Session: 05-12 Heliostat Consortium 2

Paper Number: 116766

116766 - Characterizing Heliostats at a Commercial Scale With Non-Intrusive Optics 

The performance of power tower concentrating solar power (CSP) plants is contingent on the performance of the heliostat field. Surface level deformations, errors in facet canting, and errors in heliostat pointing, called tracking errors, can cause dramatic losses to the energy reaching the receiver.

Several proposed methods use heliostat-affixed sensors, specially designed targets, or laser-based alignments, but most are designed to address overall errors in pointing only. The most commercially prevalent calibration system is the Beam Characterization System (BCS), which requires heliostats to be taken off target, and outputs tracking errors only. BCS calibration of a full heliostat field can take several months to over a year.

The Non-Intrusive Optical (NIO) methodology is a unique, drone-based approach to address the problem of heliostat calibration. Using a drone-mounted camera and pre-programmed flight paths, we capture images of a known reference, the central receiver tower, reflected in each heliostat. The reflected tower image traverses the mirror as the drone moves, collecting data on each point the tower reflection crosses. This method of data collection, combined with a deep learning and computer-vision based post-processing algorithm, allows the NIO technology to characterize optical errors due to tracking, canting, and point-wise deviations on an individual mirror facet referred to as slope error.

Recent developments have made the process automatic. The aircraft flies programmatic flight plans, a parser algorithm finds heliostat features in the collected video data, and the post-processing algorithm compares the expected features with the detected features.

NIO runs on the four software modules shown below.

1.     Field modeling – used to inform decisions on flight areas and heliostat measurability.

2.     Data collection – plans programmatic flights for an autonomous UAS over the heliostat field and exports flight-computer compatible flight plans for a variety of filetypes.

3.     Data parsing – separates video data into relevant frames, uses video metadata to assign expected orientations to heliostats, extracts reference features from the heliostat image, and produces photogrammetric camera positions.

4.     Post processing – locates the reflected tower edge via pixel gradient search and calculates a slope error from the comparison between the reflected tower edge in the image, and the expected location for an ideal geometry.

NIO requires four known quantities to calculate deviations in the mirror surface normal:

1.       The reference tower position

2.       The heliostat state, including position and expected orientation.

3.       The UAS position at the time the corresponding image is captured.

4.       The location of the reflected edge in the corresponding image

Quantities 1 and 2 can be found from heliostat field information and computed from the time of day. Quantities 3 and 4 are carefully extracted using machine learning and computer vision techniques on extracted video frames.

The NIO method has been demonstrated on dynamic heliostats in a commercial field. Using a consumer level drone, we have captured over 40 heliostats in a single, brief, 10-minute flight. Overall, we have collected NIO data on several hundred heliostats and are working to autonomously generate surface shape data for each of these datasets. In this presentation, we will show the approach behind the NIO technology, the current state of the technology, and the low-cost, rapid solution to heliostat calibration it offers the CSP industry.

Presenting Author: Devon Kesseli National Renewable Energy Laboratory

Presenting Author Biography: Name: Devon Kesseli
Institution: National Renewable Energy Laboratory