Agricultural Engineering · Engineering Sciences · also Instrumentation and Measurement Techniques, Sensors, AI and Robotics in Agriculture
Measuring temperature, pressure, flow, force and moisture, logging the data, and controlling processes automatically.
In the exams: ICAR AICE subjects 38 (Unit 7), 39 (Unit 8) and 41 (Unit 9); NCEES PE sensors and data acquisition; NCEES FE 4–6 questions
Instrumentation is the engineering of measurement: choosing, installing, calibrating and reading sensors so that the numbers can be trusted. Process control is what is done with those numbers: keeping a dryer at a set air temperature, opening greenhouse vents before the crop overheats, starting an irrigation pump when the root zone has dried to a set level, or stopping a conveyor line before grain piles up on a stalled belt. The subject joins physics, electronics and control theory, and it sits under almost every other branch of agricultural engineering, from tractor testing to cold chains.
This manual follows the measurement chain from the physical quantity to the decision. It begins with the generalised measurement system, errors and the static and dynamic behaviour of instruments; works through the transducers an agricultural engineer meets most often (temperature, pressure, flow, level, force, torque, strain and moisture); covers signal conditioning, analog-to-digital conversion, data loggers and microcontrollers; and ends with control: transfer functions and stability, on-off, P, PI and PID control, programmable logic controllers and IoT sensor networks. Greenhouse, irrigation, dryer, grain-handling and machinery-testing examples are used throughout.
Worldwide, the subject is examined in two ways. Indian PhD and scientist entrance tests (the ICAR AICE-JRF/SRF instrumentation units in dairy, farm machinery and processing engineering, and ASRB NET) ask definitions, transducer principles and short calculations. Licensing examinations such as the NCEES FE (Instrumentation and Controls) and the PE Agricultural and Biological Engineering, and confirmatory syllabi such as that of Engineers Canada, expect the student to size and interpret measurements and control loops. Where a value depends on the crop, the standard or the country, the manual says so rather than giving one number.
Every instrument, from a glass thermometer in a seed store to a wireless soil-moisture node, can be described by the same chain of functional elements. The primary sensing element is in contact with the measured quantity (the measurand) and responds to it: a thermocouple junction, a Bourdon tube, a bonded strain gauge. The variable-conversion element changes that response into a more convenient form, for example a displacement into a resistance change. The variable-manipulation element changes the size of the signal without changing its nature, as an amplifier does. The data-transmission element carries the signal to where it is used, over a cable, a 4–20 mA loop or a radio link. The data-presentation element shows, records or stores it: a dial, a display, a data logger or a database. A transducer converts energy from one form to another; a sensor is the part that responds to the measurand. Self-generating (active) transducers produce their own output energy, such as a thermocouple or a piezoelectric crystal. Passive transducers need external excitation, such as a resistance thermometer or a strain gauge, which must be supplied with a voltage or current. Analog instruments give a continuously variable output; digital instruments give a discrete, numbered output.
Static error is the measured value minus the true value; the static correction is its negative. Because the true value is never known exactly, practice uses a reference standard of much smaller uncertainty as the "conventional true value". Errors fall into three groups. Gross errors are human blunders: reading the wrong scale, transposing digits, installing a sensor in the wrong place. Systematic errors repeat in the same direction under the same conditions. They include instrumental errors (a worn spring, a zero offset), environmental errors (temperature changing the resistance of lead wires, radiation heating an unshielded air-temperature sensor) and observational errors such as parallax. A systematic error does not shrink with averaging; it is removed by calibration and correction. Random errors vary unpredictably from reading to reading because of noise, small fluctuations and resolution limits. They are treated statistically: the mean of n readings has a standard error equal to the standard deviation divided by the square root of n. Accuracy figures are stated in different ways, and the basis matters. An error "±1% of full scale" on a 0–10 bar gauge is ±0.1 bar everywhere on the scale, so at a reading of 2 bar it is ±5% of the reading. An error stated as a percentage of reading stays proportionate. This is why an instrument should be chosen so that normal readings fall in the upper part of its range.
Inserting an instrument changes the quantity it measures; this is the loading effect. A thermometer with a large mass cools a small sample of hot milk; a voltmeter with modest input resistance draws current from a high-resistance sensor and reads low. For a source of internal resistance Rs read by a meter of input resistance Rm, the meter shows Rm/(Rs + Rm) of the true voltage, so the meter resistance should be very much larger than the source resistance. Calibration is the comparison of an instrument against a standard of known, smaller uncertainty, across its range, under stated conditions. Traceability means the chain of calibrations can be followed back to a national metrology institute and the SI units. A calibration produces a calibration curve or a correction table and a stated uncertainty. A two-point calibration fixes zero and span for a linear device; several points are needed to check linearity and hysteresis, approached from below and from above. When a result depends on several independent error sources, the most probable combined uncertainty is the root sum of squares of the individual uncertainties, not their arithmetic sum. The arithmetic sum gives the worst case, which is rarely reached because independent errors seldom all take their extreme values together. For a computed result y = f(x1, x2, …), each uncertainty is first multiplied by the sensitivity of y to that variable (the partial derivative) before squaring and adding.
