3D positioning and geometric study of objects through Deep Learning. 2019-2020
System with learning and analysis capacity from some kind of 3D sensor (structured light, stereography, flight time, etc.). Starting from a 3D CAD model of a piece, the system will be able to build a virtual three-dimensional environment, and by simulating the process that we want to analyze, train a deep learning network that allows us to carry out the subsequent analysis of information in a safe and fast. We will use massively parallel hardware to accelerate the analysis process, and the software will be focused on optimizing said architecture. This software will be able to calculate a robot trajectory based on the position of the piece in space, verify the presence of an element in the study piece, check the measurements of a certain characteristic of the piece ...
Autoconfigurable metrological measurement system for production lines in automobile pieces providers. 2014
Many of the manufactured parts are large and have complicated geometries, therefore, it is impossible to measure them in the production line. Industrial measurement systems are generally large machines (coordinate system) located in temperature controlled laboratories. Current solutions are able to measure very few parts per time unit. These laboratories are typically used inspection of a part of each lot in a production, assuming that the assembly process is effective enough to not have to measure 100 % of them. But this is not this case, WIP has since found that this rule was valid in the past, but the specifications (tolerances) that automobile companies require to their suppliers (tolerances of tenths of a millimeter in large welding pieces), are destroying this measurement model.
Computational Maps in dynamic systems. 2012
The general objective is to achieve a computational model for unsupervised automatic learning. Computational maps in dynamic systems are investigated both in their theoretical aspect, related to artificial intelligence and in the field of development, since it has a direct application in the field of robotics. On the one hand, the theoretical study of the system is made to see its possible derivations and its theoretical potential, and on the other hand, simulations of the system are carried out on the computer and finally, to test its efficiency in a practical way, a final development is carried out by implementing the algorithmic of our theory in an industrial robot.
Analysis and digital image processing using biological algorithms. 2011
The acronym for this project is "Bioalgorithms". It consists of research related to the applicability of biological algorithms that allow us to have high-capacity image analysis and processing tools, it could open up a spectrum of commercial possibilities that in the near future could be decisive, acquiring a prominent position in the sector, and therefore a very important competitive advantage.
It is, therefore, to investigate new techniques for their application in artificial vision, as well as to work on the development of new algorithms for the processing and analysis of images, in order to achieve novel solutions that are reliable and functional, and that allow us to detect errors hitherto undetectable or verifiable by known artificial vision systems, and / or to optimize inspection times.
Electric power generation by means of low-temperature solar thermal energy. Stirling engine. 2008-2010
Basic investigation. The CSIC collaborates in the project.
The main objective of the project is based on the generation of electrical-thermal-solar energy, combining thermal solar technology with a low / moderate temperature Stirling machine for the generation of electrical energy. It would then be a matter of proposing an innovative renewable energy use system, economically sustainable in itself, based on a system of relatively low efficiency but very profitable, as an alternative to the photovoltaic technology used today.
In this way, we would obtain a system capable of providing renewable energy with competitive production costs in the current energy market. To achieve this, we will have to have a three-stage energy conversion system: 1. Solar energy → Thermal energy. 2. Thermal energy → Mechanical energy. 3.Mechanical energy → Electric energy.
Development of a photovoltaic solar monitoring system. 2008
Utility Model Nº 200802050.
This machine is currently running in several solar fields.
Prototype of a headlamp assembly dimensional correction system for position compensation and thermofusion fixing. 2006
The project aims to develop a prototype of a fixing system between the glass and the housing of a certain model of vehicle headlight, by checking the position of the glass and housing, correcting the geometry of the resulting assembly before curing of the adhesive based on the position taken and that of reference, and finally rigid fixing of the assembly to guarantee the position of the glass with respect to the casing during the drying time of the adhesive, in which the assembly continues its production process. A study of the lighthouse is carried out to arrive at a theoretical solution that solves:
- A system for measuring the position of the glass with respect to the housing by means of electronic probes.
- A mechanical-electrical system for rectifying the position of the glass by means of servo actuators that move it.
- A hot air deformation headlamp fixing system (floating thermofusion system).
- A control software for sealing, centering and referencing of the set.
- Elimination of position control after the total time for sealing fixation.
- The versatility of the final tooling, which allows working alternately with more than one headlight model.
- The necessary changes in the geometry of the product to guarantee the thermofusion.
Synthetic biological system for complex image treatment. 2005
Neural networks have performed well for certain issues (think of handwritten symbol recognition systems, or pattern recognition systems in general), but have been ineffective for more complex issues.
