deep learning
Improve you processes with Deep Learning
Deep Learning
Deep Learning or machine learning algorithms allow us to provide machines with the ability to learn to obtain desired information based on training with already known data. At ATRIA we design and execute the complete project, from the data acquisition system to the trainig and validation of the algorithms.
What industrial applications has Deep Learning?
Industrial Applications
Intelligent recognition and location of objects
Deep Learning algorithms allow us to identify objects, classify them and locate them within an image. This can be applied to different industrial applications: identification of parts, detection of improper parts, location of parts, among many others.
Detections of variable defects
The defects are due to various causes derived from the process itself and the conditions to which the treated material is exposed. The algorithms create models that allow detecting defects in parts, superficial defects in manufacturing, painting, detection of defects in organic products, among others.
Machine maintenance and predictive maintenance
Through the analysis of data (of any type) obtained from the different machines, models can be generated that are capable of predicting when a failure will occur. Likewise, this serves to improve processes and prevent failures before the machines break down. Avoid production stoppages and reduce preventive maintenance times.
Smart production
In production, vision systems and robotics are combined with algorithms to improve processes and increase productivity. In fact, tasks with variability that a traditional robot could not perform on its own can be automated: recognizing and locating types of parts, processes and variable trajectories, etc. For this reason, it allows in many cases to reduce costs and increase the competitiveness of companies.
How do we select the most appropiate technologies?
Design of data acquisition system
At ATRIA we mainly work on Deep Learning projects applied to images. For this reason, we design the data acquisition systems that allow us to collect images to create a dataset with sufficient quality to develop the Deep Learning algorithms. We work on the selection of the hardware, its calibration and the synchronization for the data capture. Finally, we carry out the installation of it.
Data collection and preparation
We carry out the data capture of the real process through our acquisition system. The data is the basis for creating your own dataset. Previously, it is necessary to carry out a pre-processing and labeling of the same that we also do internally.
Model creation and training
We select and implement the appropriate neural network (RRNN) architectures for each application. We perform the training of the models using the previously created dataset.
Validation and start-up
We perform off-line and on-line validation to evaluate the performance of the algorithms. Finally, we perform the productive installation of the system and its optimization.
