Remaining useful life prediction of IGBTs Date: 2018-06-14 in MRI - - PowerPoint PPT Presentation

β–Ά
remaining useful life prediction of igbts
SMART_READER_LITE
LIVE PREVIEW

Remaining useful life prediction of IGBTs Date: 2018-06-14 in MRI - - PowerPoint PPT Presentation

Sep 27, 2023 250 likes β€’487 views

Reference: P1806141249 Remaining useful life prediction of IGBTs Date: 2018-06-14 in MRI gradient amplifiers Author(s): Martijn Patelski Distribution: PE Event 2018 attendees Reference: P1806141249R01 Template PN: 6001-1246-5511 a

slide-1
SLIDE 1

Slide 1 of 22 Reference: P1806141249R01

a passion for technology Date: Reference: Author(s): Distribution:

Template date: 06-07-2017 Template PN: 6001-1246-5511

Remaining useful life prediction of IGBTs in MRI gradient amplifiers

P1806141249 Martijn Patelski PE Event 2018 attendees 2018-06-14

slide-2
SLIDE 2

Slide 2 of 22 Reference: P1806141249R01

  • Introduction to the NG2200-XP Gradient Amplifier
  • System topology
  • Remaining useful life prediction
  • Results and Accuracy
  • Recommendations & ongoing improvements
  • Conclusion

Content

slide-3
SLIDE 3

Slide 3 of 22 Reference: P1806141249R01

Prodrive Technologies designs and manufactures a wide range of electrical products for Automotive, Industrial, Energy & Infrastructure and Medical markets. Prodrive NG2200-XP is a high tech Gradient Amplifier(GA) for MRI scanners

  • Current error of <0.1%, extremely linear output
  • 3-axes system, each containing 2100V / 1200Apk / Β±360ARMS end-stages
  • Each end stage contains several H-bridges
  • 2.5MVA peak apparent power
  • Insulated Gate Bipolar Transistors (IGBT) used as power switches

Introduction

slide-4
SLIDE 4

Slide 4 of 22 Reference: P1806141249R01

  • H-bridge topology with several parallel IGBT and freewheeling diode dies
  • Several H-bridges in series in each axis
  • Large part of BOM
  • Most power is dissipated in the silicon of the power switches
  • Average 10kW per axis
  • Very local heating
  • Temperatures vary

System Topology

slide-5
SLIDE 5

Slide 5 of 22 Reference: P1806141249R01

  • Customers want to know how long amplifier will last.
  • Customers design their own current profiles using a tool which includes the

thermoelectric model

  • Use of amplifier differs greatly between (research) hospitals
  • Few hours up to 24h per day
  • Different types of scans
  • Profiles restricted only by protections such as
  • RMS & peak output current
  • Maximum junction temperature of IGBT / Diode
  • Certain sequences are very intensive for lifetime
  • Unexpected failures result in expensive downtime
  • Possible solution: GA calculates remaining useful lifetime based on electrical and

thermal measurements in real-time.

  • No extra hardware required

Why lifetime prediction?

slide-6
SLIDE 6

Slide 6 of 22 Reference: P1806141249R01

  • Temperature gradients and mismatch in thermal expansion coefficient (CTE) results in

degradation of the interconnections between material.

  • Typical failure modes are
  • Bond wire lift-off
  • Heel cracking of the wire bond
  • Fatigue of the solder connections
  • Different conditions cause different failures
  • Simplify: Regard all failure modes as one

IGBT Failure Modes

slide-7
SLIDE 7

Slide 7 of 22 Reference: P1806141249R01

Overview of lifetime prediction

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

slide-8
SLIDE 8

Slide 8 of 22 Reference: P1806141249R01

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

Overview of lifetime prediction

slide-9
SLIDE 9

Slide 9 of 22 Reference: P1806141249R01

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

Overview of lifetime prediction

slide-10
SLIDE 10

Slide 10 of 22 Reference: P1806141249R01

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

Overview of lifetime prediction

slide-11
SLIDE 11

Slide 11 of 22 Reference: P1806141249R01

Overview of lifetime prediction

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

LC

slide-12
SLIDE 12

Slide 12 of 22 Reference: P1806141249R01

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

  • To calculate Junction temperature the manufacturer

Junction-Case thermal model can be used

  • TIM (Case-Heatsink) thermal impedance still unknown

π‘ˆ

π‘˜ 𝑒 = π‘Žπ‘’β„Ž,𝐾𝐼𝑄 π‘˜ 𝑒 + π‘ˆ β„Ž

Thermal Model: Junction-Heatsink

slide-13
SLIDE 13

Slide 13 of 22 Reference: P1806141249R01

  • Relationship between voltage drop and temperature is measured at 1A.

