sensors Article

Closed-Loop Control of Chemical Injection Rate for a Direct Nozzle Injection System Xiang Cai 1, *, Martin Walgenbach 2 , Malte Doerpmond 2 , Peter Schulze Lammers 2 and Yurui Sun 3 Received: 19 October 2015; Accepted: 14 January 2016; Published: 20 January 2016 Academic Editor: Vittorio M. N. Passaro 1 2 3

*

School of Information Science and Technology, Beijing Forestry University, Beijing 100083, China Department of Agricultural Engineering, University of Bonn, 53115 Bonn, Germany; [email protected] (M.W.); [email protected] (M.D.); [email protected] (P.S.L.) College of Information & Electrical Engineering, China Agricultural University, Beijing 100083, China; [email protected] Correspondence: [email protected]; Tel.: +86-10-6233-6702

Abstract: To realize site-specific and variable-rate application of agricultural pesticides, accurately metering and controlling the chemical injection rate is necessary. This study presents a prototype of a direct nozzle injection system (DNIS) by which chemical concentration transport lag was greatly reduced. In this system, a rapid-reacting solenoid valve (RRV) was utilized for injecting chemicals, driven by a pulse-width modulation (PWM) signal at 100 Hz, so with varying pulse width the chemical injection rate could be adjusted. Meanwhile, a closed-loop control strategy, proportional-integral-derivative (PID) method, was applied for metering and stabilizing the chemical injection rate. In order to measure chemical flow rates and input them into the controller as a feedback in real-time, a thermodynamic flowmeter that was independent of chemical viscosity was used. Laboratory tests were conducted to assess the performance of DNIS and PID control strategy. Due to the nonlinear input–output characteristics of the RRV, a two-phase PID control process obtained better effects as compared with single PID control strategy. Test results also indicated that the set-point chemical flow rate could be achieved within less than 4 s, and the output stability was improved compared to the case without control strategy. Keywords: variable-rate application; closed-loop control; pulse width modulation; direct nozzle injection; thermodynamic flowmeter

1. Introduction Previous studies have pointed out that there are advantages to realizing site-specific and variable-rate application of pesticides [1–3]. Broadcast application of chemicals without concentration variance and spatial selection results in fields being sprayed without distinction. On the one hand, over-application of chemicals not only increases the cost of crop production, but also increases the risk of environmental contamination and the exposure hazards to operators. On the other hand, insufficient application of chemical may cause a decrease of crop yield, or make the weeds chemical-resistant. Conventional boom sprayers mix a measured quantity of pesticide with a defined amount of carrier in a tank. Chemical application rates can only be varied by changing the liquid delivery pressure, but the pressure changes also unexpectedly affect the droplet size spectra and spray distribution pattern of the nozzles [4]. However, by using a direct injection system, chemical and carrier are kept separately and mixed together when sprayed. The delivery pressure of carrier stream is maintained at the same level, as are the pressure at the nozzles, the droplet size spectrum and the spray distribution pattern [5]. Through changing the chemical injection volume, the chemical application rate can be changed.

Sensors 2016, 16, 127; doi:10.3390/s16010127

www.mdpi.com/journal/sensors

Sensors 2016, 16, 127

2 of 11

Sensors 2016, 16, 127

2 of 11

spray distribution pattern [5]. Through changing the chemical injection volume, the chemical application rate can be changed. GopalaPillaietetal. al.tested testedan anelectrical electricalflow flowcontrol controlsystem systemfor forsite-specific site-specificherbicide herbicideapplications applications GopalaPillai and its steady state performance [6]. Solenoid valves were used and controlled by a PWM signal in the and its steady state [6]. Solenoid valves were used and controlled by a PWM signal in system. TheThe chemical flow raterate waswas varied by changing thethe duty cycle of the PWM signal from 10% the system. chemical flow varied by changing duty cycle of the PWM signal from to 100%. However, this this was was an open-loop system. The The precision and and stabilization of chemical flow 10% to 100%. However, an open-loop system. precision stabilization of chemical rate were not discussed. Han et al. also tested a DIS with solenoids controlled by a PWM signal in an flow rate were not discussed. Han et al. also tested a DIS with solenoids controlled by a PWM signal control system [7]. They found the flow rate rate control errorerror ranged fromfrom ´15% to +12%. inopen-loop an open-loop control system [7]. They found the flow control ranged −15% to +12%. Effective flow rate control requires fast and accurate metering of chemicals for the direct Effective rate control requires fast and accurate metering of chemicals for theinjection direct sprayer, sprayer, which could accomplished by usingbya using positive driven pump a closed–loop control injection whichbecould be accomplished a positive driven or pump or a closed–loop systemsystem with the help of help flowmeters [8]. Frost al. developed a metering system in whichinawhich metered control with the of flowmeters [8].etFrost et al. developed a metering system a flow of water was used to control the flow rate of chemical [9]. That was a kind of central direct metered flow of water was used to control the flow rate of chemical [9]. That was a kind of central injection system (CDIS), wherewhere chemical was injected into the downstream from from the main direct injection system (CDIS), chemical was injected intosystem the system downstream the carrier tank tank and prior to branching of theofdistribution hoseshoses carrying the solution to different boom main carrier and prior to branching the distribution carrying the solution to different sections. The advantage of thatofsystem is that is thethat chemical was metered by a closed-loop contoller boom sections. The advantage that system the chemical was metered by a closed-loop so that it could provide a wide range aof wide chemical doseofrates precisely, while the precisely, disadvantage wasthe the contoller so that it could provide range chemical dose rates while long lag time from a change in the chemical flow rate to the corresponding changes in its concentration disadvantage was the long lag time from a change in chemical flow rate to the corresponding at the spraying points [10]. at the spraying points [10]. changes in its concentration Tosolve solvethe thelong longlag lagtime timeproblem, problem,Vondricka Vondrickaand andSchulze SchulzeLammers Lammersproposed proposedaadirect directnozzle nozzle To injection system (DNIS) concept, in which the point where the chemical was injected into the system injection system (DNIS) concept, in which the point where the chemical was injected into the system waschanged changedtoto nozzles improvement shortened the chemical flowand path and was bebe at at thethe nozzles [11].[11]. ThisThis improvement shortened the chemical flow path hence hence reduced time significantly. Subsequently, their team focused onaspects many aspects of this reduced the lag the timelag significantly. Subsequently, their team focused on many of this DNIS. DNIS. Vondricka addressed the problem of mixture homogeneity and Doerpmundassessed assessedthe the Vondricka addressed on theonproblem of mixture homogeneity and Doerpmund cleanabilityofofthis thisDNIS DNIS[12–14]. [12–14].However, However,there thereisisstill stillaaneed needtotoinvestigate investigatethe theperformance performancefor for cleanability DNIS metering and chemical flow rate control, which is the premise for realizing site-specific and DNIS metering and chemical flow rate control, which is the premise for realizing site-specific and variable-rateapplication applicationof ofchemicals. chemicals. variable-rate Theobjective objectiveof of this this article article was to test the The the injection injection uniformity uniformityof ofRRVs RRVswithin withina aboom boomsection sectionof the DNIS, to present a closed-loop control method to meter and stabilize the chemical application rate of the DNIS, to present a closed-loop control method to meter and stabilize the chemical application withwith the help of a thermodynamic flowmeter, and toand evaluate the performance of the of control system rate the help of a thermodynamic flowmeter, to evaluate the performance the control and theand DNIS an EC system theby DNIS by sensor. an EC sensor. Materials andMethods Methods 2.2.Materials and 2.1. Description of RRV 2.1. Description of RRV A new rapid reacting magnetic valve design was developed to inject pesticides. A schematic of its A new rapid reacting magnetic valve design was developed to inject pesticides. A schematic of structure is shown in Figure 1. It mainly consists of a cylindrical metal cavity, an induction copper coil its structure is shown in Figure 1. It mainly consists of a cylindrical metal cavity, an induction copper (resistance: ca. 3~5 Ω), metal ball, valve seat and rubber washers. coil (resistance: ca. 3~5 Ω), metal ball, valve seat and rubber washers.

