Automatic Laser-based Geometry Capture for Finite Element Analysis of Weld Beads

Automated welding is widely used and capable of creating uniform and reproducible welds. It, however, currently lacks the ability for adaptation during multipass welding processes, something that a skilled technician would do naturally during manual welding. By using a laser scanner, our protocol allows the progress to be monitored without causing delays unlike manual capture of the weld geometry using physical gauges.

It also produces a digital record of the deposition, which can be used for future analysis. This protocol is enabled by Meta one, which is the digital train of the welding cell. They are developed at the University of Manchester.

The digital train allows the recording and playback in a digital space of not just the build geometry, but also robot instructions, actual robot movement, and camera footage of the welding cell and the local melt pool. The laser scanner need to be calibrated to correct for any skew or tilt between it and the build plate. As part of this calibration process, the laser scanner data is brought into the coordinate system of the welding robot.

Rapid capture of weld geometry can facilitate decision-making between passes, possibly enhance through machine learning, and allow finite element models to be populated with true deposition geometry. To begin, place the calibration component inside the welding cell in the same orientation relative to the origin in the coordinate measuring machine. Set up the scan of the calibration component using the laser scanner system and record the scan output in the coordinate file containing X, Y, and Z values.

Measure each fiducial sphere using the laser scanner. Scan across each sphere to measure the calibration component. Truncate the line scan data by excluding points originating from the substrate, using circular masks in X and Y around the known positions, and applying thresholds in Z.Use the filtered point set to perform a least square sphere fitting and determine the central coordinates.

Now locate the calibration component using the welding robot equipped with a metallic tip of known dimensions. Using the touch sense routine, bring the tip into contact with the surface of each fiducial sphere at several locations. Allow the low voltage circuit to complete upon contact and record the robot's position through the system software.

Then place the substrate or parent material that will receive the deposition into the welding cell. Next, initiate a weld pass using the welding robot to deposit a bead of material along the substrate surface. Program the scanning robot to follow a path aligned with the bead location.

For simple linear beads, use the same start and end points as the welding robot and scan the entire plate area with the scanning robot for enhanced coverage. Take measurements while moving the scanner over the work piece. Record the start and end points of the scanning path, the movement speed of the robot, and the acquisition frequency.

Confirm that the frequency corresponds to a spatial interval, no smaller than the minimum element size planned for the finite element analysis. Fit the scanned bead profile with an appropriate analytical function using Python code. For single bead depositions, use a parabolic function for curve fitting.

Apply the same affine transformation matrix to the laser scan data as used during calibration. Crop the scan data to a region that covers the bead and includes a margin around it. Remove any scanning artifacts caused by reflections using height-based filters.

Then flatten the local region of the plate in the cropped scan by solving for the plate normal and apply a rotation matrix to align the plate normal with the vertical axis. Rotate the bead so that it aligns with the Y-axis to simplify further processing. Detect the bead orientation by applying a height filter just above the base plate level and fit the resulting data with a first order polynomial to generate a vector representing the bead direction.

If the bead is not straight, divide it into smaller segments that approximate straight lines. Now crop the rotated section to remove excess plate area while retaining approximately five millimeters of material on either side of the bead. Choose the central region of the bead forfeiting and execute a parabolic curve fit to the cross section of the data at this location.

Then using the fitted parabolic curve, determine the bead's lateral extents by evaluating the curve at the X coordinates where the bead meets the base plate at its maximum height. Next, evaluate the Y coordinates, which indicate the extents of the bead by using similar height thresholding techniques. Create termination profiles at both ends of the bead using the same parabolic shape transposed into the XY plane.

Now calculate the full outline of the bead by evaluating all parabolas at multiple X and Y coordinates covering the bead's entire span. Then add the base substrate geometry by specifying its actual extents or using predefined dimensions. Then use meshing tools in FreeCAD to convert the set of parabolic cross-sections into a single 3D solid geometry.

This surface should be suitable for import into a finite element analysis environment. Translate the outline points into the original welding robot coordinate system. First, reset the bead to the origin of a new coordinate system.

Then apply the transformation to restore its position in the original frame. Use macros to export the bead and substrate as STL and step files using the previously calculated geometry points. Finally, export the complete geometry into a file format appropriate for finite element analysis, such as the INP format.

Optionally, use the NETGEN mesh FEM tool in FreeCAD to create the mesh using moderate fineness with a maximum element size of 0.5 millimeters and minimum of 0.1 millimeters. The transformation matrix applied to the fiducial sphere locations resulted in overlapping, visibly distinct point sets between the coordinate systems. The measured radii of the fiducial spheres using the laser scanner showed consistent slight underestimation compared to edge finding and coordinate measuring machine techniques.

The positional error of the laser scanner in determining distances between fiducial spheres ranged from 1.30 millimeters to 2.05 millimeters, while both CMM and electronic edge finding errors remained below 0.1 millimeters. The final meshed full weld bead geometry was successfully imported into the finite element analysis software.