Motor Current Signature Analysis Implementation: RFP, FAT and SAT

Implementing motor current signature analysis (MCSA) starts with a narrower question than “Where can we fit a current transformer?” A clamp that distinguishes running from stopped equipment and a system that evaluates frequency components of current under defined operating conditions require different data, instruments and acceptance tests. This guide helps factory teams in Thailand select motors, specify an RFP, test FAT and SAT, and hand the result to maintenance without promising universal fault prediction.

Define the diagnostic use case before implementing motor current signature analysis

MCSA examines the current drawn by an operating motor for components that may reflect rotor, air-gap and driven-load conditions. The US Department of Energy (DOE) motor and drive sourcebook describes FFT and demodulated current spectra for detecting rotor, air-gap and load-related faults. ABB’s condition assessment guide discusses rotor-bar deterioration, eccentricity, phase imbalance and some mechanical problems. These descriptions establish possible applications; they do not guarantee performance on a particular motor and production process.

The first decision is the asset, failure mode and action: which failure would seriously affect production, safety or quality; which current feature might reveal it; under which speed and load; and what will maintenance do after an alert? A constant-load fan and a conveyor with varying products have different healthy spectra. A proposal that says “AI detects all anomalies” without specifying faults and operating coverage cannot be accepted objectively.

Our separate article on CT current sensors for machine run-state monitoring addresses thresholds for RUN/STOP. This guide deals with waveform and frequency analysis. The broader selection and operating process is covered in predictive maintenance case studies. Collecting low-rate RMS values for a run-state dashboard does not preserve the waveform needed for later spectral analysis; conversely, a simple run signal does not justify high-bandwidth acquisition.

What current can suggest, and what needs another measurement

Rotor, air-gap and load changes can modulate stator current in an induction motor. Repeated comparisons at equivalent operating points can therefore provide evidence of condition change. One spectral peak, however, does not uniquely identify a broken part. Speed, slip, load, supply harmonics and control mode influence the signal. ABB describes combining its current results with other diagnostic measurements. DOE’s predictive maintenance guide stresses equipment knowledge and operator training for proper application.

For each target fault, record the candidate current signature, confounding conditions, confirmation method and maintenance action. For a suspected rotor defect, preserve current, speed, load and voltage conditions, then request expert analysis and relevant inspection. For a belt or pump load change, compare vibration, temperature, flow or pressure and the actual mechanism before deciding that the motor is faulty. An alert without a feasible confirmation and action is not a maintenance outcome.

DOE’s motor-driven systems guide warns that variable-speed drives and other electronic loads introduce harmonics and that purchasers should confirm instruments can measure such electrical environments accurately. Data quality therefore precedes analytics.

Choose motors by criticality and observable operating conditions

A pilot benefits more from a small number of critical, well documented assets than from indiscriminate sensor coverage. Assemble each asset’s motor nameplate, type, power, poles, connection, starter or VFD, driven machine, protection, single-line drawing, previous replacements and failure history. Reconcile asset identifiers between the MCC, maintenance register and production system. A correct waveform assigned to the wrong machine can generate a dangerous work order.

For each candidate, document the failure mode, previous downtime, allowable outage, standby capacity, lead time for spares, inspection opportunity and period of stable loading. Do not convert one historical breakdown into an annual failure rate. If history is sparse, prioritize criticality and ease of controlled verification, then decide on wider rollout using observed data quality and alert workload.

Distinguish direct-on-line, soft-starter and VFD applications. A spectral feature described for a constant-speed induction motor cannot simply be transferred to variable frequency and PWM output. Define the measurement location, usable speed range and drive signals. Siemens’ SIMOCODE M-CP announcement describes current and voltage modules and data delivery for advanced MCSA, but this is a particular product and configuration. Check the actual model, license and sampling access rather than assuming that every drive exports the same information.

