Sensors, Instrumentation and Data Acquisition

Introduction and Context

A mechatronic system can act intelligently only when it can obtain trustworthy information about its physical environment and internal condition. Sensors convert physical variables—such as position, temperature, force, pressure, speed, acceleration, light, or chemical concentration—into signals that an electronic controller can process. Instrumentation includes the complete measurement chain: sensing, signal conditioning, conversion, transmission, computation, display, storage, and diagnostic interpretation. Data acquisition (DAQ) is the organized capture of these signals for monitoring, control, testing, and analysis.

In an advanced mechatronic product, sensor selection is not an isolated component choice. It is a systems-engineering decision involving mechanical installation, electrical compatibility, dynamic response, calibration, uncertainty, electromagnetic compatibility, sampling, software, safety, maintenance, and cost. A sensor with excellent laboratory accuracy may perform poorly if it is badly mounted, inadequately shielded, sampled too slowly, or exposed to an unsuitable temperature range.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Classify sensors and explain their operating principles.
2. Select sensors using range, sensitivity, resolution, bandwidth, uncertainty, environment, and interface requirements.
3. Design basic signal-conditioning and data-acquisition chains.
4. Analyze calibration, error, uncertainty, noise, sampling, aliasing, and filtering.
5. Explain sensor fusion and the value of diagnostic information.
6. Evaluate an integrated sensing solution for a mechatronic application.

1. The Measurement Chain

A general measurement system consists of the following stages:

Physical variable → sensor → signal conditioning → analog-to-digital conversion → computation and storage → decision, control, or display.

The measured variable is often called the measurand. The sensor responds to the measurand, while a transducer converts one form of energy or information into another. In engineering practice, the terms sensor and transducer are often used interchangeably, although a transducer may also include actuation devices.

A useful static model is:

Here, x is the true measurand, y is the indicated output, f(x) is the ideal sensor response, b is a systematic bias, and n is random disturbance or noise. This model assumes that the error sources can be represented as an additive bias and additive random component. In real systems, gain error, nonlinearity, hysteresis, drift, and cross-sensitivity may also be present.

For an approximately linear sensor:

where a is sensitivity, with units of output per unit input, and b is the zero offset. If the output is voltage, sensitivity may be expressed in volts per degree Celsius, volts per newton, or millivolts per millimetre.

A complete specification must distinguish:

• Range: minimum to maximum measurable input.
• Span: upper range value minus lower range value.
• Sensitivity: change in output per change in input.
• Resolution: smallest input change that can be distinguished.
• Accuracy: closeness of a result to a reference or true value.
• Precision: closeness of repeated measurements to one another.
• Repeatability: precision under the same conditions.
• Reproducibility: agreement under changed conditions, such as operator or instrument.
• Bandwidth: frequency range over which the response remains acceptable.
• Response time: time required to follow a change in input.
• Drift: change in output over time when the measurand is constant.
• Hysteresis: different output for increasing and decreasing input at the same input value.
• Loading: disturbance of the measurand or source caused by the measurement system.

Accuracy and precision are not interchangeable. A sensor may produce tightly clustered but biased measurements, or widely scattered measurements centered near the reference value.

2. Sensor Technologies and Selection

2.1 Common sensor classes

Sensors may be classified by the physical quantity measured, operating principle, output form, or energy requirement.

Sensors may also be active or passive. An active sensor generates an output without external excitation, as a thermocouple or piezoelectric sensor does. A passive sensor requires excitation, as an RTD, strain gauge, or potentiometer does. This distinction affects power, wiring, noise, and safety.

2.2 Selection criteria

Selection begins with the measurand and required decision, not with a preferred technology. The engineer should specify:

1. Physical range, overload, and expected rate of change.
2. Required accuracy, resolution, repeatability, and uncertainty.
3. Frequency content and allowable phase delay.
4. Environmental exposure: temperature, humidity, vibration, shock, dust, chemicals, and electromagnetic fields.
5. Mechanical constraints, mounting, alignment, and loading.
6. Electrical interface: analogue voltage/current, bridge, pulse, frequency, digital bus, or wireless link.
7. Power, isolation, intrinsic safety, and fail-safe behavior.
8. Calibration method, service interval, availability, lifetime, and total cost.

