Transcript Slide 1

Predicting Assembly Quality of
Complex Structures
Using Data Mining
Ph.D student Ekaterina Ponomareva
Supervisor Professor Terje K. Lien
Contents
•
•
•
•
•
Problem description
Data mining
Data mining techniques
Manufacturing example
Conclusions
Problem description
Quality improvement by means of:
• On-line data collection
• Establishing models of the processes that
reliable performance predictions can be
made from data
Manufacturing Example
Control Arm
Ball Joint
• 1 – ball stud
• 2 – plastic
liner
• 3 – housing
• 4 – cap
• 5 – seal
• 6 – clamping
ring
• 7 – clip ring
• 8 – sleeve
Manufacturing Example
• Assembly problem of the ball joint of
suspension system of the car is
considered
• Assembly process is automated, but there
are still many potential sources of failures
• It is important to establish a set of data,
which could possibly affect functions of the
assembly part
Manufacturing Example
• 1: Ball stud: heat treatment, surface finish, shape, and
diameter
• 2: Liner: flash, temperature resistance, grease
resistance, change in the volume of the part
• 3: Housing: heat treatment, inner diameter
• 4: Cap: properties of material, stiffness
• 5: Seal: diameter, amount of grease
• 6: Clamping ring: heat treatment, anti-corrosion
properties
• 7: Clip ring: heat treatment, anti-corrosion properties
• 8: Sleeve: properties of material, flash
Manufacturing Example
• The process data are examined by means
of data mining system to identify the cause
of the problem
• Data mining system verify the set of
“if-then” rules
• By means of these rules system indicates
significant problems in the process
Data Mining
• There is a need to turn the data into useful
information and to take steps according to
the knowledge gained
• This knowledge may include complex
relationships among process variables that
can define the optimal control settings or be
used to prevent defects
Data Mining Steps
•
•
•
•
Problem understanding
Data preparation
Pattern evaluation
Knowledge presentation
Data Mining Techniques
• The statistical techniques are regression
or clustering algorithms (require more prior
domain knowledge)
• The artificial intelligence techniques are
decision trees, neural networks and cluster
analysis (require extensive computing
recourses)
”IF-THEN” Rules
Conclusion
• Data mining can be used to extract previously
unknown manufacturing knowledge
• These knowledge can be used to discover and
analyse relationship between parameters that can
cause failures
• It also can be used to improve manufacturing
processes through more effective process and
quality control as well as safety enhancements
Acknowledgements
• Professor Terje K. Lien (NTNU)
• Professor Kesheng Wang (NTNU)
• Kristian Martinsen (Raufoss)