Specify the MCC measurement path and electrical safety

Show on the single-line diagram exactly which conductor each CT or clamp surrounds and which motor it serves. A one-phase current measurement and simultaneous three-phase current and voltage measurement provide different diagnostic evidence. ABB describes a particular high-voltage motor service using CTs and PTs at the motor control centre; the arrangement should not be generalized to every installation.

The RFP should list measurement point, current range, CT ratio, accuracy, bandwidth, phase response, isolation, installation space, cable length, clock synchronization, network path and power-loss behaviour. Specify waveform sample rate, window, overlap, FFT settings, retention, missing-data flag and export format. Preserve raw waveform or sufficient features together with operating context, configuration and alarm history so a specialist can reproduce findings and the factory can change suppliers later. A proprietary health score alone limits root-cause review.

Work inside an MCC may expose personnel to hazardous energy. Qualified electrical staff and the panel supplier must assess the installation, protection-circuit interactions and site-specific isolation, lockout, verification and PPE procedures. Do not casually share or open a protection CT secondary. Schedule an outage if safe installation cannot be completed under the approved procedure. An analysis unit is neither a protection relay nor an emergency-stop function. If alarms feed PLC logic, define authority separately and prevent unsafe automatic restarts.

Establish a healthy baseline at matched speed and load

A no-load test is not a baseline for all production. Tag normal records with speed or output frequency, product recipe, flow or pressure, process stage, load, ambient conditions and whether the motor is starting or steady. A fan may require damper position, a conveyor product presence and density, and a pump valve or flow conditions. The aim is not to log every signal at maximum rate, but to capture the variables needed to distinguish a healthy change in duty from deterioration.

Siemens’ Drivetrain Analyzer Edge manual explains a model that compares current mechanical condition with a training-phase reference. Its display thresholds belong to that product; they are not universal alarm limits. Record the baseline period, excluded setup and maintenance states, responsible approver and rules for retraining. Uncontrolled retraining may absorb gradual deterioration into “normal”.

Inspect a reference machine before treating its current as healthy. Past fault data can be replayed when diagnosis is well documented; deliberately creating dangerous faults on live equipment is unnecessary. Label cases “confirmed”, “suspected” or “unknown”, with dates, parts, inspection findings and operating conditions. Separate validation data by asset, period or production condition rather than only adjacent minutes from the training series. The apparent accuracy of a model tested on near-identical data says little about field performance.

As an illustrative assumption, not a standard operating band, one factory might compare steady runs at 30–40 Hz separately from 45–55 Hz. Actual ranges must follow the drive settings, recipe and measured distribution. Any test that deliberately changes load requires production and safety approval and must remain within equipment limits.

Account for VFD, load variation and supply quality

First, a VFD changes fundamental frequency and PWM components with speed and settings. Line-side and motor-side measurements are not interchangeable. Ask whether a drive provides raw sampled current or filtered aggregate values, and confirm the sensor, cable, converter and filters support the features of interest. A peak rule at 50 Hz cannot be assumed valid across a variable-speed range without test evidence.

Second, normal changes in cutting depth, product mass or fluid viscosity can resemble deterioration. Capture process tags and compare only equivalent operating states, or define a repeatable inspection interval. Where conditions are unknown, mark the analysis as unavailable. Do not display “100% monitoring coverage” while silently including unassessable states.

Third, supply-voltage variation or phase imbalance may affect current. Record relevant three-phase voltage and power-quality events or arrange a follow-up measurement. DOE distinguishes current-only MCSA from electrical signature analysis that uses both voltage and current. A tiered architecture can use continuous screening and portable confirmation measurements instead of permanently fitting every asset with every instrument.

Turn findings into maintenance work

Each alert should carry asset and time window, observed feature, matched baseline, operating context, candidate cause, confidence or exclusion reason, required confirmation, owner and deadline. A red icon without an owner or inspection path has little value. Show the underlying evidence when an algorithm offers a fault name. Maintenance must be able to see why the recommendation changed.

A practical workflow may use information, investigation and intervention levels, but the number of levels is a local design choice. A developing trend can be watched; an investigation may trigger vibration and inspection; a high-consequence, corroborated condition may prompt a planned outage. Safety-critical protection remains a separate engineered function. MCSA should inform maintenance planning, not substitute for a verified protective device.