A high-resolution sensor is not automatically accurate. For example, a 16-bit converter connected to a poorly calibrated sensor does not create 16-bit measurement accuracy.

2.3 Static and dynamic behavior

Many sensors can be approximated as a first-order system:

where τ is the time constant in seconds, K is the static gain, x(t) is the input, and y(t) is the output. This model assumes linearity, time invariance, and negligible higher-order dynamics. For a step input of amplitude X₀, with zero initial output:

At t = τ, the output reaches approximately 63.2% of its final value; after about five time constants it reaches more than 99%. A sensor with a small time constant responds rapidly but may pass more high-frequency noise.

A second-order model is often more appropriate for accelerometers, mechanical structures, and servo mechanisms:

Here, ω_n is natural angular frequency in radians per second, ζ is damping ratio, and s is the Laplace variable. This model applies to a linear second-order system. Resonance and phase lag must be considered when the measurement is used in feedback control.

3. Signal Conditioning

Raw sensor outputs are rarely suitable for direct conversion. Signal conditioning may provide excitation, amplification, filtering, linearization, isolation, protection, and conversion between voltage and current.

3.1 Amplification and bridges

Strain gauges and resistance temperature detectors are often used in bridge circuits. In a balanced Wheatstone bridge, small resistance changes produce a differential voltage. An instrumentation amplifier then provides high input impedance, high common-mode rejection, and accurately controlled gain.

For a differential amplifier, the ideal output is:

where G is differential gain and V_+ and V_− are input voltages. Real amplifiers also have input offset voltage, bias currents, finite common-mode rejection ratio, gain error, noise, and limited input/output range.

The common-mode rejection ratio is:

where A_d is differential gain and A_cm is common-mode gain. CMRR is expressed in decibels. High CMRR is important when a small sensor signal is superimposed on a large common voltage.

3.2 Filtering

An analogue low-pass filter suppresses high-frequency components before conversion. An idealized first-order low-pass transfer function is:

Its cutoff frequency is:

where f_c is in hertz and τ is in seconds. At the cutoff frequency, the magnitude is approximately 0.707 of the low-frequency value, corresponding to −3 dB. Real filters introduce phase shift; therefore, filtering can affect control-loop stability and event timing.

Other conditioning functions include:

• Linearization: converting a nonlinear sensor response into a more convenient scale.
• Cold-junction compensation: required for accurate thermocouple measurement.
• Galvanic isolation: preventing unwanted current paths and improving safety.
• Excitation regulation: stabilizing bridge or resistive sensor supply.
• Surge protection: limiting electrostatic discharge, transients, and overvoltage.
• Shielding and grounding: reducing electric-field interference while avoiding ground loops.

A shield should normally be connected according to the system grounding strategy, often at one end for low-frequency analogue circuits. High-frequency installations may require a different, carefully engineered termination. There is no universal grounding rule independent of frequency and topology.

4. Analogue-to-Digital Conversion and DAQ Architecture

An analogue-to-digital converter (ADC) maps a continuous input range to discrete codes. For an ideal unipolar ADC with N bits and input range from 0 to V_ref, the nominal code width is:

where q is the least significant bit size in volts. The quantization uncertainty for rounding is commonly modeled as uniformly distributed between −q/2 and +q/2, with standard deviation:

This model assumes an ideal converter and sufficiently variable signal; a constant or highly correlated signal may not behave like independent uniform quantization noise.

The effective number of bits (ENOB) is lower than the nominal resolution when noise and distortion are present. For a measured signal-to-noise-and-distortion ratio, one common relationship is:

This relationship assumes a sinusoidal test signal and the conventional ideal-ADC model.

DAQ architectures include:

• Centralized DAQ: sensors connect to one acquisition unit; simple for compact systems but vulnerable to long analogue wiring.
• Distributed DAQ: local modules digitize signals near sensors; reduces analogue cable length and can improve scalability.
• Embedded acquisition: a microcontroller or industrial computer integrates conversion and processing.
• Networked acquisition: modules communicate using CAN, EtherCAT, industrial Ethernet, RS-485, or similar buses.

Important interface choices include voltage input, 4–20 mA current loop, frequency or pulse input, thermocouple input, bridge input, and digital protocols such as I²C, SPI, CAN, or Ethernet. A 4–20 mA loop is relatively robust over long industrial cables and allows a live-zero diagnostic: a signal near 0 mA can indicate a broken wire, whereas 4 mA represents the lower measurement value.

