We Develop Vision for Machines and Robots with Deep Learning and AI
In our senswork Innovation Lab in Munich, we are continuously working on new solutions for optomechanical inspection tasks using deep learning and AI. We take up new knowledge and technologies for industrial image processing at an early stage in order to make them usable in industry and research -paving the way for the future of quality assurance in the machine vision industry.
Deep Learning Whitepaper: An Introduction to Deep Learning in the Automation Industry.
Conventional image processing systems and algorithms for image evaluation have continued to develop over the past few years. Above all, machine learning (or deep learning), a method from the field of artificial intelligence, opens up new possibilities for image data analysis.
Neuralyze® by senswork
senswork has developed the deep learning software Neuralyze® for quality tests with complex test objects. The efficient inspection software is suitable for tasks that cannot be solved with conventional image processing. This includes, for example, inspections of test objects with transparent, reflective, curved or inhomogeneous surfaces or the detection of products with a high variance of features.
A self-learning method with neural networks is implemented to assess characteristics. A large amount of image data is required for the training process, with the help of which the algorithm is then optimized.
With Neuralyze, for example, cracks, scratches, voids, fibers, grooves, hair, oxidation, delamination or inclusions can be detected.
FAQs – Frequently Asked Questions
In our Innovation Lab we focus on deep learning and image processing in order to be able to better solve the complex tasks of our customers.
Deep learning and Machine Learning (DL / ML) are methods used in the research field of artificial intelligence (AI). AI is originally a domain from computer science, the aim of which is to model software in such a way that it is capable of learning to a certain extent and, in terms of its interactions, indistinguishable from human actions.
The discipline has gone through several hype and consolidation phases over the past 65 years. In the course of the development of complex application frameworks and an enormous increase in the performance of computing hardware, a remarkable climax has been reached in recent years.
Although machine learning in image processing has been used in productive systems for more than 20 years, a number of new, methodical approaches open up possibilities here to significantly expand the scope of application and to use this technology on a broad basis.
For the use of DL / ML processes in image processing, senswork offers comprehensive support throughout the entire planning and integration process. When selecting a suitable toolkit, we are independent and orientate ourselves towards the optimal solution - this can consist of using our own deep learning software as well as the libraries of well-known manufacturers. Of course, we are also available to our customers with regard to advice, training and implementation as a single service.
A non-binding consultation is essential for inquiries about new image processing systems. From our experience, it is usually easy to assess whether the requirements can be solved with conventional methods, deep learning or machine learning, or possibly not at all.
With DL / ML systems, we always carry out a feasibility analysis in advance. For such studies, image data of the process to be checked are necessary, which either already exist at the customer's premises or which we record on site with an image processing system to be designed.
The basis of the implementation of DL / ML applications at senswork is that we carry out feasibility studies free of charge, but the customer undertakes to commission if the study proves successful implementation.
Deep learning (DL), machine learning (ML) and artificial intelligence include processes that implement self-learning methods for assessing characteristics with neural networks.
Often these are tasks that humans can solve intuitively. At the same time, however, they are difficult to describe and thus evade or have evaded the rule-based procedures that have prevailed to date.
With DL / ML there are now a number of powerful tools to deal with such applications efficiently. Many difficult questions in image processing that have not yet found a solution can now be solved very well. It is therefore advisable in any case to subject previously unsolvable inspection tasks to a feasibility analysis again.
In 2D applications, deep learning / machine learning is particularly strong for tasks on diffuse, inhomogeneous surfaces, in which the error features have a wide range. This includes:
- Die-casting technology: detection of cracks and cavities on rough surfaces
- Surface inspection: grooves, grinding marks, inhomogeneities, planarity defects
- Particle detection: detection and classification of contaminants
- Seal seam inspection: quality and tightness of weld seams
- Classification: assessment of plants and natural products
senswork has many years of experience in the industrial image processing sector, due to several employees who deal with issues in this industry, which has existed for more than 30 years, both in management and developer positions. We know the entire breadth of image processing and thus also the depths, trends and hypes.
Classification methods that have a certain overlap with the service promises of DL / ML have been used in industrial image processing for 20 years, so that the processes are not always as new as they are presented in the media.
Standardized toolkits and the enormous increase in performance in the hardware area are indicative of the strong surge in recent years. In combination, this enables things with a scope of complexity, the practical realization of which was unthinkable ten years ago. We have been accompanying this development for many years.
Find out more about deep learning in an interview with Markus Schatzl, head of the senswork Innovation Lab in Munich.
Membership in the field of AI
We are involved in the