Issue link: https://resources.mouser.com/i/1442826
AI high-end applications. However, this decade has seen three trends which now offer the opportunity to add CM capabilities to products in more general industrial, and even consumer spaces: 1. MEMS sensor technology has enabled manufacturers to drive the cost of 3-axis accelerometers and other sensors to under one dollar. At the same time, sensitivity and sampling rates have increased, and noise levels decreased. 2. Advanced CMOS technologies have had a similar effect on the microcontroller business. Today's MCUs are cheaper and more potent than ever. 3. Machine learning techniques have gone mainstream. At the same time, many have gone open source. I.MX RT1060 • High performing Arm Cortex-M7 • 3020 Core-Mark/1284 DMIPS @ 600MHz • 1MB On-Chip SRAM with up to 512KB configurable as Tightly Coupled Memory (TCM) LEARN MOREu ANOMALY DETECTION These trends make it possible to add CM capabilities cost- effectively to a wide variety of products such as white goods, home monitoring systems, or almost any product containing a motor. Moreover, if those products get connected to the worldwide web, it is possible to enable whole new revenue streams. Instead of flashing a service light at a customer, you can now send a text to the consumer saying that their washing machine would benefit from service, and "just click here" to schedule a technician visit! Inclusion of machine learning techniques means that you no longer need to have expert technicians scrutinizing your data. ML techniques can be used to learn the nominal sensor signatures expected from your product, and then raise flags when operation strays from expected values. To enable these functions, you'll need the proper components to make that happen: • Required sensors with the appropriate software drivers • A means to collect data collected from sensors for offline analysis • Software libraries for running computed models. You may also want the ability to do some retraining right in the product. Running computed models is often a task that may get performed on relatively lower cost/power MCUs. One may need additional horsepower to train in sitú. The process of integrating these components is still non-trivial, which can be a roadblock to implementation. This is especially true because you probably want to do some essential data collection and feasibility study before committing to the approach. Semiconductor and startup companies are taking notice and are beginning to offer solutions that address these problems. Pre-designed hardware modules that can get used for early data collection and product prototyping, pre- integrated drivers and machine learning libraries, high-speed data collection and cloud connectivity all make this process much less intrusive and speed time to market. One such system is the NXP Classic Machine Learning Toolbox. ■ 16