A Brief Analysis of Selection Fleas from the Industrial Control Machine

On the shop floor, a flea is any small sensing element that turns a physical value into an electrical signal. Pressure, temperature, flow, position, vibration, current. The control system reads that signal and decides what the machine does next. The name is informal, but the function is not. A flea that reads wrong makes the machine act wrong.

Most industrial control machines carry several of them at once. A spindle head may hold a temperature probe and a vibration sensor. A hydraulic power unit carries pressure and flow devices. The selection question is rarely about one flea alone. It is about which signals the controller actually needs, and where each one can physically sit.

Two families show up most often. Contact types touch the medium or the moving part: thermocouples, RTDs, strain gauges, limit switches. Non-contact types read across a gap: inductive proximity, capacitive, photoelectric, ultrasonic, eddy-current displacement. The gap decides mounting tolerance, and mounting tolerance decides how the bracket is machined.

Start with the signal the controller can accept. A 4–20 mA loop survives long cable runs and electrical noise far better than a raw millivolt output. A 0–10 V signal is common on short runs inside a cabinet. Once the output type is fixed, the rest of the selection narrows quickly.

Accuracy and repeatability are not the same thing. A device can read 0.5% off absolute truth and still repeat to 0.05% cycle after cycle. For closed-loop control, repeatability usually matters more. Ask the supplier for both numbers, not one.

Environment sets the housing and the sealing class. Coolant mist, chip spray, washdown, vibration and ambient temperature all push the choice. An IP67 body with a stainless shell survives a machining center. A bare plastic-cased unit does not.

Then check the mechanical interface. Thread size, probe length, cable exit direction and the wrench flat all have to fit the space you have. A sensor that reads perfectly but cannot be reached for adjustment is the wrong sensor.

Response time matters on fast loops. A temperature probe with a 10 s time constant cannot control a process that moves in 200 ms. Match the time constant to the loop, not to the catalog headline.

• 1
  Output type4–20 mA for long runs, 0–10 V for short cabinet runs
• 2
  RepeatabilityOften more important than absolute accuracy in closed loop
• 3
  Sealing classIP67 minimum near coolant and chip spray
• 4
  Mechanical fitThread, probe length, cable exit, wrench access

The flea is bought. The metal around it is made. That metal decides whether the sensor reads what it is supposed to read. A bracket that flexes under load changes the gap. A boss that is not square to the axis tilts the probe.

Threaded sensor ports are the most common feature. A 1/4 NPT or M12 × 1 port has to be cut concentric with the sealing face. Runout on that face leaks coolant straight into the thread. On a mill-turn center we cut the port and the face in one setup, so concentricity holds without a second op.

Flatness on a mounting pad matters more than most people expect. An eddy-current probe sitting on a pad that is out by 0.05 mm reads a gap error of 0.05 mm. At a scale of microns per volt, that is a large number. Pads that carry a displacement sensor usually get ground or fine-milled to Ra 0.8–1.6 μm.

Vibration isolation is a machining choice too. A stiff, well-damped bracket moves the resonance away from the machine's operating frequency. Thin unsupported plate does the opposite. Where a design calls for it, we machine a pocketed body that adds stiffness without adding mass.

Cable routing gets decided at the same time. A sharp internal corner wears through a cable jacket within weeks. Radiused exits, cable glands and strain-relief clamps are cheap features to cut and expensive to retrofit.

Materials follow the environment. Anodized 6061-T6 covers most cabinet and fixture work. For washdown or food-adjacent machines, 316L stainless with a bead-blasted finish holds up better. Both are standard stock for us.

• 1
  Concentric portThread and sealing face cut in one setup
• 2
  Flat padRa 0.8–1.6 μm for displacement sensors
• 3
  Stiff bracketPush resonance above the machine's operating band
• 4
  Radiused cable exitAvoids jacket wear at internal corners

Some loops do not want a sensor at all. A hard stop, a mechanical detent or a torque limiter costs less and never drifts. Add a sensor when you need to know the value, not just to prove the machine reached a position.

Non-contact sensing has a floor. Inductive devices lose range against aluminium and brass, often by 40% or more versus mild steel. If the target is non-ferrous and the gap is tight, a different principle usually wins.

Extreme temperature is another limit. Standard transducers top out well below the temperature inside a heat-treat furnace or near an exhaust manifold. Remote mounting with a capillary or a thermowell solves it, but it changes the response time.

Finally, watch the cable. Most field failures are not the sensing element. They are a chafed jacket, a loose gland or a connector full of coolant. A simpler sensor with better cable protection often outlasts a better sensor with poor routing.

We see this on machined housings every week. The pocket for the connector is often an afterthought. Give it a real dimension and a real seal, and the flea stays alive.

Most sensor hardware is not exotic geometry. It is a plate, a boss, a tube or a block with a precise hole in it. The difficulty sits in a few features: bore diameter, concentricity, face flatness and hole position.

Our standard tolerance is ±0.005 mm on critical features, with Ra 0.8–1.6 μm as a normal machined finish and Ra 0.2–0.8 μm where a sealing face or a sliding bore needs it. Position tolerance on mounting holes typically runs ±0.02 mm unless the drawing says otherwise.

Prototype volumes run from one part to a few hundred. Production runs go past 10,000. There is no minimum order quantity, so a single replacement bracket and a full production lot go through the same inspection route.

Every part is inspected before shipment. That covers incoming material, in-process checks and a final dimensional report. Reports go out on request with the shipment.