Computers and Electronics in Agriculture 52 (2006) 1–10

Adaptive modeling and control of a manure spreader for precision agriculture
Manu Krishnana , Christopher A. Fostera , Richard P. Strosserb , James L. Glanceya , Jian-Qiao Suna,∗

Department of Mechanical Engineering, University of Delaware, Newark, DE 19716, United States b CNH America LLC, 500 Diller Ave., New Holland, PA 17557, United States Received 16 May 2005; received in revised form 1 November 2005; accepted 28 November 2005

Abstract This paper describes a general modeling and control approach for automating various agricultural machines for precision farming applications. Experimental validation of control designs was performed on a modified New Holland manure spreader. An adaptive numerical modeling approach for describing the system input–output dynamics is proposed, and an optimal control that accounts for the control hardware limits is developed. Field tests have demonstrated the effectiveness of the theoretical development. © 2006 Elsevier B.V. All rights reserved.
Keywords: Adaptive control; System identification; Precision agriculture; Embedded control; Manure spreader

1. Introduction Precision agriculture may be defined as a systems based approach to manage spatial and temporal variability for sustainable profitability (Straub et al., 1998). This approach can not only decrease costs, but can also increase yields. Furthermore, accurately applying chemicals and fertilizers only where needed reduces the potential for ground and surface water pollution. Manure produced by livestock contains valuable nutrients for crops. At the present time existing commercially available non-liquid manure spreading equipment does not control the amount of manure applied per unit area in the field. Additional fertilizers are often applied to ensure the crop production. Excessive applied manure and fertilizer contributes to ground and surface water pollution and also increases the cost of crop production. There is a need to develop an automated spreader in order to achieve consistent and precise application of crop nutrients. At present, none of the major equipment manufacturers offer an automated system capable of interfacing with available non-liquid equipment for precision agriculture applications. The objective of the study presented in this paper is to develop control algorithms for automating a manure spreader made by the New Holland for precision agriculture applications. Straub et al. (1998) described a computer controlled manure spreader developed by John Deere Corporation in collaboration with the University of Wisconsin-Madison, and carried out field tests indicating that the control

Corresponding author. Tel.: +1 302 831 8686; fax: +1 302 831 3619. E-mail address: (J.-Q. Sun)

0168-1699/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.compag.2005.11.005

As discussed in Section 1. The spreader has two augers which convey material to the rear gate. Malgeryd and Wetterberg (1996) presented one of a series of studies to characterize manure as an inhomogeneous material. the control task set for the present study is to attain a specified constant discharge mass per unit area from the spreader taking into account varying speed of the tractor and the material variability of the semi-solid animal wastes. Landry et al. / Computers and Electronics in Agriculture 52 (2006) 1–10 system worked well with lighter and dryer manure.e. In order to develop control algorithms to achieve the above objective. The parameters of the numerical model are updated in real time to account for the time varying and nonlinear properties of the spreader dynamics. New Holland. analytical modeling of such a system is a difficult task. The study reported by Malgeryd (1994) has formed a basis for setting the European standard for manure and slurry spreaders. The objective of the control algorithm is to regulate these two quantities for a pre-determined spreading application density per unit area. we need a relationship between the input. As the spreading proceeds. As an example.e. A numerical regression model is designed to describe the input–output dynamics of the spreader.2 M. and varies from batch to batch. 2. There are inherent difficulties in developing an analytical model for the manure material. i. the auger speed and the gate opening size. The auger speed and the gate opening size can be controlled. the material discharge rate. In the section below. Recall that the material is a highly inhomogeneous mix of liquids and solids with unknown percentage of each phase. Fig. Pennsylvania) is shown in Fig. Magnetic inductive flowmeters are used to measure the manure’s flow rate. i. The weight and viscosity of the material affect the dynamics of the hydraulic system that drives the auger. Two spinning disks spread the material onto the field. The present study also develops an on-line system identification algorithm. Krishnan et al. 1. the weight of the spreader is measured over time and a finite difference method is used to compute the discharge rate. The present study proposes to sense the weight of the manure remaining in the tank and calculate the discharged material. we describe the objectives of the research and a description of the hardware and software system. Note that the auger speed is adjusted by varying the swash plate angle of the hydraulic pump. Specifically. The material is a highly inhomogeneous mix of solids and liquids. including a supervisory control and a control with a Kalman filter and a Smith predictor for time delay. High-performance controllers. Experimental results from field tests are presented to demonstrate the effectiveness of the proposed control system. Then we present a discussion of numerical modeling of the input–output dynamics of the spreader and an experimental validation of the model. Research objectives and system description The manure spreader considered in this study is a variable rate precision machine for animal wastes. A picture of the machine (Model 308 Spreader. The adaptive optimal control for regulating the discharge rate of the spreader is subsequently developed. In their control system. 2005) recently studied physical and rheological properties of manure and investigated the effectiveness of conveying systems for manure spreaders. and the output. 1. the amount of material remaining in the tank changes. we must first develop a dynamic model for the spreader. A picture of the spreader and tractor considered in the present study. All these factors attribute to a nonlinear and time-varying dynamics of the spreader. (2001). (2004. have been developed by Munack et al. One of the problems they faced is how to measure the manure discharge. . It is assumed that the spreading width of the material is a constant. The remainder of the paper is organized as follows.

