Appearance
A manufacturer producing 40 mm × 1,000 mm bores in 4340 steel using BTA drilling experiences intermittent spiral mark defects that reduce surface finish below Ra 1.6 µm, causing 8% scrap. Post-process inspection detects the defects but cannot prevent them. Installing a multi-sensor monitoring system with strain gauge torque measurement, piezoelectric vibration sensors, and coolant pressure transducers enables real-time detection of the chip packing events that cause spiralling. A machine learning classifier trained on vibration and pressure features predicts spiralling onset 2.5 seconds before visible surface damage occurs, enabling feed override intervention that reduces scrap from 8% to 0.5%.
Principles of In-Process Gauging
In-process gauging for deep hole drilling refers to the real-time measurement of process variables during cutting, enabling immediate detection of anomalies and closed-loop control. Unlike post-process inspection, which detects defects after the part is complete, in-process gauging provides the data needed to prevent defects from occurring.
| Measurement Type | Physical Principle | Typical Sensor | Sampling Rate | What It Detects |
|---|---|---|---|---|
| Cutting torque | Strain or Villari effect | Strain gauge bridge / magnetic sensor | 1–10 kHz | Tool wear, chip packing, edge breakage |
| Feed force | Strain or piezoelectric | Piezoelectric dynamometer / strain gauge | 1–10 kHz | Guide pad contact, chip blockage, material variation |
| Vibration | Piezoelectric acceleration | Accelerometer (ICP type) | 10–50 kHz | Chatter, spiralling, bearing wear, tool vibration |
| Acoustic emission | Piezoelectric AE sensor | AE sensor (150–300 kHz) | 1–5 MHz | Chip fracture, crack initiation, tool fracture |
| Coolant pressure | Diaphragm pressure transducer | Pressure transmitter | 100–500 Hz | Chip evacuation blockage, coolant pump fault |
| Coolant flow | Turbine or magnetic flowmeter | Flow transmitter | 10–50 Hz | Coolant passage blockage, pump performance |
| Spindle power | Motor current / power sensor | Hall effect transducer | 100–500 Hz | Overall process load, tool condition trend |
| Spindle position | Encoder / linear scale | Rotary or linear encoder | 1–10 kHz | Feed rate consistency, depth correlation |
Torque and Force Monitoring
Cutting torque and feed force are the most direct indicators of the mechanical state of the deep hole drilling process. Changes in tool condition, chip evacuation, and material properties all manifest as measurable changes in these signals.
Strain Gauge Measurement
Strain gauges mounted on the drill tube or tool holder provide direct measurement of torque and feed force. Full-bridge configurations with temperature compensation achieve 0.1–0.5% accuracy.
| Parameter | Typical Value |
|---|---|
| Gauge type | Full Wheatstone bridge (350 Ω) |
| Excitation voltage | 5–10 V DC |
| Sensitivity | 1–2 mV/V at full load |
| Bandwidth | 0–5 kHz |
| Temperature range | 0–80°C (with compensation) |
| Typical torque range | 0–500 Nm (BTA 20–60 mm dia) |
| Typical force range | 0–20 kN |
| Accuracy | ±0.5% of full scale |
The primary challenge for strain gauge measurement in deep hole drilling is signal transmission from the rotating drill tube. Slip rings require regular maintenance and suffer from noise at high rotational speeds. Telemetry systems using inductive coupling or Bluetooth transmission eliminate slip ring maintenance but introduce latency.
Contactless Torque Measurement — Villari Effect
The Villari effect (inverse magnetostriction) enables contactless torque measurement. When a ferromagnetic drill tube is subjected to torsional stress, its magnetic permeability changes in proportion to the applied torque. A magnetic sensor coil surrounding the tube detects the permeability change without physical contact.
| Parameter | Villari Sensor |
|---|---|
| Contact type | None (non-contact) |
| Gap tolerance | 2–10 mm between sensor and tube |
| Bandwidth | 0–10 kHz |
| Sensitivity | Proportional to tube material magnetic properties |
| Temperature sensitivity | Moderate (compensation required) |
| Proof of concept | TU Dortmund / Magnetic Sense GmbH (2023) |
The Villari effect sensor demonstrated linear proportionality to strain gauge reference data in BTA drilling trials at TU Dortmund, confirming reliable torque measurement without the maintenance burden of slip rings.