Static characteristics describe an instrument when the input is steady or changing slowly. Range is the span of input values from the lowest to the highest the instrument is designed to measure; span is the difference between them (a −20 to 80 °C sensor has a span of 100 °C). Accuracy is closeness to the true value; precision is closeness of repeated readings to each other. Precision is subdivided into repeatability (same observer, instrument and conditions over a short time) and reproducibility (different conditions, observers or laboratories). An instrument can be precise and inaccurate: tightly grouped readings with a constant offset. Sensitivity is the change in output per unit change in input, the slope of the calibration curve, for example 41 µV/°C or 2 mV per kN. Resolution is the smallest input change that produces a detectable output change; threshold is the smallest input from zero that produces a detectable output. Linearity is the maximum deviation of the calibration curve from a stated straight line, usually expressed as a percentage of full-scale output; the reference line may be the best-fit (independent) line or the line through the end points (terminal linearity), and the two give different figures. Hysteresis is the difference between readings at the same input when approached from increasing and decreasing directions, caused by friction, backlash, magnetic or elastic effects. Dead band is the range of input over which there is no output change. Drift is a slow change in output with time at constant input; zero drift shifts the whole curve, span (sensitivity) drift changes its slope.
Dynamic characteristics describe how an instrument follows a changing input. A zero-order instrument responds instantly; a slide-wire potentiometer is the usual example. A first-order instrument stores energy in one place, such as heat in the mass of a thermometer bulb. It obeys τ dy/dt + y = K x, where τ is the time constant and K the static sensitivity. After a step change of input, the indicated change is (1 − e^(−t/τ)) of the final change: 63.2% at t = τ, 86.5% at 2τ, 95.0% at 3τ and 98.2% at 4τ. For example, a thermometer with τ = 10 s plunged from air at 20 °C into water at 80 °C reads 66.6 °C after 15 s. For a ramp input changing at a steady rate, a first-order instrument settles to lag behind by a constant time τ, so the steady reading error equals rate × τ. A sensor with τ = 30 s in dryer air warming at 2 °C per minute reads 1 °C low throughout the warm-up. For a sinusoidal input of angular frequency ω the amplitude ratio is 1/√(1 + (ωτ)²) and the phase lag is arctan(ωτ). The time constant depends on the installation as well as the sensor: a thermocouple in a thick thermowell, or in still air rather than flowing water, responds far more slowly than its bare-bead figure.
A second-order instrument stores energy in two forms, typically a mass and a spring, as in an accelerometer, a diaphragm pressure transducer with fluid-filled tubing, or a moving-coil meter. It is described by the undamped natural frequency ωn and the damping ratio ζ. With ζ < 1 the system is underdamped and overshoots a step before settling; with ζ = 1 it is critically damped, the fastest response without overshoot; with ζ > 1 it is overdamped and sluggish. The fractional peak overshoot of an underdamped system is exp(−ζπ/√(1 − ζ²)), which is about 16.3% at ζ = 0.5. The settling time to within 2% is approximately 4/(ζωn). Many instruments are designed with ζ around 0.6 to 0.7, which gives a short rise time, small overshoot and the widest frequency range over which the amplitude ratio stays close to one. To measure a fluctuating quantity faithfully, such as vibration on a harvester frame or pressure pulsations in a pump delivery line, the natural frequency of the transducer must be well above the highest frequency in the signal. Long or air-bubbled connecting tubing lowers the natural frequency of a pressure system and can make readings ring or lag.
Liquid-in-glass thermometers and bimetallic strips use thermal expansion. A bimetallic strip bonds two metals of different expansion coefficients so the strip bends with temperature; coiled, it drives a dial, and as a snap disc it is the simple on-off thermostat in heaters, incubators and brooders. A resistance temperature detector (RTD) uses the rise in resistance of a pure metal, almost always platinum, with temperature. The industrial standard Pt100 has 100 Ω at 0 °C and a mean coefficient α = 0.00385 /°C between 0 and 100 °C, so its resistance at 100 °C is about 138.5 Ω. Over a narrow range R ≈ R0(1 + αT); over wider ranges the Callendar–Van Dusen equation adds a second-order term. The relevant international standard defines tolerance classes. RTDs are accurate, stable and fairly linear but need excitation current, which causes self-heating: the error equals the power dissipated divided by the dissipation constant (mW per °C), so the current is kept small (around 1 mA). Lead-wire resistance adds directly to a two-wire reading; a three-wire connection in a bridge cancels most of it, and a four-wire connection with separate current and voltage leads removes it.
Thermistors are semiconductor resistors with a large temperature coefficient. Negative temperature coefficient (NTC) thermistors, made from metal-oxide ceramics, fall in resistance as temperature rises, by roughly 3–5% per °C near room temperature, about ten times the change of platinum. Their characteristic is strongly non-linear. Over moderate ranges it is described by the beta equation, 1/T = 1/T0 + (1/β) ln(R/R0), with temperatures in kelvin, where R0 is the resistance at reference temperature T0 (often 25 °C) and β is a material constant in kelvin. For higher accuracy the three-coefficient Steinhart–Hart equation is used. Thermistors are small, sensitive and inexpensive, which makes them common in low-cost environmental nodes, milk-cooling tanks, incubators and soil-temperature probes, usually read through a voltage divider into a microcontroller ADC. Their range is narrower than that of RTDs or thermocouples, and interchangeability between units depends on the tolerance grade bought. Positive temperature coefficient (PTC) types switch sharply at a set temperature and are used for over-temperature protection rather than measurement.