5. Sampling, Aliasing, and Digital Filtering

A continuous signal must be sampled at discrete times. If the highest significant frequency in the signal is f_max, the Nyquist condition requires:

where f_s is sampling frequency. This condition is necessary for ideal band-limited signals, but practical systems sample substantially faster to allow filter transition bands, timing uncertainty, and non-ideal signal content.

Aliasing occurs when high-frequency content appears as a false lower-frequency component after sampling. An anti-aliasing low-pass filter must therefore be placed before the ADC. Digital filtering after conversion cannot recover information already aliased.

For a sampled sinusoid, frequencies separated by integer multiples of f_s are indistinguishable in the sampled data. A 700 Hz sinusoid sampled at 1,000 samples per second appears at 300 Hz because its discrete-time representation is equivalent to that of the lower-frequency component.

Digital filters may be finite impulse response (FIR) or infinite impulse response (IIR). FIR filters can provide exact linear phase and guaranteed stability, but may require more computation. IIR filters are computationally efficient but generally introduce nonlinear phase and require careful handling of initial conditions and numerical stability.

Moving-average filtering is simple:

where M is the number of samples, x[k] is the input sequence, and y[k] is the filtered output. For uncorrelated zero-mean noise, the noise standard deviation is reduced approximately by √M, but the filter introduces delay and attenuates genuine signal changes.

Sampling must also account for time synchronization. In multi-sensor systems, unsynchronized timestamps can create apparent phase differences and degrade sensor fusion. Hardware triggering, shared clocks, precision timestamps, or time-synchronization protocols may be required.

6. Calibration and Measurement Uncertainty

Calibration compares an instrument with a reference under defined conditions and establishes a relationship between indication and reference value. Calibration does not necessarily mean adjustment. A calibration record should identify the instrument, reference standards, environmental conditions, procedure, results, uncertainty, date, and traceability.

For a linear two-point calibration, suppose reference inputs x₁ and x₂ produce measured outputs y₁ and y₂. The corrected input estimate for an observed output y can be obtained from:

This assumes a linear relationship between the two calibration points and stable behavior between them. Multi-point calibration may use a fitted polynomial, lookup table, spline, or physical model. Extrapolation outside the calibrated range is generally unsafe.

Uncertainty describes doubt about a measurement result. If independent input quantities contribute standard uncertainties u₁, u₂, …, u_n, and the result is Y = g(X₁, X₂, …, X_n), the law of propagation of uncertainty is:

This simplified form assumes independent inputs and a locally linear model. If inputs are correlated, covariance terms must be included:

where u(X_i, X_j) is covariance. An expanded uncertainty is commonly reported as:

where k is a coverage factor selected according to the desired coverage probability and statistical assumptions. A value near 2 is often used for approximately 95% coverage when the effective degrees of freedom are adequate, but it is not universally valid.

Worked Example 1: Combined uncertainty

A temperature measurement has independent standard uncertainties of 0.20 °C from sensor calibration, 0.10 °C from electronics, and 0.15 °C from repeatability. The combined standard uncertainty is:

Thus:

Using k = 2, the expanded uncertainty is approximately ±0.54 °C, assuming the stated components are independent and the coverage assumptions are appropriate.

7. Noise, Interference, and Information Quality

Noise is unwanted variation; interference is an unwanted signal coupled from an identifiable source. Common sources include thermal noise, shot noise, flicker noise, quantization noise, switching converters, motors, relays, radio transmitters, ground loops, vibration, and mechanical resonance.

Noise may be described in the time domain by variance or standard deviation and in the frequency domain by power spectral density. The root-mean-square value of a zero-mean signal is:

For a sinusoid with peak amplitude X_p, X_rms = X_p/√2. Signal-to-noise ratio is:

when amplitudes are compared consistently. If powers are compared, the corresponding factor is 10 rather than 20.

Good practice includes separating power and signal wiring, using twisted pairs for differential signals, minimizing loop area, selecting suitable cable shields, filtering at the correct location, using differential inputs, bonding enclosures appropriately, and verifying grounding with measured current paths. Filtering should not be used to conceal a poorly designed mechanical or electrical installation.

