Skip to main content

Melt-pool physics: keyholes, balling and spatter

📍 Where we are: Part I · The Physics of the Melt — Chapter 1 of 19. The preface explained how to read this book; this chapter starts at the smallest scale where things go wrong, the liquid pool under the laser.

The simple version

Think of drawing a line across a pan of hot oil with a blowtorch pointed at a spoonful of flour. Where the flame is gentle, you get a shallow, calm puddle. Turn the flame up and it starts to dig a narrow hole, and the liquid around it sloshes, spits and sometimes traps bubbles. Laser powder bed fusion (LPBF) is that situation, shrunk to a puddle about as wide as a human hair, moving about as fast as a walking pace, and every layer of every part inherits whatever the puddle did.

What this chapter covers​

We follow the physics of the melt pool in laser-based metal additive manufacturing, mostly LPBF: the forces that stir the liquid, the two melting modes (conduction and keyhole), and the ways the pool throws away material (spatter, denudation) or traps gas (pores). We then ask what we can actually predict. Evidence levels are labelled: "lab-scale" (synchrotron or single-track experiments), "production" (seen on commercial machines in routine use) and "contested".

Problem​

A melt pool in LPBF is typically on the order of 100 µm across, lives for fractions of a millisecond to a few milliseconds, and cools very fast; reported cooling rates reach the order of a million kelvin per second [1]. You cannot see inside it with a camera: the metal is opaque and the vapour plume scatters light. Yet the pool decides whether the layer fuses, whether pores form, and how much powder is lost. The engineering problem is to hold the pool in a stable regime for hours, across thousands of layers, on powder that varies from batch to batch. Reviews cover the physics and modelling challenges of LPBF [2] and, more recently (2023), single- and multi-track melt-pool behaviour for process optimisation [3].

State of the art​

Forces that stir the pool​

Five effects dominate, and they are coupled.

  • Surface-tension gradients (Marangoni flow). Surface tension changes with temperature, so a hot centre and a cooler edge pull the surface liquid in a circulation. Experiments that tracked tracer particles inside welds and additive pools made this flow directly visible [4]. Demonstrated at lab scale.
  • Recoil pressure. When the surface approaches boiling, escaping vapour pushes back on the liquid like a rocket exhaust, depressing the surface into a vapour cavity. Mesoscale simulations used it to explain how melt flow produces pores and spatter [5].
  • Vapour-driven gas flow. The vapour jet drags the surrounding gas, and with it nearby powder, toward the pool. This "denudation" leaves bare tracks on either side of the scan line [6].
  • Buoyancy and gravity matter less at this scale but shape the pool trailing behind the laser.
  • Capillary instability. A long, narrow liquid track is geometrically unstable and can break into beads, which is the origin of balling.

Conduction mode versus keyhole mode​

At low power or high speed the laser heats the surface and heat conducts inward, giving a shallow, roughly half-round pool (conduction mode). Past a threshold, the recoil pressure digs a deep, narrow vapour cavity, the keyhole, and the laser is trapped by multiple reflections inside it, which raises absorption sharply. Keyhole-mode melting in LPBF was documented by time-resolved imaging early on [7]. The synchrotron X-ray method that made this routine for researchers, combining high-speed radiography with diffraction on a real LPBF set-up, was demonstrated in 2017 [8].

The transition is not a clean switch. Ultrahigh-speed imaging of Ti-6Al-4V measured the keyhole threshold and how the vapour-depression morphology changes with power and speed [9]. Work on an aluminium alloy found a transition regime between stable and unstable keyholes in which the cavity oscillates radially at several kilohertz and pores can start [10].

How pores are born​

Several distinct routes were identified with X-ray imaging combined with simulation [11]:

RouteWhere it happensEvidence level
Keyhole collapse traps a vapour bubbleDeep cavity, high energy densityLab-scale, imaged directly
Instability at the moving keyhole tipRear and bottom of the cavity [12]Lab-scale, Ti-6Al-4V
Gas trapped at scan-track ends or turn-aroundsLaser accelerates, decelerates or switches onLab-scale; seen in simulation and X-ray
Gas porosity from powderPre-existing gas in particlesProduction-relevant, less studied than keyhole routes

Dimensionless scaling laws that collapse keyhole stability and porosity across several metals have been proposed [13]. They are useful as screening tools, and they simplify heavily.

Spatter and balling​

Spatter is ejected material, both droplets from the pool and particles blown off by the vapour jet. Its origin and its link to the denuded zone are coupled in simulation and experiment [5][14]. Balling is the pool breaking into spheres rather than a continuous track, typically at high speed with a long, narrow pool.