A thermocouple is two dissimilar metal wires joined at the measuring (hot) junction. Because of the Seebeck effect, a net voltage appears that depends on the temperature difference between the measuring junction and the reference (cold) junction where the wires meet the copper of the instrument. Three laws govern their use. The law of intermediate metals: a third metal inserted in the circuit has no effect if both its junctions are at the same temperature, so a meter can be connected. The law of intermediate temperatures: the emf between T1 and T3 equals the emf between T1 and T2 plus that between T2 and T3, which allows reference-table corrections. The law of homogeneous circuits: a single homogeneous wire generates no net emf whatever the temperature along it. Because the output depends on the reference temperature, the reference must be fixed (an ice bath at 0 °C in the laboratory) or measured and compensated electronically; this is cold-junction compensation. The usual approximation near ambient is to convert the cold-junction temperature to an equivalent emf, add it to the measured emf, and look up the total in the reference table. Type K (nickel-chromium/nickel-aluminium) has a sensitivity of about 41 µV/°C near room temperature and a wide range; type J (iron/constantan) about 52 µV/°C; type T (copper/constantan) is good at low temperatures. Extension and compensating cables of matching alloys must run to the reference junction; ordinary copper wire in between creates new junctions and errors. Thermocouples are rugged, fast and self-powered, but give small signals that need amplification and careful noise control.
Infrared radiation thermometers infer surface temperature from emitted radiation. The reading depends on the emissivity setting; most plant leaves, soil and water have high emissivity, polished metal low, so a metal dryer duct read with an infrared thermometer can be badly in error unless painted or taped. The instrument sees a spot whose size grows with distance. Canopy temperature measured this way is the basis of crop water stress indices, because transpiring leaves are cooler than air while stressed leaves warm. Installation errors often exceed sensor error. An air-temperature sensor in sunlight reads high unless it is in a ventilated radiation shield. Grain-bin temperature cables carry several sensors hung in the grain mass to detect hot spots caused by insect or mould activity. In a dryer the plenum air temperature is controlled, while grain temperature is the quality limit; the safe limits depend on the crop and its end use (seed, milling, feed) and are given in national and crop-specific guidance. In pasteurisers the indicating and recording thermometers are part of the legal process control, and their accuracy and response time are specified by food regulators.
Pressure is force per unit area, in pascals (1 bar = 100 kPa). Gauge pressure is measured relative to local atmospheric pressure, absolute pressure relative to vacuum, and differential pressure between two points. A U-tube manometer gives Δp = (ρm − ρf) g h, where ρm is the manometer liquid density and ρf the density of the fluid above it (negligible for air). Inclined manometers increase the reading length for small pressures, such as fan static pressure across a grain bed. The Bourdon tube, a flattened curved tube that straightens under pressure, is the standard dial gauge on sprayers and pump outlets. Diaphragms, capsules and bellows convert pressure into displacement. Electrical pressure transducers sense that displacement or strain: bonded or thin-film strain gauges on a diaphragm, piezoresistive silicon diaphragms, capacitive diaphragms that change the gap between plates, and piezoelectric crystals. Piezoelectric sensors respond only to changing pressure because their charge leaks away, so they suit pulsation and combustion measurement but not steady pressure. Submersible pressure transducers measure water level in wells and tanks; vented cable or a separate barometric sensor is needed to remove atmospheric pressure changes from the reading.
Obstruction meters (orifice plate, venturi tube, flow nozzle) accelerate the flow and measure the pressure drop. From Bernoulli and continuity, the volume flow rate is Q = Cd A2 √(2Δp / (ρ(1 − β⁴))), where A2 is the throat area, β the ratio of throat to pipe diameter and Cd the discharge coefficient (close to 0.98 for a venturi and about 0.6 for a sharp-edged orifice). Because Q varies with √Δp, flow halves when Δp falls to a quarter, which limits useful turndown. The venturi recovers most of its pressure loss; the orifice is cheap but wastes energy. A Pitot-static tube gives local velocity v = √(2Δp/ρ) and is used to traverse ducts in dryers and ventilation systems. The rotameter is a float in a tapered tube; it reads flow directly and is common on fertigation and chemical dosing lines. Turbine meters produce pulses at a rate proportional to velocity, and the K-factor (pulses per litre) converts counts to volume. Positive-displacement meters trap and count fixed volumes and are accurate on clean water and fuel. Electromagnetic meters apply Faraday's law, e = B d v, to a conductive liquid flowing through a magnetic field; they have no obstruction and suit dirty water and slurries but do not work on non-conducting liquids such as oils. Ultrasonic transit-time meters compare upstream and downstream sound travel times and can clamp onto the outside of a pipe; Doppler types need particles or bubbles to reflect sound. Most meters need several diameters of straight pipe upstream for a stable velocity profile.
Irrigation canals and drains are measured with weirs and flumes, in which discharge is a fixed function of upstream head, Q = C Hⁿ: the exponent is about 1.5 for a rectangular weir and 2.5 for a V-notch. Flumes such as the Parshall flume pass sediment better and lose less head than weirs. Head is read with a staff gauge, a float in a stilling well, a pressure transducer or an ultrasonic sensor. Level in tanks is measured with sight glasses, floats, hydrostatic pressure (h = p/(ρg), so a dense fertiliser solution reads a lower depth than water for the same pressure unless the density is set), bubbler tubes, capacitance probes, and non-contact time-of-flight sensors. An ultrasonic sensor mounted above the liquid measures the echo time; distance is c·t/2, where c is the speed of sound, about 343 m/s in air at 20 °C and rising by about 0.6 m/s per °C, so temperature compensation is needed. Radar sensors are unaffected by air temperature and dust and suit grain silos and closed tanks. For bulk solids, point-level switches (rotating paddles, tilt switches, vibrating forks) signal full and empty in bins and hoppers.
