COVER STORY
Latest software technology brings the world of automated Machine Learning and Artificial Intelligence to the masses
Imagine the transformation if the next system you design could autonomously learn about its environment, predict anomalies, and understand and report causes of failure. This is already becoming a reality with thousands of pumps, motors and home appliances! In the quest for smarter equipment and infrastructure, Machine Learning (ML) and Artificial Intelligence (AI) integration are now being employed to help boost the reliability, innovation and capabilities of equipment. Machine Learning gives software applications the ability to predict outcomes more accurately without being explicitly programmed to do so using data and is a type of Artificial Intelligence.
Nick Stone, field application engineer at Anglia, introduces the NanoEdgeTM AI Studio product
from STMicroelectronics which enables developers to create an optimal ML library for their STM32 Microcontroller based projects, using a minimal amount of data in just a few steps.
Implementing Machine Learning and Artificial Intelligence
The evolution of smart devices
As technology has moved into the connected IoT era many more powerful features have been made possible by the intelligent and powerful processing power available in the Cloud, this includes ML and AI implementations. As these new powerful features have been realised, they are now often entirely depended on for the efficient functioning of equipment, an example of this is smart home devices, we know how frustrating it can be when we are unable to turn on a light by voice activation due to no internet connection. For many applications this is just a minor inconvenience, however for mission critical equipment connectivity or latency issues it can be a big problem, therefore the need to bring some of this intelligent processing including ML and AI back onto the equipment itself, via intermediate gateways or edge devices which are not reliant on Cloud connectivity, has increased.
Before we go any further, it is worth drawing the distinction between ML and AI. A full implementation of AI allows a machine to address tasks that would normally require human intelligence. ML is a form of artificial intelligence that addresses tasks by learning from reference or collected data and then makes predictions or acts based on this data. Whilst ML and AI have now worked their way into everyday vernacular, implementing them on a technical level in a real- world application is challenging for embedded developers, even those with some prior experience of AI. The investment, complexity and development time required can often be barriers to AI adoption. Fortunately, STMicroelectronics have recognised developers need a new generation of tools to allow the mass implementation of ML and AI in applications and have released NanoEdge AI Studio, a new Automated Machine Learning tool that brings true innovation easily to end-users allowing the embedding of cutting-edge ML and AI algorithms directly into the host system microcontroller.
A major advantage of NanoEdge AI Studio is that it requires no specific data science skills. Any software developer using the Studio can create optimal ML libraries and start embedding smart features into the source code from its user-friendly environment with absolutely no AI skills. The Studio can generate four types of libraries: anomaly detection, outlier detection, classification, and regression libraries. An Anomaly Detection (AD) library can be generated using a minimal amount of data examples which show normal and abnormal behaviours to train the system by example. Once created the library can be loaded into the target systems microcontroller to train and interpret directly on the device. The library learns the equipment behaviour from data acquired locally and dynamically adapts to each equipment behaviour. Once trained, the library inference compares data coming from equipment over time against the locally created models allowing it to accurately identify and report anomalies. Outlier Detection (1C) can be used to detect any abnormality with the one-class classification method. This is especially useful when no examples of abnormal behaviours can be provided. In this instance the normal expected signal is imported into the Studio and then developers can easily create an optimized outlier detection ML library. A Classification (nC) library is used to classify a collection of data, each representing different types of equipment defects (such as bearing problems, cavitation problems or others), or distinct types of events in the equipment environment. The developer imports the reference signals for each type of defect into the Studio and, in just a few steps, they can create a classification ML library that gathers all this knowledge into a single library. When this library is subsequently run on the microcontroller, the classifier analyses the live data and indicates the percentage of similarity against the static reference knowledge. The classification library is an immensely powerful tool, in a smart industrial setting it can not only provide quick fault detection but can also give an indication of
8 May 2022 Components in Electronics
www.cieonline.co.uk
https://www.anglia-live.com/
lnkt.aspx?id=101
https://www.anglia-live.com/supplier/stmicroelectronics/News?utm_source=web&utm_medium=mixed&utm_campaign=gmkt&utm_term=cie_article-01052022-ST_Nanoedgestudio
https://www.anglia-live.com/product/1951331001/STEVAL-STWINKT1B/STMICROELECTRONICS/
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