na ) and bj (j = 1. The coefficients ai (i = 1. 1985) can be used to adjust the coefficients of the ARX model and minimize the error between the prediction of the numerical model and the real measurement. A properly identified ARX model will accurately represent the dynamics of the system over a short time interval and will not be valid for the entire history of the spreading task from a full tank to empty. 1991). where dk is the measured output and yk is the predicted output. / Computers and Electronics in Agriculture 52 (2006) 1–10 3 In this study. Ljung. 3. where β is an adaptation gain parameter. nb ) are undetermined. Pennsylvania).1. straw and mud. The controller has an Intel processor 80C196CB with 100 Hz frame rate. the ARX model is valid for linear dynamic systems. Experimental validation of the LMS algorithm We have selected a simple ARX model for the spreader given by yk = a1 y(k−1) + a2 us (k−1) + a3 u(k−1) . The LMS algorithm for updating the undetermined coefficients in order to minimize J(k) is given by w(k+1) = wk + βek uk . The on-line model is obtained by using the New Holland controller (HCM. 2 and 3 show the measured data from spreading a full load of farmyard manure consisting of cow urine and feces. and found that this model describes the system with a good balance of accuracy and efficiency. 3. The online modeling algorithm determines the coefficients and approximates the numerical model to the measured system dynamics in some optimal manner.M. Efficient real-time adaptive algorithms will be needed for this task. the ARX model is also known as an infinite impulse response (IIR) filter (Haykin. Diaz and Desrochers. k wT k (2) where is a vector consisting of the undetermined coefficients at the kth time step and uk is a vector consisting of both the past history of yk and the control inputs. It communicates with a laptop computer via RS232 at 9600 baud. 1988. The present spreader system is time varying and nonlinear. 1986. We have carried out experiments to compare this simple model with more complicated ones. Adaptive algorithm In signal processing. A popular steepest gradient descent method known as the least mean square (LMS) algorithm (Widrow and Stearns. Strictly speaking. A digital second order IIR low pass filter of bandwidth 2 Hz programmed in the C language is . Figs. New Holland. A performance index can be defined as J(k) = e2 . . . (1) The current output yn is assumed to be a function of a finite history of output values y(n−1) to y(n−na ) and the delayed input u(n−nk ) to u(n−nk −nb +1) . Therefore the ARX model must be updated frequently during spreading. and is sufficient for our work.2. we propose to develop an on-line numerical model of the input–output relationship. On-line system modeling The on-line system model fits the experimental data to a pre-determined numerical model with undetermined coefficients. It is given in a general form as yn + a1 y(n−1) + · · · + ana y(n−na ) = b1 u(n−nk ) + · · · + bnb u(n−nk −nb +1) . g g (3) (4) where us (k−1) denotes the swash plate angle that regulates the auger speed and u(k−1) is the rear gate opening. . 3. This controller can interface with and control a wide range of equipment including variable rate applicators for precision agriculture. . . known as the system transfer function. The estimation error is ek = dk − yk . . . A very common numerical model that can describe a large class of dynamic systems is the autoregressive model with exogenous inputs (ARX) (Billings. We write the ARX model in a vector notation as k yk = wT uk . . 1987). Krishnan et al.