Information quality is broader than numerical accuracy. A diagnostic system should consider:

• Validity: whether the value is physically plausible.
• Completeness: whether samples are missing.
• Timeliness: whether data arrive before a decision is needed.
• Consistency: whether related measurements agree.
• Traceability: whether values can be linked to calibration and timestamps.
• Availability: whether the channel remains operational.
• Integrity: whether data have been corrupted or altered.

Range checks, rate-of-change checks, stuck-signal detection, redundancy checks, residual analysis, and model-based checks can identify failures. A value that is numerically within range may still be invalid if it is frozen, delayed, or inconsistent with system behavior.

8. Sensor Fusion

A single sensor rarely provides complete knowledge. Sensor fusion combines complementary measurements. For example, an accelerometer responds quickly but drifts when integrated, while a position sensor may be stable at low frequency but noisy or slow. Combining them can improve the estimate.

A basic weighted estimate for independent measurements of the same quantity is:

where x_i is measurement i, σ_i² is its variance, and x_hat is the fused estimate. The formula assumes unbiased measurements, known variances, and independence. Correlated errors require a covariance-based estimator.

The Kalman filter is a recursive model-based fusion method. In conceptual form, it alternates between prediction from a system model and correction using measurements. Its effectiveness depends on reasonable process and measurement-noise models, correct timing, observability, and appropriate handling of nonlinearities. Extended and unscented variants address some nonlinear systems, but they do not remove the need for validation.

Fusion can also be qualitative: agreement between an encoder, motor current, and limit switch can provide a stronger diagnostic conclusion than any one signal alone. However, redundant sensors may share common failure modes, so apparent agreement is not proof of correctness.

9. Integrated Applied Case: Conveyor-Based Precision Assembly

Background

Consider an illustrative case involving a robotic conveyor cell that places metal components into a fixture. The system must detect component presence, measure conveyor position, control gripping force, and identify misalignment before assembly.

Problem

The initial cell uses a low-cost proximity sensor, an incremental encoder, and a force sensor. Intermittent assembly failures occur when electrical noise from the motor drive causes false presence detections. In addition, the encoder count becomes unreliable after power interruptions, and force readings vary with temperature.

Analysis

The engineering team first identifies the information required by each decision. Presence detection is a discrete event, position is required for synchronization, force is needed for quality control, and temperature is needed for compensation. The team measures motor-drive switching frequency, cable routes, sensor bandwidth, sampling times, and observed failure rates rather than assuming that the sensors themselves are defective.

The encoder output is sampled at a rate sufficient for the expected maximum speed and is timestamped by the same controller clock as the presence input. The presence sensor is replaced by an inductive sensor with a shielded, differential interface and a properly designed debounce or filtering strategy. The force sensor is temperature-characterized using calibration points across the operating range.

Approach and solution

The acquisition architecture is distributed. A local module near the force sensor provides bridge excitation, instrumentation amplification, analogue filtering, and 24-bit conversion. The encoder module provides hardware timestamping and a retained reference mark for position recovery. The controller applies plausibility checks:

• A component-present signal must persist for a physically reasonable duration.
• Position must increase consistently with conveyor direction.
• Force must remain within safe mechanical limits.
• A sudden force change without corresponding motion is flagged for inspection.
• Missing, stale, or frozen data are distinguished from valid zero values.

A calibration database stores sensor coefficients, dates, reference standards, and uncertainty. The system records raw and processed values so that a failed assembly can be investigated without relying only on filtered data.

Outcome

The redesigned system reduces false presence detections by controlling cable coupling and applying filtering before threshold decisions. Timestamp alignment improves placement accuracy. Temperature compensation reduces force-measurement bias, while diagnostic checks detect encoder-reference loss and frozen signals. The principal improvement comes not from a single superior sensor, but from treating sensing, wiring, sampling, calibration, software, and diagnostics as one measurement system.

Learning points

1. Sensor selection must begin with the required decision and failure modes.
2. Analogue filtering is essential before sampling when aliasing is possible.
3. Calibration and uncertainty must include installation and environmental effects.
4. Data quality includes timing, completeness, validity, and traceability.
5. Redundancy is most useful when sensors have sufficiently independent failure modes.

