Siemens is launching onto the market a new module for the Simatic S7-1500 controller and the ET 200MP I/O system, which has a chip with artificial intelligence (AI) capability: The S7-1500 TM NPU (neural processing unit) is equipped with the Intel Movidius Myriad X Vision processing unit, thus enabling the efficient processing of neural networks. The module gets its function from the provision of a trained neural system on an SD card and is equipped with the USB 3.1 interfaces and a Gigabit Ethernet port. On the basis of the neural network, data from a connected sensor system or from the CPU program can be processed.
By using machine-learning algorithms, for example, visual quality checks in production plants or image-guided robot systems can be efficiently realized. This makes a considerably more efficient and more «human-like» behavior possible. With this module, Siemens is taking another step toward the integration of future technologies into industrial applications.
The installed VPU, Intel’s new Myriad X VPU chip, is the first in its class to have a dedicated hardware accelerator for deep neural network structures. The integrated image processing unit together with the computing unit for neural networks makes the Myriad X the trailblazer for computer vision applications.
The integral Intel chip facilitates new applications in industrial automation by speeding up the image-processing processes and fast local data evaluation by means of the trained models.
Users can connect compatible sensors such as cameras or microphones to the integrated interfaces of the newly developed S7-1500 TM NPU module. The data of the connected sensor systems, as well as information from the CPU program itself, can be processed using neural networks. The result of processing is then evaluated in the CPU program. Where the data of each workpiece must be configured most precisely for the recognition of workpieces using conventional image processing, this process can be structured with considerably more flexibility by applying learning procedures to identified image data. Open AI frameworks such as Tensorflow are used for this purpose.
The resulting advantage comes to bear, for example, during pick-and-place applications, in which a mobile robot must recognize, pick out and place components that are lying randomly in a crate. Added value can thus also be achieved during quality checks: Human expert knowledge regarding parameters such as consistency, color or quality of a product or a process can be transferred direct to the module through the continuous training of a neural network with assigned (image) data, e.g. by means of a connected camera.
Comments are closed.