Building a Controlled Machinery Fault Library with TMFSS

Field vibration data gets its label weeks after capture, from a work order or a teardown, and teardowns are ambiguous: a bearing run to failure shows spalling, cage damage and lubrication distress all at once. Nobody can say which signature the old recording actually contains. That label noise misleads trainees and puts a hard ceiling on any machine-learning model graded against it.

A fault simulator reverses the order: install a known, documented fault first, then capture. The TMFSS Macro reproduces more than 30 rotating-machinery fault conditions on a rigid steel base, so Tuesday's signature matches Monday's, and each fault ships with a technical note on the physics and expected spectrum. The label exists before the signal does.

Unbalance is the cleanest case: a heavy spot produces a rotating force at exactly one times running speed (1x), radial, growing with the square of speed. On the rig you step trial mass up and watch only the 1x line climb while 2x and 3x stay quiet; that selectivity is the diagnosis, because broadband growth is not unbalance.

Misalignment answers differently: parallel offset raises 2x radially, angular misalignment adds a strong 1x axial component, and the TMFSS runs each as a separate configuration so trainees see them unblended. Looseness breaks the clean-sinusoid picture entirely, as once-per-revolution impacts grow a long comb of harmonics, sometimes with half orders, and a visibly clipped time waveform.

Rolling-element bearing defects appear at frequencies fixed by geometry (BPFO, BPFI, BSF, FTF), never on a neat integer order, and early defects are tiny impacts best read in an envelope spectrum rather than the raw FFT. The TMFSS base library covers all four defect classes, and an accelerated-bearing-wear add-on kit lets you capture a staged degradation sequence instead of a binary good-or-bad pair.

Gears concentrate their story around the gear mesh frequency (tooth count times shaft speed). One damaged tooth disturbs the mesh once per revolution of its own shaft, throwing up sidebands spaced at exactly that shaft's speed, so the spacing names the faulty shaft and the sideband growth, not the mesh line, tracks severity.

Two rig features carry the acquisition discipline. The Macro's VFD with WiFi control software repeats an exact speed grid run after run, and the built-in tachometer's once-per-revolution analogue output is the phase reference that makes order tracking, synchronous averaging and balancing exercises possible.

PhonoVibe DAQs sample every channel simultaneously at 24-bit, power IEPE, ICP and CCLD sensors (24 V, 4 mA) and read TEDS, so cross-channel phase between a bearing sensor and the tach survives intact; every unit carries a factory calibration certificate. In TVIB, the free TSAP201 module adds FFT up to 102,400 points, harmonic, band and sideband cursors, and acceleration-to-velocity-to-displacement integration, with acquisition settings saved alongside each recording.

A representative lab workflow, not a specific customer: a reliability group wants a labelled in-house library before rolling out condition monitoring. Day one is spent entirely on healthy baselines across four speeds, with repeats, because every later comparison leans on them. Days two to four introduce one fault at a time from the Macro library, each run across the same speed grid, and nothing is captured until the fault, speed, sensor set-up and settings are written into the run record.

Day five withholds selected speeds and severity steps as hold-out data, so later algorithms and trainees can be tested on conditions they have not seen. The deliverable is not a headline accuracy number; it is an audited library where every record answers what was wrong, how badly, at what speed, and how it was measured.

The theory behind every figure above is covered in the free TIERA 101 primers at 101.tieraonline.in; they are primers, not an accredited certification. When a team needs formal, recognised competence, the TCAT programme, aligned with ISO 18436-2 Category I to IV, adds instructor-led depth, assessed practicals on simulator hardware, and proctored examinations on exams.tieraonline.in, with details on the services page at tieraonline.in/services.

The natural sequence for a new lab: read the primers, put every analyst through structured simulator hours on the fault set they will meet in the field, and let those carrying diagnostic responsibility sit the appropriate TCAT categories. The simulator turns theory into pattern recognition; certification makes that recognition accountable.

Every stage of the workflow above is equipment TIERA designs, builds and supports from Thiruvananthapuram. The fault source is the TMFSS Macro: 30-plus rotating-machinery fault conditions on a rigid base, each shipped with a technical note on the physics and the expected spectrum, with add-on kits — bent shaft, finned rotor, accelerated bearing wear, lockable bearing, mechanical rub — to grow the library over time. The Mini (7+ faults) and classroom-format Micro cover training labs that do not need the full research rig.

The capture chain is the PhonoVibe DAQ series and the TVIB TSAP 201 software bundled free with it, so the phase reference, acquisition settings and calibration record that make a library defensible are handled by the same toolchain that recorded it. Tell us the faults your team needs to recognise and we will spec the rig, the channel count and the training path around them.