Strain ε is the change in length divided by original length; it is dimensionless and is often quoted in microstrain (µε, 10⁻⁶). In the elastic range stress σ = E ε, so for steel (E about 200 GPa) a stress of 50 MPa produces 250 µε. A bonded metal-foil strain gauge is a fine grid glued to the surface; when the surface stretches, the grid lengthens and narrows and its resistance rises. The gauge factor GF = (ΔR/R)/ε is about 2 for common metal-foil gauges, so a 120 Ω gauge at 500 µε changes by only 0.12 Ω. Semiconductor gauges have far larger gauge factors but are more temperature-sensitive and non-linear. Gauges respond to strain along the grid axis; rosettes of two or three gauges at set angles are used when the principal directions are unknown. Good results depend on surface preparation, correct adhesive, protection from moisture, and matching the gauge's self-temperature-compensation to the material.
Resistance changes this small are measured with a Wheatstone bridge: four arms excited by a voltage Vs, balanced (zero output) when R1/R2 = R3/R4. With one active gauge (quarter bridge), the output for small strain is approximately Vo = Vs·GF·ε/4. Two active gauges with opposite strains in adjacent arms (half bridge), such as one on the top and one on the bottom of a beam in bending, double the output to Vs·GF·ε/2 and cancel temperature effects that act equally on both. Four active gauges (full bridge) give Vs·GF·ε. Gauges in adjacent arms subtract; in opposite arms they add. This lets the designer cancel unwanted effects: a bending beam with gauges arranged to cancel axial load, or a dummy gauge on an unstrained piece of the same metal in an adjacent arm to cancel temperature. Output signals are only millivolts, so the bridge is followed by a high-gain differential (instrumentation) amplifier, and the excitation voltage must be stable or the output measured as a ratio to it.
A load cell is an elastic element (column, ring, S-beam, shear beam or bending beam) carrying a full strain-gauge bridge. Its rated output is quoted in millivolts per volt of excitation at full capacity, commonly 1–3 mV/V; a 2 mV/V cell excited at 10 V gives 20 mV at full load, and the output at part load is proportional. Combined error, creep, temperature effect on zero and span, and safe overload are the main specifications. Load cells are used in hopper and silo weighing, livestock and bale scales, seed-drill and fertiliser-spreader metering checks, and in yield monitors that weigh grain on a conveyor or measure impact force. In machinery testing, a drawbar dynamometer is a load cell in the hitch line, and three-point-linkage dynamometers carry load cells in each link to resolve draft and vertical forces for tillage tool research. Torque on a rotating shaft is measured with gauges at ±45° to the shaft axis, where shear strain from torsion produces equal tensile and compressive principal strains; the signal is taken off by slip rings or a short-range radio telemetry module. Power is then P = 2πNT/60 with N in rev/min and T in N·m. A power take-off torque meter on a 540 rev/min shaft is a standard tool for matching implements to tractors.
Moisture content on a wet basis (wb) is mass of water divided by total mass; on a dry basis (db) it is mass of water divided by dry-matter mass. They convert by Mdb = Mwb/(1 − Mwb) and Mwb = Mdb/(1 + Mdb). Trade and storage use wet basis; drying theory uses dry basis. A 100 g sample that weighs 85 g after oven drying has 15% moisture wet basis and about 17.6% dry basis. When grain is dried, dry matter is conserved, so the final mass is initial mass × (1 − Mi)/(1 − Mf). The reference method is oven drying, but there is no single oven method: standards differ by country and commodity in temperature, time and whether the grain is ground (for example, whole-kernel methods at about 103 °C for many hours and ground-sample methods at about 130 °C for a shorter time). Results from different methods are not identical, so a reported moisture should name the method. Rapid electrical meters estimate moisture from the dielectric constant (capacitance meters) or from electrical resistance (conductance meters) of the grain. Water's dielectric constant is about 80 against a few units for dry grain, so capacitance rises with moisture; the reading also depends on temperature, bulk density, variety and crop, and each meter is calibrated per crop against the oven method and temperature-corrected. Near-infrared (NIR) analysers use water absorption bands and can measure protein and oil in the same scan. On-line capacitance sensors in dryer discharge augers are used to control dryer discharge rate.
Gravimetric water content θg is mass of water per mass of dry soil, found by oven drying at 105 °C. Volumetric content θv equals θg multiplied by the dry bulk density divided by the density of water, and the equivalent depth of water in a layer is θv × layer depth. Volumetric content is what irrigation scheduling needs. Time-domain reflectometry (TDR) sends a fast voltage pulse along a metal probe and times its reflection. The apparent dielectric constant is Ka = (c·t/(2L))², where t is the two-way travel time, L the probe length and c the speed of light. The empirical Topp equation, θv = −0.053 + 0.0292 Ka − 5.5×10⁻⁴ Ka² + 4.3×10⁻⁶ Ka³, works for many mineral soils without site calibration but less well in organic, clay-rich or saline soils. Capacitance or frequency-domain (FDR) sensors measure the same dielectric effect more cheaply but are more affected by salinity, temperature and soil contact, so site-specific calibration is advisable. Tensiometers measure soil matric potential through a water-filled porous ceramic cup and a vacuum gauge or transducer; they work only to about −80 kPa before air enters, so they suit wet-range irrigation scheduling of vegetables and orchards. Granular matrix and gypsum blocks measure potential over a drier range through resistance. The neutron probe, which counts slowed neutrons from a radioactive source, is accurate but requires radiation licensing.
