CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the...
Transcript of CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the...
![Page 1: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/1.jpg)
CLAMPSMachine Learning in Automated Composite Manufacturing
NCC and CFMS Collaborative Project
www.cfms.org.uk
![Page 2: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/2.jpg)
CFMS Business LinesModel Based Engineering – constructing efficient computational architectures for system design that provide the foundation to establish an integrated product digital twin
Advanced Simulation – performing mathematical modelling of the physical world to derive an improved understanding of the performance of industrial products
Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying artificial intelligence methods to reduce cost and enhance performance
Engineering Computing Services – the heart of CFMS capability, an HPC resource & IT Laboratory providing a secure, agile experimental platform to test industrial M&S solutions
CFMS Vision : to be the recognised, independent and trusted digital test bed for the design of high value engineering products and processes
![Page 3: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/3.jpg)
www.cfms.org.uk
Business Case
Why
Back in the 80’s when A320 was first designed, nobody had expected that the Aircraft would be such a big success in the future.Hence its design was mainly driven by Performance criteria (e.g. payload, fuel burn and range) rather than High Volume Production criteria.But given the HUGE success of single aisle aircrafts, the next generation of single aisle aircrafts need to meet both Performance and industrialrequirements.
What
Dry fiber composite manufacturing with resin infusion technology promises to have the potential to meet the industrial requirements of next generation single aisle aircrafts. BUT given that it’s a relatively novel manufacturing process a quicker route to validate the potential and maturethe process is required.
How
Combining Machine Learning with Manufacturing Simulation (as shown in subsequent slides) will lead to accelerated novel manufacturing process learning curve for shorter ‘time to market’ lead times and also provide insights into improving process and product quality.
![Page 4: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/4.jpg)
www.cfms.org.uk
Resin Flow Simulation in Composite Part
Poor Resin CoverageGood Resin Coverage
Video Source : NCC
![Page 5: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/5.jpg)
www.cfms.org.uk
Mould Schematic
@20 Sensors
Tooling
ResinInlets 1,2
ToolingDry Fibre Part
Air Vents 1,2
![Page 6: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/6.jpg)
www.cfms.org.uk
CLAMPS Overview
Machine Learning
Mfg. Process Insights
Enables Faster Mfg. Process Maturity
Resin Simulation
Mfg. Process Parameters
Enables Mfg. Process MaturityIn virtual environment
Manufacturing Engineer
Domain Knowledge SimulationNCC NCC & CFMS
Quality Prediction
Reduces Simulation Lead-time
Machine LearningCFMS
![Page 7: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/7.jpg)
• Using resin flow simulation data , develop a machine learning model that predicts the quality of the composite part using manufacturing process parameters as a leading indicator
www.cfms.org.uk
CLAMPS Objective
Machine Learning
FE Flow Simulation
New Mfg. Process Scenario Prediction
Mfg. Quality IndicatorsMfg. Process Parameters
![Page 8: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/8.jpg)
www.cfms.org.uk
Input Data For Machine Learning
16 Manufacturing Process Parameters
TCI1 TOI2 TCI2 TOV1 TCV1 TOV2 TCV2 RT1T RT2T VFZ1 VFZ2 VFZ3 PI1 PI2 PVUnfilled Nodes
Fill Time
Simulation Inputs
Mfg. Process KPIs
Simulation Outputs
1500
0
![Page 9: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/9.jpg)
www.cfms.org.uk
Simulation
• Embarrassingly parallel RTM simulation application (single threaded).• Complete DoE sweep – no ’nudges’• Run on CFMS in-house HPC cluster - ~100k core-hours
![Page 10: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/10.jpg)
www.cfms.org.uk
Manufacturing Quality Criteria
Resin CoverageAcceptable
200 unfilled nodes
Resin CoverageNot Acceptable
![Page 11: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/11.jpg)
www.cfms.org.uk
Machine Learning Data Split
Training Set70%
Test Set30%
MACHINE LEARNING DATA SPLIT
15,000 simulation data set is split into training set and test for machine learning model development and testing
![Page 12: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/12.jpg)
www.cfms.org.uk
ML Development Process
Develop ML Model
Training Data SetLearnt Model
Test Data Set Validation of Predictions
![Page 13: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/13.jpg)
www.cfms.org.uk
Machine Learning Model Prediction KPIs
Exam Question :
Given a set of manufacturing process parameters, Can we predict if the RTM process can achieve satisfactory resin coverage
(Simulation) Ground Truth
Machine Learning Prediction
Good resin Coverage Poor resin Coverage
Good resin Coverage True Positive = 90.1 % False Positive = 2.4 %
Poor resin Coverage False Negative = 9.9 % True Negative = 97.6 %
* Using basic Machine Learning model tuning
![Page 14: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/14.jpg)
www.cfms.org.uk
Execution Times Comparison
FE Flow Simulation Machine Learning Prediction
Iterations 5000 5000
Hardware HPC Cluster (~10 nodes) Standard CAD Laptop
RAM 128 GB 16 GB
CPU Intel Xeon E5-2650v4 Intel Core i7
Lead Time
100 hours 3 minutes( plus one off training lead time of 15 minutes)
![Page 15: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/15.jpg)
www.cfms.org.uk
Key Mfg. Process Parameters for Machine Learning Prediction
Mfg
. Pro
cess
Para
mete
rs
Mfg. Process Parameter Importance
![Page 16: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/16.jpg)
www.cfms.org.uk
Key Mfg. Process Parameter#1 : First Vent Timings (TCV1)
Opening the first vent (TOV1) at t=0 has a strong influence on improving the resin coverage
Insight
![Page 17: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/17.jpg)
www.cfms.org.uk
Key Mfg. Process Parameter#2 : Second Vent Opening & Closing Timings
Opening the second vent (TOV2) at t=0 has a strong influence on improving the resin coverage
Insight
![Page 18: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/18.jpg)
www.cfms.org.uk
Key Mfg. Process Parameter#3 : Race Track Thickness Tolerance Band
Lower values for RT1T and RT2T has strong influence on improving the resin coverage
Insight
![Page 19: CLAMPS Machine Learning in Automated Composite Manufacturing · Data Science – deriving the maximum value from digital simulation, physical test or process operations by applying](https://reader030.fdocuments.net/reader030/viewer/2022041015/5ec629cfae2d043997041515/html5/thumbnails/19.jpg)
CFMS Bristol and Bath Science Park // Dirac Crescent // Emersons Green //Bristol // BS16 7FR
w:e: t:
[email protected] 906 1100