“Modern agriculture is increasingly employing sensing and localization technologies to increase the efficiency of field operations and maximize crop yields by tracking local growth conditions and applying the appropriate resources such as water, pesticides and fertilizers as needed. System designers for such applications understand that satellite positioning has its limitations in terms of accuracy. However, the use of inertial measurement units (IMUs) can bridge the gap.
Author: Richard A. Quinnell
Modern agriculture is increasingly employing sensing and localization technologies to increase the efficiency of field operations and maximize crop yields by tracking local growth conditions and applying the appropriate resources such as water, pesticides and fertilizers as needed. System designers for such applications understand that satellite positioning has its limitations in terms of accuracy. However, the use of inertial measurement units (IMUs) can bridge the gap.
The IMU integrates a three-axis accelerometer with a three-axis gyroscope to measure system motion and determine system positioning through dead reckoning. By combining this data with Global Positioning System (GPS) information, designers can develop machine control systems that can accurately and continuously understand the position of equipment relative to fields and crops, taking into account terrain inclination, equipment arm movement, and other factors. Correction.
This article explores the importance and role of IMUs in precision agriculture and discusses potential sources of error when performing dead reckoning using IMUs, how to reduce these errors, and environmental and safety considerations for developers. Finally, this article will introduce precision IMUs from Honeywell Sensing and Productivity Solutions and Analog Devices, showing how these IMUs can be used to help increase accuracy to levels not possible with satellite navigation systems alone.
Why location tracking is vital to agriculture
Traditional agriculture follows a thick line. Although there are inevitable differences in soil composition, evaporation, etc. within a farmland, arable land, sowing, irrigation, fertilization, and harvesting are more or less uniformly performed on the entire farmland (usually several acres). During these activities, manipulating machinery can result in missed or overlapping areas, reducing field utilization, or wasting resources with redundant operations. While a foot or two of steering error between passes may not seem like much, when criss-crossing large fields, losses can accumulate significantly, increasing the time and fuel required (Figure 1).
Figure 1: Conventional agriculture treats the entire field as one and uses manual manipulation methods, both of which result in wasted time and resources. Precision agriculture has changed that. (Image credit: John Deere®)
Knowing the location accurately has many benefits. Not only is it possible to collect information on soil conditions at specific locations in a large area of farmland, but water, fertilizers, and pesticides can be applied accordingly to specific locations to maximize yields. The higher the positioning accuracy – preferably down to a single plant – the greater the benefit.
Precision agriculture changes the way farmers work their land. The advent of satellite navigation technology has allowed farmers to accurately map changes in growing conditions in the field, and can provide agricultural machinery with real-time information about relative locations within that space. This detailed mapping, combined with precise location information, allows farmers to tailor irrigation, fertilization and pesticide spraying to increase yields, minimize waste, and reduce environmental impact.
Real-time location information also allows farmers to avoid missing or overlapping planting and harvesting operations, maximizing field utilization, while minimizing time and fuel consumption by optimizing machine travel routes. In addition, the system can support semi-autonomous driving of agricultural machinery to reduce driver fatigue and enable efficient operation even in low-visibility conditions such as dust, fog, rain and low light. Precision farming methods are currently used on more than 50% of farmlands large and small, and the rate of adoption is increasing.
The ideal agricultural positioning system should be accurate enough to reliably locate a single plant or row of crops in a field that may extend for hundreds of acres—that is, provide an accuracy of a few inches. But the positioning accuracy provided by satellite navigation systems alone is limited. The basic receivers of US GPS can only provide an accuracy of a few meters. Dual-channel GPS receivers or real-time kinematic (RTK) systems that relay GPS signals from fixed stations can achieve accuracy well below 1 m. Even so, however, they rely on the accuracy of the satellite broadcast information, typically yielding an average accuracy of around 0.7 m. Other complicating factors in GPS positioning include the effects of nearby objects and terrain reflecting or blocking the signal, satellite constellation geometry, and time of day.
