Source: ATOLLA TECH LLC submitted to NRP
SMART SENSOR FOR A NEW METHODOLOGY OF SPRAYER CALIBRATION
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
COMPLETE
Funding Source
SMALL BUSINESS GRANT
Reporting Frequency
Annual
Accession No.
1019464
Grant No.
2019-33610-29789
Cumulative Award Amt.
$106,500.00
Proposal No.
2019-00830
Multistate No.
(N/A)
Project Start Date
Jul 15, 2019
Project End Date
Jan 14, 2021
Grant Year
2020
Program Code
[8.12]- Small and Mid-Size Farms
Recipient Organization
ATOLLA TECH LLC
184 MAPLE AVE
ROCKVILLE CENTRE,NY 115704373
Performing Department
(N/A)
Non Technical Summary
Inefficient crop spraying causes drift which can lead to crop damage or unwanted pesticide use in neighboring non-target crop areas. It may contaminate nearby bodies of water or be an issue for human health. Our company can significantly reduce this through the implementation of a sensor system which would produce a quick calibration analysis by determining the efficiency of sprayer machines and the spray pattern of the material in real time. Small farmers tend to skip the calibration process due to the time and labor of the task. However, by completing a full calibration procedure in the frequency recommended by the cooperative extension experts, they will essentially reduce the costs of wasted materials and increase guarantee of uniform spray to avoid pest infestations. An owner of a small farm that uses our calibration service will save a significant amount on time and labor. In the long run, the farmer would also save on chemical costs and avoid the consequences due to drift of chemicals onto neighboring farms, public roads, or non-agricultural areas; all due to non-calibrated machines.This project will prove the potential of our technology as a competitive player in the rapidly growing ag-tech field. These types of sensors have been previously built for complex and sophisticated applications in mostly academia and government sectors. We look to prove a new functionality while reducing the size and cost of the technology. The sensor would be used to replace a tedious and often ignored task by small farmers. This would revolutionize the way calibration has been performed throughout the past, and thereby changing the perspective of the way farmers view calibration.
Animal Health Component
50%
Research Effort Categories
Basic
0%
Applied
50%
Developmental
50%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
40253102020100%
Knowledge Area
402 - Engineering Systems and Equipment;

Subject Of Investigation
5310 - Machinery and equipment;

Field Of Science
2020 - Engineering;
Goals / Objectives
Project Goal: Our goal is to prove the feasibility of using our proposed sensor to replace current techniques for calibration of airblast machines. Innovating the calibration methodology will lead to better spraying practices in small and medium sized farms. Calibration of the sprayer is necessary for checking the performance of the machine and the spray pattern in relation to the canopy. The efficiency of a spray application is dependent on a properly calibrated sprayer machine.The current method for calibration, if done correctly, is both time consuming and labor intensive. The initial step for calibration of an airblast sprayer is to check for the actual speed of the tractor since many older tractors are known for having inaccurate speedometers. Next, the air direction needs to be adjusted to the canopy height. This is done by tying ribbons to the ends of each nozzle and turning the fan on while the tractor is situated between two canopy rows. Air flow differs between the right and left side and therefore needs to be adjusted differently on each side. The following step is to match the air volume and speed to the canopy. To do this, ribbons need to be tied to the top, bottom and middle of the canopy and as the tractor passes by, a viewer needs to be on the other side of the row to check the ribbon pattern. If for example, the ribbon blows straight out, then the amount of spray needs to be decreased.The next step is to calculate the expected nozzle output in gallons per minute (GPM). This calculation is for the total GPM of all the nozzles on a single side of the sprayer. The values for this can be found in the specs of the nozzle manufacturer. There are also apps available that will do this calculation for the farmer. Next, the nozzle output needs to be measured and compared to the calculated value. This step is conducted by filling the sprayer with water and attaching a hose to each nozzle. The sprayer needs to run for 60 seconds while the water is collected into a measuring container. Any nozzle that is more than 10% off from the calculation needs to be replaced. The final step is to verify the spray coverage with water sensitive paper (WSP). Squares of the WSP should be sporadically stapled onto the leaves in the canopy row. Using the sprayer set up determined from the previous calibration steps, run the sprayer filled with water down the row where the WSP is set up. The droplet pattern on the cards is used to determine the spray density and spread and help prevent future over and under application. These last two calibration steps mentioned can be replaced using lidar. To derive the mass density and spread of the spray using lidar, the lidar image of the plume needs to be characterized.For this proposal we look to explore a methodology for determining a rough estimate of the mass distribution and spray direction through the use of a non-scanning lidar sensor. Small farms are unfortunately the group that skips calibration altogether due to the lack of time and labor. Offering an affordable service that provides better precision and quicker results than the current methodology would undoubtedly cause an increase in well-functioning and more efficient airblast machines; thereby reducing over-spraying and drift of spray materials that may negatively affect the environment or neighboring crops.Technical Objective 1: Through a numerical analysis, determine the feasibility of using a single channel, elastic, non-scanning lidar to derive the characteristics of a spray plume needed for calibration. Demonstrate that the lidar can provide sufficient data to estimate the mass density and spread of a spray plume.Technical Objective 2: Demonstrate that the soft target non-scanning lidar can be used experimentally to determine mass distribution of a controlled plume. This technical objective establishes whether the numerical objective can be experimentally verified.
