A Framework for Weld Fault Classification Module has been developed and shown in Fig. 1.
Figure 1. Framework for event and process driven analysis methods in RLW
The framework involves the following steps: (i) data cleaning: (ii) fault classification; (iii) event driven
fault classification; (iv) parameter adjustment.
The data cleaning is performed using Low pass filter which eliminates the high variance noise from the
signal obtained. The cleaned data is then passed through the fault classification module which uses
Hotelling’s T2 based approach for monitoring the process. The approach seems to work very well in
steady state, however too many Type I errors are observed in the transient state. To further improve
37 Final Report
the accuracy, a sensor fusion approach based on Dempster-Shafer Theory is implemented which
utilises the sensor data distribution to detect the fault.
The experimental study has shown that the sensor fusion approach is robust for Type II errors,
however leads to higher Type I errors. The event driven approach causes lower Type I errors with
higher Type II error rate. The video of the developed tool are shown below.
What is it?
- A RLW process monitoring software that is capable of identifying weld faults and estimating part-to-part gap by real-time analysis of in-process monitoring signals (e.g. plasma, temperature, reflection)
What does it do?
- Advanced statistical off-line analysis of weld defects in laser welding
- Online laser welding process monitoring based on multi-sensor information
- Process parameter (laser power and feed rate) adjustment in accordance with estimated part-to-part gaps
- Rapid and seamless weld defect identification
- Guarantee joining quality and further reduce costs for post-weld analysis and treatment
- Weld fault prevention in a batch of parts by adaptive process adjustment