Figure1.1.Structure Structureofofthe theRRV. RRV. Figure

Sensors 2016, 16, 127 Sensors 2016, 16, 127

3 of 11 3 of 11

When magnetic field is generated generated inside inside When electric electric current current flows flows through through the the copper copper coil, coil, aa strong strong magnetic field is the principle, then then the the metal the metal metal cavity cavity based based on on the the electromagnetic electromagnetic induction induction principle, metal ball ball is is pushed pushed aside aside by the magnetic field. Thus an access for the chemical from the inlet to the orifice of the valve seat by the magnetic field. Thus an access for the chemical from the inlet to the orifice of the valve seat is is formed, formed, through through which which pressurized pressurized chemical chemical can can be be injected. injected. If If there there is is no no electric electric current current in in the the coil, coil, the pressurized chemical valve seat seat causing causing aa tight tight contact contact between between the pressurized chemical pushes pushes the the metal metal ball ball onto onto the the valve the ball and the rubber washer, which finally shut down the flux of chemical. In this case, the ball and the rubber washer, which finally shut down the flux of chemical. In this case, RRV RRV is is switched off. switched off. The is powered The RRV RRV is powered by by aa PWM PWM signal signal at at aa frequency frequency of of 100 100 HZ HZgenerated generatedby byaaFPGA FPGAcontroller. controller. The pulse width during which the RRV remains open in one PWM period determines the injection injection The pulse width during which the RRV remains open in one PWM period determines the time and hence hence the theinjection injectionvolume volumeofofchemical. chemical.Thus, Thus, adjusting length of the pulse width, time and byby adjusting thethe length of the pulse width, the the chemical injection rate could be varied. chemical injection rate could be varied. 2.2. Experiment Setup The proposed DNIS was 21 m of length, and was equipped with 42 direct injection units which were divided into seven boom sections. Within one section, six direct injection units were controlled by a same PWM signal as shown in Figure 2a. Figure 2b shows the structure of one direct injection unit, which includes a carrier valve, a RRV, RRV,aamixing mixingchamber, chamber,an anEC ECsensor, sensor,and andaaspray spraynozzle. nozzle.

(a)

(b)

Figure Schematic structure of the setupsetup with awith boom (b) one of injection Figure 2. 2.(a)(a) Schematic structure of experimental the experimental a section; boom section; (b)the one of the units. injection units.

Figure 3 shows test bench to implement our closed-loop control strategy for metering the Figure 3 shows test bench to implement our closed-loop control strategy for metering the chemical chemical injection rate of the RRV. It is composed of a direct injection unit, flowmeter, two pressure injection rate of the RRV. It is composed of a direct injection unit, flowmeter, two pressure regulators, regulators, and a controller. The EC sensor mounted between mixing chamber and nozzle in injection and a controller. The EC sensor mounted between mixing chamber and nozzle in injection unit unit was used to indicate the concentration of the spraying solution, it could also be used to verify was used to indicate the concentration of the spraying solution, it could also be used to verify the the control effect. control effect. The flowmeter is based on the thermodynamic principle. When a liquid medium flows through, The flowmeter is based on the thermodynamic principle. When a liquid medium flows through, the sensor generates a heat pulse internally. The heat is then conducted away by the medium flux the sensor generates a heat pulse internally. The heat is then conducted away by the medium flux resulting in a cooling down of the sensor. The temperature within the sensor is measured and resulting in a cooling down of the sensor. The temperature within the sensor is measured and compared compared with the temperature of the medium, and the flow rate can be derived from the with the temperature of the medium, and the flow rate can be derived from the temperature difference. temperature difference. Theoretically, the flowmeter is independent of the viscosity of the Theoretically, the flowmeter is independent of the viscosity of the agricultural chemicals. agricultural chemicals. The chemical applied in the experiments was not real, and it was simulated by a solution consisting The chemical applied in the experiments was not real, and it was simulated by a® solution of (in proportion by weight) 10% LUVITEC® powder, 3% salt and 87% water. The LUVITEC powder consisting of (in proportion by weight) 10% LUVITEC® powder, 3% salt and 87% water. The was used ®as additive to adjust the solution viscosity in the range of 230–240 mPa¨ s. Salt made the LUVITEC powder was used as additive to adjust the solution viscosity in the range of 230–240 mPa·s. Salt made the solution electrical conductive, so the EC sensor could detect the concentration

Sensors 2016, 16, 127

4 of 11

Sensors 2016, 16, 127

4 of 11

solution electrical conductive, so the EC sensor could detect the concentration of the solution. The of the solution. The more electroconductive the solution at the nozzle was, the more chemical the more electroconductive the solution at the nozzle was, the more chemical the RRV injected. Moreover, RRV injected. Moreover, the stability of EC values indicated the stability of the chemical flow rate. the stability of EC values indicated the stability of the chemical flow rate.

