sponsored

Five best practices for remote asset management with edge analytics

Here’s a practical guide to edge computing strategies, using a powerful use case.

gettyimages 618343826
Dell EMC

Across a wide range of industries, the emergence of the industrial Internet of Things (IIoT) has ushered in many new capabilities for increasing the performance, reliability and efficiency of remote assets while optimizing operational intelligence and predictive maintenance. Increasingly, these use cases leverage analytics at the network edge, where the data is generated.

Leveraging real-time sensor data with increasingly powerful on-site analytics and management, edge intelligence solutions are ideally suited for a broad range of remote industrial applications. Among many other use cases, these solutions can be used effectively in manufacturing, power and water, oil and gas, mining, transportation, smart grid, and smart building use cases.

In this blog, we explore an example use case that is, in so many words, blowing in the wind. This use case focuses on improving the accuracy and speed of wind energy forecasting. While the subject here is relevant to the operation of wind farms and the systems on them, the principles, processes and best practices for leveraging edge analytics span the range of industries.  

Let’s get on to our example use case. Power generated by wind turbines is highly dependent on weather conditions affecting wind speed and direction. Due to the unpredictable nature of wind energy, most grid operators have to supplement wind energy with power from other sources, such as coal, hydro or solar.

In many countries, federal laws mandate that wind power operators predict their output to ensure consistent power at all times across the electrical grid. Analyzing sensor data collected from wind turbines in real time — such as nacelle wind speed, wind deviation, nacelle position and blade pitch — provide the necessary data to train a predictive model that can forecast power output with a high degree of accuracy.

Most wind farms are located in remote areas where there may be network bandwidth and reliability issues. An edge intelligence solution can provide the advantage of being able to analyze data locally in real time without relying on continuous network availability. It can manage all applications autonomously at the edge of the network and communicate with a central location when network communication is available.

Follow these five best practice steps to utilize edge computing for remote asset management in your industry.

1. Establish business objectives

Businesses need to stay competitive by leveraging technology to improve profit margin and increase productivity, efficiency, product quality and customer satisfaction. This is also true for renewable energy companies. Key success factors include:

• Enabling effective management of production capabilities

• Minimizing the cost of ongoing maintenance

• Delivering salient benefits to both partners and customers

Due to the uncertain nature of weather and wind conditions, wind energy companies face both long-term and short-term challenges of maintaining reliable operations at low cost. For example, most commercial wind energy operations need to commit to feeding a certain amount of energy into the power grid. If actual production exceeds the commitment in a given period, the operator may risk overloading the system. Delivering less than the committed amount may require expensive substitutions from conventional sources, such as coal.

Like all IoT projects, well-defined business objectives are essential for project success. For example, a 50-MWh wind farm can save up to a million dollars per year if it can increase the accuracy of 24-hour energy forecasts by 10 percent.

2. Identify data sources

Choosing the right data sources is also essential for successful pilot IoT projects. The process should be agile, flexible and iterative. With wind energy forecasting, for example, the first source of data should always come directly from the control and data acquisition systems at each of the turbines in the wind farm. Typically this data will include wind speed and direction, blade pitch, energy generation, rotor RPM and more. Next are statistics and reference data, usually coming from centralized databases, such as information on historical power curves, location and turbine height. Each turbine is a separate variable data source for the solution. In addition, weather forecasts from advanced models based on atmosphere, terrain and historical weather conditions play a key role in the solution.

3. Leverage both edge and big data analytics

In order to produce the most accurate forecasts, both edge and big data analytics should be leveraged together. Data enrichment, complex event processing and real-time anomaly detection can be used for just-in-time forecasting at the edge. Big data analytics on private or public clouds are optimal for building and enhancing highly accurate machine learning models to predict wind conditions at precise locations and heights using hundreds of terabytes of data from historical and environmental simulations.

4. Train a model to improve forecast accuracy

The next step should be to build and train a forecasting model by developing and running a pilot operation, which consists of the major building blocks for the use case. During the pilot phase, operators, planners, data engineers and data scientists should work together on milestones to meet this objective. The following is an example showing key data sources and processing components for a wind energy forecasting system that should be established, constantly evaluated and enhanced with additional data and advanced simulations:

• Real-time turbine, weather and simulated data: In addition to data from controller systems, evaluate additional real-time data from sensors, such as precipitation, temperature, pressure and air density.

• Forecasting rules with known conditions: Evaluate conditions that affect conversion rates, such as change of wind direction.

• Forecasting predictive models: Use both public forecast systems and simulated data to build and train a fine-grain forecast model at the proximity of each individual turbine location. Use machine learning techniques to determine unknown conditions that affect conversion rates.

5. Deploy solutions and establish five procedures to meet or improve forecasts

Finally, the solution, which consists of edge analytics for each turbine, can be placed into production. At any given time, analytics at the edge computes the forecast of power generation and raises any alert conditions. The system will send the forecast and alerts to a centralized processing and reporting service in the cloud. Operators will be able to see a real-time dashboard and generate reports that can be used to communicate with utility companies that manage the power grid. Information from the edge can also help operators adjust settings to achieve forecasted output, adjusts forecasts if needed and plan for system maintenance and upgrades.

The overall solution enables the operators to detect conditions at least 90-minutes ahead of time when any individual turbines cannot meet 24 hour energy generation forecasts. With this information, operators can either attempt to optimize turbine settings in order to increase power generation or revise the 24-hour forecast.

The following are important steps to put the solution in production:

• Analyze more than six months of historical and real-time data collected from controller systems on each turbine, augmented with weather, atmosphere and terrain data.

• Train models based on more than 20 attributes to predict power generation over 15-minute intervals.

• Apply models at the edge to produce real-time scores on power generation. Use analytical expressions to compare performance against forecasts to generate alerts.

• Establish procedures for operators to either fine tune turbine settings to achieve forecasted output or to revise the forecast.

Key takeaways

While the use case explored here focused on the wind energy industry, the edge computing strategies and best practices we outlined are applicable in virtually any industry that wants to drive a new level of efficiency into the management of remote assets.

For a deeper dive into edge analytics, visit the Dell EMC IoT Analytics site.

Kevin Terwilliger is the IoT Solutions Director at Dell EMC.