The aim of the project is to investigate, develop and implement a set of software tools capable of solving complex problems, based on the topology and functioning of natural biological systems (synthetic biology and complex engineering). The entire project is oriented towards artificial vision, with the intention of solving and implementing solutions for some of the needs that the industry raises in the areas of production and quality, and which have no solution with the existing systems at present.
Non-contact joint laser tracking system in welding processes. 2003
When considering the possibility of including a robot in an automatic welding process, the repeatability of the joints within half the diameter of the wire, the complexity of the joint configuration and the thickness and type of material must be taken into account.
Systems that address online repositioning in the torch space through contact with the joint are being used successfully, but there are a large number of applications where the touch sensor produces unexpected results. Traditionally this is resolved with an operator in front of a monitor who takes the necessary corrective measures.
The system that we managed to develop with this project in robotic welding equipment managed to track the welding joint based on laser triangulation and artificial vision, without requiring contact with the part and working in real time.
It was made up of a reading head that was integrated with the welding gun and that included the camera, lasers and optics necessary for real-time image acquisition, and an external module for capturing, communicating and processing data.
Recognition of non-rigid surfaces in space by artificial vision, using neural networks and genetic algorithms. 2002
Many techniques in pattern recognition have been used successfully, from sophisticated statistical methods to example-based learning systems. The vast majority of these techniques have been designed to generically address the problem of 2D pattern recognition.
Currently, and limited almost entirely to the scope of laboratory research, methods and techniques are being studied to solve the problem in space (3D). These techniques, for the most part, are based on the use of laser triangulation or stereo vision methods.
Having a general method of flexible 3D recognition would mean having the necessary techniques, methods and mathematical tools to face a huge number of automation processes, hitherto unapproachable in the industrial environment in which we operate.
With this Research project it was possible to design a set of mathematical algorithms and describe a method to solve the problem of recognition and positioning in space of flexible surfaces and by means of a single camera, aimed at treating the generic case of manipulation of objects in the industry. To do this we had to investigate neural networks such as calibration and projection methods in the coordinate space, and genetic algorithms to find the minimum error in the correlation space.
Orientation by rotating magnetic fields and artificial vision inspection of the toric ferrite adhesive. 2002
The GAP of the toric ferrites (0.1mm ferrite groove) was being sealed and inspected by manual procedure. This procedure involves the application by hand of an adhesive in the groove and the subsequent visual inspection by the operator. This leads to excessively long production times and very high error probabilities.
In this project we find the technical problem of the mechanical positioning of the part in industrial automations, its orientation and its fixing to a turning table when the turning movement is taking place. Due to the small dimensions of the part and the material it is made of, it is necessary to resort to a mechanical-electrical solution that combines rotating magnetic fields with vacuum. In this project we had to cover very different disciplines to achieve our objective: industrial artificial vision, manipulation of parts using magnetic fields, decision-making software based on neural networks and high-precision mechanics.
Sheet metal feeding installation with artificial vision forming. 2000
The developed project responded to the needs of a client, for whom a standard vision system was not sufficient, since it was not possible to achieve a binarization of the captured image.
The solution included a specific camera, image acquisition card, lighting, and software.
Claxon plastic cover inspection. 2000
The development consisted of a robotic installation that, starting from an assisted manual loading of horn threads, is capable of dosing them one by one, shaping them for loading in robot clamps and manipulating these using this robot to load the threads in an injection mold.
Installation is complex insofar as the wires are flexible wires that do not have a constant geometry, so their feeding and dosing require special precautions. Similarly, control and detection of threads is key, since a double injection mold load can destroy it in the closure.
Acquisition and image processing of tissue culture samples. 2000
The development consisted of developing specific software for a high-definition camera that is attached to a commercial microscope and that through a PC and the corresponding Frame Grabber allows images to be captured that the researcher may need for tissue culture.
The software developed is prepared to, once a tissue sample image has been captured, treat the image by applying a series of digital filters, file it in a database, count the number of particles in a marked area, determine areas in marked areas, etc. . This software was developed in Builder C ++.
Universal software platform that solves image processing applications obtained with artificial vision. 2000
In this project we had to solve the difficulty inherent in the industrial environment that arises from the poor quality of the images captured by the artificial vision cameras, mainly due to environmental disturbances or confusing environments.
In addition, we had to respond to the challenge that always arises in industrial environments related to achieving short process times. For all these cases, commercial standard tools do not work, and a program tailored to each application must be designed. In order to profitably face any industrial application of a certain complexity with an imaging process, it would be absolutely essential to have a set of software tools specifically designed to work in these conditions. In this project we develop this universal software platform that solves applications for image processing obtained with artificial vision.