Feature: IIOT
AI-enabled sensor fusion improves industrial processes
By Suad Jusuf, Senior Manager, Renesas Electronics S
ensors are increasingly being used in a growing number of everyday applications, collecting data for analysis and process improvements. Tis is particularly applicable
in industrial automation, access control, security systems and healthcare, among many others. Gathering data from different sensors and
then combining it to generate an accurate representation of a setup or a network of systems is called “sensor fusion”. To increase its intelligence and reliability, deep learning algorithms are applied, to accurately predict abnormal behaviour in industrial machinery, tools and processes and hence anticipate and prevent failures and damage, saving time and cost and improving safety. At the heart of sensor fusion networks
(Figure 1) are smart sensors, which can fit into one or more category: • Redundant sensors: All sensors give the same information to the world.
• Complementary sensors: Tese provide separate information about the world.
30 March 2023
www.electronicsworld.co.uk
• Coordinated sensors: Sensors that collect information about the world sequentially.
All these sensors are linked via a communications backbone, which could be connected in any one of the following configurations: • Decentralised: No communication between sensor nodes.
• Centralised: All sensors provide measurements to a central node.
• Distributed: Te nodes exchange information at a given communication rate (e.g., every five scans, known as one-fiſth communications rate). Te centralised scheme can be considered
a special case of the distributed scheme, where sensors communicate every so oſten with one another; see Figure 2. For a sensor fusion network to be effective,
precisely calibrated and synchronised sensors are crucial.
Deep learning Performing late fusion allows for interoperable solutions, whereas early fusion gives AI-rich data for predictions.
Latest techniques involve time and space
synchronisation of all onboard sensors before feeding synchronised data to a neural network for predictions. Tis data is then used for AI training or Soſtware- In-the-Loop (SIL) testing of real-time algorithms that receive just a limited piece of information. Deep learning aims to present
complicated data simply. It relies on neural networks to enable computational platforms such as Renesas RA MCU and RZ MPU to train and execute tasks. Te neural networks consist of many processing layers, arranged to learn data representations from sensor fusion with varying levels of abstraction. Te more layers in the deep neural network, the better the training of the network and the more accurate the learned representations become. Multi-stream approaches are successful
in neural-network-based multi-modal data fusion. Tis way, neural networks can be trained and used on MCU-based endpoint applications, helping simplify setups and reduce costs, which in turn will accelerate
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
