Deploying Wireless Vibration Sensors on 50 Pumps: A Field Guide (i-ALERT)

i-ALERT is a battery-powered sensor that measures tri-axial vibration, surface temperature and run time on the bearing housing of a pump, motor, fan or gearbox, mounts with a magnet, and sends data by gateway or phone to the i-ALERT AI platform, where each machine's baseline is learned and alarms are raised on thresholds and trends. PlusSmart engineers then diagnose alarms with remote FFT analysis. The technology is simple; the deployment is where programmes succeed or fail. This is the sequence we use for a typical 50-pump rollout, drawn from pilots where five of nine assets needed immediate action and a misalignment was diagnosed remotely four days after installation.

Week 0: choose the machines

Do not instrument everything. Rank by consequence and failure history and take the top 20 % first — the pumps whose failure stops production, the ones with repeated work orders, and the ones in locations nobody visits. A list with motor kW, speed, bearing type (rolling or sleeve), foundation (rigid or flexible per ISO 20816-3) and the last three failures is the deployment document. Sleeve-bearing machines and very low-speed machines (below ~120 rpm) need a different approach; put them aside for now.

Week 1: mounting

Rule	Why
One sensor on the motor drive-end bearing, one on the pump bearing housing (or outboard bearing)	Distinguishes motor-side from pump-side faults; two points per machine is the minimum for diagnosis
In the bearing load zone, horizontal or at 45°, as close to the bearing as the housing allows	High-frequency bearing signals attenuate quickly with distance and through joints
Flat, clean, bare-metal magnetic surface; stud or adhesive on curved, painted or non-magnetic housings	A rocking magnet filters out the frequencies that reveal bearings
Never on guards, fan covers, thin sheet or pipework	They resonate and carry no bearing information
Mark the position and orientation; the sensor never moves	Trends are only meaningful from one fixed point
Check the sensor is away from steam, direct washdown spray and radiant heat	Battery life and temperature readings

A two-person team mounts 50 sensors in two or three days including the asset data entry on the platform (machine type, speed, bearing numbers if known). Gateways go where they see the sensors and have network or cellular access; walk-test the radio in metal-dense areas before fixing the gateway positions.

Weeks 2–4: baseline

Let the platform learn. Each machine needs two to four weeks of data across its normal operating states — loaded, unloaded, different speeds — before alarms are meaningful. During this period, review the first data for installation problems (a sensor on a resonant bracket, a loose magnet) and for machines that are already in trouble: a pump at 25 mm/s is not a baseline, it is a work order. In the water-treatment pilot above, the realignment that followed brought the motor below 3 mm/s.

Week 4: alarm settings

• Trend alarms at 1.5 × baseline (alert) and 2–2.5 × baseline (alarm) on velocity RMS, capped by the ISO 20816-3 zone C/D limit for the machine group and foundation (4.5 mm/s for 15–300 kW rigid; 7.1 mm/s flexible or above 300 kW rigid; 11 mm/s large flexible).
• Bearing indicators (high-frequency acceleration, crest factor or the platform's bearing index) trended separately; alert on rate of change.
• Temperature as a confirming alarm, not a primary one.
• Rate-of-change alarm: a rise of 50 % in two weeks at any absolute level.
• State awareness: suppress during start-up and known transients; use run-time detection to separate running from stopped.

The i-ALERT AI platform learns baselines and raises threshold and trend alarms; PlusSmart's alarm management layer on top reduces false alarms and routes the ones that matter to an engineer for FFT diagnosis.

Months 2–3: diagnosis and action

The first alarms arrive. Typical first-quarter findings on a 50-pump fleet: three to five misalignments, one or two unbalances from impeller build-up, two bearings in early failure, one pump cavitating because a suction condition changed, and one sensor that was mounted on a guard. Each needs a diagnosis (remote FFT: 1× radial = unbalance; 1×/2× with high axial = misalignment; non-integer bearing frequencies = bearing; broadband = cavitation), a recommended action and a date. The plant's job is to do the action and let the sensor prove the result — vibration before and after is the report that justifies the programme. Do not skip this step: the programmes that fail are the ones where alarms were acknowledged and nothing changed.

What the programme costs and returns

Hardware and platform for 50 pumps is a modest fraction of one unplanned failure on a critical pump including lost production; pilots typically show a return above five to one. The ongoing cost is the engineer's time to read the data — which is why PlusSmart's model includes remote diagnosis rather than leaving the plant with a dashboard. Combine with ElectroSine for motors the sensors cannot reach and with a Pump Health Check from the historian to decide which fifty pumps come next.

Quick answers

How many sensors per pump?

Two is the standard — motor drive-end and pump bearing housing. Large pumps with two pump bearings and critical machines get a third; small close-coupled pumps can manage with one on the bearing bracket.

How long until we get useful alarms?

Two to four weeks for baselines; the first real diagnoses typically within the first month. Machines already in trouble are visible on day one.

Who reads the data?

The platform raises alarms; PlusSmart engineers perform the FFT diagnosis remotely and recommend the action. The plant team receives a prioritised list, not raw spectra, and can escalate any machine for a closer look.