Feed Force Monitoring
Feed force in BTA drilling is typically 30–60% of the cutting torque component. The feed force signal contains distinct signatures for:
- Normal cutting: Steady force with ±5–10% variation
- Guide pad contact fluctuation: Cyclic variation at rotational frequency
- Chip packing: Gradual force increase followed by sudden release
- Tool edge breakage: Step change in force level
- Material hardness variation: Correlated slow drift in force and torque
Vibration and Acoustic Emission Monitoring
Vibration monitoring is the most widely implemented in-process gauging method for deep hole drilling because accelerometers are robust, inexpensive, and easily mounted on the machine structure or workpiece.
Vibration Monitoring
| Sensor Type | Frequency Range | Mounting Location | Typical Application |
|---|---|---|---|
| Piezoelectric accelerometer (ICP) | 1 Hz–10 kHz | Workpiece fixture, spindle housing | Chatter, spiralling, bearing condition |
| Piezoelectric accelerometer (high-freq) | 1 Hz–50 kHz | Near cutting zone (if accessible) | Edge chipping, high-frequency torsional oscillation |
| MEMS accelerometer | 0–5 kHz | Machine base, coolant supply | Low-frequency vibration, imbalance |
Chatter detection: Chatter in deep hole drilling produces a characteristic vibration signature at specific frequencies determined by the drill tube natural frequencies (typically 80–300 Hz for BTA tubes above 1 m length). The onset of chatter is detectable as a rapid amplitude increase in these frequency bands.
Spiralling detection: Spiral marks on the bore surface produce a vibration signature at the rotational frequency with harmonic content. The amplitude of the rotational frequency harmonic increases from baseline by 3–10× during spiralling events.
Acoustic Emission Monitoring
Acoustic emission (AE) sensors detect high-frequency stress waves (150–300 kHz) generated by material deformation and fracture. In deep hole drilling, AE provides unique sensitivity to:
| AE Phenomenon | Frequency Range | Detection Significance |
|---|---|---|
| Chip fracture | 100–300 kHz | Normal chip formation signature |
| Built-up edge fracture | 150–250 kHz | Indicates unstable cutting conditions |
| Tool edge micro-chipping | 200–400 kHz | Early warning of impending tool failure |
| Crack initiation in workpiece | 100–500 kHz | Subsurface damage detection |
| Guide pad galling | 50–150 kHz | Pad surface degradation |
| Chip packing / jamming | 50–200 kHz (bursts) | Erratic chip evacuation |
AE monitoring requires higher sampling rates (1–5 MHz) than vibration monitoring, which demands more capable data acquisition hardware. However, AE provides earlier detection of chip evacuation problems — typically 0.5–3 seconds before vibration sensors detect the resulting force change.
Coolant Pressure and Flow Monitoring
Coolant pressure and flow measurements provide indirect but valuable information about the chip evacuation state — a critical process condition that is otherwise invisible during drilling.
Pressure Monitoring
| Location | Typical Pressure Range (BTA) | What It Indicates |
|---|---|---|
| Pump outlet | 20–150 bar | Overall system pressure; pump condition |
| Drill tube inlet | 15–120 bar | Pressure drop across rotary union |
| Boring head (estimated) | 10–100 bar | Pressure at cutting zone (calculated from flow) |
| Chip tank return | 0.5–2 bar | Chip separation system condition |
Chip packing signature: When chips begin to pack in the annular gap between the drill tube and bore wall, the back-pressure increases. A characteristic sawtooth pressure pattern — gradual rise of 5–20 bar followed by a sudden drop — indicates intermittent chip packing and clearance events.