Project Methods
The methodology for a new calibration technique using lidar involves characterization of the lidar image of a spray plume to determine its mass distribution. However, this becomes a complex inverse problem for a single channel elastic lidar. The lidar output, post-calibration, is a matrix of a single variable, backscatter coefficient, and a solution needs to be derived with several unknowns. For short range lidar measurements, we can assume a homogenous atmosphere, negligible molecular contributions in the IR, and a uniform plume of spherical particles. We can also assume that we have a plume with a known size distribution based on the nozzle used and a known complex refractive index based on the material sprayed. With these assumptions, the slope method for the Fernald inversion can be applied to retrieve the extinction coefficient. Once the extinction coefficient is determined, the number density can be simulated using Mie theory and the Fredholm integral equations which relates the optical properties with the aerosol quantities.Technical Objective 1:To determine the feasibility of calibration using lidar, the mass distribution of the plume must be understood through a numerical study. This initial test of feasibility assumes a simple case scenario of a perfect spray plume with no influence from external or systematic variables and noise. The study assumes perfectly functioning nozzles and flow. Values given by the nozzle manufacturer will be used to simulate the lidar image of the spray pattern. These values will represent the ground truth for the numerical analysis.Task 1.1: Simulate a lidar image of the spray plume in terms of range over time using MATLAB. A stationary lidar produces a measurement of intensity over range at a single point in space. The range in our case is the depth of the plume looking horizontally. The image will be composed with reference to the spray shape provided by the nozzle manufacturer. The intensity of the signal return will be in units of backscatter coefficient and will be determined based on a monodisperse plume with a size distribution prescribed by the nozzle manufacturer. The number concentration of particles is determined by the nozzle used.Task 1.2: Determine the mass distribution of the spray plume; with a known complex refractive index and particle size range, the mass density can be modeled using the method of inversion from backscatter coefficient. Several assumptions can be made when applying the inversion method using the proposed lidar. For IR wavelengths and short distances between plume and instrument, we can assume negligible molecular contribution and set molecular coefficients to zero. We assume a homogenous atmosphere, giving us a uniform distribution of particles outside of the plume. Using the slope-method we can solve for the aerosol extinction coefficient using the lidar signal profile at a reference range at the boundary of the far side of plume. Once the extinction coefficient is calculated, the number density profile of the point on the plume being analyzed, n(z), can be determined using Mie theory for spherical particles.Task 1.3: Determine accuracy by comparing collected data to ground truth. Ground truth is the estimated mass density based on the information provided by the nozzle manufacturer. This becomes complex when looking at points on a plume without accurate location data linked to it. The data points being analyzed are taken at predetermined distances away from the nozzle for as accurate of a comparison as possible. The error caused by location uncertainty would need to be added into the simulation.Evaluation: For the adoption of a simple, new and affordable methodology for calibration, the mass density needs to be determined using a non-scanning lidar. The numerical analysis should model the mass distribution with an accuracy of greater than 60%. A general understanding of the spray spread should be presented in the analysis.A potential pitfall is that a collection of single points in space (measured as a function of depth) will not provide sufficient data to characterize plumes. The determination of a fail of the objective would be an error between the model and the ground truth of greater than 60%. This is likely caused by a need for data in a 2-D space for a more complete visualization of the plume. In that scenario, a scanner would need to be implemented so that a 2-D image analysis could be performed.We expect to reach our goal, which is determining mass distribution with an accuracy of greater than 60% based on our knowledge of the capabilities of lidar technology.Technical Objective 2:A stationary portable sprayer producing a water plume from a single nozzle would be set up in a controlled space. The experimental study introduces new variables to the model such as systematic and external noise. A controlled environment limits the variables that further complicate the signal so that they could be ignored until phase II. Such variables would drastically change the shape and dynamic of the plume; i.e. influence of more than one nozzle, wind, a moving sprayer and noise from the field environment. The portable sprayer will need to produce a steady flow rate. The nozzle will be purchased by a manufacturer that has tested the spray pattern characteristics, such as Teejet. The spray plume will be measured in two ways simultaneously; using the lidar as well as with water sensitive paper, WSP. The WSP will be analyzed using current and accepted techniques that determine the pattern and mass distribution from the cards.An imaging software will be used to process the cards. A model-based approach will be implemented to