Industrial temperature measurement and thermowells

Why it matters

Temperature is the most measured variable in process plants: reactor control, distillation, furnaces, steam systems and product quality all depend on it. In the plant, the sensor (thermocouple or RTD) is only part of the job: it sits in a thermowell, connects through lead wires or a head-mounted transmitter, and lags behind the process. Thermowell failure from flow-induced vibration has caused serious leaks and fires, so its mechanical design matters as much as its accuracy.

Key ideas

Sensor choice.

• Thermocouples (types J, K, T, E, N; R, S, B for very high temperatures): Seebeck emf depends on the hot- and cold-junction temperatures. Wide range (about −200 °C to 1700 °C depending on type), rugged, fast, cheap; need cold-junction compensation and matching extension or compensating cable; accuracy typically ±1–2 °C or a fraction of a per cent.
• RTDs (Pt100 to IEC 60751, α = 0.00385 /°C): more accurate and stable (class A about ±0.15 °C at 0 °C), nearly linear, but limited to roughly −200 °C to 600 °C and slower and more fragile.
• Thermistors: very sensitive over a narrow range; used in equipment rather than process lines.
• Filled-system and bimetal thermometers: local indication without power. Radiation pyrometers: non-contact, for furnaces, moving or very hot targets.

Wiring and transmitters.

• A 2-wire RTD adds both lead resistances to the reading; 3-wire connections cancel equal lead resistances in a bridge; 4-wire connections eliminate them completely (laboratory and custody use).
• Thermocouples must be extended with matching extension or compensating cable; ordinary copper wire creates new junctions and errors. The cold junction is measured and compensated at the transmitter terminals.
• Head-mounted or rail-mounted transmitters convert to 4–20 mA (often HART), with linearisation, sensor-break detection and isolation.

Thermowells.

• A closed-end tube (drilled bar stock for pressure service) welded, screwed or flanged into the pipe or vessel. It protects the sensor from pressure, flow, corrosion and erosion and lets the sensor be removed without shutting down.
• Shapes: straight, tapered (stiffer, higher natural frequency, better response at the tip) or stepped.
• Immersion length: the tip must be deep enough that heat conducted along the well to the cooler pipe wall does not bias the reading. A common rule is at least about 10 tip diameters, more for gases and low velocities; insulating the external part reduces stem conduction error.
• Response: the well adds thermal mass and contact resistance, so the time constant may rise from a few seconds (bare sensor) to tens of seconds. Spring-loaded sensors, tight bores and thermal paste or oil reduce it. A first-order sensor following a ramp lags by τ·(rate of rise).
• Flow-induced vibration: vortices shed from the well at f_s = St·V/d with St ≈ 0.22 over the usual Reynolds-number range. If f_s approaches the well's natural frequency f_n, transverse resonance can fatigue and snap the well. The classic rule kept f_s below about 0.8·f_n; the current ASME PTC 19.3 TW code also guards against in-line resonance near f_s ≈ f_n/2 and adds stress checks, so use the code (or the vendor's calculation) for real designs.
• The natural frequency of a cantilevered well falls as 1/L², so long wells in high-velocity flows are the dangerous combination; shorter, thicker or tapered wells, or a velocity collar, are the fixes.
• Material is chosen for corrosion and temperature (316 stainless, Inconel, Hastelloy, titanium; ceramic sheaths in furnaces).

Formulas

R_T = R₀·(1 + α·T) (Pt100, linear approximation; R₀ = 100 Ω, α = 0.00385 /°C)

ΔT_lead ≈ (2·R_lead)/(R₀·α) (2-wire RTD error)

E(T, 0) = E(T, T_ref) + E(T_ref, 0) (thermocouple law of intermediate temperatures)

T_sensor(t) lag on a ramp = τ·R (first-order sensor, steady ramp rate R)

f_s = St·V/d (vortex shedding frequency)

f_n = (1.875² / (2π·L²))·√(E·I/m′) (uniform cantilever, first mode)

I = π·(D⁴ − d_b⁴)/64, m′ = ρ_w·π·(D² − d_b²)/4

Symbols: R_T = resistance at T (Ω); T = temperature (°C); R_lead = resistance of each lead (Ω); E(T₁, T₂) = thermocouple emf (V); τ = time constant (s); R = ramp rate (°C/s); f_s = shedding frequency (Hz); St = Strouhal number; V = fluid velocity (m/s); d = well tip diameter (m); f_n = natural frequency (Hz); L = unsupported length (m); E = Young's modulus (Pa); I = second moment of area (m⁴); m′ = mass per unit length (kg/m); D, d_b = outside and bore diameters (m); ρ_w = well material density (kg/m³). The f_n formula ignores the sensor mass, fluid added mass and mounting flexibility, which all lower f_n in practice.

