2-D quality control has been widely used for solving tons of industrial applications where the position of the piece was previously known, and thus, a precise picture of the desired plane could be taken. Several 2-D tools are able to align defined regions to the processing images, however, this is only possible in cases of flat objects always located in the same plane. But, what happens with complex 3-D objects with 6 degrees of freedom in the object position? In these situations, 2-D techniques fail at providing a precise solution, hence bringing 3-D into play.
Until now, due to speed limitations, 3-D scans have been done outside the production line on random samples. But thanks to 3-D software1, 100% of the production can be analyzed at production time, directly on a conveyor belt.
Let’s take, for example, a production line-any industry will do-where parts of the produced objects must comply with demanding dimensional accuracy standards. The objects are coming on a conveyor belt in random positions and, therefore, they can be tilted, shifted or rotated in any axis. Knowing where the critical parts are in order to ensure their dimensions is not an easy task and can hardly be done by means of 2-D techniques. In this article, we will elaborate on how this can be quickly and easily done by using the 3-D software library together with any standard 2-D analysis tool.
Solution OverviewThe proposed solution will entail the combined use of 3-D and 2-D vision tools to achieve a system that can be used on a production line for the automated inspection of the 100% of production without the need of human intervention.
As a summary of the process, we will first scan a reference object, rotate this 3-D representation to the desired position and then project it into a 2-D representation, which will be processed by a standard 2-D analysis tool where the regions of interest (ROI) will be defined.
The subsequent objects will then be scanned, no matter their position, and their cloud of points aligned with this first reference object to represent both surfaces on the same reference coordinate. Once we have these two clouds of points aligned, we will be able to compute the differences or automatically create the 2-D representation so that the ROI can be analyzed with the 2-D tools.
The AcquisitionThe first step when working with 3-D models will involve obtaining the 3-D representations of the objects under scan. An easy way of doing this would be to use a laser triangulation system where the part to be scanned moves under a laser line and a camera records the shape of the reflected laser over the surface of the object. While there are commercial 3-D cameras that can work with this system, simple 2-D cameras also can be used, thanks to software tools used to obtain the maximum intensity point of a laser stripe.
However, the use of 3-D software is not restricted to laser-triangulation systems and any other 3-D source can be used, for example, coded structured light acquisition sensors.
But, as important as obtaining the 3-D representation is, having it in real metric units is more, otherwise, it would be impossible to perform any metric measurement of it. There are input sources that come with a generic pre-calibration to directly return a cloud of points with metric coordinates. But to provide better results, more flexibility or precision of a previous calibration step would be needed. Here again, 3-D software can provide several calibration tools for laser triangulation systems, which adapt to different acquisition set-ups (linear, angular, multicamera) to provide a cloud of points with real metric units.
The Reference ModelOnce we have obtained a first cloud of points, we will use it as a reference (golden model) for the others and, this way, automate the quality control process. As stated, we want to accurately analyze some parts of this object to control measurements on several reference points (gages, diameters). Until now, these measurements have been successfully obtained by 2-D tools when the object position was known. So, we will use 3-D tools to generate the correct view of the object and then take advantage of the 2-D tools by exporting our 3-D representation to a 2-D image, which can be understood by any standard 2-D analysis tool.
In 3-D software, this step can be performed by a module and consists on the generation of a Zmap. This Zmap object is a two-dimensional array representing a projection of a cloud of points on the X-Y plane, with every value of that array representing the Z coordinate at that point (if we take the plane Z = 0 as the floor, we can think of it as the height at that point). In addition, from every array on the Zmap it is possible to compute the corresponding X and Y coordinates, thanks to a known and fixed ratios pixel/metric.
Two factors must be taken into account in order to ensure that the exported 2-D object will fulfill our purposes. The first one is to ensure that the 3-D object representation has no perspective or distortion that would lead to incorrect measurements. If the data was calibrated and we had real metric units, this perspective distortion would have already been avoided. But to remove the lens distortion, another tool would be required, like the one that can be provided in 3-D software package.
The second consideration to be taken into account prior to export would be ensuring that this reference cloud of points is in the desired position so that the projection from 3-D to 2-D is done in the most appropriate plane. Imagine, like the example image, that we want to measure the diameter of a circle. If the cloud of points was tilted, the resulting 2-D image would not be a circumference but an ellipsis instead. This rotation can be done manually or by means of some helping tools like 3-D software’s geometric tool, which mathematically moves a cloud of points so that the selected plane is parallel to the Z plane.
Once the correct Zmap is generated, any 2-D tool would be able to select the regions of interest and perform the desired measurements. It is important to mention that this process must be done only during the project setup. As we will see, the following scanned objects will be mathematically aligned to match this reference object and thus the following Zmaps will be on the same position. This way, we will assure that the 2-D tool automatically measures on the expected position.
Quality Values on the Production LineOnce the reference model has been obtained and the 2-D tool configured, we are ready to perform measurements on any object coming on the production line (distances between reference parts, diameters, gauging), no matter which position it comes from.
This tool allows the fast matching of two clouds of points, based on the best fit approach, so that the scanned object is mathematically aligned to the reference object in hardly 100 ms. Once we have ensured that the scanned part is on the chosen position, a Zmap can be obtained and processed with the desired 2-D tool obtaining the wanted, highly accurate measurements in metric units in less than half a second.
Comparing ObjectsAnother processing option, instead of applying the several measures over a part, would be obtaining the differences between the scanned part and the reference model so that any malformation could be detected.
This easily can be obtained once the two objects have been aligned with a match tool. By means of such a tool, the differences between the two objects can be computed and represented on a disparity map, an object similar to the Zmap but containing the metric differences between two clouds of points. This data can be further processed with any 2-D tools, as it was done when measuring the Zmap, obtaining the differences between both objects in accurate metric units. Here, the 2-D tool can be used to define the most appropriate acceptation criteria. Q
1. SAL3D from the software company, AQSense.