Relative humidity (RH) is the ratio of actual water-vapour pressure to the saturation vapour pressure at the same temperature. Saturation vapour pressure rises steeply with temperature; one common form (used in FAO-56) is es = 0.6108 exp(17.27T/(T + 237.3)) kPa, with T in °C. Dew point is the temperature at which the air becomes saturated when cooled at constant pressure. Vapour pressure deficit (VPD) = es − ea is the drying power of the air and is used in greenhouse climate control because it relates directly to plant transpiration. A psychrometer compares dry-bulb and wet-bulb temperatures; the wet bulb must be ventilated at an adequate air speed and kept with clean, wet wicking, otherwise it reads high and RH is overestimated. Thin-film capacitive polymer sensors are now standard in data loggers and greenhouse controllers; their dielectric changes as the polymer absorbs water. They drift, especially after long periods near saturation or exposure to chemicals, so they need periodic checks against saturated-salt references or a calibrated instrument.
Raw transducer outputs are rarely ready for display or digitising. They may be millivolts (thermocouples, strain bridges), resistances (RTDs, thermistors), currents, charges (piezoelectric) or frequencies (turbine meters). Signal conditioning supplies excitation, converts the signal to a voltage or current, amplifies it to match the input range of the converter, removes noise, linearises, isolates and protects. The operational amplifier (op-amp) is the building block. With negative feedback, the non-inverting amplifier has gain 1 + Rf/Rin and very high input resistance, so it does not load the sensor; the inverting amplifier has gain −Rf/Rin and an input resistance equal to Rin; the voltage follower (gain 1) is a buffer between a high-impedance source and a load. Bridge and thermocouple signals are differential: the wanted signal is the small difference between two wires that both sit at some common voltage. An instrumentation amplifier, typically three op-amps with gain set by one resistor (for the classic three-op-amp form, G = 1 + 2R/RG), amplifies the difference and rejects the common-mode voltage. Common-mode rejection ratio (CMRR), usually quoted in decibels, measures how well it does so; a high CMRR is essential where long cables pick up mains interference equally on both conductors.
Noise enters by electrostatic (capacitive) coupling, magnetic (inductive) coupling from motors and power cables, ground loops when a signal is earthed at two points of different potential, and thermal noise within resistors. The remedies are practical: twisted-pair cable to cancel magnetic pick-up; a shield earthed at one end only; differential inputs; separation of signal cables from motor and variable-speed-drive cables; and single-point grounding. Filters remove frequencies outside the band of interest. A first-order RC low-pass filter has a cut-off (−3 dB) frequency fc = 1/(2πRC); above it the response falls at 20 dB per decade. Before an analog-to-digital converter, a low-pass anti-aliasing filter removes components above half the sampling rate, which would otherwise fold back as false low frequencies. Slow agricultural variables such as soil water, air temperature or grain temperature change over minutes, so heavy filtering and averaging are acceptable; vibration or draft-force fluctuation needs a wide bandwidth. Integrating converters and averaging over a whole number of mains cycles strongly reject 50 or 60 Hz interference.
Over long cables in farms and processing plants, voltage signals suffer from cable resistance and noise. The 4–20 mA current loop is the industrial standard: the transmitter controls the loop current, so cable resistance does not change the signal as long as the supply has enough voltage. The live zero of 4 mA means 0 mA indicates a broken wire or dead transmitter, and the 4 mA also powers two-wire (loop-powered) transmitters. The measured value is lower range + (I − 4)/16 × span. A 250 Ω precision resistor converts 4–20 mA to 1–5 V for an ADC. Voltage outputs (0–5 V, 0–10 V) suit short runs. Isolation (optical or transformer) breaks ground loops and protects equipment from surges, important on long outdoor cables exposed to lightning. Linearisation corrects non-linear sensors, either by circuit (a resistor in parallel with a thermistor straightens its curve over a limited range) or, now usually, in software by lookup table or polynomial. Cold-junction compensation for thermocouples, bridge completion resistors for quarter-bridge gauges and constant-current sources for RTDs are also signal-conditioning functions.
A data acquisition (DAQ) system samples conditioned signals and converts them to numbers. The sampling theorem requires a sampling frequency greater than twice the highest frequency present (the Nyquist rate); in practice several times higher is used. If a signal contains frequencies above half the sampling rate, they appear as false lower frequencies: a component at frequency f sampled at fs appears at |f − k·fs| for the nearest whole k. This is aliasing, and it cannot be removed after sampling. An n-bit converter divides its full-scale range into 2ⁿ steps. The least significant bit (LSB) is FS/2ⁿ, and the quantisation error is ±½ LSB. A 12-bit converter on 0–5 V has an LSB of about 1.22 mV. The ideal signal-to-quantisation-noise ratio of an n-bit converter is about 6.02n + 1.76 dB. The digital code for an input voltage V is the whole number part of V/FS × 2ⁿ (for a unipolar converter). Resolution is not accuracy: offset, gain error, non-linearity and noise in the analog front end often limit a cheap 12-bit system to fewer effective bits. The main converter types are successive approximation (SAR: fast, moderate resolution, common in microcontrollers), sigma-delta (high resolution at modest speed, used for bridges and thermocouples), integrating or dual-slope (slow, very good noise and mains rejection, used in digital multimeters), and flash (fastest, low resolution). A multiplexer switches several channels into one converter, and a sample-and-hold circuit freezes the input during conversion.
A data logger is a self-contained, battery-powered DAQ system with a clock and memory, designed to run unattended for weeks or months. Planning a logging campaign means choosing for each channel the sensor, the excitation and wiring, the scan interval (how often the logger reads) and the storage interval (how often it saves). Storing averages, maxima and minima over each storage interval keeps files small while preserving information. Memory needed = channels × bytes per value × records per day × days; power needed is estimated from the average current, as for wireless nodes. Good practice includes time synchronisation and a stated time zone, field metadata (sensor serial numbers, calibration dates, depths and positions, photos), regular downloads and checks, desiccant and sealed enclosures, surge protection on long cables, and sensor maintenance such as cleaning rain gauges and radiation shields. Data should be checked for range, rate-of-change and flat-lining, and flagged rather than silently edited.