Measure duplicate notifications, time to first review, proportion investigated, unassessable operation, resulting actions and unconfirmed causes. A headline “99% accuracy” is misleading when actual faults are rare. During a pilot with no true failure, assess repeatability, operating coverage, noise resistance, data availability and maintenance workload; do not claim a measured fault-detection rate.

Put comparable requirements in the RFP

Supply asset lists, consequences of failure, intended faults, electrical drawings and OT/network constraints. Ask bidders to identify excluded motor types and VFD configurations, measurement point, current and voltage channels, sampling specification, required speed/load tags, baseline method, diagnostic assumptions, retraining, evidence display and data export. Quotes based on different equipment scopes cannot be compared fairly.

Compare installed cost rather than hardware price alone: CTs, panel work, outage coordination, networking, edge or cloud fees, licensing, analysis service, calibration, training, storage, support and exit-data rights. If an external specialist interprets alerts, include response hours and fees. For a subscription, check covered motors, waveform retention and chargeable alerts.

Illustrative arithmetic only; these are assumptions, not measured savings or a guarantee. If a critical motor failure causes eight hours of lost production at THB 50,000 per hour and THB 100,000 of extra emergency repair, the simple incident cost is 8 × 50,000 + 100,000 = THB 500,000. The investment case must still account for the probability that MCSA would detect that fault in time, costs that remain under planned maintenance, false-alert effort and failures outside the method’s scope. One historical incident should not be multiplied into a confident annual benefit.

FAT: reproduce acquisition and processing before delivery

In factory acceptance testing, verify instrument model, firmware, licensed functions, channels, CT-ratio setting, time synchronization, network loss buffering, recovery, roles, audit trail and export. Feed a synthetic or properly anonymized measured waveform through the pipeline and check that the same input yields the expected features and alarm. Use a held-out test set rather than only the data on which the model was trained.

Test healthy speed and load changes, plus a disconnected sensor or missing phase signal where simulation is safe. The system should flag a data-quality problem rather than report a motor defect. An alarm should preserve source interval, operating context and reason for any inapplicable diagnosis. Contractual acceptance thresholds for alarm count, retransmission delay and data-gap notification must be set for the site; there is no single numerical FAT limit for every factory.

If representative fault recordings are unavailable, record fault-classification performance as unverified and carry it into the field evaluation. FAT completion must not turn an untested capability into a proven one. Define supplier support and retest after model changes.

SAT: prove the application on the actual machine

Site acceptance testing checks the physical circuit and operational coverage: MCC feeder versus asset ID, CT orientation and ratio, phases, VFD, communication, local time and process tags. Include outages, network interruption, product change, speed change, cleaning and manual modes. Show which intervals receive valid diagnostics and which are excluded. Use language-independent asset and fault codes where Thai operator names, English supplier screens and Japanese management reports coexist.

A “30-day pilot” is not representative if only one recipe ran. Record planned and observed speed, load and product cases side by side; mark missing cases “not tested”. A motor change, belt adjustment or VFD setting change may invalidate comparison with the old baseline. Obtain production approval before changing load for a test.

Have the actual maintenance team receive an alert, identify the asset, compare matched historical waveforms, order a vibration or physical inspection, and close a CMMS work order. Verify night-shift escalation, deputies, suppression of false alerts, supplier support and separation from emergency protection. Hand over ownership and decision rights alongside the hardware.

Data governance and the deployment boundary

Choose edge or cloud processing based on network, data volume, retention and sharing rules. Continuous high-rate waveform upload is different from sending locally calculated features and keeping short raw recordings around alerts. Document retention, time zone, clock drift, gaps, backup, deletion and transfer when hardware is replaced. A practical design might retain short raw windows at the site and long-term event records, if that meets diagnostic needs.

