Transcript
  • 1

    Detecting Overheating in AM Parts using Computationally Efficient Thermal Models

    Rajit Ranjan, Matthijs Langelaar, Can Ayas, Fred van Keulen

    Structural Optimization and Mechanics (SOM)Delft University of Technology, The Netherlands

  • 2

    Local Overheating in Additive Manufacturing

    • Fusing successive layers of material on top of each other

    • Certain geometries accumulate more heat

    • Low local conductivity: improper heat evacuation

    [1] Image source: http://www.mkstechgroup.com[2] Mertens et. al. Optimization of Scan Strategies in Selective Laser Melting of Aluminum Parts With DownfacingAreas.

    Heat accumulation

    Dross formation [2]

    Powder

  • 3

    [1] https://www.kmwe.nl/upload/file[2] Darya et.al 2017 Reduction of local overheating in selective laser melting[3] Adam et.al 2014 Design for Additive Manufacturing—Element transitions and aggregated structures

    • Some more examples

    Local Overheating in Additive Manufacturing

    Vacuum seals manufactured by SLM [1]

    Same overhang angle but different

    overheating [3]

    Upward curling due to longer thermal path to the substrate [2]

    Thin section Thermal bottleneck

    https://www.kmwe.nl/upload/file

  • 4

    To develop an AM process model which can

    1. Detect features causing heat accumulation in a given part geometry and is

    2. Computationally efficient.

    Objective

  • 5

    Overview

    Conceptual Understanding:1D Analytical study

    Layer-by-layer part scale thermal model

    (Computationally expensive)

    Simplified models for estimating heat accumulation behaviour

    Performance evaluated based on• Accuracy• Computational effort

  • 6

    Overview

    Conceptual Understanding:1D Analytical study

    Layer-by-layer part scale thermal model

    (Computationally expensive)

    Simplified models for estimating heat accumulation behaviour

    Performance evaluated based on• Accuracy• Computational effort

  • 7

    Layer-by-layer thermal model• Element birth-and-death based lumped ‘layer-by-layer’ model [1]

    • Exposure Time per layer = Laser scanning time• Heat source calculated based on energy equivalence

    0T T

    Maximum temperature for all time steps

    ‘Hotspot map’

    • Temperatures indicates heat evacuation behaviour: Indicative fields

    • Simplifications• Only conduction heat transfer• Conduction through powder not considered • Phase transformation not considered• Constant room temperature thermal properties for Ti-6Al-4V are used [2]

    [1] Chuimenti et.al. 2017. Numerical modelling and experimental validation in Selective Laser Melting[2] Mujhkerjee et.al. 2018. Heat and fluid flow in additive manufacturing-Part I: Modelling of powder bed fusion

    913

    0

    K

  • 8

    Layer-by-layer thermal model

    • Hotspot map for a complex part geometry

    • Features with constant overhang angles• Different thermal behaviour!!

    • Wall clock time: 43m 23s

    Laser Power [W] 280

    Absorption Coefficient 0.45

    Scanning speed [m/s] 1.2

    Recoater time [s] 10

    Real layer thickness [mm] 0.1

    Lumping factor 10

    Key parameters [1]

    180 mm

    60

    mm

    [1] Chuimenti et.al. Numerical modelling and experimental validation in Selective Laser Melting

  • 9

    Overview

    Conceptual Understanding:1D Analytical study

    Layer-by-layer part scale thermal model

    (Computationally expensive)

    Simplified models for estimating heat accumulation behaviour

    Performance evaluated based on• Accuracy• Computational effort

  • 10

    1D Heat Equation

    Conceptual Understanding: 1D Analytical study

    2

    2

    1T T

    x t

    Lx

    2

    2

    1

    21

    ( 1)( , ) ( , ) 2 sin

    n tnh h L

    n

    n n

    x xT x t T x e

    L L

    Method of separation of variables

    2 2

    2 2

    1

    21

    ( 1)( , ) 2 ( , ) (1 )sin

    m h mt tmc h L L

    m

    m m

    xT x t T x e e

    L

    Q

    (2 1)

    2n

    n

    (0, ) 0T t ( ,0) 0T x( ,0) ( , )h

    hT x T x t

    ht

  • 11

    Observation 1: Only local domain matters

    2

    2

    21

    1( , ) ( , ) 1 2

    n ht

    h h Lh

    n n

    T L t T L e

    • Only relevant domain can be considered if top(=maximum) temperatures arerequired

    Q Q Q

    1L

    2L

    3L

    htSame and

    2c hL t

    1n

    Graph created using 1000n

  • 12

    Observation 2: Cooling is fast

    x L• Cooling at topmost point i.e.

    • Normalization ( , )ˆ ( , )( , )

    c

    c

    h

    h

    T L tT L t

    T L t

    ht

    2L

    • Factors that influence cooling• Heating time• Characteristic time

    2 2

    2 2

    2

    2

    ( )

    21

    21

    12

    ( , )ˆ ( , )( , ) 1

    1 2

    m m h

    m h

    t t t

    L L

    cm mc

    h th L

    m m

    e eT L t

    T L tT L t

    e

    3max 10

    • For our case, •

    • Cools to 30% of max within first 2s

    max 1ht

    1000, 1 hn t

    3

    4

    1

    10

    10

    10

    ht

  • 13

    Observation 3: Steady state as an indicatorQ

    L

    ssQL

    Tk

    0T

    • Steady state temperature also pick up low conductivity• Caution: no regard for local vicinity

  • 14

    Overview

    Conceptual Understanding:1D Analytical study

    Layer-by-layer part scale thermal model

    (Computationally expensive)

    Simplified models for estimating heat accumulation behaviour

    Performance evaluated based on• Accuracy• Computational effort

  • 15

    Concept extension: Decoupling

    • Layer-by-layer model

    • Fast cooling between the layers• Therefore, we consider each layer addition starting at initial temperature

    Maximum across all time

    steps

    Maximum across all geometry instances• Reduced wall clock time

    • No cooling• Parallel computation

    0T T 0T T 0T T

  • 16

    Concept extension: Localization and Steady State

    0T T 0T T 0T T

    • Decoupled geometries (same as previous slide)

    • Only domain close to heat deposition is relevant• Therefore, we only consider local domain given by

    cL

    • Transient• Steady State analysis

    Maximum across all geometry instances

    0T T0T T

    0T TcL

    cL

    cL

  • 17

    Results

    Layer-by-layer

    Decoupled layers

    Maximum % err = 1.2Mean % err = 0.19

    Steady state local analysis

    Max = 11 %Mean = 1.8 % Decoupled and local layers

    Layer-by-layer

    Wall clock time 43m 23s

    Qualitative match

    Decoupled layers

    6m 30s

    Decoupled and localize

    3m 38s

    Steady state

    43s

    Maximum % err = 10.38 Mean % err = 1.4Max = 10.38 % Mean = 1.4 %

  • 18

    Conclusions and Future work

    • Developed a layer-by-layer model for detecting heat accumulation in AM parts.

    • Significant computational gains achieved using 1D conceptual understanding• Valid for high fidelity part scale models as well

    • Computational gain makes it possible to integrate with Topology Optimization method.

    • In Progress:• Integration with Topology Optimization• Extension to 3D


Recommended