Choosing flow and level measurement, signal type, and control tuning — and beating loop instability and drift
How to measure and control a process — flow and level instruments, signals, control tuning — with the reasoning, the risks, and a confidence level. Source instrumentation from the Process Control & Instrumentation supplier directory, and see the full engineering guides index.
Use a magnetic (mag) meter for conductive liquids and slurries — no pressure drop, no moving parts. Use a Coriolis meter for the highest accuracy and direct mass flow (custody transfer, dosing), accepting cost. Use a vortex meter for steam and clean gas/liquid. Use differential-pressure/orifice for a low-cost, well-understood standard on clean fluids. Use ultrasonic (clamp-on) where you can't cut the pipe.
Flow-meter choice follows the fluid (conductive? clean? gas/steam? slurry?), the accuracy needed, and whether mass or volume matters. A mag meter needs a conductive fluid but is unobstructed and slurry-tolerant. Coriolis measures true mass flow and density with the best accuracy but costs the most. Vortex handles steam well; orifice plates are cheap and standardized but add permanent pressure loss and dislike dirty or two-phase flow.
A mag meter on a non-conductive fluid (oil, pure water) reads nothing. An orifice on dirty or two-phase flow plugs and errs. Coriolis where a simple mag meter would do overspends. Ultrasonic on a pipe with scale or bubbles loses accuracy.
Use guided-wave or non-contact radar as the versatile default — accurate, largely unaffected by density, vapor, and temperature. Use hydrostatic (DP) level for simple, robust liquid level in open or pressurized tanks. Use ultrasonic for low-cost non-contact on clean, calm surfaces. Use capacitance for interfaces and some solids, and a float/switch for simple point (high/low) alarms.
Level technology follows the medium and conditions. Radar reflects off the surface regardless of density and tolerates vapor, foam (guided-wave), and temperature — which is why it has become the general workhorse. Hydrostatic infers level from head pressure (simple, but density changes bias it). Ultrasonic is cheap and non-contact but confused by foam, vapor, dust, and turbulence. Match the sensor to whether the surface is clean and calm, foaming, under vapor, or a solid.
Ultrasonic on foam/vapor/turbulence gives false echoes. Hydrostatic on a variable-density fluid reads the wrong level. A float in a sticky or solids service fouls. Picking by price instead of the medium is the common mistake.
Use 4–20 mA analog (optionally with HART digital overlay) as the robust, universal default for a single process variable per loop — simple, noise-immune, and easy to troubleshoot. Use a digital fieldbus / industrial Ethernet (Foundation Fieldbus, Profibus/Profinet, EtherNet/IP, IO-Link) when you need many variables per device, diagnostics, and reduced wiring on a larger installation.
4–20 mA is a current loop, and current doesn't drop with wire resistance, so it's immune to voltage-drop error over long runs and a broken wire reads 0 mA (an unambiguous fault) — that robustness and simplicity is why it endures. Fieldbus trades that simplicity for rich digital data (multiple variables, device diagnostics, remote config) and less wiring per point, at the cost of more complex commissioning and a bus that can take multiple devices down at once.
Fieldbus on a tiny, simple install adds commissioning complexity for little benefit. 4–20 mA where hundreds of points and diagnostics are needed means a wiring and data bottleneck. Also mind sensor scaling — map the sensor's range to the 4–20 mA span correctly or every reading is off.
Start with PI control for most loops (flow, level, pressure): proportional for responsiveness, integral to remove steady-state offset. Add derivative only for slow loops with significant lag (temperature) where it helps damp overshoot — and avoid D on noisy or fast loops (flow), where it amplifies noise. Tune with a method (e.g. Ziegler-Nichols as a starting point, then refine), not by random adjustment.
Each term does a distinct job: proportional reacts to the current error (but alone leaves an offset), integral accumulates past error to drive offset to zero (but adds lag and can wind up), and derivative reacts to the rate of change to anticipate and damp (but multiplies measurement noise). That noise sensitivity is why fast, noisy loops like flow use PI and leave D off, while sluggish temperature loops benefit from D.
Derivative on a noisy flow loop makes the valve chatter. Too much proportional/integral gain makes the loop oscillate; too little makes it sluggish and slow to reject disturbances. Integral windup (during saturation) causes large overshoot on recovery unless anti-windup is used.
The controlled variable oscillates or hunts around setpoint, the valve cycles continuously, or the loop overshoots and rings after a disturbance. Product quality and equipment wear both suffer.
Gains tuned too high (or derivative on a noisy signal), measurement noise fed straight to the controller, valve stiction (the valve sticks then jumps, so the loop can never settle — a very common and misdiagnosed cause), and integral windup. An oversized control valve (controlling near the seat) also destabilizes.
1) Check the valve for stiction first — a limit-cycling loop with a healthy tune is usually a sticking valve, not a tuning problem. 2) Reduce gains / remove derivative on noisy loops; filter the measurement. 3) Add anti-windup on the integral term. 4) Re-tune systematically for adequate stability margin. 5) Right-size the control valve so it modulates in mid-travel.
A reading that slowly departs from reality — the process is fine but the instrument says otherwise, or vice versa. Control holds the wrong value, alarms misfire, and batch quality drifts, often undetected until a manual check disagrees with the sensor.
Sensor aging and fouling (coating, corrosion, buildup on the element), temperature and pressure effects, zero/span drift in the transmitter, and simply a missed calibration interval. Some technologies (pH, some analytical sensors) drift much faster than others.
1) Calibrate on a schedule matched to the sensor's drift rate (fast-drifting types more often). 2) Keep the sensing element clean — self-cleaning, retractable, or scheduled cleaning where fouling is the cause. 3) Use a technology less prone to drift for the service where possible. 4) Cross-check against an independent measurement or manual sample periodically. 5) Trend the reading — a slow, steady departure flags drift before it causes off-spec product.