Chip Evacuation Monitoring Using Pressure
| Pressure Pattern | Interpretation | Corrective Action |
|---|---|---|
| Stable, within ±2 bar | Normal chip evacuation | None |
| Gradual increase (>10 bar/min) | Progressive chip accumulation | Increase coolant flow; reduce feed |
| Sawtooth pattern (±5–15 bar) | Intermittent chip packing | Adjust feed rate; check chip breaker |
| Sudden spike (>20 bar in <1 s) | Complete chip blockage | Immediate retract drill; clear bore |
| Slow drift upward over hours | Filter loading or coolant degradation | Check filtration; replace coolant |
| Pulsation at rotational frequency | Coolant hole partial blockage | Check drill head coolant passages |
Multi-Sensor Fusion Systems
No single sensor provides complete process visibility. Multi-sensor fusion combines complementary measurements to build a comprehensive picture of the drilling state.
Typical Sensor Fusion Configuration
| Sensor | Signal | Sample Rate | Feature Extracted |
|---|---|---|---|
| Strain gauge torque | Torque (Nm) | 5 kHz | Mean, trend, RMS, peak frequency |
| Accelerometer (workpiece) | Vibration (g) | 20 kHz | RMS, FFT bands (80–300 Hz), kurtosis |
| AE sensor | AE RMS (V) | 3 MHz (AE) / 10 kHz (RMS) | RMS, count rate, burst duration |
| Coolant pressure | Pressure (bar) | 500 Hz | Mean, trend, sawtooth amplitude |
| Spindle power | Power (kW) | 100 Hz | Mean trend, load change rate |
Signal Processing Pipeline
Raw sensor data must be processed to extract meaningful features for process monitoring:
- Anti-aliasing filtering: Low-pass filter at 0.5× sampling rate
- Periodic signal removal: Subtract rotational-frequency components to isolate anomaly signals
- Time-domain features: Mean, RMS, variance, skewness, kurtosis, peak-to-peak
- Frequency-domain features: FFT magnitude at key bands (rotational freq, chatter freq, natural frequencies)
- Time-frequency features: Short-time Fourier transform (STFT) or wavelet packet decomposition for non-stationary signals
- Feature selection: Principal component analysis or ReliefF to reduce dimensionality
- Classification: Support vector machine, random forest, or neural network for state identification
Decision Framework
| Observed Feature Combination | Most Likely Condition | Recommended Action |
|---|---|---|
| Torque ↑ + Vibration ↑ (broadband) | Tool wear progression | Regrind tool |
| Torque stable + Vibration ↑ (chatter band) | Chatter onset | Adjust spindle speed 15–20% |
| Torque ↓ + Vibration ↑ (rotational freq) | Spiral marking | Reduce feed; check chip breaker |
| Pressure sawtooth + AE burst | Chip packing | Reduce feed; increase coolant flow |
| Torque spike + AE burst | Edge chipping / fracture | Retract; inspect cutting edge |
| Force ↑ + Pressure ↑ | Guide pad buildup | Increase coolant flow; check pad condition |
| All signals stable + Ra degradation | Gradual pad wear | Regrind or replace guide pads |
Signal Processing and Feature Extraction
Time-Domain Analysis
Time-domain features are computed from the raw or filtered sensor signal x(t) over a sliding window (typically 0.1–1.0 seconds):
| Feature | Formula | Process Information |
|---|---|---|
| Mean | μ = (1/N) Σ xᵢ | Steady-state load level |
| RMS | RMS = √((1/N) Σ xᵢ²) | Signal energy — increases with tool wear |
| Variance | σ² = (1/N) Σ (xᵢ − μ)² | Process stability |
| Kurtosis | K = (1/N) Σ ((xᵢ − μ)/σ)⁴ | Impulsiveness — chip fracture events |
| Peak-to-peak | xₘₐₓ − xₘᵢₙ | Maximum load variation |
| Trend slope | dμ/dt over multiple windows | Wear progression rate |
Frequency-Domain Analysis
The FFT transforms the time-domain signal into frequency components. Key frequency bands for deep hole drilling monitoring:
| Frequency Band | Source | What It Monitors |
|---|---|---|
| 1–10 Hz | Rotational frequency (at typical speeds) | Spindle balance, runout |
| 20–200 Hz | Drill tube bending modes | Chatter in deep bores |
| 80–300 Hz | Drill tube torsional modes | Spiral marking, stick-slip |
| 500–2000 Hz | Cutting edge passing frequency | Edge condition, micro-chipping |
| 1–10 kHz | Machine structural resonances | Overall stability |
| 100–300 kHz | Acoustic emission | Chip fracture, crack initiation |
Wavelet packet decomposition provides superior time-frequency resolution for non-stationary signals. For deep hole drilling, wavelet analysis can detect the exact moment of chip packing onset with 50–100 ms resolution.