Minimize write access to the PLC network, separate appropriate OT segments, and agree remote-support accounts, authentication, logs, updates and recovery with plant IT/OT staff. Automatic motor shutdown from an analysis score requires separate risk assessment and change control. A failed monitoring device must not compromise protective circuits.

Common mistakes and a sensible rollout sequence

The most common error is buying “predictive maintenance” from low-rate current values without retaining waveform or operating context. The opposite error is adding expensive analyzers to every motor before identifying detectable failure modes. Another is training the baseline on an already degraded machine. Check maintenance history and physical condition, then compare post-repair current with the previous reference. An alert followed by a replacement does not alone prove that the alert diagnosed the part; preserve inspection findings and the change after repair.

Proceed through asset and fault selection, panel survey, RFP for measurement and data, FAT, planned installation, SAT, matched baseline, operational evaluation and wider rollout. Set a gate at each stage. Even when the pilot contains no genuine fault, it can still show data quality, verified operating coverage and whether maintenance can close the response loop.

The handover package should include the single-line drawing and photographs of each measurement point, CT ratio and orientation, instrument model and settings, time-sync source, data dictionary, baseline period, excluded modes, alarm meaning, conditions where analysis is unavailable, inspection procedure and contacts. Assign owners for retraining after motor replacement, calibration due dates and communication failures. Where maintenance, electrical and production teams have separate roles, identify the primary reviewer, the person authorized to stop the machine, the parts decision maker and their deputies. At an agreed review interval, compare alerts with physical findings, false alarms, missing operating states, new product recipes and VFD changes. A quiet dashboard may mean a healthy machine, missing coverage or suppressed alerts. More alarms do not necessarily mean better detection. Use these findings to adjust measurement, state segmentation, thresholds and response work before contract renewal or rollout.

FAQ about motor current signature analysis implementation

Is one CT enough for predictive maintenance?

One-phase current may be enough for a RUN/STOP threshold. MCSA requirements depend on fault mode and may call for suitable waveform quality, frequency resolution, speed/load context, three-phase current or voltage. Compare the required diagnostic with the CT specification and validate at FAT and SAT.

Can MCSA be used on a VFD-driven motor?

Some product and operating combinations support it, but there is no universal answer. Measurement location, speed range, PWM content, drive data access and target fault all matter. State the exact model in the RFP and test representative speeds and loads. Show excluded transients or low-speed states explicitly.

Can current alone identify a bearing failure?

Current sometimes provides a supporting clue, but observability changes with bearing, speed, load and noise. Use vibration, temperature, lubrication records and inspection to confirm. Do not promise detection of every bearing fault using current alone.

How should cost and ROI be compared?

Include panel work, outage, communications, subscriptions, specialist interpretation, maintenance response and data exit along with sensor cost. Estimate benefit as a range based on actual downtime and the chance of converting an unplanned failure into planned work. Revise assumptions with pilot measurements.

What is the difference between FAT and SAT?

FAT checks configuration, data processing and reproducibility with controlled inputs before delivery. SAT verifies actual CT connection, drive behaviour, representative production conditions and the maintenance workflow on site. A pilot without a true fault cannot establish a field fault-detection rate.

Conclusion

Motor current signature analysis is a conditional diagnostic method, not a universal sensor claim. Define faults, build the electrical measurement path, compare matched speed and load baselines, confirm candidates with physical evidence, and turn alerts into maintenance decisions. Put exclusions and export rights in the RFP, verify reproducibility at FAT and operational scope at SAT.

If your Thai factory is evaluating an existing motor, MCC or VFD for this method, contact TOMAS TECH with the asset register and single-line drawing. We can help define the measurable fault modes and field acceptance tests before a wider purchase.

Primary references

• DOE, Operations & Maintenance Best Practices Guide, Chapter 6
• DOE, Improving Motor and Drive System Performance
• DOE, Continuous Energy Improvement in Motor Driven Systems
• ABB, Condition Assessment for HT Motors and Generators
• Siemens, SIMOCODE M-CP announcement
• Siemens, Drivetrain Analyzer Edge operating manual