Deep Hole Drilling Simulation: Chip Formation and Torque

Deep hole drilling has historically been an empirical process — tool geometries, feed rates, and coolant pressures are selected based on handbooks and past experience. Simulation and digital twin technologies are changing this. By predicting chip formation, thrust force, torque, and tool deflection before the first chip is cut, engineers can optimize parameters, reduce tool tryout, and detect anomalies in real time.

Overview ​

The deep hole drilling process involves complex interactions between cutting geometry, material deformation, coolant hydrodynamics, and structural dynamics. Unlike conventional drilling, where chip evacuation is straightforward and tool access is simple, deep hole drilling presents:

• Constrained chip evacuation through narrow flute or bore annulus
• High length-to-diameter ratios that amplify tool deflection and vibration
• Guide pad interaction with the bore wall for self-piloting
• High coolant pressures (up to 35 MPa) that affect chip formation mechanics

Simulation addresses these challenges at three levels:

Why Simulation Matters for Deep Hole Drilling ​

The cost of trial-and-error in deep hole drilling is high. A single BTA drill head can cost several hundred dollars, test pieces require material and setup time, and a tool breakage at depth can scrap an entire workpiece worth thousands.

Simulation reduces this cost by:

• Predicting optimal feed and speed without cutting tests
• Validating tool geometry (point angle, chamfer, guide pad position) before manufacturing
• Identifying chip evacuation problems before they cause jamming
• Forecasting tool life through stress and temperature distribution

Validation studies report that well-calibrated FEM simulations predict torque within 5–12% of experimental values and thrust force within 12–20%, which is sufficient for process design and parameter optimization.

Finite Element Methods for Deep Hole Drilling ​

Three FEM approaches dominate deep hole drilling simulation, each with distinct advantages.

Deform 3D — Lagrangian with Adaptive Remeshing ​

Deform 3D uses a Lagrangian formulation with continuous adaptive remeshing to handle the large plastic deformation of chip formation. It is the most widely reported FEM tool for deep hole drilling simulation.

Capabilities:

• Chip formation morphology (continuous, segmented, or ribbon chips)
• Temperature distribution at the cutting edge (critical for tool wear prediction)
• Stress and strain distribution in the shear zone
• Thrust force and torque prediction

Reported applications:

• Gun drilling of AISI 316L — chip geometry and force validation against experiment
• BTA drilling — 3D model with guide pad interaction
• Low-frequency vibration-assisted drilling of Ti6Al4V — showed 30–50% reduction in axial force

Limitation: Adaptive remeshing is computationally expensive. A single drilling simulation to 10 mm depth can take 24–72 hours on a workstation.

Abaqus CEL — Coupled Eulerian-Lagrangian ​

The CEL method models the workpiece material as an Eulerian (fluid-like) domain while the tool is a Lagrangian (solid) body. This avoids mesh distortion problems entirely.

Advantages:

• No mesh distortion — suitable for large-deformation chip formation
• Handles complex tool geometries (multi-insert BTA heads, indexable drills)
• Separately models central and peripheral inserts, reducing computational cost

Validation results:

• Torque prediction: within 5% of experiment for indexable drilling of AISI 4140
• Thrust force: within 12–20% of experiment
• Chip geometry: good qualitative agreement with high-speed video

Limitation: Does not capture chip breakage or segmentation well — continuous chip assumption limits accuracy for materials that form segmented chips (titanium alloys at high speed).

ANSYS — Structural and Modal Analysis ​

ANSYS is used primarily for structural analysis of the drill rod and tool body rather than chip-scale simulation.

Applications in deep hole drilling:

• Static FEA: stress and deformation at cutting edges and guide pads under service loads
• Modal analysis: natural frequencies and vibration modes of the drill rod (twist and bending modes dominate)
• Buckling analysis: critical axial load for drill rod stability under WOB

Liu and Feng (2019) used ANSYS to model a BTA drill and found that the most concentrated stress occurs at the cutting edge and tool block cutting surface, and the drill stem vibrates primarily in twist and bending modes.

Chip Formation Simulation ​

Chip formation in deep hole drilling differs from conventional drilling because the uncut chip thickness varies along the cutting edge, and the confined space affects chip curling and evacuation.