average a1 = 1. The parameters a2 and a3 are negative. The prediction and the measurement data are overlapped. This is physically reasonable since in the absence of the control. max(a2 ) = −0.0111.e. 3 and 4.6952 kg/bit.0094. During spreading. 2. Time history of the mass remaining in the tank during the test for system identification. Time histories of the swash plate opening (top) and the rear gate opening (bottom) during the test for system identification. 2. The parameter a1 is nearly equal to one.4 M. used to block noise in the weight signal. min(a1 ) = 0. The weight signal is sampled at a rate of 100 Hz in real time. The ranges of these coefficients are as follows: max(a1 ) = 1. average a2 = −0. and the resolution of the data acquisition is 26. i. min(a2 ) = −0. 3.0083.0011. The model prediction is seen to be quite accurate. . The purpose of doing so is to create a set of data from one test run that excites as much of the system dynamics as possible. / Computers and Electronics in Agriculture 52 (2006) 1–10 Fig. the material remaining in the tank yk decreases.0063. min(a3 ) = −0. Fig. the material remaining in the tank is unchanged. When the control inputs are greater than zero. Four different gate openings at 25.0080. Krishnan et al.1828 mV/kg. when the auger speed is zero and the gate is closed. average a3 = −0. 50. max(a3 ) = −0. The weight sensitivity is 0. It is again physically reasonable that a2 and a3 are negative. the gate positions are fixed for a period of time during which the swash plate is swept from being completely closed to being fully open to adjust the auger speed. 75 and 100% are used as shown in Fig. The results of the model prediction and the parameter adaptation with the LMS algorithm are shown in Figs.0047.0045.9881.

4. 4. 1991). we have ref ref T ek = a1 e(k−1) + a2 us (k−1) + a3 u(k−1) + a1 y(k−1) − yk = a1 e(k−1) + c(k−1) u(k−1) + Gk−1 ≡ f (k−1) (e(k−1) . (4) and (5). The integration of the control loop and the parameter updating loop is natural and is coded in the software. 1991) . Gk−1 ). The control is calculated based on the numerical model as if it were the true plant. (5) g From Eqs. the parameters of the numerical model and the plant output. We developed an optimal control that could account for the time-varying parameters of the model and the control hardware limits in a systematic way. A block diagram of such a controller is shown in Fig. . Krishnan et al. the resulting controller that contains the updating algorithm is called a self-tuning controller (Slotine and Li.1. u(k−1) . The estimator for updating the model parameters needs two inputs. This section focuses on a discussion of the control algorithm design. Control algorithm The control algorithm design depends on the system model. The block diagram for the self-tuning control. Variation of the parameters of the numerical ARX model with time. / Computers and Electronics in Agriculture 52 (2006) 1–10 5 Fig. The controller takes three inputs: the reference signal. 5. 5. Unconstrained optimal control ref Let yk denote the desired discharge time history. This is called the “certainty equivalence” principle (Slotine and Li. 4. Define an error such that ref ek = yk − yk . namely the control and the plant output. (6) Fig.M. Since the IIR filter for modeling the system transfer function is updated as the spreading goes on.

g s. However. Let Uk denote the swash plate and gate opening control elements of the vector in Eq. (12) It can be shown that the uk that minimizes w also minimizes the left hand size of the inequality (11) (Lewis and Syrmos.g uk = Uk . The optimal control can be found from the inequality by considering an auxiliary problem of minimization of the following quadratic form: 1 w = 2 (uk + R−1 ck λk+1 )T (uk + R−1 ck λk+1 ). respectively. the matrix R cannot be zero. uk ).g s. we need to use the Pontryagin’s minimum principle (Lewis and Syrmos. uk ) + λk+1 f k (ek . Uk ≥ ukmax (14) The middle branch of the solution is the same as that in Eq.g ∗s.g s. ⎪ ⎪ ⎪ s. 4.e. We have found that this so-called one step optimal control is sufficient for our purpose. Let the lower and upper bounds of the control be denoted by ukmin and ukmax . known as the terminal time. The k unconstrained optimal solution (10) is valid when the bound is not exceeded. s. k+1 g (11) The inequality holds for all admissible values of uk . Recall that the range of us and uk is finite.g . eN ) + k=i Lk (ek .g s. To account for the bounds on the controls.g ukmax . The target set (eN . more research needs to be done to study the effect of multiple step control designs. / Computers and Electronics in Agriculture 52 (2006) 1–10 where c(k−1) = {a2 . 1995) . the solution given by Eq. .6 M. T u(k−1) = {us (k−1) .g s. (10) is optimal. Range limited optimal control We shall continue the study with the one step optimal control. we obtain the range constrained optimal control as ⎧ s. This leads to the following inequality ∗ for determining the optimal control uk : 1 ∗T ∗ 2 uk Ruk 1 T T ∗ T + λ∗ ck uk ≤ 2 uk Ruk + λk+1 ck uk . (10). when the system operates within the physical limits of the controls. In other words. (10) More complicated and effective controls can be obtained by taking N > k + 1 at the price of more computational effort.g ⎪ ukmin . ukmin < Uk < ukmax . The scalar q and the 2 × 2 matrix R weigh the state and N k the control effort. N) for the control is defined as eN = 0.2.: T Uk = [−(R + ck ck sN )−1 ck sN (a1 ek + Gk )]s.g 1995). and has to be positive definite.g s. Lewis and Syrmos (1995) have shown that after going through the steps of optimal control solutions and taking i = k and N = k + 1.g (13) After several algebraic steps. (10). g ref ref Gk−1 = a1 y(k−1) − yk . a3 }T . (9) where λk+1 is a Lagrange multiplier. s. Note that when the number of inputs is greater than the number of outputs. The Hamiltonian of the system is given by H k = Lk (ek . uk ) = 2 (qe2 + uk Ruk ). u(k−1) } . we obtain the unconstrained optimal control solution as ∗ T uk = −(R + ck ck sN )−1 ck sN (a1 ek + Gk ). eN ) = 2 sN e2 and Lk (ek .g s. (8) 1 1 T where φ(N. i.g s. uk . The initial time step is denoted i and N is the number of steps to control the system. Gk ).g ⎩ s. Uk ≤ ukmin ⎪ ⎪ ⎨ s. Krishnan et al. (7) Define a performance index with a terminal cost: N−1 Ji = φ(N.

