Managing solar ­asset performance with connected analytics

Subscribe to ABB Review

ABB recently executed several research initiatives capitalizing on digital twins for asset performance management. Approaches for assets with short histories and long histories were developed. These analytics and tools are now being productized and deployed in managing asset health and performance.

Karen Smiley, Xiao Qu, Travis Galoppo, K. Eric Harper, Alok Kucheria, Mithun P. Acharya, Frank Tarzanin ABB Corporate Research Raleigh, NC, United States,,,,,,,

This article highlights selected elements of the work ABB has recently conducted on several related research initiatives that capitalize on digital twins for asset performance management. Discussed first is an “agile analytics” approach that jump-starts the creation of useful health and performance analytics for new devices or materials that do not yet have long monitoring, operational or environmental data histories. The second topic of discussion is an ongoing initiative to leverage a long history of intracloud data collection for solar inverters and plants to create new analytics for event and weather correlation as well as for benchmarking, forecasting and self-service business intelligence (BI). These analytics and tools are now being productized and deployed for internal and customer use in managing asset health and performance.

Industrial asset performance management (APM)
Efficiently keeping critical long-lived assets healthy and performant is essential to success. Analytics play an increasingly important role in understanding and optimizing asset health and performance. Active APM enables customers to increase operational awareness of the health and performance of enterprise assets. Heightened health awareness empowers customers to move from costly reactive maintenance towards risk-based management techniques that optimize performance and maximize return on net assets (RONA). While there are many ways to quantify the health of an asset, most algorithms reflect its risk of failure (RoF) and its remaining useful life (RUL). However, typically, these health key performance indicators (KPIs) alone do not reflect performance degradation, which can impair production and RONA long before actual failure or end of life. Accordingly, comprehensively managing asset performance also requires analytics that quantify productivity and degradation.

Digital twins for APM
Connected assets underpin a coordinated approach that marries operational technology (OT) and information technology (IT). Digital avatars or digital twins that blend OT and IT data can be thought of as reflections of real-world objects for combining models of devices. This does not mean they need to be exact clones; they merely need to reflect the most important asset behaviors to be explored. These digital avatars or twins can empower faster creation of analytic models that integrate asset metadata and operational data (eg, telemetry and events) with data on the asset’s environment (such as weather, irradiance, lightning, operational conditions), for characterizing and managing asset performance at the device and plant levels. The avatars or twins provide value in several scenarios, including planning, deployment, operation and maintenance. One major benefit is safe and nonintrusive exploration of “what-if” scenarios – eg, to simulate how possible actions to tune health or production would affect the devices and system.

Asset performance analytics for solar
Penetration of renewable energy is increasing around the world, as is demanded for a reliable and sustainable energy supply. For over 13 years, ABB has designed and manufactured a broad range of solar photovoltaic (PV) inverters spanning the needs of residential, commercial, utility-scale and microgrid applications [1]. As of 2017, ABB’s installed base of solar inverters worldwide exceeded 26 GW. To support connected solar analytics, these inverters are becoming more intelligent and more easily integrated with complex and smart environments →1.

01 Digital-twin-based analytics can detect gradual or storm-related degradation in solar plants and prevent lost power production long before it would otherwise be noticed and addressed. Such asset health and performance management analytics and tools fully exploit the availability of large quantities of data.
01 Digital-twin-based analytics can detect gradual or storm-related degradation in solar plants and prevent lost power production long before it would otherwise be noticed and addressed. Such asset health and performance management analytics and tools fully exploit the availability of large quantities of data.

The ABB Ability™ Aurora Vision Plant Management Platform [2] complements ABB’s inverter portfolio with a cloud-based solution for next-generation management of solar PV plants →2.

02 ABB Ability Aurora Vision Plant Viewer for Mobile.
02 ABB Ability Aurora Vision Plant Viewer for Mobile.

Aurora Vision automatically collects monitoring data from solar inverters and other devices, and provides highly interactive, real-time access to key performance and operations metrics to optimize inverter performance and inform business decisions. The monitoring data collected over the past 13 years spans more than 250,000 devices in solar plants worldwide →4.

04 Some solar plants include on-site environmental stations that record weather and solar irradiance.
04 Some solar plants include on-site environmental stations that record weather and solar irradiance.