Tool Condition Monitoring
Tool condition monitoring (TCM) uses in-process sensor data to predict remaining tool life and signal when regrinding is needed.
Wear Progression Signatures
| Wear State | Torque Signal | Vibration Signal | Pressure Signal | Action |
|---|---|---|---|---|
| Sharp (new) | Baseline level | Low RMS | Stable | None |
| Initial wear (first 10–20%) | +5–10% from baseline | +10–20% RMS | Stable | None |
| Steady wear (20–70% life) | Gradual increase (+10–30%) | Gradual RMS increase | Stable | Schedule regrind |
| Accelerated wear (>70% life) | Rapid increase (>30%) | High RMS + chatter onset | Possible fluctuation | Regrind required |
| End of life (failure imminent) | Erratic fluctuations | High amplitude bursts | Pressure spikes | Stop immediately |
Tool Life Prediction Methods
| Method | Input Features | Prediction Horizon | Accuracy |
|---|---|---|---|
| Threshold-based | Torque level or RMS | Immediate (alarm) | Basic |
| Trend extrapolation | Slope of torque over time | Next 10–50 holes | ±20% |
| Machine learning (regression) | Multi-sensor features + cycle count | Remaining tool life | ±10% after training |
| Neural network (LSTM) | Time-series from multiple cycles | Remaining tool life | ±5% after sufficient training |
In-Process Bore Diameter Gauging
Direct measurement of bore diameter during drilling is technically challenging but has been demonstrated using:
Ultrasonic Gauging
An ultrasonic transducer mounted behind the cutting head emits pulses through the coolant and measures the time-of-flight to the bore wall. The round-trip time gives the instantaneous radial clearance between the tool and bore wall.
| Parameter | Specification |
|---|---|
| Sensor type | Immersion ultrasonic transducer |
| Frequency | 5–20 MHz |
| Accuracy | ±0.01 mm (with temperature compensation) |
| Couplant | Coolant (must be bubble-free) |
| Max coolant temperature | 50°C (for stable speed of sound) |
| Demonstration | Tool Joint Products patent AU2011227425 |
Post-Cycle Fast Profilometry
The Boeing patent (EP2047207B1) describes a method for measuring bore diameter profile immediately after drilling by retracting a diametric probe at controlled speed while recording diameter vs. depth. While not strictly in-process (measured after cutting but before workpiece removal), this method provides complete bore profile data within 4 seconds per hole.