Worked examples

Example 1 (standard): 2-wire RTD lead error and ramp lag. Given: Pt100 (α = 0.00385 /°C) connected with 2 wires, each of 5 Ω. The sensor sits in a thermowell with τ = 20 s, and the process temperature rises at 2 °C/min.

1. Lead resistance adds 2 × 5 = 10 Ω.
2. Sensitivity = R₀·α = 100 × 0.00385 = 0.385 Ω/°C.
3. Lead error = 10/0.385 = 26 °C (reads high). A 3-wire connection cancels this if the leads are equal.
4. Ramp lag = τ·R = 20 s × (2/60) °C/s = 0.667 °C (reads low during the ramp).
5. Answer: lead error ≈ +26 °C (2-wire); dynamic lag ≈ 0.67 °C.

Example 2 (GATE level): thermowell wake-frequency check. Given: straight 316 SS thermowell, OD D = 20 mm, bore d_b = 6.5 mm, unsupported length L = 0.25 m, E = 193 GPa, ρ_w = 8000 kg/m³; water at V = 5 m/s; St = 0.22.

1. I = π(0.020⁴ − 0.0065⁴)/64 = 7.77 × 10⁻⁹ m⁴.
2. Area = π(0.020² − 0.0065²)/4 = 2.81 × 10⁻⁴ m²; m′ = 8000 × 2.81 × 10⁻⁴ = 2.248 kg/m.
3. √(E·I/m′) = √(193 × 10⁹ × 7.77 × 10⁻⁹ / 2.248) = √666.8 = 25.82 m²/s.
4. f_n = (1.875²/(2π × 0.25²)) × 25.82 = (3.516/0.3927) × 25.82 = 8.952 × 25.82 = 231 Hz.
5. f_s = 0.22 × 5/0.020 = 55 Hz; f_s/f_n = 0.24.
6. Answer: f_n ≈ 231 Hz, f_s = 55 Hz, ratio ≈ 0.24, below both 0.8 and the stricter in-line limit of about 0.4–0.5, so this well is acceptable on frequency. If the same well were 0.40 m long, f_n would fall to 231 × (0.25/0.40)² ≈ 90 Hz and the ratio would rise to 0.61 — no longer safe from in-line resonance.

Example 3 (thermocouple cold junction). A type K thermocouple reads 11.209 mV with its terminals at 25 °C, where E(25, 0) = 1.000 mV (table value). E(T, 0) = 11.209 + 1.000 = 12.209 mV, which the table gives as 300 °C. Adding 25 °C to the temperature for 11.209 mV would be wrong because the emf–temperature curve is not linear.

Common mistakes

• Using copper wire to extend a thermocouple, or reversing the polarity of compensating cable.
• Adding the cold-junction temperature instead of the cold-junction emf.
• Ignoring lead resistance on 2-wire RTDs over long runs.
• Choosing a well that is too short (stem conduction error) or too long and thin (vibration failure).
• Forgetting that the thermowell slows response, so control loops tuned with a bare sensor become sluggish.
• Calculating wake frequency with the insertion length instead of the tip diameter.

For GATE IN

Expect numericals on thermocouple emf with cold-junction correction, RTD resistance and lead-wire error, bridge circuits, first-order response (time constant, ramp lag, step response) and, occasionally, vortex shedding frequency. Conceptual questions ask about 3-wire versus 4-wire RTDs, compensating cables, and why thermowells change accuracy and response. Practise the law of intermediate temperatures with table values.

Quick check

1. Why does a 3-wire RTD connection reduce lead-wire error?
2. A well with τ = 30 s follows a ramp of 3 °C/min. What is the lag?
3. What happens to a thermowell's natural frequency if its length doubles?
4. Tip diameter 25 mm, St = 0.22, V = 2 m/s: what is f_s?
5. Name two reasons for using a thermowell.

Answers: 1. One lead is in each adjacent bridge arm, so equal lead resistances cancel. 2. 30 × 0.05 = 1.5 °C. 3. It falls to one quarter. 4. 0.22 × 2/0.025 = 17.6 Hz. 5. Protection from pressure, flow and corrosion, and sensor removal without shutdown.