How Do Industrial Sensors Acquire Data Acquisition?

• 1
  Key takeaways
• 2
  How industrial sensors acquire data acquisition: the six-stage chain
• 3
  Analog and digital sensors acquire data differently
• 4
  Sampling rate, filtering, and resolution: picking numbers that work
• 5
  Why acquired data drifts, spikes, or reads zero
• 6
  How to set up sensor data acquisition in 7 steps
• 7
  Signal type comparison for industrial sensors acquire data acquisition

Every pressure, temperature, flow, or vibration reading travels the same path. The sensing element changes a physical quantity into an electrical one. A bridge circuit outputs a few millivolts. A thermocouple outputs microvolts per degree. A photodiode outputs nanoamps. None of these is a number yet.

Stage two is signal conditioning. The circuit amplifies, filters, isolates, and linearizes the raw signal. A load cell at 2 mV/V excitation might produce 20 mV at full scale. A precision instrumentation amplifier with gain 100 brings that to 2 V, which an ADC can resolve properly. Skip the amplifier and you digitize mostly noise.

Stage three is sampling. The ADC takes a snapshot at fixed intervals. The interval must be at least twice the highest frequency you care about, and industrial practice uses five to ten times for margin. A 100 Hz vibration component sampled at 200 Hz is technically legal but the waveform looks rough.

Stage four is conversion. The ADC turns the sampled voltage into a binary integer. A 16-bit converter over a 0–10 V range has a resolution step near 153 µV. A 24-bit converter in the same range steps near 0.6 µV. That extra resolution only helps if the front end is quiet.

Stage five is the protocol. The reading leaves the module over 4–20 mA, 0–10 V, Modbus RTU, IO-Link, EtherNet/IP, or MQTT. Stage six is storage and time stamping. Without a stable clock, two sensors on the same machine cannot be correlated later.

• 1
  Analog pathSensor, amplifier, filter, ADC, controller
• 2
  Digital pathSensor, bus transceiver, protocol stack, gateway
• 3
  Common weak pointGround loops and unshielded cable runs

An analog sensor hands you a continuous voltage, current, or resistance. You own the accuracy of everything downstream, including the cable. A 4–20 mA loop is the workhorse in plants because current does not drop over 100 m of copper the way voltage does.

A digital sensor does part of the work internally. It samples, linearizes, and often calibrates inside the housing, then sends a packet. IO-Link, for example, carries the measured value plus device status and parameter data over a standard three-wire cable.

The trade-off is latency and update rate. A digital sensor reports at its own cycle time, often 1–10 ms. If you need microsecond timing, an analog front end with a dedicated ADC still wins.

Mixed systems are normal. Vibration goes analog into a fast card. Temperature goes digital onto a bus. The controller merges both streams. Decide per channel, not per machine.

One more point that buyers miss: digital does not automatically mean accurate. A digital sensor with poor internal compensation sends a precise wrong number.

• 1
  Choose analog whenFast transients, tight timing, simple loops
• 2
  Choose digital whenMany points, long cable runs, device diagnostics
• 3
  AvoidMixing floating and grounded supplies on one ADC card

Sampling errors are the most common reason a system looks fine on the bench and fails on the machine. The rule is simple. Sample at least ten times faster than the fastest event you must catch. For a 50 Hz motor fault, that means 500 Hz or more.

Aliasing is the trap. If a 900 Hz noise signal enters an ADC sampling at 1,000 Hz, it appears as a 100 Hz ghost in your data. No filter in software can remove it afterwards. You need an anti-aliasing filter before the converter.

Resolution is not accuracy. A 24-bit ADC on a noisy 5 V supply may deliver 14 useful bits. Read the effective number of bits from the datasheet, not the marketing headline. In practice, ±0.005 mm position feedback and thermocouple readings near ±0.5 °C both depend on the front end, not the bit count alone.

Filtering should match the signal. A moving average over 100 samples costs you response time. A single-pole RC at 10 Hz before the ADC costs almost nothing and kills radio-frequency pickup. Use hardware first, software second.

Record the sample rate, filter setting, and clock source in your documentation. When a trend looks wrong six months later, that record is the only way to prove the data is trustworthy.

• 1
  Temperature1–10 Hz, 12–16 bit, 0.1 °C target
• 2
  Pressure100–1,000 Hz, 16 bit, 0.1% full scale
• 3
  Vibration10–50 kHz, 24 bit, AC coupled

A reading that wanders slowly is usually a ground loop. Two grounded points on the same shield create a current path, and that current appears as an offset. Break the loop with an isolated transmitter or an isolated amplifier.

A reading that spikes when a motor starts is inductive coupling. The fix is physical: separate signal and power trays, use twisted pair, and add a ferrite on the signal cable near the cabinet entry. Check the shield termination before touching the software.

A reading stuck at zero often means the loop lost excitation. Measure voltage across the transmitter terminals with the loop powered. On a 4–20 mA device, anything below 3.6 mA signals a fault, and the controller should be configured to flag it rather than clamp to zero.

Random noise across the whole range points at the ADC reference or the supply. A reference that shares a rail with relay coils will move. Give the analog section its own linear regulator and a star ground point.

Finally, check the clock. Two data loggers without synchronized time produce trends that cannot be compared. Use NTP or a hardware trigger when channels must align.

• 1
  Slow driftGround loop or thermal EMF at junctions
• 2
  Motor-start spikesCable routing or missing ferrite
• 3
  Stuck at zeroExcitation loss or broken loop
• 4
  Broadband noiseReference or supply contamination