Figure3.3.Setup Setup controllingpesticide pesticideinjection injectionrate ratebybyclosed-loop closed-loopmethod. method. Figure forfor controlling

2.3.Uniformity UniformityTest Testofof6 6RRVs RRVsininOne OneBoom BoomSection Section 2.3. Withina aboom boomsection, section,six sixinjection injectionunits unitsare arecontrolled controlledbybyone onePWM PWMsignal signal(see (seeFigure Figure2a), 2a),soso Within theyare areexpected expectedtotoinject injectchemical chemicalinina asame samerate, rate,but butbecause becausethe thechemical chemicalloses losespressure pressurealong alongthe the they fluxdirection directionin inthe the boom, boom, this results units being different, andand there exist flux results in inthe thepressures pressuresatatinjection injection units being different, there individual differences among thesethese RRVs,RRVs, so thesochemical injection rates are notare consistent in practice. exist individual differences among the chemical injection rates not consistent in Therefore experiments to assessto the uniformity of six RRVs in one boom section when theywhen workthey in an practice. Therefore experiments assess the uniformity of six RRVs in one boom section openincontrol loop wereloop conducted. In this experiment, water andwater chemical adjusted work an open control were conducted. In this experiment, andpressure chemicalwere pressure wereat 3 bar and and the pulse of thewidth PWMofsignal was set at 1000 volumes adjusted at 73 bar, bar respectively, and 7 bar, respectively, andwidth the pulse the PWM signal wasµs. setThe at 1000 μs. of chemical by theinjected six injection units 3 min were three times repeatedly, the The volumes injected of chemical by the six in injection unitssampled in 3 min were sampled three and times the averageand injected volume per minute the standard deviation calculated as the injection repeatedly, the the average injectedand volume per minute and were the standard deviation were rate and error respectively. calculated as thebar, injection rate and error bar, respectively. 2.4.Calculating Calculatingthe theRange RangeofofChemical ChemicalInjection InjectionRate Rate 2.4. Assumethat thatthe theDNIS DNIS(with (witha aboom boomofof2121mmininlength lengthand andwith with4242injection injectionunits) units)was wasused usedfor for Assume spraying in a field, and each injection unit was supposed to inject chemical in a same rate of r mL/min. spraying in a field, and each injection unit was supposed to inject chemical in a same rate of r mL/min. thetractor tractordrove drovewith withspeed speedofofv vm/h, m/h, after t hours,the thesprayed sprayedarea areawould wouldbebeS Shahaand andthe thetotal total IfIfthe after t hours, volumeofofchemical chemicalsprayed sprayedwould wouldbebeWWmL. mL.The Thethe thefollowing followingequations equationscould couldbe bederived: derived: volume

4242ˆrtrtˆ60 60=“WW

(1)(1)

21  vt = 10000  S

(2) (2)

21 ˆ vt “ 10000 ˆ S Equation (1) divided by Equation (2) gives the chemical injection rate r:

Sensors 2016, 16, 127 Sensors 2016, 16, 127

5 of 11 5 of 11

r=

vW

Equation (1) divided by Equation (2) gives 1200000 the chemical S injection rate r:

(3)

Empirically, the tractor speed v ranged from 6vW km/h to 10 km/h, chemical per hectare needed in (3) r“ the field (W/S) was assumed from 20 mL/ha to 4000 mL/ha. Accoording to Equation (3), the injection 1200000S rate of each RRV should range from 0.1 to 33.3 mL/min, which could cover most spraying Empirically, the tractor speed v ranged from 6 km/h to 10 km/h, chemical per hectare needed in applications. the field (W/S) was assumed from 20 mL/ha to 4000 mL/ha. Accoording to Equation (3), the injection rate of each RRVControl should range from 0.1 to 33.3 mL/min, which could cover most spraying applications. 2.5. Closed-Loop 2.5. Closed-Loop Control The chemical injection rate from the RRV is susceptible to factors such as the delivery pressure of both carrier and chemical, chemical viscosity, and any unexpected external disturbances, so the The chemical injection rate from the RRV is susceptible to factors such as the delivery pressure of actual flow rate injected by the RRV deviates from the desired value while spraying because it is not both carrier and chemical, chemical viscosity, and any unexpected external disturbances, so the actual possible to keep the operation conditions constant. The deviation would not be corrected flow rate injected by the RRV deviates from the desired value while spraying because it is not possible automatically in a system with an open-loop control strategy, but a closed-loop controller could to keep the operation conditions constant. The deviation would not be corrected automatically in compensate for the deviation and keep the flow rate stable, thus realizing a more precise application a system with an open-loop control strategy, but a closed-loop controller could compensate for the of chemical. deviation and keep the flow rate stable, thus realizing a more precise application of chemical. Figure 4 shows the block diagram of a closed-loop control system for the chemical injection rate, Figure 4 shows the block diagram of a closed-loop control system for the chemical injection rate, based on the PID method. Symbol E represents the deviation between the set-point flow rate Qt and based on the PID method. Symbol E represents the deviation between the set-point flow rate Qt and the actual flow rate Q measured by a flowmeter. The controller adjusts the pulse width P of the PWM the actual flow rate Q measured by a flowmeter. The controller adjusts the pulse width P of the PWM signal to minimize E. signal to minimize E.