10. Safety, Ethics, Governance, and Contemporary Development

Sensor data may influence safety functions, worker monitoring, product acceptance, and automated decisions. Safety-related measurements require defined failure responses, diagnostic coverage, proof testing, and appropriate standards. A non-safety sensor must not be assumed to provide a safety function merely because it appears reliable.

Data governance includes access control, secure communication, retention, audit trails, and protection against unauthorized alteration. In collaborative robots or connected factories, sensor streams may reveal worker behavior, production rates, or proprietary processes. Collection should be limited to legitimate purposes, and personally identifiable information should be protected where applicable.

Artificial intelligence is increasingly used for anomaly detection, predictive maintenance, and soft sensing. AI can identify patterns in vibration, current, temperature, and acoustic data that are difficult to model explicitly. Nevertheless, training data may be biased, sensor drift can invalidate a model, and an unexplained classification may be unsuitable for a safety-critical decision. AI outputs should therefore be accompanied by confidence information, data-quality checks, drift monitoring, and a defined fallback strategy.

Current developments include event-based cameras, fibre-optic sensing, MEMS sensor networks, time-sensitive networking, edge computing, digital twins, self-calibrating sensors, and integrated photonic or wireless sensors. These technologies increase capability but also introduce challenges in cybersecurity, synchronization, maintainability, power management, and lifecycle verification.

11. Engineering Best Practices

A robust sensing design should:

1. Define the measurand, decision, acceptable uncertainty, and failure response.
2. Create a signal-chain budget for range, gain, noise, bandwidth, and delay.
3. Select a sensor with adequate—not merely maximum—resolution and range.
4. Match sampling frequency and anti-aliasing to the physical dynamics.
5. Specify calibration, traceability, recalibration intervals, and environmental limits.
6. Design mechanical mounting and cable routing as carefully as electronics.
7. Record raw data, processed data, timestamps, status flags, and calibration metadata.
8. Test normal operation, overload, disconnection, drift, interference, and power loss.
9. Validate diagnostic thresholds using representative faults rather than arbitrary values.
10. Document assumptions so future engineers can interpret measurements correctly.

Chapter Summary

Sensors and DAQ systems form the information foundation of mechatronic control, monitoring, and diagnosis. Effective design treats the measurement chain as an integrated system comprising the measurand, sensor, conditioning, conversion, communication, computation, and decision function. Sensor choice depends on range, sensitivity, resolution, accuracy, uncertainty, dynamics, environment, interface, cost, and failure behavior.

Signal conditioning provides excitation, amplification, filtering, isolation, protection, and linearization. ADC resolution is limited by quantization, noise, distortion, and reference quality. Sampling must satisfy the Nyquist condition for the relevant signal bandwidth, and anti-aliasing must occur before conversion. Calibration establishes the relationship between indication and reference, while uncertainty quantifies remaining doubt.

Noise reduction requires attention to mechanical, electrical, analogue, digital, and software causes. Information quality includes validity, completeness, timeliness, consistency, traceability, availability, and integrity. Sensor fusion combines complementary data but depends on assumptions about uncertainty, correlation, timing, and system models. Finally, safe and ethical implementation requires diagnostic coverage, cybersecurity, privacy protection, traceability, and explicit fallback behavior.

Key Terms

Measurand: Physical quantity intended to be measured.

Transducer: Device that converts energy or information from one form to another.

Sensitivity: Change in output per unit change in input.

Resolution: Smallest distinguishable input change.

Accuracy: Closeness of a result to a reference or true value.

Precision: Closeness of repeated results to one another.

Hysteresis: Dependence of output on the direction of input change.

Drift: Change in indication over time for a constant measurand.

Bandwidth: Frequency range over which a measurement response is acceptable.

Signal conditioning: Processing that prepares a sensor signal for conversion or use.

Common-mode rejection ratio: Measure of an amplifier’s ability to reject equal signals at both inputs.

ADC: Analogue-to-digital converter that maps analogue values to digital codes.

Quantization: Representation of a continuous amplitude by discrete code levels.

Aliasing: False low-frequency representation caused by inadequate sampling.

Calibration: Comparison with a reference to establish measurement relationship and performance.

Uncertainty: Quantified doubt associated with a measurement result.

Sensor fusion: Combination of measurements to produce a better estimate or diagnosis.

Traceability: Ability to relate a measurement to stated references through documented comparisons.