For solar, the PV panels, inverters, meters, environmental units, energy storage, plant and power grid can all be modeled with digital twins. These facades provide access to measurements, estimates and analytics results based on an understanding of the physics and engineering of the underlying systems, and operating experience.

Understanding asset condition and performance
Predicting asset behavior, production, events, RoF, RUL, etc. are all becoming increasingly important in understanding the true condition of an asset and how best to maximize its value. The three key, interrelated, concepts are degradation, RoF and RUL →3. Degradation in the physical or logical condition of an asset decreases RUL and increases RoF but can be mitigated by maintenance actions to enhance production and performance prior to failure or end of life.

03 Relationships among degradation, RoF, RUL and maintenance.
03 Relationships among degradation, RoF, RUL and maintenance.

Using a digital twin, one can analyze asset data to characterize asset production, condition and degradation; estimate RUL; quantify RoF; and assess the potential impact of maintenance actions. Analytics can be applied predictively to drive proactive preventive maintenance decisions or to gain a better understanding of failures that have already occurred.

Classifying analytics algorithms
To aid in selecting analysis approaches depending on the available data, algorithms relevant to each concept have been classified. For example, RUL algorithms were grouped under the headings of directly observed, indirectly observed, and state processes and machine-learning, while RoF algorithms were grouped as failure tracking, symptom monitoring and detected error reporting. Many other types of analytic algorithms can also be useful: For example, anomaly detection is a key aspect in augmenting failure prediction models and forecasting can be highly valuable for proactive asset load management. These algorithm types have also been classified. Some of the detailed classifications are shown in →5,6.

05 Classification of RoF algorithms.
05 Classification of RoF algorithms.

Classifying and mapping available data
Data types, volumes and varieties are critical to applying these algorithms effectively. For instance, if telemetry or event data are not available with sufficient time resolution, certain RoF algorithms based on early detection of anomalies are not applicable. Assessment of “data readiness” for an analytics application can provide valuable insight for choosing which algorithms to apply. →6,7 show an example from a tool ABB developed to characterize data availability with respect to data requirements for a class of analytic algorithms. Understanding these relationships can guide efficient development of new asset analytics in an agile, iterative way that maximizes the benefit of the available data.

06 Classification of RUL algorithms.
06 Classification of RUL algorithms.

The analytics collaboration also cataloged various data imputation strategies for handling missing values and borrowed from previous forecasting experience using neural networks. These methods help offset some concerns relating to limited data completeness or quality when developing new algorithms for analyzing and managing asset health and performance.

07 Example spider diagram illustrating data readiness.
07 Example spider diagram illustrating data readiness.

Intracloud analytics architecture
Identifying and blending many additional types of data proved highly beneficial for creating useful digital twins for solar inverters and plants. The Aurora Vision solar monitoring platform already provided a rich set of metadata about solar plants and the equipment used in them, as well as streaming telemetry and event data sent to the cloud [3]. Some plants also include on-site environmental stations that record weather and solar irradiance →4. This data was supplemented by weather, lightning and irradiance measurements gathered by off-site sources [4 – 6]. Using these off-plant sources extended the plant coverage of weather data: One off-site weather source provided useful data for 50 percent of the plants where on-site weather stations were absent.

Complementary textual, numeric and complex data was also acquired from other systems available inside and outside ABB [7]. The inclusion of free-form text enabled the use of natural language processing (NLP) techniques to build a better understanding of sequences of events, whether those events were reported by customers or logged automatically by the Aurora Vision system.

Combining this data with the telemetry and event data enabled new analytics to benchmark and predict inverter performance, and led to a better understanding of the impact of environmental conditions on solar inverter reliability, availability and yield. Similarly, the augmented digital twin for the solar plant helped to drive new analytics that deliver fresh insights into plant performance. As an example, mixing failure modes from field engineering reports with telemetry data like device alarms, temperatures and environmental conditions allows ABB to correlate categorized component failures with the signals seen in the field before devices fail. This intelligence provides early warnings or predicts when a failure rate threshold may be exceeded.

Weather analytics
The blending of this diverse data amplified the variety and power of the performance analytics. For instance, weather data from off-plant sources enhanced detection of anomalous readings from in-plant environmental stations. →8 shows a comparison of total daily irradiance in 2016 between two data sources, an in-plant environmental unit and an off-plant sensor. This anomaly detection analysis improved the accuracy of cloud, edge, or hybrid analytics relying on weather readings as inputs, such as estimating or/and predicting power generation for inverters. In turn, this enhanced the ability to detect gradual or sudden degradation of panel or inverter performance that could impair solar plant production.