Figure 4. 4. Block Block diagram diagram of of the the closed-loop closed-loop control. control. Figure

2.6. Experimental Procedures 2.6. Experimental Procedures The The following followingexperiments experimentswere wereperformed: performed: 1.

2. 3. 4.

1. RRV was tested, and relationshipbetween betweenthe theinput inputpulse pulsewidth widthPPand and the the output output TheThe RRV was tested, and thethe relationship chemical injection investigated; in this experiment, water pressure set3 at 3 chemical injection raterate waswas investigated; in this experiment, the the water pressure waswas set at bar, air pressure for pushing the chemical adjusted to 76,and 7 and bartotoobtain obtain the the andbar, theand air the pressure for pushing the chemical was was adjusted to 6, 8 8bar input-output characteristics of RRV the under RRV under three pressure differences and to input-output characteristics of the three pressure differences and hence to hence compare the optimal operatingfor conditions thecompare optimal operating conditions the RRV. for the RRV. 2. The uniformity of six RRVs controlled bysame the same was 1000 μs) The uniformity of six RRVs controlled by the PWMPWM signalsignal (pulse(pulse width width was 1000 µs) within boom section onewithin boom one section was tested.was tested. 3. The flowmeter was calibrated. The amountofofinjected injectedchemical chemicalinin33min minwas was weighed weighed three three The flowmeter was calibrated. The amount times, an the average value per minute was calculated as the injection rate. times, an the average value per minute was calculated as the injection rate. 4.A single A single module applied in control the control system at first. Subsequently a two-phase PIDPID module waswas applied in the system at first. Subsequently a two-phase PID PID control strategy was used that managed to improve the control effect in accordance control strategy was used that managed to improve the control effect in accordance withwith the the input-output characteristic of the RRV. input-output characteristic of the RRV.

Sensors 2016, 16, 127 Sensors 2016, 16, 127

6 of 11 6 of 11

Sensors 2016, 16, 127 6 of 11 3. Results and Discussion 3. Results and Discussion 3.1. Tests3.ofResults the RRV and Discussion 3.1. Tests of the RRV Figure 5 shows the injection rate of the RRV as a function of the pulse width under three pressure 3.1. Tests of the RRV Figure 5 shows the injectiontests, rate of the RRV as a function of the width three pressure differences. Through experimental it was determined that an air pulse pressure of 7under bar for pushing showsexperimental the injection of the RRV asoptimal a function of thean pulse Through it was determined that air pressure of three 7 barpressure forofpushing the differences. chemicalFigure and a5 water pressure of rate 3tests, bar were the choice for thewidth bestunder performance the differences. Through experimental tests, it was determined that an air pressure of 7 bar for pushing chemical waterdemonstrate pressure of 3that barRRV werehas thea optimal choice rate for the best performance of the RRVthe and nozzle. and The acurves large injection range from less than 0.1 the chemical and a curves water pressure of 3 bar were thehas optimal choice for therate bestrange performance of the RRV and nozzle. The demonstrate that RRV a large injection from less than mL/min to more than 30 mL/min, that meets the requirements as calculated previously. The curve 0.1 RRV and nozzle. The curves demonstrate that RRV has a large injection rate range from less than mL/min to more than 30ofmL/min, that meets thethose requirements as 5calculated previously. curve under a pressure difference 4 bar is smoother than of 3 bar and bar, which means the The input0.1 mL/min to more than 30 mL/min, that meets the requirements as calculated previously. The under a pressure difference of 4inbar is case smoother than thosethan of 3 the bar and 5 bar, which means the inputoutput characteristics of the RRV this linear andwhich the normally curve under a pressure difference of 4 barisismore smoother than those ofother 3 bar cases, and 5 bar, means output characteristics of the RRV of inthe thisRRV case is more linear than the other cases, and the normally linear system is easier tocharacteristics control. the input-output in this case is more linear than the other cases, and the linear system is easier to control. normally linear system is easier to control.

Figure 5. Injection rate of RRV under three air pressures, calibrated by using 10% LUVITEC® solution. ® solution. Figure 5. Injection rate RRVunder underthree three air air pressures, by by using 10%10% LUVITEC ® solution. Figure 5. Injection rate of of RRV pressures,calibrated calibrated using LUVITEC

3.2. Uniformity Test of Six RRVs within One Boom Section 3.2. Uniformity Test SixRRVs RRVswithin within One One Boom 3.2. Uniformity Test of of Six BoomSection Section Figure 6 shows that, under the given experimental conditions: (1) the injection rates of six RRVs Figure 6 shows that, under the given experimental conditions: (1) the injection rates of six RRVs Figure 6 shows under the given experimental conditions: the injection of six RRVs within one boom sectionthat, are roughly equal, and around 14 mL/min; (2) (1) however, alongrates the direction within one boom section are roughly equal, and around 14 mL/min; (2) however, along the direction within one boom section are equal, and around 14 mL/min; (2) however, along the direction of chemical flow inflow the boom theroughly injection raterate of each RRV has a trend which of chemical in the boom the injection of each RRV has a trendtotoslightly slightly decrease, decrease, which isis of chemical flow in the boom the injection rate of each RRV has a trend to slightly decrease, which probably due to due the to loss chemical pressure at the inlet of of thethe RRVs, the RRVs RRVsnot not is probably theofloss of chemical pressure at the inlet RRVs,which whichleads leads to to the probably due to the lossrecommended of chemical pressure at the theerror inlet ofrepresenting the RRVs, which leads todeviations, the RRVs not being operated at the recommended pressure; (3) the error bars the deviations, being operated at the pressure; (3) bars representing thestandard standard being operated the recommended pressure; (3) error bars representing the standard (which range from 0.2 to 0.2 0.77tomL/min), emphasise thethe unstable flows injected the RRVs. (which rangeat from 0.77 mL/min), emphasise the unstable flows injectedby by the RRVs. deviations, (which range from 0.2 to 0.77 mL/min), emphasise the unstable flows injected by the RRVs.

Figure 6.Figure Injection rates ofrates 6 RRVs within one boom section, when PWM width 6. Injection of 6 RRVs within one boom section, when PWMsignal signalwas waswith with pulse pulse width Figure 6. Injection rates of 6 RRVs within one boom section, when PWM signal was with pulse width 1000 μs.1000 µs. 1000 μs.

