Laser Metal Deposition (LMD) is an additive laser deposition technique focused on direct manufacturing of metal parts, through the successive deposition of melted tracks layer by layer, building the component directly from its CAD model.

Unlike Selective Laser Melting (SLM), mainly used for rapid prototyping, consisting on the deposition of powder layers and their subsequent laser melt, it only adds the required material. Moreover, this additive manufacturing (AM) technique works on current processing technology, known as Laser Cladding, commonly used for repair of metal components. This makes LMD a promising technique for retrofitting of current industrial facilities towards the digital factory (Industrie 4.0).

Existing dimensional monitoring approaches are not suitable for on-line path control of LMD, because they do not resolve measurements in working coordinates without imposing restrictions in the movements. This enables the adaptation of laser paths to get the expected geometry. The geometrical control envisaged to guarantee LMD parts manufacturing requires two main developments:

• To develop a 3D geometrical monitoring system, intended to identify geometrical distortions and some superficial defects, where the deposited tracks have to be measured on-line.
• To develop an on-line path planner and control system, intended to fix geometrical part distortions, adapting the movement of the robot and the parameters of the process.

A cost effective 3D robotized scanning solution can be successfully used for on-line LMD monitoring. The solution provides measurements resolved in the robot coordinates with no motion constraints (6-DoF). It is flexible in terms of setup, because a novel self-calibration method is provided, which simultaneously estimates all the calibration parameters. Moreover, this unified solution is suitable to automate the reparation of damaged parts.

This means a step forward on the state of the art, since to our knowledge existing methods not solve the point cloud in working coordinates with 6 DoF
(degrees of freedom). Besides the self-calibration method solves the tedious task required for a flexible setup. Such capabilities are crucial for the on-line path control of the LMD process, and to adapt the solution to different requirement.

This meta-package currently contains packages that provide nodes and libraries for on-line 3D scanning and off-line 3D path programming for ABB industrial robots. It also includes the required utility for the scanning calibration, which was presented in IEEE INDIN 2015.

INDIN 2015 presentation

It contains four packages:

• etna_workcell: contains the working cell description files.
• etna_scanning: contains the on-line scanning tool for LMD monitoring.
• etna_calib: contains the 3D calibration tools required for scanning.
• etna_planning: contains the off-line path planning tool for LMD automation.

The video shows the system working in a typical industrial robot, with a clear
relaxation of requirements on the setup (without the need of complex and accurate fixtures) while keeping a good accuracy.

The solution is based on the 3D triangulation principle, which uses a laser stripe and a camera to determine the 3D points where the laser intersects the piece. The arrangement of laser and camera determines the resolution and accuracy of the system. This means, the system is able to be adapted for different part sizes with a simple setup readjustment.

The 3D triangulation method uses point correspondences to map from 2D image points to 3D coordinates. First, the projected laser stripe is detected using a center of gravity peak detector to determine the points of maximum light intensity in the image. Second, each laser peak detected is transformed to a 3D point in the camera coordinates. Third, the 3D points are translated from camera coordinates to the robot coordinates. Thus, since the camera is attached to the robot, the robot pose is used for reconstructing the 3D point cloud of the piece.

To launch the on-line 3D scanning system:

roslaunch etna_workcell etna_workcell.launch

A calibration procedure is required to perform the 3D point cloud reconstruction. The method uses an inexpensive planar checkerboard pattern. It uses the same set of images (shown in the next figure) to perform both camera and laser calibration. In this way, the system is calibrated in the robot coordinates, ensuring a proper correspondence between camera and robot coordinates and avoiding complex and accurate mechanical couplers.

Automatic checkerboard detection in different robot positions for system calibration.

The calibration procedure consists in placing fixed the checkerboard and moving the robot at arbitrary orientations in front of the pattern. In each position, an image is acquired and its corresponding robot pose is recorded. For each image, the checkerboard pose is estimated and the laser intersection is determined. Thus, the estimated point cloud is used to find the best-fit laser plane. Different least-squares algorithms have been used to minimize errors in the calibration estimation problem.