Thermal Model: Junction-Heatsink

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

slide-14
SLIDE 14

Slide 14 of 22 Reference: P1806141249R01

  • Relationship between voltage drop and temperature is measured at 1A.
  • Can be used to measure junction temperatures

Thermal Model: Junction-Heatsink

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

slide-15
SLIDE 15

Slide 15 of 22 Reference: P1806141249R01

  • Power and temperatures are known

π‘Žπ‘’β„Ž,𝐾𝐼 𝑒 = π‘ˆ

π‘˜ 𝑒 βˆ’ π‘ˆ π‘˜(∞)

𝑄

π‘˜

π‘Žπ‘’β„Ž,𝐾𝐼(𝑒) =

𝑗=1 𝑂

𝑠𝑗(1 βˆ’ 𝑓

βˆ’ 𝑒 πœπ‘—)

Thermal Model: Junction-Heatsink

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

slide-16
SLIDE 16

Slide 16 of 22 Reference: P1806141249R01

  • Only interested in dies which fail first  only interested in hottest dies
  • Create simple model from NTC to temperature to hottest part of case:

π‘Žπ‘’β„Ž,πΌβˆ’π‘‚π‘ˆπ·(𝑒) = 𝑠 (1 βˆ’ π‘“βˆ’π‘’

𝜐)

Thermal Model: Heatsink-NTC

R=0.017 K/W Ο„=9s C=517.2 J/K

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

slide-17
SLIDE 17

Slide 17 of 22 Reference: P1806141249R01

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

  • Cycle counting principle not error free
  • Simplification of reality: up to 32% error
  • Rainflow best performing counting model
  • Up to 15% error compared to FEM simulation

Rainflow Cycle Counting

slide-18
SLIDE 18

Slide 18 of 22 Reference: P1806141249R01

𝑂

𝑔 = 𝛽 βˆ— π›¦π‘ˆ π‘˜ βˆ’π‘œ βˆ—

𝐹𝑏 π‘“π‘™π‘βˆ—π‘ˆπ‘˜π‘›

  • Where Ξ±, n and Ea are found experimentally.

Kb is the Boltzmann’s constant and Ea represents the activation energy of the deformation process.

  • Model is simplified:
  • Heating time (ton) effect unknown for this particular

IGBT

  • Short heating time stresses bond wires
  • Long heating time stresses solder layers
  • In reality both cycles occur
  • More testing required

Reliability data IGBT module

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

slide-19
SLIDE 19

Slide 19 of 22 Reference: P1806141249R01

Dissipation model Thermal Model Cycle Counting (Rainflow method) Reliability Model Accumulate degradation (Palmgren- Miner) Current Profiles Lifetime Consumption

  • All (half) cycles add up to the total degradation, inversely proportional to Nf
  • Palmgren-Miner rule:

𝑀𝐷 = π‘ˆπ‘£π‘žπ‘’π‘—π‘›π‘“

𝑗

π‘œ 𝑗 𝑂

𝑔(𝑗)

  • Industry standard method to combine degradation contributions
  • Simple method with acceptable accuracy
  • Disadvantages:
  • No (simple) way to relate predicted lifetime to probabilities
  • Does not take into account that fatigue under different circumstances could be

independent

  • SEMIKRON research indicates it might even be completely independent under certain

circumstances

  • More testing is required

Palmgren-Miner rule & lifetime consumption

slide-20
SLIDE 20

Slide 20 of 22 Reference: P1806141249R01

  • Additional temperature sensors
  • Eliminate 5% NTC error
  • In progress: Improve reliability data
  • Power cycle tests
  • Compare simulation results with actual failure data
  • Vce/Vf measurement at known temperature and current
  • Eliminate model and reliability error
  • Failure occurs at 5% voltage increase (~30mV)

Recommendations & ongoing improvements

slide-21
SLIDE 21

Slide 21 of 22 Reference: P1806141249R01

  • Accurate model is now available
  • Thermo electrical model is verified and implemented
  • Also important for other implementations such as maximum junction temperature protection
  • Relatively easy to implement in FPGA
  • Much uncertainty in reliability data (extrapolation of data)
  • Online condition monitoring useful addition
  • Tracking of thermal cycles will give valuable Prodrive information for product

development

Conclusion

slide-22
SLIDE 22

Slide 22 of 22 Reference: P1806141249R01

a passion for technology

Prodrive Technologies

T +31 40 2676200 E contact@prodrive-technologies.com I www.prodrive-technologies.com