Intro to aerial thermal inspections

Why do we physically  inspect PV sites? 

Before diving into the details of aerial  inspection tools in PV systems, it is important to  consider why we perform physical inspections  of PV sites in the first place. With the rise of  advanced data analytics, improved sensor  networks, and seemingly endless Big Trends  (Big Data! Machine Learning! Blockchain?),  why are physical site inspections required, and  will they be required in the future? 

Historically, the majority of operational plans  for PV assets include physical site inspections,  and broadly speaking the motivation for these  inspections is driven by two main factors: 

1. Underlying uncertainty of system analytics
2. Detection of non-energy risk factors 

Academic literature has established that the  accuracy of PV system modelling is between  5%-10%.** This uncertainty can be decreased  by looking at peer to peer inter-comparisons  such as a ranking of normalized combiner  outputs. In this case, the uncertainty can be  decreased to around 3-4% which translates  to a probability of false negative (i.e. missing  a single string out) of approximately 60%  for a site with 28 strings per combiner. This  measurement is also susceptible to shifts in  global mean production, or in other words,  faults which occur in the majority of combiners  would not be detected.  

In addition to underlying uncertainty,  production data analytics alone also cannot  identify non-energy risk factors. For example,  groupings of hot-spots or sub-string failures on a site will not cause a near-term appreciable  energy loss, but can be signals of potentially  serious serial defects on a site. 

Because of these issues, physical site  inspections are a standard part of the  operating plans for PV assets. Traditionally,  these inspections have been performed by  an I-V curve trace of all or a subset of strings  on a site, and this data is used to answer two  questions: 

1. Is a string or module broken? 
2. Is the system degrading over time? 

There is a growing recognition, however, that  this tool is not an effective tool for achieving  either of these goals. I-V trace is generally too  expensive and labor intensive to be applied  to an entire site, and exposes technicians to high voltage DC components during the testing  procedure, posing an arc-flash hazard.

Because of these deficiencies there has been  a significant shift in the industry over the past  few years to move towards aerial inspections in  lieu of traditional manual physical inspections.. 

What are aerial Thermal inspections? 

PV aerial thermal inspection refers to the collection of  high resolution infrared and visible imagery  of a site from an aerial platform. The  fundamental concept being used is simple:  All modules are receiving the same amount  of energy for the sun, and those that are not  converting that energy into electricity will turn  it into heat. Therefore, energy losses will show  up as module heating. 

Aerial thermal inspections can be used to detect a wide  variety of site defects, from common fault  modes such as hot-spots, module breakage,  sub-module faults (i.e. diode engagement) and  string outages, to more subtle defects such as  Potential Induced Degradation (PID), MPPT  issues and junction box resistance.  

Typically, these inspections are performed  annually as part of site preventative  maintenance in lieu of traditional string  testing. Ideally, the scans are performed in  advance of regular preventative maintenance  work, especially for unmanned sites, as this  ensures that technicians can arrive with the  knowledge, tools, and equipment needed to  quickly remediate issues.  

Aerial inspection process

The aerial inspection process can be considered  in three stages: Acquisition, Analytics and  Remediation.  

Acquisition

Acquisition is performed from an aerial  platform, either UAS (drone) or manned  aircraft. In the case of a drone, the drone is  transported to the site, and a flight team will  perform initial site safety checks, set up defined  landing zones and proceed with the inspection.  In the case of aircraft inspection, the aircraft  will begin from a nearby airfield and fly the site  as part of a regular flight pattern. 

Both platforms will fly a designated pattern of  successive scans over the site, building up an  imagery database covering all modules in the  system. For drones, the IR cameras which can  be used have a longer integration time, meaning  flight speed is limited to reduce the possibility  of motion blur. In addition, the system must  land to change batteries every 15-20 minutes  of operation. This results in a capture rate for  standard resolution inspections of around  20MW/day, though operations at lower  resolution are able to cover more than 40MW/ day. 

Aircraft inspections are able to fly faster and  with fewer passes due higher resolution and  integration times in their camera systems,  and have endurance for full site inspections.  Therefore, aircraft based systems able to  achieve a scanning rate of up to 150MW/hr.  

It is important that proper irradiance is  maintained during the period of inspection in  order to ensure all possible fault modes are  detectable. IEC 62446-3^* specifies a minimum  irradiance threshold of 600W/m2 in the plane  of array of the PV modules, and this threshold  is mirrored in the NREL O&M best practices  guide. There is an additional advantage to  performing these inspections over a short time  interval, as it allows for intercomparing of the  thermal properties of different portions of an  array while the system is at a relative thermal  equilibrium.  

Analytics

Once data is collected for a site, it must be  analyzed to produce an actionable report.  This is the most critical stage of inspection, as  a poor analysis can lead to missed faults and  poor localization leading to inefficiency during  remediation. A typical site inspection can  yield anywhere from from multiple gigabytes  to multiple terabytes of raw sensor data, and  advanced systems must be put in place to  reduce this data into a usable output for site  operations.  

For most aerial applications, commercial  tools are available using a process called  photogrammetry to generate stitched  composite imagery of a site by finding  unique “keypoints” in successive frames.  Unfortunately, this tool cannot be used  effectively when applied to IR aerial inspections  of PV systems, due to the regularity of the  imagery collected. When keypoint matching is  attempted, multiple matches will occur causing  image distortion and missed modules. Because  of this, standard aerial post-processing tools  cannot be utilized, and software utilizing a  fundamentally different methodology for  analysis must be utilized. 

The methodology employed by Heliolytics  utilizes Artificial Intelligence (AI) processes  to analyze these complex datasets. AI tools  provide a powerful way to classify a defect  according to not only the distribution of  thermal deviations within the module but also  to correlation to collected visible imagery, and  information from the faults around it. 

These inspections can find a wide variety of issues, including:

• Hot Spots
• Sub-Module (Bypass Diode) issues
• String outages
• Potentially Induced Degredation (PID)

Remediation

Once the data has been analyzed, it must be  passed to stakeholders in the correct format  to allow for actionable remediation. For field  personnel, this means accurate site maps  which clearly identify the location and type of  defect to be remediated. For site operators, it is  important that the results from inspections can  be integrated into existing digital workflows,  to make sure that faults are tracked through  their remediation phase, and records can be  kept of the process.  

Summary