Optimize Maintenance With NOAA Data

The purpose of this blog is to exhibit 60Hertz’s groundbreaking research, funded by NOAA’s 2022 SBIR Phase I grant program. This exploration focuses on developing a Clean or Wait predictive model based on novel datasets including NOAA Aerosol Optical Depth data, SCADA and expected performance.

The 60Hertz team successfully developed a powerful analytic algorithm that stands to revolutionize clean or wait decisions. With further plans to extend the scope beyond this research in 2023 and commercialize predictive models.

60Hertz is transforming our ability to access and make actionable information that impacts maintenance strategy and your bottom line.

In this age of data abundance, actionable accessible data from SCADA to NOAA can be made possible with our differentiated CMMS, that layers logic of complex information so that renewable energy asset owners can take swift action before a costly failure occurs.


Project Objectives: Gain a Deeper Understanding of Environmental Conditions That Impact Performance of PV Systems

Weather plays a significant role in solar panel performance and can cause a decrease in efficiency if not managed properly. Renewable energy developers strive to increase generation efficiency by 1-2% across their portfolio. Better maintenance with intelligent foresight is critical to long-term sustainability, operability, and CO2 reduction.

Poor weather conditions such as dust, snow, hail and wind can all take their toll on the panels leading to premature degradation. To achieve optimum productivity, companies must ensure regular cleaning and maintenance operations; this often comes with a considerable cost – sometimes upwards of $15,000 – $250,000 per event depending on the size of your generation.

These costs can quickly add up depending on the frequency and mileage of service required, the size of the solar farm and the location. Ignoring data analysis can be costly – like cleaning panels just before the neighboring farmer sprays fertilizers or washing just before precipitation.

Cleaning costs vary depending on design, remoteness and size as well as low maturity levels.

“…The service pricing is defined by companies with poor and low maturity levels. For example [a] site [in remote California] was one of the poorest in terms of being “Solar Panel Cleaning friendly”… this job cost 400% more than a standard carport. When you make a design mistake on a small installation it is a small problem. When you make a design mistake on large installations, then the problem is big. This site had 14,000 solar panels in a very remote location in central California.”
– Carla Dawson, Soliar Technologies CEO.


How Is Cleaning Deployment Currently Determined?

  • Human/visual inspection
  • Design cleaning schedule
  • Solar guestimate tools (Excel)
  • Inverter calculates and informs cleaning based on performance
  • Or they just don’t clean at all

The last bullet was the most surprising. During our research, it was painfully obvious the industry has an Achilles heel, but market demands are forcing change. Asset owners are realizing their systems are not being optimized, and money is being left on the table.

“The personality of the Director of O&M and his/her problems at a specific time (aka time & circumstances) define the threshold. In general terms, 2% is the minimum value.”
– Carla Dawson, Soilar Technologies CEO

According to the NREL, performance loss due to soiling is estimated to be between 2-25% annually. Making use of reliable and wider-scoped environmental data will minimize or avoid losses and/or critical failures. While the industry chases 1% to 2% gains in efficiencies, there are up to 25% gains at our fingertips.

The NOAA SBIR Phase I program is the first step in achieving these gains. In partnership with Novel Sciences, Sunshare and ScadaSolar, 60Hertz laid the foundation necessary to build a weather-informed maintenance module, which will support operations in a couple of ways.

  • Near-term response to ensure maintenance interventions are timely and optimized
  • Long-term support planning decisions that will optimize design and budget decisions
“If the cost of having this data is less than the money lost due to poor management, then a cleaning algorithm would be successful.”– Carl Dawson Soilar Technologies CEO

“Carla has presented a standard we intend on meeting as we move toward commercialization during 2023”

– Tonya James, COO 60Hertz.




Project Findings: Unveiling The Impact Of Environmental Conditions On the “Clean or Wait Decision”

60Hertz’s analysis of the effects of particulate matter on photovoltaic (PV) arrays reveals that extended exposure to soiling caused by adverse weather events such as:

  • Fire – ash
  • Heavy precipitation
  • Atmospheric dust
  • Bird droppings
  • Vehicle or agricultural emissions
  • Volcanic ash
  • Algae or fungi growth
  • Pollen

These weather events pose a serious risk to solar-generated power. Since PV cells require direct sunlight to convert light into energy efficiently, the accumulation of caustic and benign atmospheric particulates can further impede the efficiency of electricity generation from solar energy by blocking sunlight from reaching the panels.

Increased exposure to weather-inflicted soiling accelerates solar equipment degradation. Which significantly impacts reliability and usable asset lifespan, as well as incurs high costs for mobilization of maintenance interventions.

With solar service companies periodically cleaning acres of modules, every speck can have a significant effect on the business’s bottom line. However, making the decision to clean is more complex than you would think. Today, directors make their decision economically, but they are missing data.