Cross-section of a moving melt pool contrasting a shallow conduction-mode pool and a deep keyhole-mode pool with a vapour cavity, recoil pressure, Marangoni flow, spatter, a denuded zone and a trapped pore. Left, conduction mode: shallow pool with surface Marangoni flow. Right, keyhole mode: deep vapour cavity under recoil pressure, with spatter, denudation and a pore left behind by keyhole collapse. Original diagram by the authors, created with AI assistance.

Why it is hard​

  • It is opaque and fast. Direct observation needs a synchrotron (or similar) beamline, with frame rates from tens of kilohertz up, so each experiment is expensive and covers a few tracks.
  • Many coupled physics. Heat, fluid flow, phase change, vapour dynamics, laser absorption and powder motion all interact, and the key quantities (absorptivity inside a keyhole, vapour-plume effects, material properties near boiling) are poorly known.
  • Transfer is not guaranteed. Much of the X-ray work uses thin plates and specially built chambers. Whether a threshold measured there holds in a full-size production machine, with different gas flow, scan strategy and powder, is contested in the sense that it is rarely tested directly.
  • Alloy-dependence. Scaling laws and thresholds are fitted per material; extrapolation to a new alloy needs data.

Open questions​

  1. Can a keyhole threshold be predicted for a new alloy from thermophysical properties alone, without a parameter study? Scaling approaches exist [13] but are not yet a validated, general design tool.
  2. Which signals that are cheap to measure on a production machine (light, sound) carry enough information to detect pore-forming events in time to act? Recent work trained a model with X-ray-observed events as ground truth, first using thermal imaging [15] and then using sound and light sensors with sub-millisecond detection [16]. Both are demonstrated at lab scale; generalisation across machines is open. Chapter 6 on in-situ monitoring takes this up.
  3. Can beam shaping or modulation suppress spatter and keyhole pores without losing build rate? Simulation and experiment point toward interdependent controls [14], but production evidence is thin.
  4. How much of this carries over to directed energy deposition (DED), whose pools are larger and slower? Treat any transfer from LPBF as unproven until imaged in DED.

Industry impact​

Melt-pool understanding already pays off in production in a modest way: machine vendors supply parameter sets, validated for specific alloys and powders, that keep the pool in conduction or a stable shallow keyhole mode, and process engineers keep inside those windows by running density cubes, not by solving the physics. The frontier payoff is different: if the pool could be predicted and sensed well enough, parameter development for a new alloy would shrink from months of trial builds to a handful, and defect detection could move from after-the-fact CT to real-time. We flag that this payoff is not yet demonstrated in production.

Why it matters​

Almost every later chapter in this book (defects, microstructure, simulation, monitoring and control, alloy design, qualification) starts here. A pore that forms in a microsecond ends up as a fatigue-crack origin years later, and the only point at which it can be prevented cheaply is while the metal is liquid.

In the real world​

  • Process windows for commercial alloys are set empirically: vendors vary power, speed and hatch spacing until density cubes are dense, then fix them in the machine's parameter set.
  • Research groups use beamline X-ray imaging at national synchrotron facilities to study what happens inside the pool, and the resulting datasets are used to train and validate detection models [15][16].
  • Powder reuse and spatter deposits on the bed are a practical quality concern, because spatter lands on the next layer and can seed defects there (the mechanism is documented in the literature above; quantitative control limits vary by company and are rarely published).

Key terms​

  • Melt pool — the small volume of liquid metal under and behind the heat source.
  • Conduction mode — melting in which heat flows inward from the surface and the pool is shallow and stable.
  • Keyhole mode — melting in which a deep vapour cavity forms and traps the laser, giving deep, narrow pools.
  • Recoil pressure — the pressure exerted on the liquid by evaporating metal leaving its surface.
  • Marangoni flow — fluid motion driven by surface-tension differences caused by temperature differences.
  • Denudation — removal of powder near a scan track by vapour-driven gas flow.
  • Spatter — droplets and particles ejected from or near the pool.
  • Balling — breakup of a melt track into spheres by capillary instability.
  • Keyhole porosity — pores formed when the vapour cavity collapses or becomes unstable.
  • Synchrotron X-ray imaging — high-speed radiography using a synchrotron beam to see inside the melt pool.

Where this leads​

The pool freezes into a solid carrying its pores, cracks and locked-in stress. Next: Defects, residual stress and distortion.