In addition, satellite navigation has other limitations. The position provided by the system is only one point – the phase center of the receiver antenna. GPS does not provide direction information; for example, the facing direction can only be inferred by determining the direction vector between successive points. Similarly, GPS is insensitive to pure rotation and therefore cannot determine any tilt relative to vertical GPS.
This antenna-centric positioning and insensitivity to rotation can create positional errors in agricultural applications. For example, on a GPS-equipped tractor, the antenna might be mounted on top of the cab, perhaps 10 feet above the ground, which is the central location for GPS positioning. It is reasonable to assume that the position of the tractor or any attached equipment on the ground can be reliably determined from the antenna position by simple geometric operations. The problem is that since the GPS system cannot determine the direction, such as when the tractor is driving on a slope (Figure 2), the position predicted by the rigid geometry will deviate from the actual ground position. Even an inclination as small as 5 degrees (°) would result in a ground position error of more than 10 inches (25.4 cm) in this case.
Figure 2: GPS cannot determine direction, so slopes can cause errors in determining the actual ground position of the device. (Image credit: Richard A Quinnell)
One solution to these problems is to use sensors that measure system motion for dead reckoning, complementing GPS navigation with inertial navigation. Inertial dead reckoning can continue to provide accurate position information when GPS signals are weak or absent, while also providing a “truth check” for spurious results that may arise from multipath or other signal distortions. In addition, inertial navigation sensors can fill in the directional information that satellite navigation cannot. For example, by simply measuring the direction of gravity, inertial sensors allow the system to correct for tilt errors in GPS ground positioning and can improve operator safety by supporting rollover warnings.
In practice, such inertial measurement devices rely on two types of microelectromechanical systems (MEMS) sensors: accelerometers and gyroscopes. An accelerometer can measure changes in linear motion along three orthogonal axes, and since gravity is an acceleration, it can also indicate its direction. A gyroscope measures angular motion (ie, rotation) about each of three identical linear axes. Combined, the change in motion of the system along six degrees of freedom can be measured (Figure 3).
Figure 3: Inertial navigation uses sensors to measure changes in motion along six degrees of freedom (three linear and three angular) to support dead reckoning of position. (Image credit: Honeywell Sensing and Productivity Solutions)
However, these inertial sensors do not directly Display position. The accelerometer only measures the system’s front-to-back, up-and-down, and side-to-side roll. These values must be integrated over time to get the system velocity and then integrated again to get the position. Similarly, gyroscopes measure roll, pitch, and yaw, which must be integrated over time to obtain angular orientation.
These integrals may help reduce the effects of random motion noise in sensor measurements, since such signals tend to average out. But integration can exacerbate some of the major sources of systematic error inherent in inertial sensors. If left uncorrected, these errors can add up and destroy the accuracy of dead reckoning positions, limiting the method’s effectiveness in compensating for missing GPS information. Generally, the smaller the sensor measurement error and the longer the dead reckoning, the more likely it is to provide a position with the required accuracy.
Error sources in the IMU
Bias Error: In MEMS inertial sensors, whether it is an accelerometer or a gyroscope, one of the main sources of error is the bias error. Zero bias error is the residual signal produced by the sensor in the absence of rotational or linear acceleration. This error tends to be deterministic, unique to each device, and usually a function of temperature. Integrating this signal over a period of time quickly reaches unacceptable levels, but with a proper calibration test, the sensor’s bias error can be determined and excluded from calculations.
Zero bias instability: Zero bias instability is related to zero bias error, which means that the zero bias error of the device changes randomly over time. This source of error cannot be eliminated by calibration, so developers must evaluate how much variation their designs can tolerate and find a sensor with a low enough bias stability specification to meet their needs.
