Columbus, Ohio -- A trio of sensing techniques combine to form a single, reliable system that can inspect the quality of high-power laser welding, according to a study at Ohio State University.
While each individual technique inspects welds with only modest accuracy, combining the three can detect faults with significantly better accuracy.
Dave Farson, assistant professor of industrial, welding, and systems engineering, explained that manufacturers employ high-power laser welding to join thick sheet metal parts for automobiles and airplanes. Airplane manufacturers can afford to invest time and money in expensive technology such as ultrasound to inspect the quality of every weld in a structure.
Makers of welded auto parts, on the other hand, typically destroy a handful of parts from each new batch to gauge overall quality quickly and inexpensively.
Increasingly, General Motors Corp., Ford Motor Co., and DaimlerChrysler are demanding that parts suppliers present computerized data proving the quality of their manufacturing processes. But laser welding is a complicated process, fraught with a host of potential pitfalls, so monitoring is difficult.
Conditions as simple as the natural wear of equipment or the presence of dirt skew sensor readings. To make matters worse, scientists don't yet understand how weld conditions lead to certain sensor readings.
Farson and his colleagues wanted to create a simple system that would gather reliable weld information from a distance, without interfering with parts or welding machinery. They used all commercially available equipment.
"Generally, this combination of sensors is going to require little maintenance and fuss on the part of the operators who load parts into the machines and take them out," said Farson. Farson conducted a study with Ohio State graduate student Afsar Ali and X.C. Li, a graduate student at Stanford University. They tested optical, acoustic, and charged-particle sensors alone and in concert to inspect welds at the Edison Welding Institute (EWI), an engineering consulting company in Columbus, OH. Some results appeared in a recent issue of the Journal of Laser Applications.
The researchers mounted the sensors unobtrusively away from the welding equipment. They created welds in low-carbon steel sheet material with deliberate defects such as gaps and insufficient fusion, then measured how well the three sensors recognized the defects. These welds were intentionally designed to be difficult to classify as good or defective to test the capability of the monitoring techniques.
Validation test results showed that the three sensors together performed much better than any individual sensor by itself.
Alone, the optical sensor correctly classified 61 percent of weld defects. The acoustic sensor correctly classified 75 percent, and the charged-particle sensor 69 percent. The combination of all three sensors correctly classified 83 percent of the welds.
Farson said that industry calls this approach 'sensor fusion' -- when different sensors measure the same thing, but they each have different shortcomings, such as noise or errors that corrupt the signals and make them inexact.
"The idea is that the three different sensors are corrupted by noise and errors in different ways," he said. "So by putting those multiple measurements together we get a better measurement of the information we're really looking for."
Farson and his colleagues based the design of their weld quality monitor on a common statistical technique called linear discriminant analysis (LDA), an approach which is similar to, but simpler than, neural network classification. In order to detect weld defects, LDA and neural network classifiers must first learn what good and bad weld looks like by examining examples of sensor signals from each.
Farson estimates that purchasing the components of an automated quality monitoring system with all three sensors would cost a manufacturer around $7,500, which he characterizes as not very expensive for the job. The computer data acquisition system makes up the majority of that cost.
Since writing the paper, Farson has been using the sensor combination to help industrial partners of EWI test manufacturing quality. He is also working with makers of commercial sensors like the ones he used in order to help them refine their systems.
In the future, Farson wants to simplify the LDA processing system, and simulate on computer how the signals received by the sensors relate to weld quality.
Parts of this work were funded by the Edison Welding Institute and Worthington Industries, a steel-processing company in Columbus.
Contact: Dave Farson, 614-688-4046; Farson.4@osu.edu
Written by Pam Frost, 614-292-9475;:Frost.18@osu.edu
Journal
Journal of Laser Applications