08 Comparison of total daily irradiance measures between in-plant and off-plant weather stations for anomaly detection.
08 Comparison of total daily irradiance measures between in-plant and off-plant weather stations for anomaly detection.

These findings can provide valuable guidance to solar plant operators or service personnel in assessing RoF and RUL and taking appropriate actions. Short-term actions might include preparing for storm impacts. In the longer term, such findings can justify preventive actions such as installing additional lightning protection in the plant or creating more accurate capital and maintenance plans.

Visual self-service analytics
In visual analytics, self-service BI tools are employed to illustrate various data – including metadata, telemetry data and event data – using figures, maps and other charts. The BI tools can automatically fuse multiple data sources and types via common fields. Information is easily tailored for different user requirements and interests through interactive filters and selections, and all selections made to one chart are applied to other artifacts automatically, in real time. These features help owners and operators of solar plants to visually identify anomalies and obtain other insights efficiently and effectively.

Concrete benefits from the solar APM work
Achievements from the solar collaboration included novel algorithms for benchmarking and forecasting solar inverter performance and reliability; prototypes for visual self-service BI applications; edge algorithms for real-time estimates of AC output and DC input power; automated diagnostic tools for service engineers for analyzing events, telemetry and free-text customer correspondence; and new KPIs that can help ABB’s customers to better understand (and gain more value from) their solar plants. The project also demonstrated the synergistic value of an intracloud analytics architecture aligned with ABB Ability that leverages complementary internal and external data sources. Environmental data was also shown to augment the business value of the analytics.
 These capabilities are now being integrated into ABB’s portfolio of solar monitoring and asset performance solutions.

As an example of the potential business value of these analytics, consider an industrial solar plant with 10 TRIO 60 kW solar inverters. Digital-twin-based analytics can detect gradual or storm-related degradation varying across the plant (which may be addressable by simple cleaning), preventing lost power production over the many days or weeks before it would otherwise be noticed and addressed. Unexpected failure of one inverter in a plant with 10 inverters means losing about 10 percent of the plant’s solar power production, which could subject the plant operator to regulatory penalties for failing to meet renewable power production commitments. Early prediction of failures with analytics provides valuable lead time for performing maintenance and, if necessary, for acquiring and installing a replacement device or the right parts. Actual monetary benefits are situational and can be calculated based on days of lead time, whether hot spares are available in the plant, avoidance of multiple service trips by bringing the most-likely failed components, and factors such as regulatory penalties and the value of electricity in the region.

When vast quantities and diversities of data are available, much can be achieved. However, data is not free: Adding instrumentation and collecting, storing and transmitting data have capital and operational costs. When much less historical data is available, creation of digital twins for new types of industrial assets in an APM system can be accelerated by leveraging the algorithm and data categories and associated agile analytics tools. These tools also support the cost/benefit assessment of proposed analytical strategies for newer asset types.

In all, the use of digital avatars or digital twins for solar APM brings a wealth of analytics and tools that provide multiple benefits for operators of solar PV plants.

The authors gratefully acknowledge the valuable contributions of research team members James Ottewill, Marcin Firla, David Cox, Karl Severin, Imtiaz Ahmed, Hang Xu, Melwin Jose and Rohini Kapoor. We also note with appreciation the strong support of Filippo Vernia, Emanuele Figliolia, Antonio Rossi, Ronnie Pettersson, Siri Varadan, CJ Parisi and many others. Additionally, the authors warmly thank the ABB Ability Velocity Suite team, Earth Networks and The Weather Company for generously sharing supplemental weather data used in this research.

[1] ABB solar inverters. Available:
[2] ABB Ability™ Aurora Vision Plant Management Platform. Available:
[3] M. P. Acharya et al., “Real-time AI powered by edge-deployed ­digital twins”, pp 14–19.
[4] Earth Networks. Available:
[5] The Weather Company. Available:
[6] ABB Ability™ Velocity Suite. Available:
[7] Salesforce. Available:


Contact us


Share this article

Facebook LinkedIn Twitter WhatsApp