A microcontroller is a single chip containing a processor, program memory, data memory and peripherals. The peripherals that matter for sensing are general-purpose input/output (GPIO) pins for switches and relays; an on-chip ADC (often 10 or 12 bits); timers and counters for frequency and pulse inputs such as rain-gauge tips or turbine-meter pulses; pulse-width modulation (PWM) outputs to drive fans, valves and motors through a driver; interrupts that let a pulse wake the processor; and low-power sleep modes. Many modern sensors contain their own converter and send digital data. Inter-integrated circuit (I²C) is a two-wire addressed bus for short distances on a circuit board; serial peripheral interface (SPI) is a faster four-wire bus; the universal asynchronous receiver-transmitter (UART) gives simple serial links. For field cables, SDI-12 is a low-power serial standard widely used by environmental sensors, letting many addressed sensors share one data wire, and RS-485 differential signalling carries Modbus or similar protocols over hundreds of metres. In agricultural machinery the CAN bus, and the ISO 11783 standard built on it, connect tractor and implement controllers.
A control system manipulates an input to a process so that an output (the controlled variable) follows a desired value (the setpoint). In open-loop control the action does not depend on the output: an irrigation timer runs a valve for 40 minutes regardless of how wet the soil is, and a dryer burner set to a fixed fuel rate gives whatever air temperature results. Open loop is simple and stable but cannot correct for disturbances such as rain, wind or changing ambient temperature. In closed-loop (feedback) control a sensor measures the controlled variable, the controller compares it with the setpoint to form the error, and the controller output drives a final control element: a valve, a variable-speed drive, a heater, a vent motor or a fan. Soil-moisture-triggered irrigation, thermostatic greenhouse heating and dryer air-temperature control are closed-loop systems. Feedback corrects for disturbances and for changes in the process, but it can oscillate if badly tuned. Feedforward control measures a disturbance and acts before it affects the output, for example increasing vent opening when a solar-radiation sensor shows a sudden rise; it is usually combined with feedback. Cascade control uses an inner loop around a fast secondary variable (burner air temperature) commanded by an outer loop on a slow primary variable (grain outlet moisture or temperature).
A transfer function is the ratio of the Laplace transform of the output to that of the input, with zero initial conditions. A first-order process such as a greenhouse air mass or a heated tank is G(s) = K/(τs + 1), with steady-state gain K and time constant τ. Dead time (transport delay) θ, such as the time for air to travel from a burner to a sensor or for grain to move through a column dryer, multiplies the transfer function by e^(−θs). Blocks in series multiply; blocks in parallel add. For a forward path G and feedback path H with negative feedback, the closed-loop transfer function is G/(1 + GH). The final value theorem gives the steady-state value of a stable system: y(∞) = lim s→0 of s·Y(s). For a unit step setpoint change and a unity-feedback loop whose open-loop transfer function has a finite steady-state gain K0 (a type-0 system), the steady-state error is 1/(1 + K0). A proportional controller Kp on a process of gain K gives K0 = Kp·K, so the error shrinks as gain rises but never reaches zero. Integral action adds a pole at the origin (a type-1 system), which drives the steady-state error to a step to zero. Closing a proportional loop around a first-order process also speeds it up: the closed-loop time constant becomes τ/(1 + KpK).
A linear system is stable if every pole of its closed-loop transfer function (every root of the characteristic equation 1 + GH = 0) has a negative real part, that is, lies in the left half of the s-plane. Poles on the imaginary axis give sustained oscillation; poles in the right half give growing oscillation or runaway. The Routh–Hurwitz criterion tests stability from the coefficients without solving for the roots. A necessary condition is that all coefficients are present and have the same sign; for a cubic s³ + a2 s² + a1 s + a0 with positive coefficients the condition is a2·a1 > a0. A second-order characteristic equation s² + 2ζωn s + ωn² gives the natural frequency and damping ratio directly; peak overshoot, rise time and settling time follow as for second-order instruments. In agricultural processes the main threat to stability is dead time combined with high controller gain: a grain dryer whose outlet moisture sensor sees the effect of a burner change only after the grain has travelled through the column will oscillate if the controller acts aggressively on the delayed reading. Detuning the controller, cascade loops and model-based control are the remedies.
The on-off (two-position) controller switches its output fully on or fully off. To prevent rapid switching (chatter) around the setpoint it has a differential, or hysteresis band: a greenhouse heater set to 18 °C with a 2 °C differential might switch on at 17 °C and off at 19 °C. The controlled variable therefore cycles continuously. For a process that heats at rate rh and cools at rate rc between switching points separated by a differential D, ignoring lags, the cycle time is D/rh + D/rc; lags in the process and sensor make the actual swing larger than the differential. A wider differential reduces switching frequency and wear on contactors, valves and burners at the cost of a larger swing. On-off control is cheap and adequate where a swing is acceptable: stock-tank heaters, solenoid irrigation valves triggered by a soil-moisture threshold, cold-store compressors and simple ventilation fans. A proportional (P) controller makes its output proportional to the error: u = Kp·e + u0, where u0 is the bias output when the error is zero. Proportional band (PB) is the percentage of the measurement span over which the output moves from 0 to 100%; for a controller with input and output both in percent, Kp = 100/PB. A PB of 20% on a 0–100 °C range means the output goes from full to zero as the temperature moves 20 °C. P control gives a smooth, stable response with a narrow enough band, but it leaves a permanent offset whenever the load differs from the condition at which the bias was set, because a non-zero error is needed to produce the changed output. Raising the gain reduces the offset but brings oscillation closer.