The error bars increase from injection unit 1 to 6 illustrates that the further away the injection Thethe error bars increase injection unit 1unstable to 6 illustrates that away injection unit is from chemical inlet of from the boom, the more the flow ratethe is.further Based on this,the it could unit is from thewith chemical inletcontrol of the strategy, boom, theit more unstable flow is.and Based on chemical this, it could be concluded that, an open is hard for the the DNIS to rate meter inject be concluded that, with an open control strategy, it is hard for the DNIS to meter and inject chemical

Sensors 2016, 16, 127

7 of 11

The error bars increase from injection unit 1 to 6 illustrates that the further away the injection unit is from the chemical inlet of the boom, the more unstable the flow rate is. Based on this, it could Sensors 2016, 16, 127 7 of 11 be concluded that, with an open control strategy, it is hard for the DNIS to meter and inject chemical precisely, even under stable conditions, but metering and injecting the chemical precisely is the premise precisely, even under stable conditions, but metering and injecting the chemical precisely is the forpremise site-specific and variable-rate application of chemicals, so closed-loop control is necessary for site-specific and variable-rate application of chemicals, so closed-loop control is necessaryfor this DNIS. thisfor DNIS. 3.3.3.3. Calibration of of thetheThermodynamic Calibration ThermodynamicFlowmeter Flowmeter The thermodynamic flowmeter a wide measurement is determined by The thermodynamic flowmeter has has a wide measurement range,range, whichwhich is determined by adjusting the of coefficient of amplification manually. In accordance of RRV injection rates,the theadjusting coefficient amplification manually. In accordance with with rangerange of RRV injection rates, the measurement specified calibrationcurve curvewas was obtained obtained as 7. 7. measurement scale scale was was specified andand thethe calibration asshown shownininFigure Figure Although this thermodynamic flowmeter was independent of chemical viscosity, it was affected by Although this thermodynamic flowmeter was independent of chemical viscosity, it was affected before a different type of of chemical cancan bebe bythe the thermal thermalconductivity conductivityofofthe theliquid liquidmedium. medium.Thus, Thus, before a different type chemical applied, a specific calibrationisisrequired requiredin inadvance. advance. applied, a specific calibration

Figure7.7. Calibration Calibration of Figure of the the flowmeter. flowmeter.

3.4. Tuning the PID Parameters 3.4. Tuning the PID Parameters According to the Ziegler-Nichols method introduced by Frost [9], the transfer function of RRV According to the Ziegler-Nichols method introduced by Frost [9], the transfer function of RRV could be simplified as below: could be simplified as below: Q( s ) ke ke´ sTsT1 1 Qpsq (4) (4) “ P ( s ) 11  Ppsq `sT sT2 2

The time constants obtained experimentally experimentallyfrom fromananopen open loop The time constantsT1T,1,TT22 and and the the gain gain kk were were obtained loop response of the system to a typical step input (input pulse width is 1200 µs, magnitude of flow response of the system to typical step input (input pulse width is 1200 μs, magnitude of flow rate rate change is 19.21 mL/s,) as shown in 8.Figure 8. The are Tas = 0.105 change stepstep is 19.21 mL/s,) as shown in Figure The values are values as follows: 1 =follows: 0.105 s, TT 21 = 0.28 s, k s, ´1 ¨ s¨ V. mL The Nichols method suggested the best averagethe settings for a PIDsettings controller T2 ==0.117 0.28 s, k −1 =·s·V. 0.117 mLZiegler The Ziegler Nichols method suggested best average for a with transferwith function: PID controller transfer function: 1 Gpsq ` 1  ` T sq G( s“ ) KKc p1 ( 1 c Ti s Tdds)

Ti s

(5)

(5)

1.2T2 1.2,TT = T2T1 ,2T where Kc “ T1d =T0.5T  01.5T1 where K c  , d kT 1 2i , i kT The above values of T1 , T2 , k gave approximatively: 1 The above values of T1, T2 , k gave approximatively: Kc “ 27, Ti “ 0.21 s, Td “ 0.057 s Kc = 27, Ti = 0.21 s, Td = 0.057 s Figure 9 shows the performance of the closed-loop control to a step input for four different groups of PID parameters when the set-point output of the flowmeter was 8 V (equivalent to a set-

Sensors 2016, 16, 127

8 of 11

Figure 9 shows the performance of the closed-loop control to a step input for four different groups Sensors 2016,16, 16,127 127 when the set-point output of the flowmeter was 8 V (equivalent to a set-point of11 11 2016, 88of ofSensors PID parameters flow rate = 11.8 mL/min). The Ziegler Nichols controller design method gave a reasonable first estimate pointflow flowrate rate==11.8 11.8mL/min). mL/min).The TheZiegler ZieglerNichols Nicholscontroller controllerdesign designmethod methodgave gaveaareasonable reasonablefirst first point of controller coefficients, through slight adjustment, PID parameters (Kc = 25, Ti = 0.6, Td = 0) was estimateof ofcontroller controllercoefficients, coefficients,through throughslight slightadjustment, adjustment,PID PIDparameters parameters(K (Kcc==25, 25,TTi i==0.6, 0.6,TTdd==0) 0) estimate verified as most suitable for the system, they caused a smallest overshoot as well as least oscilation was verified verified as as most most suitable suitable for for the the system, system, they they caused caused aa smallest smallest overshoot overshoot as as well well as as least least was of oscilation the flow rate (the solid than others whoses parameters were tuned although ofthe the flow rateline) (thesolid solidline) line)than than othersPID whoses PIDparameters parameters wererandomly, tunedrandomly, randomly, oscilation of flow rate (the others whoses PID were tuned each group of PID parameters could ultimately stablize the flow rate. Comprehensively comparing although each each group group of of PID PID parameters parameters could could ultimately ultimately stablize stablize the the flow flow rate. rate. Comprehensively Comprehensively although the index of these four transient responses, PID parameters (K = 255, T = 0.3, T 0.075) c i d comparing the index of these four transient responses, PID parameters (K c = 255, T i==0.3, 0.3,TTddcaused 0.075)the comparing the index of these four transient responses, PID parameters (Kc = 255, Ti = ==0.075) transient flow rate to exceed the measurement range of the flowmeter, so it can not be used caused the the transient transient flow flow rate rate to to exceed exceed the the measurement measurement range range of of the the flowmeter, flowmeter, so so itit can can not notin bethe caused be controller. The parameters (K = 25, T = 0.6, T = 0) were preferred, as it only took roughly 1.3 s for c usedin inthe thecontroller. controller.The Theparameters parameters (Kcc==25, 25, 0.6,TTdd==0) 0)were werepreferred, preferred,as asititonly onlytook tookroughly roughlythe i (K d TTi i==0.6, used system to the achieve a target flowaarate with a very small 1.3ssfor for thesystem system toachieve achieve target flow ratewith withaaovershoot. verysmall smallovershoot. overshoot. 1.3 to target flow rate very

