Development and validation of a low cost technique to predict spread patterns of centrifugal fertiliser spreaders

In the pursuit for a sustainable agricultural production respecting ecological, social and economic values, modern farmers are well aware of the importance of a correct and precise fertiliser application. Indeed, non-uniform spread patterns can cause extra pressure on the environment and might result in economic losses for the farmer. Because of the higher price of pneumatic fertiliser spreaders and the limited working width of pendulum spreaders, centrifugal spreaders are most commonly used in Europe. Although the centrifugal spreading mechanism is simple, the process is hard to control and can lead to a deviation between the desired and the actual fertiliser distribution in the field. This non-uniform application can have several causes: the behavior of the operator, the properties of the fertiliser, the settings of the machinery and external conditions like the unevenness of the field or the wind.
Even though farmers realize the importance of a precise fertilising, most of them cannot adapt the settings of the spreader to improve the pattern uniformity because they do not have the proper tools to determine and evaluate the spread pattern at farm level.
Therefore, the general objective of this dissertation is to explore and develop a fast and accurate technique for measuring the spread pattern of conventional centrifugal spreaders. This method aims to be low cost and applicable on farm level. Also, it has to be mobile to the extent that the device can be built up on site to test several machines. The device should enable the adjustment of the spreader in such a way that a uniform spread pattern is obtained.
Three existing approaches for evaluating the spread pattern are available. For the collector tray method, the grains are collected in trays and weighted in spread halls or on the field. The tests are time and fertiliser consuming. By using a full modeling approach, like the Discrete Element Method (DEM), this can be avoided. DEM takes all possible interactions between the grains into account and performs well in predicting spread pattern. The hybrid approach combines measurements and modeling. These measurements involve the measurement of the initial speed, direction and sometimes also the diameter of the grains. With a ballistic flight model the landing points are calculated.
In this thesis the hybrid approach is applied: the spread pattern was predicted with a ballistic flight model based on the measurement of the horizontal outlet angle, the vertical outlet angle, the grain diameter, the grain density and the initial velocities. The horizontal outlet angle is the angle between the trajectory of the fertiliser grain and the radial line at the outlet position of the grain on the edge of the disc. The vertical outlet angle is the angle between the trajectory of the grains and the horizontal plane.
Two dimensional imaging techniques were used to measure the horizontal outlet angle and the speed of the grains. First, a virtual camera was developed to assess the deviation caused by the perspective projection. It was found that in the center of the image the deviation on the horizontal outlet angle was below 1° leading to an acceptable influence on the landing point in the field.
Then a test spreader was used to make blurred images of the trajectories and to measure the horizontal outlet angle of two different spreader set ups: short vanes and long vanes. This method with one low cost camera mounted in the test spreader led to useful results for predicting the spread pattern: the mean horizontal outlet angle of the grain trajectories could be measured at different camera positions. The cylindrical spread pattern, measured with thetest spreader, provided the mass distribution and allowed one to calculate the vertical outlet angle.
However, the applied method did not give any information on the speed and the processing was partially manual and thus time consuming. Moreover, because the test spreader was a closed set the full spread pattern could not be measured which is necessary to validate the simulated spread patterns. Therefore, a new modular measurement device that tackled these shortcomings was proposed. This device consisted of two main parts: a test spreader and a measurement unit. As the arm with the measurement unit pivots on the center of the test device, the measurement unit traces a path around the circumference of the disc. The grains flying under the measurement unit were imaged using two different techniques: the high speed and a developed stroboscopic imaging technique. For the high speed technique a camera, type MotionXtra HG 100K (Roper Scientific, New Jersey, USA), was mounted in the measurement unit. The stroboscopic technique used a specially designed LED stroboscope combined with the Nikon D 100 camera.
Both techniques were able to visualize the grain flow but to extract information from the images, specific image processing was necessary. Filtering and morphological operations removed noise and made the segmentation, i.e. the recognition of the grains as objects in the image, possible. Several approaches were compared to develop a robust algorithm. Next, an efficient algorithm to detect the trajectories of the grains was developed. Finally, their direction and speed were deduced.
A comparison between the more expensive high speed technique and the developed low cost stroboscopic technique was made. Also the differences between the short and long vanes were studied. Qualitatively, the strobe and the high speed technique gave very similar results for the angles and for the speeds. To quantitatively investigate the effects of the measurement technique, vane length and outlet positions, a regression analysis was performed. Using the short vanes increased the outlet angle with 1%. The outlet speed of grains ejected by the short vanes was significantly lower than for the long vanes. Only for the speed measurement with the short vanes, both techniques gave statistically the same results. For the other cases, the high speed camera measured slightly higher values. The error for both techniques was the same.
Overall the strobe technique and the high speed technique were capable of measuring the outlet angle and the outlet speed. Small differences between the measurements existed, but ultimately we were interested in the resulting spread pattern in the field. In this phase, all necessary inputs for simulating the spread pattern were available i.e. mass of the grains, vertical and horizontal outlet angle and outlet speed.
These inputs were used to simulate four spread patterns: two spread patterns for the short vanes and two spread patterns for the long vanes based on the measurements of stroboscopic and the high speed technique, respectively. Next the real spread pattern of our set up was determined in a spread hall.
The best correspondence between the measured and the simulated spread pattern was found when the regression function was used to generate the speed and the angles for each outlet positions on the disc in incremental steps of one degree. The relative error was 29.2% for the high speed technique and 30.6% for the strobe technique with the short vanes. Looking at the spread pattern of the long vanes, the relative errors for the strobe and the high speed techniquewere 29.2% and 29.7% respectively. The simulated spread patterns between the two techniques differed little with relative errors of 13.2% and 17.0%.
The results were slightly better than the differences reported by Reumers et al. (2003a), but clearly better than the differences Olieslagers (1997) found. Their spread patterns were calculated for a grid size of 1m x 1m. The DEM simulation of Van Liedekerke (2007) performed better with the same grid size as ours, but with a smaller total transverse spread distance. The transverse spread patterns however corresponded much better. This is an important result since the transverse spread pattern is generally used to evaluate spreaders in the field. Indeed, it means that the transverse pattern of our measurement technique could replace the method with the collector trays as it exists in the field.
Two things will have to occur before farmers can use our development as a tool to adjust a spreader. First, we will need a new test set-up that allows a real spreader to be placed under the rotating arm, and second, a cylindrical collector will need to be designed.
Replacing the collector tray method by our technique has some important advantages. The collector tray method is very time consuming and expensive. At least two, but preferably four people are needed to perform a test. A large number of collectors with inserted grids are needed. To achieve the desired distribution pattern in the field, several tests can be necessary and hence, lots of fertiliser is spilled. With our technique one person can perform the test, less fertiliser is used and all fertilizer is collected in the trays rather than spread out in the field. A cylindrical collector is needed to measure the mass distribution and the vertical angle. This is an important shortcoming of our method. Furthermore, the full spread pattern will result from the superposition of the spread pattern of both discs. Possible effects that arise from two discs that spread simultaneously are not yet investigated. Our set up preferably operates indoors. The set up can be disassembled and moved from one location to another, but is not fully mobile. A possible solution could be to equip the set up with wheels and a pull bar to make it transportable by car.
Our technique could also be considered as a possible on board sensor to measure the spread pattern online, but this has not yet been investigated. The sensor should then also measure the size and the distribution of grains. The ambient light will certainly disturb the lighting and the image processing will have to be thoroughly adapted. Also important difficulties will arise from the vibrations and the movement of the spreader. Practical problems like dust and corrosion will also have to be solved.