Integral (reset) action adds an output proportional to the time integral of the error. As long as any error remains, the integral keeps changing the output, so offset is removed. In the standard (ideal) PID form, u = Kp[e + (1/Ti)∫e dt + Td de/dt], where Ti is the integral (reset) time and Td the derivative time. Smaller Ti means stronger integral action. Integral windup occurs when the output saturates (a valve fully open, a burner at maximum) while error persists; the integral keeps growing, and when the process finally responds it overshoots badly. Anti-windup logic stops or limits the integral while the output is saturated. Derivative (rate) action responds to the rate of change of the error, anticipating where the process is heading and adding damping, which is useful on slow processes with several lags such as large dryers and heated tanks. It amplifies measurement noise, so it is filtered, and it is often applied to the measured variable rather than the error to avoid a spike when the setpoint is changed. Derivative is little used on noisy, fast loops such as liquid flow. Most industrial loops run as PI; PID is reserved for slow, smooth processes. Tuning chooses Kp, Ti and Td. The Ziegler–Nichols ultimate-cycle method sets integral and derivative off, raises the proportional gain until the loop oscillates with constant amplitude, and records the ultimate gain Ku and the ultimate period Pu. The recommended settings are: P only, Kp = 0.5 Ku; PI, Kp = 0.45 Ku and Ti = Pu/1.2; PID, Kp = 0.6 Ku, Ti = Pu/2 and Td = Pu/8. These settings give a quarter-amplitude decay, which is quite oscillatory, so they are a starting point that is usually softened. The process reaction curve (open-loop step test) method fits a first-order-plus-dead-time model and sets gains from the gain, time constant and dead time. Driving a live process to sustained oscillation is not always acceptable, for example on a dryer full of seed, so step tests or relay auto-tuning are preferred.
Greenhouse climate control combines several loops. Heating is often P or PI on air temperature, through a mixing valve on a hot-water pipe system. Ventilation opens roof vents in proportion to the excess of temperature over a ventilation setpoint, with limits for wind speed and rain. Humidity is managed by VPD targets using vents, heating and fogging, and the loops interact, so the controller sets priorities. Sensors sit in an aspirated screen at crop height, not in direct sun. Irrigation control ranges from open-loop timers, through on-off control triggered when soil water falls to a set allowable depletion, to feedback on soil-water tension with flow meters confirming delivered volume. Fertigation dosing uses PI control of the electrical conductivity and pH of the mixed solution, with dosing pumps or venturi injectors as final elements and alarms for out-of-range values. A continuous-flow grain dryer controls plenum air temperature with a modulating burner (fast loop, usually PI), and controls grain outlet moisture through the discharge rate. The moisture loop has long dead time, so it is detuned or uses a model that predicts outlet moisture from inlet moisture and air temperature. Safety interlocks (flame failure, high-limit thermostat, fan airflow proof) are separate hardwired functions and must never depend on the control loop.
A programmable logic controller (PLC) is a ruggedised industrial computer for sequential and continuous control. It has a power supply, a processor (CPU) with memory, input modules (digital inputs from switches, level probes and motor-overload contacts; analog inputs, typically 4–20 mA or 0–10 V, from transmitters), output modules (relay, transistor or triac digital outputs to contactors, solenoid valves and indicator lamps; analog outputs to valve positioners and variable-speed drives) and communication ports. PLCs are built for electrical noise, wide temperature ranges and vibration, and their I/O is isolated. The PLC runs a repeating scan: it reads all inputs into an input image table, executes the user program from top to bottom using that image, then writes the results to the outputs, and performs housekeeping and communication. Scan time is typically a few milliseconds to tens of milliseconds, depending on program size. An input pulse shorter than one scan can be missed, so fast pulses (flow-meter pulses, shaft encoders) go to high-speed counter modules or interrupt inputs. Because outputs update only at the end of the scan, the order of rungs can matter.
The international standard IEC 61131-3 defines PLC programming languages: ladder diagram (LD), function block diagram (FBD), structured text (ST), sequential function chart (SFC), and instruction list (IL), which the third edition of the standard deprecated. Ladder diagram imitates relay wiring: rungs between two vertical power rails, with normally open and normally closed contacts in series (AND) or in parallel (OR) driving output coils. A seal-in (latching) circuit puts a normally open contact of the output in parallel with the momentary start button and a normally closed stop contact in series, so the motor stays on after the start button is released and drops out on stop or loss of power. Timers include on-delay (TON: output turns on after the input has been on for the preset time), off-delay (TOF) and pulse timers. Preset values are often entered as a count of the time base, so 30 s on a 0.1 s base is a preset of 300. Counters count up or down and close a contact when the preset is reached, for example counting bags filled. Analog values are scaled from raw counts to engineering units, and PID function blocks are standard.
Grain handling plants use PLCs for start-up and shut-down interlocks. Conveyors are started in the reverse direction of grain flow, the last (downstream) machine first, so nothing is fed onto a stopped conveyor; they are stopped in the direction of flow, the first (upstream) machine first, so each conveyor empties before it stops. A fault on any machine (motor overload, belt-misalignment or speed switch, choke-up sensor) stops all machines feeding it. Pump stations for irrigation use PLCs to sequence pumps, protect against dry running and low or high pressure, and log hours. Dairy plants use PLCs and SCADA for clean-in-place sequences and pasteuriser flow diversion. A supervisory control and data acquisition (SCADA) system or a human–machine interface (HMI) gives operators displays, alarms and trends on top of PLCs. Emergency stops and other safety functions must not rely on ordinary PLC software; they use hardwired circuits or certified safety controllers designed to functional-safety standards, and the emergency-stop button uses a normally closed contact so that a broken wire stops the machine (fail-safe). The same fail-safe idea underlies wiring stop buttons and high-limit switches as normally closed contacts.