Figure Openloop loopresponse responseto toaaastep stepinput inputPP P== =1200 1200 μs, to obtain the transfer function coefficients. Figure Open loop response to the transfer function coefficients. Figure 8.8.8. Open step input 1200μs, µs,to toobtain obtain the transfer function coefficients.

Figure9.9.9. Variation Variation in in output ofof flowmeter as aas a function function of time time as flow flow rate controlled with four four Figure flowmeter as of as with Figure Variation inoutput outputof flowmeter a function of time as rate flowcontrolled rate controlled with groups ofdifferent different PIDparameters, parameters, theset-point set-point flowmeter (equivalent tochemical chemicalflow flow rate== groups of PID the flowmeter ==88VV(equivalent rate four groups of different PID parameters, the set-point flowmeter = 8 V to (equivalent to chemical 11.8 mL/min). 11.8 mL/min). flow rate = 11.8 mL/min).

3.5.Two-Phase Two-PhaseControl Control 3.5. 3.5. Two-Phase Control The PID PID controller controller worked worked well well for for linear linear systems, systems, however, however, due due to to the the fact fact the the input-output input-output The The PID controller worked well for linear systems, however, due to the fact the input-output characteristicsof ofthe theRRV RRVare arenot notstrictly strictlylinear linearas asshown shownin inFigure Figure10, 10,only onlyone onePID PIDcontrol controlmodule module characteristics characteristics of the RRV are not strictly linear as shown in Figure 10, only one PID control module not enough enough for for precisely precisely metering metering chemical chemical flow flow rates rates in in the the entire entire range. range. The The system system could could be be isis not is not enough for precisely metering chemical flow rates in the entire range. The system could be treated as as aa partial partial linear linear system system with with two two sections, sections, where where each each section section has has different different input-output input-output treated treated as a partial linear system with twothe sections, where each section has different input-output characteristics.Section Section11isisthe thescale scalewhen whenthe pulsewidth widthisisbelow below850 850μs, μs,under underwhich whichconditions conditions characteristics. pulse the injection injection rate rate isis lower lower than than around around 4.5 4.5 mL/min, mL/min, and and Section Section 22 isis when when the the pulse pulse width width isis above above the 850 and and the the corresponding corresponding injection injection rate rate isis higher higher than than around around 4.5 4.5 mL/min. mL/min. In In each each section, section, the the 850

Sensors 2016, 16, 127

9 of 11

Sensors 2016, 16, 127 characteristics. Section Sensors 2016, 16, 127

9 of 11 1 is the scale when the pulse width is below 850 µs, under which conditions the 9 of 11 injection rate is lower than around 4.5 mL/min, and Section 2 is when the pulse width is above 850 and relationship between input pulse width and output injection rate was simplified as linear. By this the corresponding injection rate iswidth higherand thanoutput around 4.5 mL/min. In each section,asthe relationship relationship between input pulse injection rate was simplified linear. By this simplification, the efficient control performance in entire range could be maintained through adding between input pulse width and output injection rate was simplified as linear. By this simplification, simplification, the efficient control performance in entire range could be maintained through adding another PID module to the controller for Section 1. the efficient control performance in entire range1.could be maintained through adding another PID another PID module to the controller for Section The tuned PID module with parameters (Kc = 25, Ti = 0.6, Td = 0) was adequate for Section 2, and module the controller for with Section 1. Theto tuned PID module parameters (Kc = 25, Ti = 0.6, Td = 0) was adequate for Section 2, and it worked well when the set-point output of the flowmeter was above 7 V (equivalent to a set-point The tuned PID module with parameters (K 25, Ti = 0.6, Tdabove = 0) was for Section 2, and it worked well when the set-point output of thec = flowmeter was 7 Vadequate (equivalent to a set-point chemical flow rate higher than 4.5 mL/min), but it was not suitable for Section 1. As the target output it workedflow wellrate when the set-point output ofbut theitflowmeter was above 7 V (equivalent to a set-point chemical higher than 4.5 mL/min), was not suitable for Section 1. As the target output of the flowmeter was below 7 V, the time needed to achieve the target flow rate was much longer. An chemical flow rate higher than 4.5 mL/min), but it was not suitable for Section 1. As the target of the flowmeter was below 7 V, the time needed to achieve the target flow rate was much longer. An example for the set-point output of the flowmeter at 5 V (equivalent to a set-point chemical flow rate output of the flowmeter was below 7 V, the time needed to achieve the target flow rate was example for the set-point output of the flowmeter at 5 V (equivalent to a set-point chemical flowmuch rate = 0.17 mL/min) is shown in Figure 11 (dashed line), and the time delay is ca. 15 s. An example for the of line), the flowmeter at 5 V (equivalent =longer. 0.17 mL/min) is shown in set-point Figure 11output (dashed and the time delay is ca. 15to s. a set-point chemical flow rate = 0.17 mL/min) is shown in Figure 11 (dashed line), and the time delay is ca. 15 s.

Figure 10. Sectional linear input-output characteristic of RRV. RRV. Figure10. 10.Sectional Sectionallinear linearinput-output input-outputcharacteristic characteristicof of Figure RRV.