| Feature | Specification |
|---|---|
| Measurement time | 4 seconds per bore |
| Depth resolution | 0.1 mm (continuous) |
| Diameter accuracy | ±0.005 mm |
| Probe orientations | 2 (90° rotation) |
| Data output | Diameter vs depth; roundness per section |
| Application | Quick quality check on every part |
Implementation Considerations
Sensor Selection Criteria
| Criterion | Recommendation |
|---|---|
| Most informative single sensor | Coolant pressure (chip evacuation) |
| Best early warning sensor | Acoustic emission (AE) |
| Most robust for production | Accelerometer on workpiece fixture |
| Most direct process measurement | Strain gauge torque (or Villari effect) |
| Best combination for production | Pressure + accelerometer + spindle power |
| Best combination for research | Torque + 3-axis vibration + AE + pressure |
Data Acquisition Requirements
| Sensor Type | Minimum Sample Rate | Recommended Rate | ADC Resolution |
|---|---|---|---|
| Torque / force | 1 kHz | 5 kHz | 16-bit |
| Vibration | 10 kHz | 20–50 kHz | 24-bit |
| Acoustic emission (raw) | 1 MHz | 3–5 MHz | 14–16-bit |
| Acoustic emission (RMS) | 10 kHz | 50 kHz | 16-bit |
| Coolant pressure | 100 Hz | 500 Hz | 12-bit |
| Spindle power | 50 Hz | 100 Hz | 12-bit |
Environmental Challenges
| Challenge | Effect | Mitigation |
|---|---|---|
| Coolant flooding sensors | Short circuit; false signals | Sealed connectors (IP67+); inductive sensors |
| Chip debris on sensors | Mechanical damage; signal blockage | Protective shrouds; air purge |
| Cable flexing in drag chain | Signal noise; cable fatigue | Dynamic-rated cables; cable break detection |
| Electrical noise from VFD | 50–500 Hz harmonics in signals | Shielded cables; differential inputs; notch filter |
| Temperature drift | Zero shift in strain gauges | Temperature compensation; auto-zero at cycle start |
| Coolant temperature variation | Speed of sound change (ultrasonic) | Temperature measurement + compensation |
Troubleshooting In-Process Gauging Problems
| Problem | Likely Cause | Corrective Action |
|---|---|---|
| False chatter alarms | Vibration sensor loose or resonating | Check sensor mounting torque; verify frequency band |
| Torque signal drift | Temperature affecting strain gauge bridge | Implement auto-zero between cycles |
| AE sensor no signal | Air gap or couplant loss at sensor-workpiece interface | Apply coupling grease; verify sensor spring loading |
| Pressure signal shows no chip packing | Pressure transducer too far from cutting zone | Install sensor closer to rotary union |
| High-frequency noise on all channels | Ground loop between machine and DAQ system | Use isolated signal conditioners; check ground bonding |
| Intermittent signal dropout | Cable connector corrosion in coolant | Replace with IP67-rated connectors; use dielectric grease |
| False tool wear alarm | Feed rate variation from axis not sensor | Correlate torque with actual feed rate from encoder |
| Vibration sensor overload at bore exit | Drill breakthrough transient | Implement exit zone blanking in analysis software |
| Cannot differentiate chip types from AE | Insufficient sample rate | Increase AE sample rate to 3+ MHz |
| Model accuracy degrades over time | Tool material or coating change | Retrain model after process changes; use transfer learning |
FAQ
What sensors are used for in-process monitoring in deep hole drilling?
The most common sensors are piezoelectric accelerometers (vibration monitoring), strain gauges or Villari-effect sensors (torque measurement), coolant pressure transducers (chip evacuation status), and acoustic emission sensors (tool condition and chip fracture detection). Spindle power monitoring via motor current sensors provides a lower-cost alternative that captures overall process trends.
How does coolant pressure indicate chip evacuation problems?
When chips begin to pack in the annular gap between the drill tube and bore wall, coolant back-pressure increases. A characteristic sawtooth pattern — gradual pressure rise of 5–20 bar followed by a sudden drop — indicates intermittent chip packing and clearance. A sudden spike exceeding 20 bar in less than one second signals complete chip blockage requiring immediate drill retraction.
Can in-process gauging predict tool failure before it happens?
Yes. Tool failure prediction uses trend analysis of torque and vibration signals. As the tool wears, torque increases gradually (10–30% above baseline). When torque begins to fluctuate erratically and vibration RMS increases sharply, the tool is in the accelerated wear zone — regrinding should be scheduled immediately. Machine learning models can predict remaining tool life within ±5–10% after sufficient training data.