An Internet of Things (IoT) system for agriculture has three layers. The perception (device) layer is the sensor node: sensors, a microcontroller, a radio, a power source and an enclosure. The network layer carries data from nodes to a gateway, and from the gateway over cellular, satellite or wired internet to a server. The application layer stores, analyses and presents data and sends commands back, for example to open a valve. Nodes are arranged in a star (each node talks directly to a gateway), a mesh (nodes relay for one another, which extends range but costs power in relaying nodes), or a tree combining both. Edge computing processes data on the node or gateway, sending only summaries or alarms, which saves energy and bandwidth and keeps local control working when the internet link fails. Typical farm uses are soil-water profiles for irrigation scheduling, microclimate and leaf-wetness stations for disease forecasting, grain-bin temperature and humidity cables, water-tank and trough levels, cold-room and milk-tank temperatures, animal-house air quality (ammonia, carbon dioxide), and machine telematics.
Radio technologies trade range, data rate and power. Short-range personal-area radios and IEEE 802.11 wireless LAN give high data rates over tens of metres but use too much power for long battery life. IEEE 802.15.4-based mesh networks use low power and relay over hundreds of metres. Low-power wide-area networks (LPWAN) send very small packets over several kilometres in open country at a very low data rate, either in licence-free sub-gigahertz bands with a private gateway or over the public cellular network using the 3GPP narrowband IoT and LTE-M standards. Range in the field is reduced by crop canopy, terrain, buildings, and a node antenna close to the ground, so antenna height matters greatly. Licence-free bands have national rules on frequency, transmit power and duty cycle, which differ by region. Above the radio, lightweight messaging protocols carry the data. MQTT uses a publish–subscribe model through a broker: a node publishes readings to a topic and any subscribed application receives them. CoAP is a lightweight request–response protocol for constrained devices.
Battery life is set by the average current. A node that sleeps at a few microamps and wakes for a few seconds to measure and transmit at tens of milliamps has an average current = (I_active·t_active + I_sleep·t_sleep)/period; battery life is capacity divided by average current, reduced in practice by self-discharge, cold and battery ageing. Small solar panels with rechargeable cells extend life but must be sized for cloudy periods and shading by crops. Cheap sensors make dense networks possible, but data quality limits their value. Low-cost soil-moisture and humidity sensors need calibration against reference methods, checks for drift, and installation records. Automated quality control flags out-of-range values, sudden steps, flat lines and gaps. A decision such as "irrigate now" should be checked against more than one sensor or a water-balance model, because a single failed probe can drown or starve a field. Security is part of design: unique credentials per device, encrypted links, signed firmware updates, and no default passwords, especially where the network can actuate pumps, valves or ventilation. Data ownership and sharing terms between farmers, service providers and equipment makers vary by country and contract and should be agreed in writing.
A B.Tech student uses this subject in every laboratory: wiring a thermocouple with proper cold-junction compensation, building a strain-gauge bridge on a cantilever, calibrating a capacitance grain-moisture meter against an oven, and writing a small microcontroller program that logs soil water and air temperature. In exams the same student must convert between wet and dry basis, compute a bridge output or an ADC step, state the Ziegler–Nichols settings, and explain why a P controller leaves offset. Learning to state the basis and uncertainty of every measured number is the habit that matters most.
An engineer five years into practice specifies and commissions instruments: choosing a flow meter for a dirty-water irrigation main (electromagnetic rather than turbine), siting soil-moisture probes and setting the irrigation trigger, setting dryer burner PI loops and high-limit interlocks, writing PLC sequences for grain-handling conveyors, and checking that a greenhouse humidity sensor has not drifted. Typical decisions are which sensor survives the environment, how often it must be recalibrated, which loops can be on-off and which need PI, and what alarms must be hardwired.
A senior engineer with 25 years' experience designs the measurement and control strategy for whole facilities: grain terminals, dairy plants, protected-cropping estates and district irrigation schemes. The work is about reliability and life-cycle cost: standardising sensors and protocols, deciding where redundancy is needed, setting calibration and maintenance schedules, defining cybersecurity rules for remote-controlled pumps and vents, and judging vendors' claims about low-cost sensor networks against measured field performance. Such an engineer also signs off on safety functions and investigates incidents such as dryer fires or failed pasteurisation records.
A researcher with a PhD and long experience develops and validates new sensing and control methods: dielectric and spectroscopic moisture sensing for new crops, soil-sensor calibration across soils and salinity, model-based and predictive control of dryers and greenhouses, uncertainty analysis of yield monitors and draft dynamometers, and the design of field IoT networks whose data quality can support scientific conclusions. The core skills are rigorous calibration against reference methods, uncertainty propagation, system identification of processes with dead time, and honest reporting of where a sensor fails.
The manual gives working principles and the standard equations but does not reproduce thermocouple reference tables, standard RTD tolerance-class values, or crop-by-crop oven times and temperatures for grain moisture; use the relevant IEC, ISO, ASABE or national standard for those. It does not cover machine vision, spectral imaging, drones or robotics (see the precision agriculture and drone subjects), frequency-domain controller design (Bode and Nyquist plots) beyond a mention, digital control and z-transforms, model-predictive control, or the national radio-licensing rules for IoT bands, which vary by country. Safe grain temperatures for drying are left to crop-specific guidance because they vary by crop and end use.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.