Feature: AI
The sparse-modelling platform integrates Congatec’s Qseven computer-on-modules for most flexible performance scaleability
50 pictures for initial model creation, considerably fewer than the thousands needed by conventional AI. Tis is an essential benefit, as it enables engineers to build next-generation inspection systems that don’t always need best-in-class setup, e.g. constant lighting conditions. Tey also gain greater flexibility to adapt to changing production processes, which is essential on the way toward industrial IoT/Industry 4.0-driven lot-size-one production.
Executable on embedded edge devices From a hardware point of view, a sparse- modelling platform is lightweight and resource-efficient, and can be embedded into almost any edge device. It can run on embedded x86 computing platforms, further poised for implementation on platforms such as Xilinx, ARM, Altera and RISC-V. Compatible with both mainstream x86 processors and emerging open-source options, its design principles make it future-proof. However, as the final footprint depends entirely on the task and the complexity of the model required, a modular hardware platform based on computer-on-modules (COMs) is recommended. A pioneer in sparse modelling in
the manufacturing and medical fields is Hacarus. The company focuses on helping customers in industrial and medical applications, where rare conditions do not produce the big data required to train a deep-learning-based AI model.
Another huge application field
is precision manufacturing, where edge nodes lack the compute power to perform inference and training in parallel, and where sending data to the cloud is not feasible due to confidentiality and/or connectivity concerns.
Energy-friendly The company’s achievements with sparse modelling are highly convincing: As part of a project for an industrial customer, Hacarus was asked to perform a comparison of its sparse-modelling- based AI tool with a conventional deep- learning-based technique. A data set of 1,000 images was used
by both models to create predictions for this sample study. The customer had defined the accepted model prediction probability as 90%. Both approaches produced comparable results, but the required effort differed significantly: The sparse-modelling model was trained five times faster than the deep-learning one, despite the fact that the sparse- modelling-based tool ran on a standard x86 system with Intel core i5-3470S processor and 16GB RAM, whereas the deep-learning model required an industrial-grade Nvidia DEVBOX-based development platform with four TITAN X GPUs, each with 12GB of memory, 64GB DDR4 RAM, an Asus X99-E WS workstation-class motherboard with 4-way PCI-E Gen3 x16 support and core i7-5930K six 3.5GHz desktop processors. Ultimately, the sparse-modelling-
56 November/December 2020
www.electronicsworld.co.uk
based approach consumed just 1% of the energy needed by the deep-learning one, with the same level of accuracy – a compelling result. Aspiring data scientists should therefore add this method to their portfolio and consider using it in future AI applications to execute both training and inference at the edge.
Only 50 pictures needed Te small required footprint and performance efforts make it easy for vision system OEMs to implement AI. Existing-platform solutions can oſten be re-used, and system integration is relatively straightforward, since the Hacarus+ SDK (soſtware development kit) logic adapts to common vision- inspection systems without major changes to the setup. While existing systems can continue to perform their primary inspection, the soſtware takes care of only those images identified as ‘not good’, which means ‘possibly defect’. With around 50 or fewer such images,
sparse modelling can begin building a new inspection model. Once validated by human instpectors, it is ready to run as a second inspection loop beside the existing platform, and will deliver its inspection results back to the established system via its APIs. An optional HTML-based user interface is available for monitoring, as well. It is simple to see that such a logic can also run standalone; but since vision data pre- processing is not the core competence of sparse modelling, connectivity to
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68
