ANN Screw Fastening Summary

Bibliography:
Althoefer, Kaspar, Bruno Lara, and Lakmal D. Seneviratne. "Monitoring of self-tapping screw fastenings using artificial neural networks." TRANSACTIONS-AMERICAN SOCIETY OF MECHANICAL ENGINEERS JOURNAL OF MANUFACTURING SCIENCE AND ENGINEERING 127.1 (2005): 236.

Transcript

1. Monitoring of Self- Tapping Screw Fastenings Using Artificial Neural Networks Presented by Rahul Bali
2. Thread Fastening • Most popular assembly method. • Over a quarter of all assembly operations. Advantage of threaded fastenings • Assembled components can be disassembled • Relatively easily for repair, maintenance, relocation, or recycling.
3. Self-Tapping Screw • Torque – Insertion Depth Signature Signals • Analytical torque–insertion depth signature signal depends on the geometrical and mechanical properties of the parts. • Screw cuts the thread in the part to be fastened. • Avoiding the need for pre-tapped holes. • Handheld Insertion tools need to be handled by experts’ and their experience. • Automated Screw insertion is the Solution for manufacturing industry. Geometrical and Mechanical Properties in Action • Elastic Modulus • Coefficient of Friction • Strength of Materials • Thickness of parts to be joined • Hole Diameters • Dimensions of Screw
4. ‘Torque – Insertion Depth’ Signature Signals
5. Screw Insertion is divided into 5 distinct stages: -Begin Insertion -Screw Goes into Lower Plate -Screw Breakthrough -Only Thread Friction forces are present -Tightening Stage Screw Insertion Process
6. Neural Network Theory • RBF-ANN is used by providing training signals • Weights are optimized using the standard gradient descent approach • sum-squared error SSE, between the actual and desired outputs for given inputs was minimized Three different tests a) single insertion case, b) multiple insertion cases, c) multiple output classifications, are performed
7. NEURAL NETWORK DESIGN RBF-ANN MODEL INPUT LAYER HIDDEN LAYER OUTPUT LAYER MODEL 1 60 NODES (30 PAIRS) 15 1 MODEL 2 60 NODES (30 PAIRS) 20 4
8. MODEL 1
9. MODEL 2
10. The screwdriver, torque sensor, and instrumentation used for the acquisition of experimental data
11. Output for training set: Single Case Experiment Network is trained over 22 cycles
12. Output for training set: Variable Screw Size Experiment Network is trained over 52 cycles
13. Aim to separate Successful from Failed insertions. Further, classify successful into 3 classes A, B, C. Four-Output Classification
14. Output for test set: After 200 training cycles Four-Output Classification Experiment
15. Conclusions Provides RBF(Radial Basis Function) based ANN approach Capability to generalize and to correctly classify unseen insertion signals Ability of the network to classify into multiple categories
16. Thank You !
17. Althoefer K, Lara B, Seneviratne LD. Monitoring of Self-Tapping Screw Fastenings Using Artificial Neural Networks. ASME. J. Manuf. Sci. Eng. 2005;127(1):236-243. doi:10.1115/1.1831286. Bibliography