Based on the analysis above, the control strategy was modified to a two-phase controllor with Based on on the the analysis analysis above, above, the the control control strategy strategy was was modified modified to to a two-phase two-phase controllor controllor with with Based two different PID modules for the two sections, respectively. Section 2 continued to use the tuned two different PID modules for the two sections, respectively. Section 2 continued to use the tuned two different PID modules for the two sections, respectively. Section 2 continued to use the tuned PID module, and another PID module with parameters (Kc = 70, Ti = 0.2, Td = 0) tuned by the Ziegler PID module, module, and another PID module with 0.2, TTdd ==0) 0)tuned tunedby bythe the Ziegler Ziegler PID with parameters parameters (K (Kcc = 70, Tii == 0.2, Nichols method as well was used for Section 1. As a result, the flow rate control in the range of less Nichols method methodas aswell wellwas wasused usedfor forSection Section a result, control in range the range of Nichols 1. 1. AsAs a result, the the flowflow raterate control in the of less than 4.5 mL/min was significantly improved. Figure 11 also shows an example of the set-point less than 4.5 mL/min significantly improved. Figure also shows exampleofofthe the set-point set-point than 4.5 mL/min was was significantly improved. Figure 11 11 also shows ananexample flowmeter output at 5 V (equivalent to a chemical flow rate of 0.17 mL/min) metered by the twoflowmeter output toto a chemical flow raterate of 0.17 mL/min) metered by the flowmeter output at at55VV(equivalent (equivalent a chemical flow of 0.17 mL/min) metered bytwo-phase the twophase controller (solid line). It took less than 4 s to achieve a steady state, which is a great controller (solid line). took less thanless 4 s to achieve a steady state, which state, is a great improvement phase controller (solidItline). It took than 4 s to achieve a steady which is a great improvement compared with the dashed line obtained from a single PID mode. compared with the dashed line obtained a single PID improvement compared with the dashed from line obtained frommode. a single PID mode.

Figure 11. 11. Time Figure Time domain domain responses responses of of the system, when when the set-point flowmeter output is 5 V Figure 11. Time domain responses of the system, when the set-point flowmeter output is 5 V (equivalent to to aa chemical chemical flow flow rate rate of of 0.17 0.17 mL/min). mL/min). (equivalent (equivalent to a chemical flow rate of 0.17 mL/min).

Table 1 shows indexes of transient responses regulated by the two-phase PID controller. Overall, Table 1 shows indexes of transient responses regulated by the two-phase PID controller. Overall, the steady-state errors are less than 0.021 V, meaning that flow rates can be metered and stabilized at the steady-state errors are less than 0.021 V, meaning that flow rates can be metered and stabilized at the targeted value precisely. The settling times are less than 3.17 s, indicating the controller could the targeted value precisely. The settling times are less than 3.17 s, indicating the controller could

Sensors 2016, 16, 127 Sensors 2016, 16, 127

10 of 11 10 of 11

Table indexes of transient responses regulated by the two-phase regulate the1 shows flow rate in a short time. Associating the response index with thePID twocontroller. sections, itOverall, can be the steady-state errors times are less 0.021 V, meaning flowthan ratesthose can be stabilized found that the settling forthan Section 2 (0.98~1.31 s) that are less formetered Section 1and (2.58~3.32 s), at thethe targeted valueovershoot precisely.values The settling times are for lessSection than 3.17 s, indicating controller could and maximum (4.13%~6.48%) 2 are less than the those for Section 1 regulate the flowas rate in aThis short time. Associating themodule response with the twothe sections, it can be (20.77%~36.88%) well. illustrates that the PID forindex Section 2 meters flow rate more found thatthan the settling times Section s) arethe less than those for Section (2.58~3.32 efficiently the module forfor Section 1. 2It (0.98~1.31 does not mean module parameters (Kc =170, Ti = 0.2, Ts), d theSection maximum overshoot valuesbecause (4.13%~6.48%) for Section 2 are with less than those for Section =and 0) for 1 is not the best fitted, many other PDI modules various parameters had1 (20.77%~36.88%) asparameters well. This illustrates theTdPID been tried, and the (Kc = 70, Tthat i = 0.2, = 0)module workedfor theSection best. 2 meters the flow rate more efficiently than the module for Section 1. It does not mean the module parameters (Kc = 70, Ti = 0.2, Table 1. Indexes of the transient responses. Td = 0) for Section 1 is not the best fitted, because many other PDI modules with various parameters had been tried, and the parameters (Kc = 70, Ti = 0.2, Td = 0) worked the best. Maximum Rise Set-Point Equivalent Settling Phase Steady-State Overshoot Time Voltage of Flow Rate Time Table 1. IndexesError of the(mL/min) transient responses. NO. (%) (s) Flowmeter (V) (mL/min) (s) 4 Voltage Equivalent 0.129 Flow Settling 3.09 0.011 24.65 1 Set-Point Steady-State Maximum Rise 0.42 Time Phase NO. of Flowmeter (V) Rate (mL/min) Time Error0.016 (mL/min) Overshoot (%) (s) 5 0.174 3.17 (s) 36.88 0.40 1 4 0.129 3.09 0.011 24.65 0.42 1 6 0.277 2.58 0.037 20.77 0.42 5 0.174 3.17 0.016 36.88 0.40 6.5 2.611 3.32 0.073 30.98 0.44 111 6 0.277 2.58 0.037 20.77 0.42 7 4.647 1.31 0.059 6.48 0.38 21 6.5 2.611 3.32 0.073 30.98 0.44 7 4.647 1.31 0.059 6.48 0.38 8 11.809 0.98 0.054 4.13 0.38 22 8 11.809 0.98 0.054 4.13 0.38 2 99 23.367 1.26 0.063 5.36 0.36 22 23.367 1.26 0.063 5.36 0.36