| Ui.M. 6. a rate saturation occurs. / Computers and Electronics in Agriculture 52 (2006) 1–10 7 4.g s. When the controller requires the system to change faster than the physical rate limit.g . we obtain the optimal control under both range and rate saturation limits. a proper choice of the reference input will keep the control less saturated.g (17) (18) where uk max denotes the physically allowable maximum rate of change of the swash plate and rear gate controls over one sample interval. we once again invoke the Pontryagin’s minimum principle and consider the increment u(k−1) such that uk = u(k−1) + u(k−1) as the control variable.k | ≤ uk max ∗s. (a) Unfiltered weight readings from the sensor. Upon finishing the controller design. To account for the rate limits. Krishnan et al.g s.g u(k−1) = . Fig. (b) Low-pass filtered weight signal. (15) The optimal control increment can be found from the inequality by considering an auxiliary problem of minimization of the following quadratic form: 1 w = 2 (u(k−1) + u(k−1) + R−1 ck λk+1 )T · (u(k−1) + s. (16) Define an increment by using Eq. By minimizing w with respect to u(k−1) . How to specify the proper reference is another technical issue not addressed in the paper. By combining Eqs. Applying the Pontryagin’s minimum principle in terms of the control increment. we have another inequality 1 ∗ 2 (u(k−1) + ∗ ∗ u(k−1) )T R(u(k−1) + ∗ T ∗ u(k−1) ) + λ∗ ck (u(k−1) + k+1 ∗ u(k−1) ) 1 ≤ 2 (u(k−1) + u(k−1) )T R(u(k−1) + T u(k−1) ) + λk+1 ck (u(k−1) + u(k−1) ).g u(k−1) + R−1 ck λk+1 ). A discussion The preceding optimal control solutions are a function of the reference input and the actual output of the system.k k k max k max s. (13) as Uk = Uk − U(k−1) . Time history of the mass remaining in the tank during a constant discharge rate spreading. (10) and (18). Rate limited optimal control The range limited optimal control problem implies that the controls can be instantly switched from one level to another.3.g ) us. This is of course not realistic since a physical device always takes a finite time to change and has inherent delays. Given the hardware limitations. we obtain the optimal control increment as ⎧ s. More discussions of such optimal control problems can be found in Kobs and Sun (1997).g ⎨ Uk . . | U | > us.4.g s. 4.g i. The top branch of the solution matches the range limited optimal control and the lower branch is the rate saturated control. ⎩ sgn( U s. Dashed lines in both figures indicate the desired discharge when the tractor speed is constant.

8 M. . The delay can make the control saturation worse. Measured discharge rate obtained by differentiating the weight signal from the constant discharge density spreading experiment. Fig. Krishnan et al. More results and discussions can be found in Krishnan (1999). The spreading width of the machine is 15 m. the speed of the tractor v (kph) and DA can be found as Dt = 4. A relationship between the discharge rate Dt per unit time (kg/s). Time history of the swash plate and the rear gate opening during a constant discharge density spreading experiment. Fig. / Computers and Electronics in Agriculture 52 (2006) 1–10 we should also point out the factors that affect the real-time performance of the system. Experimental results Many tests have been done to validate the theoretical development of this work. 5. These include the delay in actuation and the finite sample time. Future studies will examine the effect of these factors. Let DA denotes the required mass per unit area (kg/m2 ). 7. 8.167 × v · DA (kg/s). This section presents the results of one typical spreading experiment. (19) Since the discharge rate is constant when the tractor speed is constant. the material remaining in the tank is a linearly decreasing function of time. 6 shows the time history of the mass remaining in the tank during a constant discharge Fig.