profile the mass distribution using lidar. The lidar results will be compared to the accepted method of calibration using WSP.Task 2.1: Set up the experiment by fixing in place a portable sprayer and controlling the volume of liquid sprayed through the nozzle. Face the nozzle towards the WSP. The lidar will be set up perpendicular to the direction of spray.Task 2.2: Run a series of 30 experiments without changing anything in the setup. Using the lidar, collect several points on the plume for each experiment. Repeat the location for each experiment. This data should be sufficient for training the model; if not, then more runs are necessary.Task 2.3: Calculate the mass distribution of the spray application using the plume data collected by the lidar and the WSP. The mass density will be calculated for the lidar results using the methodology presented in Technical Objective 1.Task 2.4: Assuming the algorithm is sufficiently trained, check for repeatability by running the experiment another 10 times. Compare the results of the mass distribution measured from the lidar data to both the WSP analysis and with the result attained in Technical Objective 1. Since this is a non-scanning lidar and location measurements can only be so precise, there will be error bars that will need to be considered when determining mass densities in the spray pattern.Evaluation: Results obtained using a lidar need to be compared to those using WSP. The accuracy of results from the experiments in Technical Objective 2 need to be greater than 65% compared to the results from the WSP.We believe we can attain a rough calibration to replace the tedious, and time-consuming calibration procedure using WSP through the use of a non-scanning lidar. However, if this objective fails due to insufficient data, then implementing a scanner for 2-D cross-sectional imaging would be necessary. The next step would therefore require us to repeat Technical Objective 2 based on 2-D imaging of the plume. It is an iterative process between quantifying the plume and building the scanning algorithm.We expect to reach our goal, which is determining the mass distribution with an accuracy of greater than 65% compared to WSP methodology based on our knowledge of the capabilities of lidar technology.

Progress 07/15/19 to 01/14/21

Outputs
Target Audience:One of our target audiences are the small fruit farms that use airblast sprayers. Those farmers understand the inefficiencies that come with that type of sprayer; however, they prefer it due to its cost and durability. The airblast sprayer has no frills like the John Deere sprayers that feature built-in sensors and autonomous features. Therefore, to ensure the most efficient spray application, calibration of the machine must be done at the beginning of the season and once every time a new material is to be sprayed. Calibration of an airblast sprayer, using today's recommended standard technique, takes several hours, a complicated analysis and at least two people to achieve. For a small farm with an endless list of tasks to complete to ensure a successful crop and therefore a profitable season, this task becomes their last priority. Therefore, the farmer of a small farm stands to benefit the most from a service that they can afford and that completes a task that up until now has been ignored. They will be provided with user-friendly results that show the benefits of servicing their machine. Agricultural academic groups also stand to benefit. These groups replicate the tasks of the farmer and run studies on different materials used and new processes for spraying. They are forced to analyze the spray application using water sensitive paper, WSP, just as the farmers are expected to. The task requires either creating your own image processing analysis (not feasible in most circumstances, since the group is composed of agriculture focused majors) or using an image processing software that is both user-friendly and provides decent results. Unfortunately, the method of measuring the density and spread of the spray plume using WSP and an accompanying software has been shown in several research studies to have high error bars; it is simply the only available method today. Considering that academics are working in proving efficiencies or inefficiencies with high accuracies, a better method for the measurement of the spray plume is needed. Furthermore, the agtech segment is growing in many ways and one of those ways is the development of new sprayer machines. These can be ground based tractors or aerial drones. No matter what kind of sprayer is developed, the startup developing it needs to show the efficiency of the device in order to get the acceptance of adoption by the agricultural community and the approval of agencies such as the NJ pesticide control agency. These are small companies without the resources to run an in-house test that would greatly benefit from outsourcing their efficiency assessment of their newly developed sprayer. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?The employee hired under this project was hired as young graduate with a bachelors degree. After working one-on-one with the PI, he was able to advance tremendously and now has an expertise in this field. We hope to continue working with him in Phase II, but either way, after this project, he is a much stronger candidate for work in the industry. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?Under the phase I, we proved that a non-scanning LiDAR was capable of determining the mass distribution of a spray plume on a consistent basis with a high level of accuracy. For our next steps going forward, there are a number of issues we still need to address. The area of our experiment that could benefit the most from further development would be addressing our LiDAR system's limits. Currently we are using a very budget friendly LiDAR developed by Garmin. We originally wanted to build our own LiDAR system as this would give us complete flexibility in our design and allow us to have complete control over the data we collected. Unfortunately, due to budget concerns and the lab restrictions imposed by the Covid-19 lockdown situation, we needed to pivot to a system that suited our new circumstances but also still managed to accomplish our goals. As a result, we pivoted to using the Garmin LiDAR as it was small enough to be easily moved from location to location, as well as being eye safe but powerful enough to accomplish our objectives. However, the over engineered LiDAR from Garmin also proved to be a nuisance on a number of occasions, and we often found ourselves working long hours to reverse engineer the black box system. Having said that, we still managed to complete our objectives under these difficult circumstances and proved that a non-scanning LiDAR could indeed be used to determine the mass density of a spray plume. While this Garmin LiDAR worked well for Technical Objective 2, it's lack of image resolution meant that we were limited in what we could analyze. While we could determine the mass density as a function of distance from the nozzle of a spray plume, we failed to capture the spray plume in 3D (the depth dimension was lacking), limiting our analysis to 2D with a scanner due to a low resolution of depth. Yet, the most significant thing that our work with the Garmin lidar proved was that even when using a very low-cost lidar system, we could extract enough data to improve current calibration techniques. This shows the potential of this device in agriculture, an industry that is constrained by costs. During phase II, we will design and build a higher-grade LiDAR system to our exact specifications. Our custom-built LiDAR system will provide more details for analysis of a spray plume. The system would prioritize a greater depth resolution determined by the specs of the detector and the laser. During phase II we will run outdoor experiments which will introduce more variables such as wind, outdoor background, and a dynamic spray plume created from a number of nozzles simultaneously. These new variables would represent challenges and noise that we would experience in real-world situations. We would experiment with isolating the different factors (i.e., wind) that affect the signal and adjusting our algorithm to account for their affects. Controlled scenarios for determining effects of isolated factors include but are not limited to; an indoor experiment that includes a fan for controlling wind, an outdoor experiment using a single nozzle sprayer in a field on a non-windy day, an experiment indoors using a sprayer with more than one nozzle. In phase II, we are aiming to capture the spray plumes in far greater detail by creating a detailed 3D scan that could determine the spray plume's shape, size, density and spread. Based on our phase I results, we are confident that a non-scanning lidar can provide enough data to determine nozzle efficiency. However, a scanning lidar has capabilities of determining a dynamic spray application's efficiency.

Impacts
What was accomplished under these goals? A detailed final report is available upon request. These are highlights of the phase I work accomplished. TO1 - Numerical model: We simulated a water spray plume emitted from a hollow cone pressure swirl nozzle under ideal conditions (i.e., brand-new nozzle and no wind). We were able to create an accurate simulation of this spray plume using the CFD simulation software, Ansys Fluent. Ansys Fluent has proved to be a useful tool for our purposes, having built in functions to recreate the complex physics that surrounds the fluid dynamics of sprays. We specified the simulation environment to be a non-constraining chamber of air and the spray material to be water to mimic the environment of our experiments in TO2. Our domain was designed as a cylinder with a height of 1m, and a radius of 3m. The parameters used in our simulation are based off the nozzle and backpack sprayer specs that are used in TO2. We compared our simulation results to those developed by Teejet using a laser sheet imager under ideal conditions. The hollow cone spray pattern is clearly visible in both. We attempted to receive the raw data from Teejet to be able to compare, however that would have cost us around $10K. We therefore relied on the capabilities of Ansys, a trusted and widely used tool in the field of aerospace. Using the analysis reports from our simulation, we extracted the values of mass density (number density) of our spray plume. Our model simulated the dispersion of the spray plume, where the number density of the spray decreases as the droplets are dispersed over the area. The surface report of our cross section at 30 cm for example, shows the mass density to be 2.4 ×1012 m-3. The flow rate of the water at the Nozzle was measured from the simulation as 0.00275 kg/s. The simulation results will be used as our ground truth for mass density as we move into technical objective 2. Under TO1, we confirmed that our spray simulation aligns with the spray lab results determined by the manufacturer, thereby ensuring that it is an accurate model to use as our ground truth. We also confirmed that we can produce the mass density and flow rate from the simulation. These tasks showed positive results for moving onto technical objective 2. TO2 - Experiment: While we expected to