ML / AI Algorithm Development for Predictive Maintenance

Your model is only as good as your training data. Most predictive-maintenance teams face the same bottleneck: real fault data is rare, unlabelled, or captured on equipment you cannot instrument properly. The TIERA lab stack removes that bottleneck.

Controlled fault generation. The TMFSS Macro reproduces 30+ fault types on a rigid, repeatable benchtop rig. You control the fault type, severity, shaft speed, and load. You run the same fault at 1200 RPM and at 2400 RPM. You run inner-race defect with one bearing and then with two. Every run is tagged with a known ground-truth label — no ambiguity, no need to infer the fault post hoc from field data.

High-channel-count capture. The PhonoVibe HD gives you 16 simultaneous 24-bit channels at up to 128 kHz per channel. Mount accelerometers at the drive end, non-drive end, housing, and baseplate simultaneously. Your model sees the full vibration signature, not just one measurement point.

Ground-truth feature extraction. TVIB computes the spectral features your model needs — envelope spectrum, narrowband energy, sideband ratios, crest factor — and exports them. Use these as reference features when validating what your neural network has learned.

Analyst-in-the-loop validation. ToLearnVibe presents fault scenarios to a human analyst and records their diagnosis. Use the analyst’s labels as a second opinion alongside your model’s predictions. Where they disagree, the raw waveform is there for inspection.

The result: a documented dataset with controlled fault conditions, multi-point capture, ground-truth spectral features, and analyst-confirmed labels. Enough to train, validate, and publish.

Products in this workflow

• TMFSS — Machinery Fault Signature Simulator
• PhonoVibe Series DAQ
• TVIB Analysis Software
• ToLearnVibe Training Software