Technical Approach

60Hertz utilized NOAA Aerosol Optical Depth (AOD) measurements as an early warning system, to cue on-site validation of the soiling station and any available meteorological (MET) station data to validate if a work order to clean deposited particulate matter is necessary.

This layer strengthens the decisions and moves from reactive maintenance – often weeks delayed, to proactive maintenance – getting ahead of the weather impact if possible.

Through this research, we developed a suite of algorithms converted into modules within our software tools that fuse multiple public and private (GOES-R ABI AOD and local site MET/soiling station) datasets with features like project economics, asset location, weather forecast, etc. to hasten solar service company’s decision times concerning cleaning and/or protecting renewable generation assets.

Our approach fused the data and explored feature relationships (correlative, causal) in the data using Machine Learning pattern identification and relationship learning algorithms along with Machine Reasoning visual pattern detection and tracking and then developed models to predict behaviors.


Technical Findings

Optimize Maintenance With NOAA Data


Site Sensor Behavior

Our quality check of site sensor behavior revealed a strong correlation between total output and both irradiance and albedo from the METs with values exceeding 0.8. However, our findings reported a definite drop in output during a fire, thus indicating the potential for soiling on the sensors that may inhibit their ability to capture accurate readings.

As we do not have access to a soiling meter, it is necessary to build an effective proxy to assess quality levels of the data effectively. Further investigation is required into appropriate quality control protocols and practices.


Behavior Pre-Fire

The behavior of cyclical AOD behavior is consistent behavior before fires. Data shows correlation between power output and AOD, ACM, CPS, Aerosols, Smoke, and Dust is low (0.11, -0.19, 0.14, 0.68, 0.65, 0.75), so these particles are not present to such an extent that it impacts production levels.

This data means that there may be other underlying factors at play with fire behavior and raises the question of how fire behavior can be better managed despite the presence or absence of these substances in its surroundings pre-fire.


Optimize Maintenance With NOAA Data


Behavior During Fire

Our research has shown a behavior reversal in the relationship between dust, smoke, and release during fire-prone conditions; the correlation decreases from 0.75 to 0.7 and even further to 0.65 to 0.54, likely leading to a production drop in these areas of up to 43%.

This behavior is further evidenced by an increase from a gray-scale reading of AOD (Absorbing Optical Depth) readings that jump from 0.11 – 0.34 due to cloud cover as particulate matter is being deposited, indicating a greater presence of both dust and smoke in the air at any given time.

Solar installers should monitor behavior and production capabilities carefully – it’s important to take precautionary steps before costly damage is done.


Disassociating Aerosols and Particulate Matter from Cloud Cover

Using a Random Forest Regressor, the team found a way to remove cloud cover from the proxy and achieve a good model of the complex relationship between atmospheric particles and soiling.


Promising Predictive Link Between AOD/Aerosol/Dust/Smoke and PV Production

The predictor built on the regression model allows us to predict generation nuance and lays the foundation for early warning and cleaning deployment modules within the 60Hertz platform.


Optimize Maintenance With NOAA Data




Our research and analysis of the impact of airborne particles on generation yielded extremely valuable insights. Our work enabled us to identify patterns in the data and draw meaningful correlations between AOD behaviors and their impact to generation.

ML pattern identification and relationship learning algorithms have been used to explore various features of the data related to the fires, such as patterns pointing towards established correlative, or even causal relationships.

Visual pattern detection and tracking can help to identify and quantify certain behaviors and performance signals seen with the fires. This tool can be important for recognizing any potential shifts, changes, or trends in observed behaviors for better risk management.

In order to aid in this analysis, ML algorithms continue to be refined and improved, providing more accurate insights that can alert us when unexpected events or signs occur.


Next Steps: Expand R&D Scope – Environmental Informed Predictive Models to Optimize Maintenance Strategies

The 60Hertz team appreciates the value of incorporating NOAA data into our decision support tools. Further work will be conducted to improve existing models used for planning and prototype early warning models for operators to assess the impact and optimize production through data-driven maintenance and operations.

By bringing attention to this problem, 60Hertz has committed to applying for NOAA’s SBIR PHASE II grant program to broaden the scope of this research to explore the impact of environmental and weather conditions beyond panels to inverters, battery container units and the like; all aimed to allow operators to make data-driven decisions that optimize performance.

The 60Hertz team will continue to drive this research to an MVP that supports improved production, reduced maintenance costs, and warranty life for renewable generation companies, ultimately improving resiliency in the grid and the market.

Interested in joining us in this research? Together, we can help build the future of maintenance. Give us a call: (844) 977-4499.


Supported by the National Oceanic and Atmospheric Administration data sets and grant funds from the Small Business Innovation Research program, 60Hertz is delivering groundbreaking research for market innovation in the realm of predictive maintenance based on weather data.


Subscribe to Our Newsletter