Since the lateral spreading width is 15 m..M. 9. S. 277–288. Krishnan. Agnew. The adaptive self-tuning optimal control algorithm can cope with various hardware limits. Adaptive Filter Theory. M. J. Seymour.894 m/s. Prentice Hall. 21 (2). Landry.. New York. Shaun A.2 kph. The authors are deeply indebted to Walter V. The present approach provides a promising methodology for automating machines for precision agricultural applications. Eng. Performances of conveying systems for manure spreaders and effects of Hopper geometry on output flow. 2005. Krishnan et al. 1999. T.L. New Jersey. Many individuals of New Holland North America have contributed significantly to the success of the project. Agric.A. rate test. Optimal Control. Sun. 243–249. On average...M. Automatica 24.. H. Haykin.Q. J. Piron. Non-linear adaptive control of precision farming machines. 2004.. which is 0. Fig.. Int. the tractor speed is 3. 159–166. Agric.. System Identification—Theory for the User.41 m2 . The actual measurement is in good agreement with the reference input. Roberge. H. Eng.. Acknowledgements The authors would like to acknowledge the support from the Delaware Research Partnership program and New Holland North America. The theoretical development has been validated by extensive experimental results. Lewis. Pype... 8 shows the discharge rate obtained from the weight sensor. 1995. Berger for all the help and support. Physical and rheological properties of manure products. Adams.. F. 1986. E. J. John Wiley and Sons. . Prentice Hall. University of Delaware. Syrmos.783 kg/m2 with a fluctuating error partly due to the oscillations in the discharge rate. Richard K. Inc. James R. Ljung.. References Billings. L. Control 44. V. Conclusions We have presented a general modeling and control approach for precision agricultural applications by using a New Holland spreader as an example. Note that there is significant noise in the measurement due to vehicle dynamics and electronic disturbances.. Englewood Cliffs. 1987. Lague. H. M.L. 20 (3).783 kg/m2 . Lague.783 kg/m2 . 1997. Diaz. Discharge density per unit area averaged over 30 s from the constant discharge density spreading experiment. J. Fig. The setpoint is 0. Modelling of nonlinear discrete-time systems from input–output data. 803–822. The system tracks the setpoint of 0. 6. Fig. A predicition-error and stepwise-regression estimation algorithm for non-linear systems. A nonlinear variable stiffness feedback control with tuning range and rate saturation... Kobs. New York. Landry.. Day. 1991. The fast oscillations in the discharge rate are due to differentiation of noisy data. S. 205 (2). / Computers and Electronics in Agriculture 52 (2006) 1–10 9 Fig.A. 9 shows the averaged mass per unit area over 30 s. A. Appl. Sound Vib. 1988. C. Roberge. 7 shows the time history of the rear gate opening and the swash plate angle during a constant discharge density experiment with DA = 0. the average area covered in 1 s is 13. The low pass digital filter designed for the weight sensor is quite effective in reducing measurement noise. Appl.. 629–641.. M. C. Desrochers. The numerical input–output modeling approach can handle a wide range of variations in manure materials and the complicated nonlinear dynamics of the machine. MS Thesis. and John G.

Technical Report..10 M. Wetterberg. Widrow. W. Li. Agrophys. / Computers and Electronics in Agriculture 52 (2006) 1–10 Malgeryd. J.. Computer controlled manure spreader. Adaptive Signal Processing.. R. C. 8. Speckmann. Slotine... New Jersey.E. 9. H. Prentice-Hall. B. Res. C. A high-performance control system for spreading liquid manure.-J. Munack. University of Wisconsin-Madison. .. Silbernagel. B. 1998.. Straub. S. Control Eng. Malgeryd. Holmes... Applied Nonlinear Control. E. Physical properties of solid and liquid manures and their effects on the performance of spreading machines. College of Agricultural and Life Sciences.. Prentice Hall. 64 (4). Pract. 2001. Agric. Int. Krishnan et al. 1985. 387–391.. J.. Stearns. J. Manure characterization. A. 289–298. 1996. 1994.D.. Inc. J. Buning. Englewood Cliffs.. Englewood Cliffs. 1991. New Jersey. 93–101. Eng.

Sign up to vote on this title
UsefulNot useful