Key Modeling Parameters ​

Johnson-Cook Constitutive Model ​

The Johnson-Cook model is the standard for drilling simulation:

$$\sigma = (A + B\varepsilon^n)(1 + C\ln\dot{\varepsilon}^*)(1 - T^{*m})$$

where:

• A = yield stress at reference strain rate
• B = strain hardening modulus
• n = strain hardening exponent
• C = strain rate sensitivity coefficient
• m = thermal softening exponent
• ε = equivalent plastic strain
• ė* = normalized plastic strain rate
• T* = homologous temperature

For deep hole drilling of steels (AISI 4140, 1045), typical Johnson-Cook parameters are calibrated from Split-Hopkinson Pressure Bar tests at strain rates up to 10⁴ s⁻¹ and temperatures up to 600°C.

Chip Morphology Prediction ​

Simulation predicts three chip types relevant to deep hole drilling:

For BTA drilling, where chips must pass through the interior of the drill tube, chip morphology prediction is critical — a continuous ribbon chip will jam, while well-broken chips evacuate reliably. Recent work using SPH (smooth particle hydrodynamics) coupled with DEM (discrete element method) has shown promise for modeling chip breakage and evacuation simultaneously.

Force and Torque Prediction Models ​

Predicting thrust force and torque is the primary goal of most deep hole drilling simulations. Three modeling approaches exist.

Analytical Force Models ​

Based on oblique cutting theory, analytical models decompose the cutting edge into infinitesimal elements and integrate the force contributions:

$$F_t = \int_0^b (F_c \sin\gamma + F_t \cos\gamma) db$$ $$M = \int_0^b r(F_c \cos\gamma - F_t \sin\gamma) db$$

where:

• F_t = thrust force
• M = torque
• F_c = cutting force component in cutting direction
• F_t = thrust force component perpendicular to cutting edge
• γ = inclination angle of the cutting edge element
• r = radius of the cutting edge element
• b = width of cut

Li et al. (2018) developed a staggered BTA drill force model using constraint relationships between cutting force components and cutting angles, calibrated using measured forces during drill entrance. Model-predicted thrust and torque showed good agreement with experiment across multiple feed rates and cutting speeds.

Advantages: Fast computation, no mesh required, provides insight into physical mechanisms. Limitations: Requires calibration for each tool-workpiece combination; does not capture transient effects like chip jamming.

Empirical Models ​

Empirical models relate thrust and torque to cutting parameters through power-law or polynomial fits:

$$F_t = C_F \cdot f^{a_F} \cdot v_c^{b_F} \cdot d^{c_F}$$ $$M = C_M \cdot f^{a_M} \cdot v_c^{b_M} \cdot d^{c_M}$$

where f = feed, v_c = cutting speed, d = diameter, and C/a/b/c are empirically fitted constants.

Furness et al. (1992) used discrete-time dynamic models (first-order with time delay) relating feed to thrust and torque, estimated via recursive least squares from transient drilling data. These models are suitable for closed-loop control.

Hybrid Physics-ML Models ​

The newest approach combines physics-based force models with machine learning:

• A physics model provides the baseline force prediction
• ML (Gaussian Process Regression, Support Vector Machines, neural networks) learns the residual error between physics prediction and actual measurement
• The hybrid model outperforms either approach alone

Just-in-Time Learning (JITL) frameworks for real-time torque prediction have demonstrated 23% lower RMSE compared to sliding window methods, making them suitable for digital twin deployment.

Digital Twin for Deep Hole Drilling ​

A digital twin is a virtual representation of the drilling process that is continuously updated with sensor data from the physical operation. For deep hole drilling, the digital twin concept is emerging as a tool for real-time process monitoring and anomaly detection.

Architecture ​

A deep hole drilling digital twin typically comprises:

1. Physical layer — sensors on the machine: torque, thrust force, coolant pressure, flow rate, vibration, acoustic emission
2. Data layer — time-series data acquisition at 1–10 kHz sampling
3. Model layer — a hybrid physics-ML model that predicts expected torque and force for the current cutting conditions
4. Decision layer — compares predicted vs actual values; flags anomalies when deviation exceeds a threshold

Prediction Capabilities ​

Practical Implementation ​

For deep hole drilling operations, a practical digital twin implementation follows these steps:

1. Instrument the machine with torque and force sensors on the spindle and feed drive
2. Build a baseline model by running controlled cutting tests at known feed and speed
3. Calibrate the model to match measured torque within ±10%
4. Deploy the twin — compare real-time sensor data against model prediction
5. Set thresholds — torque deviation > 15% triggers inspection; > 30% triggers retract

Tip: Start with torque monitoring — it requires only existing spindle load sensors on most CNC machines. Thrust force monitoring typically requires a dynamometer or instrumented tool holder, which is a higher investment.