Scale factor error: This is another deterministic error in inertial sensors. The scale factor, also known as sensitivity, is the best-fit linear relationship that maps sensor input to output. The scale factor error of a sensor is the deviation of its output from this linear relationship, usually expressed in percentages or parts per million. This can also be temperature dependent and can be compensated for with proper calibration.
g-sensitivity: One source of error specific to a gyroscope is its sensitivity to linear acceleration, also known as g-sensitivity (g is an abbreviation for acceleration from gravity, typically 9.8 m/sec2). In a MEMS gyroscope, this linear acceleration error occurs due to the asymmetry of its test mass.
A MEMS gyroscope works by vibrating a test mass in one direction while sensing any motion in an orthogonal direction. The Coriolis effect causes a detectable lateral movement of the test mass when the sensor is rotated about an axis orthogonal to the other two directions.
The sensor produces linear acceleration in a direction orthogonal to the vibration of the test mass, and this lateral movement is also caused by the inertia of the test mass. The sensitivity of the gyroscope to this acceleration depends on the design and manufacturing precision. However, using data from an independent accelerometer, the system can compensate for this error.
Vibration Rectification Error (VRE): This is another source of error specific to gyroscopes, also known as g-squared error. It is the accelerometer’s response to AC vibration (rectified to DC), which manifests as an abnormal excursion of the accelerometer’s offset. There are multiple mechanisms by which VRE occurs and cannot be compensated in real-time because it is highly application-specific. The developer should determine if the sensor’s VRE is within acceptable limits. Some vibration problems can be helped mitigate by using vibration-damping sensor mounting techniques.
Cross-Axis Sensitivity: At the system level, mechanical misalignment of sensors can also introduce errors. One of these errors is cross-axis sensitivity. This occurs when the actual sensing axis deviates from the intended direction, resulting in a signal from quadrature motion that the sensor should not detect. For example, a sensor that is expected to remain level may still detect gravity if it is not aligned. Misalignment between the accelerometer and gyroscope axes affects the system to compensate for gyroscope g-sensitivity errors.
Off-axis error: Mechanics is also one of the causes of off-axis errors in accelerometers. If the point of impact to the sensor is not at the center of the accelerometer’s test mass, the sensor detects additional acceleration due to the slight rotation of the test mass around the impact line.
Integrated IMU Mitigates Sensor Error Issues
For developers trying to build an IMU with discrete sensors, so many sources of error present a huge challenge. Fortunately, pre-integrated IMUs with six degrees of freedom are widely available, which greatly simplifies the work. Some of these are available in modules, such as the ADIS16465-3BMLZ precision IMU module from Analog Devices and the 6DF-1N6-C2-HWL from Honeywell (Figure 4). Developers simply bolt these devices to the chassis to incorporate them into system designs.
Figure 4: Integrated IMUs such as Honeywell’s 6DF-1N6-C2-HWL help simplify system design by eliminating alignment issues and many other sources of error. A board-mounted BGA IMU is also available. (Image credit: Honeywell Sensing and Productivity Solutions)
Chip and board-mounted precision IMUs are also available on the market, such as the ADIS16500/05/07 series from Analog Devices. These devices are suitable for integration with other sensors and GPS receivers as all-in-one components.
These two types of IMUs help reduce the development effort by eliminating or mitigating many of the potential errors in IMU development. For example, Analog Devices’ ADIS16500/05/07 family integrates a three-axis accelerometer, a three-axis gyroscope, and a temperature sensor in a single BGA package. These devices have built-in calibration and filtering that, in combination with other features, help reduce many sources of IMU error (Figure 5).
Figure 5: An integrated IMU, such as the Analog Devices ADIS16505 shown here, can help simplify system design by reducing many potential sources of error through on-board calibration, filtering, and calibration functions. (Image credit: Analog Devices)
Errors such as cross-axis sensitivity can be addressed in device fabrication. For example, the ADIS16505 limits shaft-to-shaft alignment errors to within 0.25°. This careful alignment and the use of a common sensor reading clock simplifies the process for designers to use accelerometer readings to correct for linear acceleration errors in gyroscopes. A built-in temperature sensor provides support for mitigating the temperature dependence of many error sources.
