This manuscript focuses on methodological and technological advances in the field of health assessment and predictive maintenance for industrial robots. We propose a non-intrusive methodology for industrial robot joint health assessment. Torque sensor data is used to create a digital signature given a defined trajectory and load combination. The signature of each individual robot is later used to diagnose mechanical deterioration. We prove the robustness and reliability of the methodology in a real industrial use case scenario. Then, an in depth mechanical inspection is carried out in order to identify the root cause of the failure diagnosed in this article. The proposed methodology is useful for medium and long term health assessment for industrial robots working in assembly lines, where years of almost uninterrupted work can cause irreversible damage.

Flatness sensors are required for quality control of metal sheets obtained from steel coils by roller leveling and cutting systems. This article presents an innovative system for real-time robust surface estimation of flattened metal sheets composed of two line lasers and a conventional 2D camera. Laser plane triangulation is used for surface height retrieval along virtual surface fibers. The dual laser allows instantaneous robust and quick estimation of the fiber height derivatives. Hermite cubic interpolation along the fibers allows real-time surface estimation and high frequency noise removal. Noise sources are the vibrations induced in the sheet by its movements during the process and some mechanical events, such as cutting into separate pieces. The system is validated on synthetic surfaces that simulate the most critical noise sources and on real data obtained from the installation of the sensor in an actual steel mill. In the comparison with conventional filtering methods, we achieve at least a 41% of improvement in the accuracy of the surface reconstruction.

Surface flatness assessment is necessary for quality control of metal sheets manufactured from steel coils by roll leveling and cutting. Mechanical-contact-based flatness sensors are being replaced by modern laser-based optical sensors that deliver accurate and dense reconstruction of metal sheet surfaces for flatness index computation. However, the surface range images captured by these optical sensors are corrupted by very specific kinds of noise due to vibrations caused by mechanical processes like degreasing, cleaning, polishing, shearing, and transporting roll systems. Therefore, high-quality flatness optical measurement systems strongly depend on the quality of image denoising methods applied to extract the true surface height image. This paper presents a deep learning architecture for removing these specific kinds of noise from the range images obtained by a laser based range sensor installed in a rolling and shearing line, in order to allow accurate flatness measurements from the clean range images. The proposed convolutional blind residual denoising network (CBRDNet) is composed of a noise estimation module and a noise removal module implemented by specific adaptation of semantic convolutional neural networks. The CBRDNet is validated on both synthetic and real noisy range image data that exhibit the most critical kinds of noise that arise throughout the metal sheet production process. Real data were obtained from a single laser line triangulation flatness sensor installed in a roll leveling and cut to length line. Computational experiments over both synthetic and real datasets clearly demonstrate that CBRDNet achieves superior performance in comparison to traditional 1D and 2D filtering methods, and state-of-the-art CNN-based denoising techniques. The experimental validation results show a reduction in error than can be up to 15% relative to solutions based on traditional 1D and 2D filtering methods and between 10% and 3% relative to the other deep learning denoising architectures recently reported in the literature.

In this paper, an automatic sampling-based view planning algorithm is proposed, for accurate 3D reconstruction of complex geometry parts present in manufacturing. The initial viewpoint sampling method is able to lower the complexity of the algorithm by creating a sparse visibility bipartite graph relating the targeted surface patches, with the potential viewpoints (camera poses defined in SE(3)), which are contained in the surroundings of the object. This graph is used to sample and simulate a subset of viewpoints, employing an iterative greedy parallel set cover which weights the coverage of the sparse and simulated visibility. This method prematurely rejects poor candidates and prioritize the viewpoints providing an increased coverage, with no expensive preprocessing of the 3D models. A randomized Greedy Heuristic with local search has been proposed to maximize the coverage, while minimizing the total inspection time, first with the set cover of the simulated viewpoints, and secondly with the sequencing of the viewpoints and robot positioning with obstacle avoidance. Furthermore, the performance of the system is demonstrated on a set of complex benchmark models from the Stanford and MIT repositories, yielding a higher coverage with a lower computational runtime compared to existing sampling-based methods. The validation of the full system has been carried scanning a Stanford Dragon positioned with a 12 axes kinematic chain composed by two robots.