Validation Methods ​

Simulation and digital twin models must be validated against experiment. The standard validation approach for deep hole drilling simulation involves:

1. Cutting tests at 3–5 feed/speed combinations within the recommended range
2. Torque and thrust measurement using a rotating dynamometer (e.g., Kistler 9123C) or instrumented tool holder
3. Chip collection and analysis — compare simulated chip morphology with micrographs
4. Surface finish measurement — Ra profile correlated with simulated temperature and stress
5. Model calibration — adjust friction coefficient and Johnson-Cook parameters to minimize prediction error

Acceptance criteria for industrial use: torque prediction within ±15%, thrust force within ±25%, and chip morphology matching qualitatively.

Summary ​

FAQ ​

What simulation software is best for deep hole drilling? ​

Deform 3D is the most widely reported for chip-scale simulation of gun drilling and BTA drilling. Abaqus CEL is preferred for multi-insert indexable drills and tools where mesh distortion is problematic. ANSYS is used for structural and modal analysis of the drill rod.

How accurate are FEM predictions of drilling torque? ​

Well-calibrated FEM simulations predict torque within 5–12% of experimental values. The Coupled Eulerian-Lagrangian (CEL) method in Abaqus has demonstrated torque prediction within 5% for indexable drilling. Accuracy depends on correct Johnson-Cook parameters and friction coefficient calibration.

Can simulation predict chip jamming in BTA drilling? ​

Traditional FEM has limited capability to predict chip jamming. Newer approaches using SPH (smooth particle hydrodynamics) coupled with DEM (discrete element method) show promise for simulating chip evacuation and blockage. This is an active research area.

What is a digital twin for deep hole drilling? ​

A digital twin is a virtual model of the drilling process that receives real-time sensor data (torque, force, pressure, vibration) and compares measured values against predicted values. Deviations trigger alerts for tool wear, chip jamming, or imminent tool breakage. It enables condition-based maintenance and process optimization.

Do I need a dynamometer to validate simulation results? ​

For research-level validation, a rotating dynamometer (e.g., Kistler) is required. For industrial adoption, spindle load monitoring provides sufficient torque data. Thrust force is harder to measure without a dynamometer, but can be estimated from feed motor current in many CNC machines.

What is the Johnson-Cook model and why is it used? ​

The Johnson-Cook model defines material flow stress as a function of strain, strain rate, and temperature. It is the standard constitutive model for machining simulation because it captures strain hardening, strain-rate sensitivity, and thermal softening — all of which are significant at the high strains (2–5), strain rates (10⁴–10⁶ s⁻¹), and temperatures (300–900°C) in deep hole drilling.

How long does a deep hole drilling FEM simulation take? ​

A single simulation to 5–10 mm drilling depth typically takes 24–72 hours on a multi-core workstation, depending on element size and model complexity. This limits FEM simulation to offline tool design and parameter optimization, not real-time process control.

Can simulation replace cutting tests entirely? ​

Not yet. Simulation reduces the number of required cutting tests but does not eliminate them. Material property characterization (Johnson-Cook parameters) and friction coefficient calibration require at least some experimental data. The goal is to reduce tryout from 20+ tests to 3–5 validation runs.

What is the difference between Lagrangian and CEL methods for drilling simulation? ​

Lagrangian (Deform 3D) tracks the material as it deforms and uses adaptive remeshing to handle distortion. CEL (Abaqus) models the workpiece as an Eulerian domain through which material flows, avoiding mesh distortion. Lagrangian is better for chip morphology; CEL is better for tools with complex geometry and multi-insert configurations.

How is vibration-assisted drilling modelled in simulation? ​

Vibration-assisted drilling is simulated by superimposing an oscillatory displacement (typically sinusoidal, 10–300 Hz, 0.05–0.5 mm amplitude) on the tool feed motion. FEM models show that this oscillation reduces continuous chip length, lowers cutting forces by 30–50%, and improves chip evacuation — all of which are validated by experiment.