictive methods (Figure 1).
However, this potential is only fully realised if the results of such analysis are properly integrated into operational processes. The main obstacles to achieving the transition to a methodology such as predictive maintenance appear to be of an organisational nature, rather than technical. The wind turbine service market is complex and dynamic, with turbine manufacturers, owners and independent service providers jostling for position. Responsibilities and contractual motivation appear to change faster than technology providers and change management teams can implement new solutions.
Figure 1: Optimisation potential through implementation of predictive maintenance
Traditional Approach to Process Digitalisation
Given these complex conditions, it is relevant to critically review the approach traditionally taken in the introduction of new digital processes. As illustrated in Figure 2, typically a multi-layer model is applied. Data acquisition and data management solutions are first generated, and then the gathered data may be used for real time fleet monitoring and visualisation of the past and current status. More mature systems will then evolve to include advanced analytics, the results of which are translated either manually or partially automatically into advisory information.
This generally consists of recommended actions to be coordinated by asset managers, passed to field service technicians and finally translated in corrective or preventative field actions. Although this approach is valid in principle, full implementation is costly and time consuming, often requiring several years to reach maturity. Within this time frame, market conditions or available technologies often change to the extent that the project becomes at least partially obsolete prior to completion.
Figure 2: Traditional process for converting wind turbine operational data into value
Modular System Design
It should be recognised that such difficulties are likely to intensify at least in the near-term as the rate of development continues to accelerate, both in the software technology domain and the wind energy market. Therefore, in order to achieve the desired transition to more proactive and predictive strategies, a modified approach is required. Rather than attempting to achieve large-scale and complex solutions as described above, it is more effective to implement relatively simple, end-to-end solutions for specific, high priority topics.
Individual software applications, or “Apps” can be rapidly implemented within complex organisations and used to convert available data into immediate value. Over time, the system may be expanded by adding new Apps, therefore gradually evolving into a more comprehensive solution. In case a specific App becomes outdated, incremental improvements or upgrades may be performed with relatively little disruption to the overall system, resulting in a highly robust design.
Containerisation, Cloud Servers and Edge Computing
This modular design philosophy is enabled and supported by several complementary technologies that are also improving rapidly. Containerisation is overtaking virtualisation as the preferred solution for deployment. The huge flexibility and relatively low complexity mean that almost any application can be executed in a range of environments with minimal effort. Apps may be developed on one of the commercially available cloud platforms, benefiting from the various advantages such as integrated development, build, test and deploy pipelines as well as immediate resource scalability.
Once complete, the App may be packaged into a container and deployed at any preferred location, for consistency with requirements such as company data security policy or budgetary constraints. In case the volume of data to be processed for a specific task is particularly large compared to the available network bandwidth, it may be preferable to deploy the App on an edge processing device and run analytics directly at the data source (Figure 3).
Figure 3: Analytics solutions designed to operate in any environment, portability provided by containerisation
Apps with Augmented Intelligence
An effective predictive maintenance solution should be capable of predicting, detecting and diagnosing a wide range of failure modes and for all subsystems within the wind turbine. Results should be delivered in the form of advisory information which can be translated as directly as possible into specific service tasks. Given the aforementioned motivation to design a modular solution, an ideal approach is to develop individual applications that can be used to monitor specific systems and components in the wind turbine.
Components for which historical failure rates and costs have been highest may be addressed first. Based on a detailed understanding of the technical domain, the relevant failure mechanisms may be defined in detail. Apps may then be developed that require only very specific input data to produce precise results and recommendations. Figure 4 shows an example of this approach applied to the main bearing, with an assessment of relevant failure modes leading to the design of multiple Apps, each of which with a specific analytic capability.
Figure 4: Main bearing failure modes analysed; applications designed with built-in domain knowledge
An Agile Approach to Process Transformation
The complexity of introducing a digital process such as predictive maintenance is significantly reduced if the transformation is managed incrementally. The structure and philosophy of wind turbine operators and service organisations vary greatly, despite the commonality of the overall business activities performed. Therefore, there is certainly no “one size fits all” solution.
However, by developing a range of Apps which can be quickly integrated within existing infrastructure and which provide immediate added value, a more agile approach is possible. Through the utilisation of the latest software technologies to create a flexible, scalable architecture, traditional barriers can be removed, and the previously elusive predictive maintenance revolution becomes a reality.
Article published in Windtech International, October 2019 http://www.windtech-international.com
i4SEE TECH GmbH | www.i4see.com | [email protected]