What is the Villari effect and how is it used for torque measurement?
The Villari effect (inverse magnetostriction) describes the change in magnetic permeability of a ferromagnetic material under mechanical stress. In deep hole drilling, a magnetic sensor coil surrounding the drill tube detects the permeability change caused by torsional stress, providing contactless torque measurement. This eliminates the maintenance problems of slip rings while providing bandwidth up to 10 kHz.
What causes spiral marking defects and how can they be detected in-process?
Spiral marks are caused by intermittent chip packing that pushes the drill against the bore wall, creating a helical groove. They are detectable in-process through a combination of vibration increase at the rotational frequency harmonics and a sawtooth pattern in coolant pressure. Multi-sensor fusion with machine learning can predict spiralling onset 2–3 seconds before visible surface damage, enabling feed override to prevent the defect.
How is acoustic emission different from vibration monitoring?
Acoustic emission (AE) detects high-frequency stress waves (100–300 kHz) generated by material deformation and fracture, while vibration monitoring measures lower-frequency mechanical motion (0–50 kHz). AE provides earlier detection of chip evacuation problems (0.5–3 seconds before vibration sensors), greater sensitivity to edge micro-chipping, and the ability to distinguish different chip formation states through burst analysis. However, AE requires higher sampling rates (1–5 MHz) and more sophisticated signal processing.
What is the minimum sensor configuration for production monitoring?
The minimum practical configuration for production deep hole drilling is coolant pressure monitoring plus spindle power monitoring. Coolant pressure detects chip evacuation problems directly, while spindle power captures overall process load trends including tool wear progression. This two-sensor configuration costs approximately $2,000–$5,000 per machine and can detect approximately 70% of process anomalies. Adding a single accelerometer on the workpiece fixture increases detection capability to approximately 90%.
How do you implement closed-loop control from in-process gauging?
Closed-loop control requires real-time feature extraction (typically using a DSP or FPGA), a decision algorithm (thresholds or trained classifier), and a machine interface (override of feed rate or spindle speed via CNC). The simplest implementation automatically reduces feed rate by 20–50% when vibration amplitude exceeds a threshold, preventing spiral marking. Advanced systems use model predictive control to optimise feed and speed continuously based on sensor features.
What signal processing methods are used for deep hole drilling monitoring?
Time-domain features (RMS, kurtosis, peak-to-peak) provide simple process indicators. Frequency-domain features (FFT magnitude at key bands) identify chatter, spiralling, and rotational anomalies. Wavelet packet decomposition provides time-frequency resolution for detecting the exact onset of chip packing events. Machine learning feature extraction (autoencoders, principal component analysis) reduces high-dimensional sensor data to compact state indicators.
How often should sensor data be analysed for production monitoring?
For production monitoring, sensor data should be analysed continuously at the machine controller level with a response time of less than 100 ms for critical alarms (chip blockage, tool breakage). Trend data should be logged for every hole cycle and reviewed at daily or weekly intervals for tool wear progression. Machine learning model retraining should occur whenever process parameters (tool material, coating, workpiece alloy) change significantly.
Summary
In-process gauging for deep hole drilling uses real-time sensor measurements to monitor cutting torque, feed force, vibration, acoustic emission, and coolant pressure during the drilling cycle. Multi-sensor fusion provides comprehensive process visibility, enabling early detection of chip evacuation problems, tool wear progression, chatter onset, and spiral marking defects. The minimum practical sensor configuration (coolant pressure plus spindle power) detects approximately 70% of process anomalies at $2,000–$5,000 per machine. Adding vibration and acoustic emission sensors increases detection to over 95% and enables closed-loop control interventions — automatic feed reduction or drill retraction — that prevent defects before they occur. The Villari effect offers contactless torque measurement without slip ring maintenance. Machine learning classifiers trained on multi-sensor features predict spiralling onset 2–3 seconds before visible damage and estimate remaining tool life within ±5–10%.