3.6. 3.6. EC EC Measuring Measuring Method Method to to Evaluate Evaluate the the Effect Effect of of the the Closed-Loop Closed-Loop Control Control The The performance performance of of the the closed-loop closed-loop controller controller also also can can be be verified verified from from the the output output of of the the EC EC sensor as shown in Figure 12, in which the readings are from the same test displayed by a solid line sensor as shown in Figure 12, in which the readings are from the same test displayed by a solid line in in Figure 9. Comparing these two curves, the flowmeter output and EC sensor change Figure 9. Comparing these two curves, the flowmeter output and EC sensor change simultaneously. simultaneously. Although they have opposite trends (the of the EC sensor isthe negative Although they have opposite trends (the calibration of the ECcalibration sensor is negative correlative, higher correlative, the higher the flow rate and therefore the conductivity, the lower the output voltage), the flow rate and therefore the conductivity, the lower the output voltage), both could be used to both could used to indicate theflow flowrate rate. After the flow rate achieved steady were state,analyzed. the EC values indicate thebeflow rate. After the achieved a steady state, the ECa values The were analyzed. The consistency and stability of the EC values indicate the concentration of the salt consistency and stability of the EC values indicate the concentration of the salt solution was stable, solution was stable, hence the flow rate metered by the closed-loop controller was constant. hence the flow rate metered by the closed-loop controller was constant.

Figure 12. Output Outputofofthethe sensor when the desired of the flowmeter set at 8 V Figure 12. ECEC sensor when the desired outputoutput of the flowmeter was set atwas 8 V (equivalent (equivalent a desired chemical ratemL/min). set at 11.8 mL/min). to a desired to chemical flow rate setflow at 11.8

4. 4. Conclusions Conclusions An basis of of the the site-specific, site-specific, variable-rate variable-rate An accurate accurate control control of of chemical chemical injection injection rates rates is is the the basis application of chemicals. chemicals.This Thisarticle article tested a feasible method to control the rate flowfor rate for a nozzle direct application of tested a feasible method to control the flow a direct nozzle injection system. From the results, the following conclusions can be drawn: injection system. From the results, the following conclusions can be drawn: 1.

The RRV used in the DNIS could inject chemical with flow rates ranging from 0.1 mL/min to more than 30 mL/min, which covers most spraying operations in the field.

Sensors 2016, 16, 127

1. 2.

11 of 11

The RRV used in the DNIS could inject chemical with flow rates ranging from 0.1 mL/min to more than 30 mL/min, which covers most spraying operations in the field. The thermodynamic flowmeter met the requirement of measuring wide range of flow rate, and it was independent of chemical viscosity. However, it was affected by the thermal conductivty of the liquid medium, so calibration for different chemicals was necessary before application.

The closed-loop controller (PID) worked well for metering and controlling flow rates precisely, but the tuning of PID parameters should be carefully considered, as it was highly related to the input-output charicteristics of the RRV. If the input-output characteristics of the injection unit are non-linear, a two-phase controlling method was an optimal solution. Acknowledgments: We wish to thank the support of the Fundamental Research Funds for Central Universities (NO. BLX2013033), National Natural Science Foundation of China under Project No. 31400621 and Chinese-German Center for Scientific Promotion (Chinessich-Deutsches Zentrum fuer Wissenschaftsfoerderung) under the project of Sino-German Research Group (GZ494). Author Contributions: Xiang Cai designed the controller and wrote the paper; Martin Walgenbach and Malte Doerpmond designed the experimental setup; Peter Schulze Lammers and Yurui Sun proposed the programme. Conflicts of Interest: The authors declare no conflict of interest.

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.

Gerhards, R.; Oebel, H. Practical experiences with a system for site-specific weed control in arable crops using real-time image analysis and GPS-controlled patch spraying. Weed Res. 2006, 46, 185–193. [CrossRef] Lutman, P.J.W.; Miller, P.C.H. Spatially Variable Herbicide Application Technology; Opportunities for Herbicide Minimisation and Protection of Beneficial Weeds; Home-Grown Cereals Authority: London, UK, 2007. Mohammadzamani, D.; Rashidi, M. Generating a Digital Management Map for Variable Rate Herbicide Application Usingthe Global Positioning System. Am.-Eurasian J. Sustain. Agric. 2009, 3, 101–106. Miller, P.C.H.; Butler Ellis, M.C. Effects of formulation on spray nozzle performance for applications from ground-based boom sprayers. Crop Protect. 2000, 19, 609–615. [CrossRef] Deng, W.; He, X.; Ding, W. Droplet size and spray pattern characteristics of PWM-based continuously variable spray. Int. J. Agric. Biol. Eng. 2009, 2, 8–18. GopalaPlillai, S.; Tian, L.; Zheng, J. Evaluation of a flow control system for site-specific herbicide applications. Trans. ASAE 1999, 42, 863–870. [CrossRef] Han, S.; Hendrickson, L.L.; Ni, B.; Zhang, Q. Modification and testing of a commercial sprayer with PWM solenoid s for precision spraying. Appl. Eng. Agric. 2001, 17, 591–594. Gebhardt, M.R.; Kliethermes, A.R.; Goering, C.E. Metering concentrated pesticides. Trans. ASAE 1984, 27, 18–23. [CrossRef] Frost, A.R. A pesticide injection metering system for use on agricultural spraying machines. J. Agric. Eng. Res. 1990, 46, 55–70. [CrossRef] Lammers, P.S.; Vondricka, J. Precision Crop Protection—The Challenge and Use of Heterogeneity; Springer: Dordrecht, The Netherlands, 2010. Vondricka, J. Study on the Process of Direct Nozzle Injection for Real-Time Site-Specific Pesticide Application. Ph.D. Thesis, University of Bonn, Bonn, Germany, 2008. Vondricka, J.; Lammers, P.S. Measurement of mixture homogeneity in direct injection systems. Trans. ASABE 2009, 52, 61–66. [CrossRef] Doerpmund, M.; Cai, X.; Walgenbach, M.; Vondricka, J.; Shuzle Lammers, P. Assessing the cleanability of a direct nozzle injection system. Biosyst. Eng. 2011, 110, 49–56. [CrossRef] Doerpmund, M.; Cai, X.; Walgenbach, M.; Vondricka, J.; Shuzle Lammers, P. The challenge of cleaning direct-injection systems for pesticide application. Trans. ASABE 2012, 55, 1643–1650. [CrossRef] © 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).