Diagnostic coverage: Ability of a system to detect specified faults.

Information integrity: Assurance that data remain correct, complete, and unaltered.

Review Questions

Short Answer Questions

1. Distinguish accuracy, precision, resolution, and sensitivity.
2. Why is an anti-aliasing filter placed before an ADC?
3. State two differences between active and passive sensors.
4. What information is provided by a calibration record?
5. Why can a high-resolution ADC fail to provide an accurate measurement?

Long Answer Questions

1. Explain the complete measurement chain in a mechatronic system and discuss the purpose of each stage.
2. Compare centralized, distributed, embedded, and networked DAQ architectures.
3. Explain calibration and uncertainty propagation, including the assumptions behind the root-sum-square method.
4. Discuss the effects of sensor dynamics, filtering, delay, and sampling on a feedback-control system.
5. Explain how electrical interference can enter a measurement system and propose engineering countermeasures.

Analytical / Application Questions

1. A temperature signal contains useful content up to 120 Hz. Select a practical sampling frequency and explain how you would design the anti-aliasing filter, considering that the ideal Nyquist limit is only a minimum condition.
2. Two independent position sensors report 50.0 mm and 50.8 mm with standard uncertainties of 0.2 mm and 0.6 mm, respectively. Derive the inverse-variance weighted estimate and explain the assumptions behind it.
3. Design a diagnostic strategy for a robotic joint using an encoder, motor current sensor, temperature sensor, and limit switch. Identify at least three independent fault indicators and specify safe responses to missing or contradictory data.

Multiple Choice Questions with Answers

Question 1

Which quantity describes the change in sensor output per unit change in input?

A. Sensitivity

B. Resolution

C. Hysteresis

D. Drift

Correct Answer: A. Sensitivity

Question 2

For an ideal unipolar ADC with N bits and reference voltage V_ref, what is the nominal code width?

A. V_ref × 2^N

B. V_ref/2^N

C. 2^N/V_ref

D. V_ref/N

Correct Answer: B. V_ref/2^N

Question 3

A first-order sensor reaches approximately what percentage of its final step response after one time constant?

A. 25.0%

B. 50.0%

C. 63.2%

D. 99.0%

Correct Answer: C. 63.2%

Question 4

A signal contains significant frequency content up to 400 Hz. Which sampling frequency satisfies the ideal Nyquist minimum?

A. 200 samples/s

B. 400 samples/s

C. 600 samples/s

D. 900 samples/s

Correct Answer: D. 900 samples/s

Question 5

What is the principal purpose of an anti-aliasing filter?

A. Increase ADC bit depth

B. Suppress out-of-band content before sampling

C. Eliminate all sensor bias

D. Synchronize independent clocks

Correct Answer: B. Suppress out-of-band content before sampling

Question 6

If three independent standard uncertainties are 0.20 °C, 0.10 °C, and 0.15 °C, what is their combined standard uncertainty?

A. 0.15 °C

B. 0.27 °C approximately

C. 0.45 °C

D. 0.0725 °C

Correct Answer: B. 0.27 °C approximately

Question 7

Which interface commonly uses a live zero and is robust for long industrial cable runs?

A. 4–20 mA current loop

B. Unshielded thermistor wire

C. Potentiometer output only

D. Single-ended microphone input

Correct Answer: A. 4–20 mA current loop

Question 8

Which statement about digital filtering is correct?

A. It can recover information already lost through aliasing

B. It always eliminates phase delay

C. It may reduce noise but can introduce delay or attenuation

D. It makes analogue signal conditioning unnecessary

Correct Answer: C. It may reduce noise but can introduce delay or attenuation

Question 9

The inverse-variance weighted fusion estimate assumes, in its basic form, that measurements are:

A. Biased and perfectly correlated

B. Unbiased, independent, and of known variance

C. Binary and asynchronous

D. Uncalibrated but high resolution

Correct Answer: B. Unbiased, independent, and of known variance

Question 10

Which is an example of a data-quality diagnostic rather than merely a numerical range check?

A. Detecting a frozen signal that remains constant despite known motion

B. Increasing the ADC reference voltage

C. Replacing a low-pass filter with a high-pass filter

D. Changing millimetres to metres

Correct Answer: A. Detecting a frozen signal that remains constant despite known motion