The internal signal chain of these integrated IMUs provides additional error rejection (Figure 6). The raw sensor information is first passed through a digital filter to remove noise and then through a user-configurable Bartlett window filter. The Bartlett window is a finite impulse response (FIR) averaging filter that uses two cascaded stages.
Figure 6: Integrated IMU devices provide built-in filtering and can compensate for many system sensor errors by applying factory-set calibration parameters. (Image credit: Analog Devices)
The signal then goes through a calibration phase where device-specific corrections are applied based on factory calibration tests run at multiple temperatures spanning the entire operating temperature range of the device. This stage uses matrix multiplication for all six sensor samples simultaneously, and is capable of compensating for accelerometer and gyroscope bias, scale factor, and alignment errors. It also corrects for linear acceleration errors in gyroscopes and axis offset errors in accelerometers.
In addition, a user-selectable tap alignment correction point is provided to adjust the accelerometer output so that it behaves as if it were all located at the same fiducial point in the package. All other factory calibration functions are generally unavailable, but these devices do allow the user to adjust the factory sensor offset compensation with an additional value of their choice.
After calibration correction, the signal is passed through a second digital filter. The decimation filter averages multiple samples to produce the final output for additional noise reduction. The number of samples averaged together depends on the user’s choice of sampling and register update frequency.
One of the few sources of error that the integrated IMU cannot correct is the VRE. Strong vibrations are unavoidable for agricultural machinery, so designers must carefully evaluate the system’s requirements on this issue. Many low-cost IMUs have very poor VREs; some are so bad that the manufacturers are reluctant to state them. To be fair, VRE is not a significant issue in the intended application of these low-cost IMUs. However, devices used in high-vibration environments such as precision agriculture require the lowest possible VRE. For example, the VRE for the ADIS16500 family is approximately 4 x 10-6 (°/sec)/(m/sec2)2. So a sustained 1 g vibration (strong enough to bounce the driver off the seat) would only result in a rotation error of about 1 degree per hour.
An important step in achieving an effective system is that there are no installation, alignment and calibration issues, but this is just the beginning. Developers must still translate inertial measurements into position tracking, resolve differences between dead reckoning and GPS positioning, and understand and mitigate application-specific factors such as the amount and frequency of system shock and vibration during daily use.
If the positioning system is used for automatic or even semi-automatic control of mobile machinery, safety factors are also considered. MEMS sensors can be overwhelmed by excessive shocks. While these devices are usually able to withstand large shocks without damage, if the shock pushes the sensor beyond its limits, it may cause the sensor to temporarily shut down, or to fix the output at the maximum value upon recovery. Systems need to be designed so that this momentary shock does not inadvertently lead to dangerous or annoying system behavior, such as sudden changes in direction or falsely triggering a system safety shutdown.
So it’s best to start with an evaluation board like the Analog Devices EVAL-ADIS2Z (Figure 7). The evaluation board enables developers to use a PC to access device registers and data, and is small enough to easily mount on a representative target machine to collect vibration and motion statistics.
Figure 7: Evaluation boards such as the EVAL-ADIS2Z simplify the experimental phase and are small enough to be mounted on the side of a machine to collect data. (Image credit: Analog Devices)
The evaluation board supports application software for basic demonstration, single register access, and high-speed data acquisition.
Precision agriculture based on satellite navigation offers farmers higher productivity while reducing resource usage. By adding inertial positioning functions, designers can greatly improve the accuracy of positioning, helping farmers achieve plant-level accuracy in farmland management. However, to do this, developers need to address sensor and system error sources in the design. The integrated 6DOF precision inertial measurement unit greatly reduces the development burden by providing careful alignment, filtering, and built-in calibration error correction.