build our own LiDAR for this project, the covid-19 shutdowns forced us to seek a smaller, more budget friendly, and portable LiDAR system that could be used from home and transported easily. As a result, we compromised for an off-the-shelf system with lower resolution but still capable of capturing the necessary spray plume data for verifying our phase I proof-of-concept (POC). We therefore purchased a Garmin LiDAR and an Arduino Computer. The limitations to using the Garmin include its low spatial resolution which limited our data output in terms of depth. The Garmin also processes the data onboard as a black box leaving no possibility of attaining raw data nor any information on what the software is doing to the data. It took a lot of work to reverse engineer the Garmin and its output, however, it certainly carried advantages during the shutdown due to its size and ease of transport. After figuring out the Garmin and its output, we began the collection of a spray plume using a Chapin backpack sprayer and new Teejet nozzle. Our experimental setup was in a closed environment, where we could ignore variables such as wind and other potential noise adding elements. The sprayer was set in a fixed position and the nozzle used was a hollow cone spray tip, TX-VK3 from Teejet. Based on manufacturer details, we assumed a spray pressure of 40 PSI, and a flow rate of 0.00275 kg/s. From our simulation, we were able to determine the number density (mass distribution) of our spray to be approximately at 30 cm from the nozzle. We use these values from the simulation to train our algorithm to determine what constitutes an efficient, working nozzle. Initial experimental tests to determine the spray signal were conducted with a brand new Teejet nozzle, with spray parameters matching our Ansys model. We can therefore assume that the signal peak from the experimental results using the new nozzle identifies with the number density from our Ansys simulation. We compared between the simulation and experimental results which is represented in Figure 2. The width of the signal (width of the wall of the cone spray) from the experiment is approximately 14cm while the width from the model is approximately 16cm wide. The small margin of error between the two can be attributed to the Garmin's low-resolution and/or to the fact that the simulation assumes an ideal environment. The experimental results show a single peak compared to two peaks shown in the simulation for a spray plume with a hollow cone shape. The first peak represents the laser return from the front of the spray cone, and the second peak represents the backscattered light from the back of the spray cone as seen in the simulation. Since the Garmin lacks the power and resolution capabilities of more sophisticated LiDAR designs, it fails to fully penetrate the spray plume resulting in only one peak in the output data. For our POC analysis we determined that due to a homogeneous spray pattern this would suffice. To validate our results, we repeated this experiment over fifty times and received consistent results. We then repeated our experiments with nozzles that had been artificially degraded to mimic real world wear and tear from erosion by pesticides and other chemicals. For this, we took a Teejet nozzle and placed it in a jar with acetone to soak for 24 hours. Figure 3 shows images of a brand-new nozzle next to a damaged nozzle that was submerged in acetone. Note the clear damage and degradation. As a result of internal damage to the nozzle, we expected intensity values to rise based on other published work that show increased flow rate with increased degradation of a nozzle. Our results show this and are represented in Figure 4 which shows the intensity values from a new nozzle compared to a degraded nozzle. One can see immediately that the degraded nozzle produced a more intense amplitude (as expected), increasing by about 31% due to the damages. We therefore assumed that the number density also increased by 31% for the degraded nozzle compared to the new nozzle. By being able to determine the level of intensity that has increased we can train our algorithm to determine whether a nozzle has become too worn and inefficient and therefore needs to be replaced. While our experimental assumptions may have some error, it still proves to be a reliable measure of efficiency, especially considering that current methods for calibration have large error bars themselves. Based off our analysis we can see that even our simple, low budget LiDAR system could effectively determine whether a nozzle was under performing, perhaps due to a clogged nozzle, or whether it was spraying with increased intensity due to being worn from overuse and needs to be replaced. Our goal was to establish whether a non-scanning LiDAR could accurately, reliably, and consistently, determine the mass density of a spray plume. With this information we could show that it is possible to accurately determine using lidar whether a nozzle adhered to its expectations outlined by the manufacturer, or whether it needs to be replaced. We were successfully able to accomplish this goal and our results showed us that a new low-budget method for nozzle calibration was indeed possible. It showed us that a new method might not only prove to be time and cost saving, but that it could also prove to be less prone to the human error seen in current calibration methods. We believe that our new method for calibration will lead to more efficient spraying, reducing the harmful environmental impact of spray drift caused by poorly calibrated nozzles, as well as saving money and time to farmers adopting our approach.

Publications