Automatic Machine Fault Diagnosis

The module produces a complete diagnosis in one refresh. The main window has an information area at the top, parameters on the left and four linked plots.

The aim of condition monitoring is not an impressive waveform but the right maintenance at the right time. The module turns vibration values into three kinds of maintenance information.

• Detect problems before failure. In early bearing spalling the fault energy may be only a few times the background noise, inaudible and impossible to feel, but it leaves a stable signature at the bearing characteristic frequencies. Checking inner race, outer race, rolling element and cage frequencies with signal-to-noise grading can flag faults at an early stage and give weeks or months for spare parts and planning.
• Base decisions on data. Each conclusion carries a confidence level and severity grade, so the maintenance priority of several machines can be ranked directly, avoiding both unnecessary maintenance and running with faults.
• Use a common language. RMS velocity is classified to ISO 10816-3 into A (new machine condition), B (unrestricted long-term operation), C (short-term operation only) and D (dangerous, stop the machine), so machine condition can be compared across suppliers and industries.

The diagnosis chain has five links: preprocessing and frequency-domain integration, automatic speed extraction, ISO evaluation, low-order harmonic diagnosis, and bearing and gear characteristic frequency diagnosis.

Preprocessing and frequency-domain integration

The raw acceleration is band-pass filtered from 10 to 1000 Hz to match the ISO evaluation band and suppress low-frequency drift and high-frequency interference. Spectra use an FFT with a Hanning window and amplitude correction. Velocity is obtained by frequency-domain integration — velocity amplitude = acceleration amplitude ÷ (2π × frequency) — which avoids the drift and trends of time-domain integration; unreliable lines below 10 Hz are set to zero.

Automatic speed extraction

Magnitude spectra are calculated frame by frame from 10 to 1000 Hz, significant peaks are searched and refined by three-point parabolic interpolation. A peak must exceed a set multiple of the band average to be accepted. During tracking, the speed may not change by more than ±30 % between frames, preventing jumps caused by occasional interference; frames without a significant periodic peak are treated as stopped, with speed zero.

Low-order harmonic diagnosis: unbalance, misalignment, looseness

The signatures of these three faults lie in the low-order harmonics of running speed. The 1× (running speed), 2× and 3× amplitudes are read from the velocity spectrum and classified:

The rules are applied only when the data is reliable: if the 1× velocity is below 0.5 mm/s, no harmonic diagnosis is made, avoiding misdiagnosis from distorted ratios of weak signals.

Bearing diagnosis

From the bearing geometry — number of rolling elements Z, rolling element diameter d, pitch diameter Dp and contact angle θ — the classical rolling bearing kinematics give four defect frequencies: BPFI (inner race defect), BPFO (outer race defect), BSF (rolling element spin) and FTF (cage).

BPFO = (Z/2) · fr · (1 − (d/Dp) · cos θ) BPFI = (Z/2) · fr · (1 + (d/Dp) · cos θ)

BSF = (Dp / 2d) · fr · (1 − ((d/Dp) · cos θ)²) FTF = (fr/2) · (1 − (d/Dp) · cos θ)

The amplitude at each characteristic frequency is read from the acceleration spectrum, which is more sensitive to high-frequency bearing impacts, and divided by the local background mean near that frequency to give a signal-to-noise ratio (SNR). Only energy clearly above the background is accepted as a fault signature, which excludes environmental interference and neighboring machines.

Two sideband checks follow. Inner race faults are usually accompanied by running-speed sidebands (each impact is modulated once per revolution) and rolling element faults by cage-frequency sidebands; when sidebands are detected, the confidence is raised.

Gear diagnosis

The mesh frequency is calculated from the number of teeth Zg (mesh frequency = running frequency × number of teeth), and the energy at the mesh frequency and its sidebands is checked to identify tooth wear, broken teeth and other anomalies.

ISO 10816-3 zone evaluation

RMS velocity in the 10–1000 Hz band is compared with the ISO 10816-3 limits for the machine power and support; eight machine groups are built in.

Built-in test signal

Test mode synthesizes a signal from the current panel parameters with 1×/2×/3× harmonics, an inner race defect frequency with running-speed sidebands, an outer race defect frequency and light noise: 1× at 0.5 g, 2× at 0.15 g and 3× at 0.05 g simulate unbalance, and inner and outer race components of 0.3 g and 0.25 g simulate bearing faults. The diagnosis table then reports rotor unbalance, inner race fault and outer race fault, using exactly the same rules as for real signals.

• Established diagnostic practice in the algorithm: harmonic ratio rules, bearing SNR grading and sideband checks give new staff the same basis as experienced engineers.
• Automatic speed detection without accessories: no reflective tape or tachometer; a normal vibration record is enough to read and track the speed.
• Conclusions ready for work orders: each conclusion gives the fault location, severity, confidence and measured characteristic frequency, with Excel export.
• Complete demonstration without a machine: the built-in test signal runs through the real diagnosis chain for demonstration, training and teaching.
• One system, several functions: shares acquisition hardware with FFT analysis, shaft orbit, order tracking and acoustic functions of the T8 platform.

Preparation

1. Select the channels to analyze (CH1–CH8, several allowed).
2. Set the analysis band: default 10–1000 Hz, matching the ISO evaluation band; normally no change is needed.
3. Set the speed source: automatic extraction from the spectrum, or the nameplate value when the speed is known and stable.

Parameters

Reading the diagnosis

1. Zone: the ISO zone at the right of the top table gives the overall condition.
2. Characteristic frequencies: check the extracted running frequency and the calculated inner race, outer race, rolling element and cage frequencies, to confirm that the bearing parameters are correct.
3. Conclusions: each line gives fault name, confidence, severity, measured characteristic frequency and SNR, sorted by severity, ready for the work order.

Reading the plots

Export and demonstration

Export to Excel saves the indicators and conclusions of the current channel for inspection records. Without a machine, enable test mode: demonstration data with unbalance and bearing fault components is synthesized from the current parameters, and the four plots and conclusions appear at once. Example with demonstration data: a speed of 1484.9 rpm is extracted automatically, and the conclusions list rotor unbalance, inner race fault, outer race fault and the ISO zone.