🎓 Lesson 18
D5
Acoustic Emission Signal Processing for Wear Detection
Acoustic emission signal processing is like listening to the tiny 'cracks' and 'pops' a drill bit makes while cutting rock, then using those sounds to figure out when the bit is wearing out.
🎯 Learning Objectives
- ✓ Analyze time-domain AE features (RMS, hit count, amplitude distribution) from raw sensor data to classify wear stages
- ✓ Design a band-pass filter and feature extraction pipeline for AE signals acquired from rotary percussive drill rigs
- ✓ Explain how AE event rate and frequency centroid shift correlate with cutter degradation in hard-rock tunnel boring
- ✓ Apply k-means clustering to unsupervised AE feature space to distinguish between normal operation, incipient wear, and critical failure modes
📖 Why This Matters
In deep mining and TBM operations, unplanned tool changeouts cost $15k–$50k/hour in downtime—and premature changes waste 30–40% of usable cutter life. Acoustic emission monitoring enables predictive maintenance: by listening to the tool’s ‘voice’, engineers detect wear onset *before* performance drops or catastrophic failure occurs. This transforms reactive maintenance into adaptive control—directly feeding into Module 10’s core theme: closing the loop between real-time sensing and automatic parameter adjustment (e.g., reducing RPM or advancing thrust when wear accelerates).
📘 Core Principles
AE signals originate from rapid microstructural changes (e.g., carbide fracture, matrix fatigue, grit pull-out) during rock cutting. Unlike vibration, AE is broadband (20 kHz–1 MHz), transient (<1 ms), and highly sensitive to early-stage damage. Signal processing follows four layers: (1) Transduction—using resonant piezoelectric sensors (e.g., PAC R15α) mounted on the drill head; (2) Acquisition—high-sample-rate (>2 MS/s), low-noise digitization with anti-alias filtering; (3) Feature extraction—time-domain (RMS, counts, rise time), frequency-domain (spectral centroid, band energy ratio 100–400 kHz / 20–100 kHz), and time-frequency (wavelet packet energy entropy); (4) Classification—supervised (SVM, RF) or unsupervised (DBSCAN, Gaussian Mixture Models) mapping to wear states defined by SEM/EDS post-mortem analysis. Critically, AE amplitude is not directly proportional to wear volume—but its *statistical dispersion* and *burst repetition rate* show strong monotonic trends across wear life.
📐 Rise-Amplitude (RA) Ratio for Wear Regime Discrimination
The RA ratio (Rise Time / Average Amplitude) helps distinguish brittle fracture (low RA, sharp bursts) from plastic wear (higher RA, longer duration). A rising RA trend over time indicates transition from initial abrasion to macro-fracture-dominated wear.
RA Ratio
RA = \frac{\text{Rise Time } (\mu s)}{\text{Amplitude } (dB)}Discriminates wear mechanisms: low RA (<0.03) suggests brittle fracture; rising RA (>0.04) indicates progressive plastic deformation and grain pull-out.
Variables:
| Symbol | Name | Unit | Description |
|---|---|---|---|
| RA | Rise-Amplitude ratio | μs/dB | Dimensionless indicator of burst waveform shape; sensitive to crack propagation mode. |
| Rise Time | Signal rise time | μs | Time for AE signal to rise from 10% to 90% of peak amplitude—shorter in brittle events. |
| Amplitude | Peak amplitude | dB (re 1 V) | Logarithmic measure of maximum voltage output from AE sensor. |
Typical Ranges:
New PDC cutter (granite): 0.028 – 0.035 μs/dB
Moderate wear (flank wear 0.2–0.5 mm): 0.038 – 0.045 μs/dB
Critical wear (flank wear >0.7 mm): >0.052 μs/dB
💡 Worked Example
Problem: A PDC cutter on a raise-bore reamer produces AE hits with average rise time = 1.8 μs and mean amplitude = 42 dB (re 1 V). Calculate RA ratio and interpret against baseline (new cutter: RA = 0.035 μs/dB).
1.
Step 1: Convert amplitude to linear voltage scale: V = 10^(42/20) ≈ 126.5 V
2.
Step 2: Compute RA = rise_time / amplitude_in_V = 1.8e-6 s / 126.5 V ≈ 1.42e-8 s/V
3.
Step 3: Express in conventional μs/dB unit: RA = (1.8 μs) / (42 dB) = 0.0429 μs/dB
Answer:
The RA ratio is 0.0429 μs/dB — a 23% increase over baseline (0.035), indicating progression into moderate wear where micro-fracture events are lengthening relative to amplitude — consistent with carbide grain boundary decohesion observed in SEM.
🏗️ Real-World Application
At the Boliden Aitik copper mine (Sweden), Sandvik’s iSURE™ system integrated AE sensors on production raise-bore rigs. By tracking cumulative AE hit count >200 dB and spectral centroid drop below 220 kHz over 4-hour windows, operators reduced unplanned cutter changes by 37% and extended average cutter life from 112 to 158 hours—validated by post-run profilometry showing <0.3 mm flank wear at intervention vs. >0.8 mm in historical failures. The AE-derived wear index triggered automatic feed reduction (−18%) and RPM modulation (+5%)—a direct implementation of adaptive control per Module 10.