RECOM and Artificial Intelligence in Power Electronics

Feb 23, 2024
Fiber Optic - High Speed Data Transfer
At first, it might seem that Artificial Intelligence (A.I.) and power supply design have very little in common. However, RECOM is already actively using A.I. in at least three separate areas to improve our products and services. Read on to see how artificial intelligence, power, and RECOM are becoming synonymous.

Table of contents

Intelligent datasheet comparison

RECOM manufactures some 33 000 standard products. The reason why there are so many products in our portfolio is the number of permutations and combinations needed to meet the needs of our customers.

Consider a simple 5W DC/DC converter – RECOM’s REC5 series. Six different input voltages are offered as standard and nine different output voltages. That is 54 different input/output combinations straight off. Additionally, the series is available with four different isolation voltage ratings and six different pinning options. Include the options of a metal instead of a plastic case, remote control on/off function, and plastic tube or tape and reel packaging, then the total number of standard product options quickly exceeds three thousand – and that is just one converter series!

All these permutations are necessary to meet the various requirements of our customers – some may need SMD pinning so that a single production process can be used for all components on the board, and another might demand through-hole pinning for extra robustness against shock and vibration and to be pin-compatible with a second source from a different manufacturer. So, all these different combinations are needed and necessary.
Parametric search function on RECOM's website
Fig. 1: Example of a parametric search function on a manufacturer’s website (RECOM)
The problem is the complexity of such a wide choice. If a customer has an existing DC/DC converter solution and is seeking to replace it with a RECOM product, it can be daunting to search through all the options available; a parametric search on RECOM’s website for a 1W converter with 5V input and 5V output returns 5 pages of results (Figure 1). A competitor shows 19 matches from their website search engine, another shows 122 results. Are they equivalent? What is the difference between them?

This is where artificial intelligence can help. RECOM is actively working with a local university to help develop an A.I. system that can extract data from different manufacturers’ PDF datasheets so that they can be compared. This is not a trivial task. Each manufacturer has a different style, text font, and datasheet layout.

Additionally, the numerical data may be presented in different ways (for example, 1A or 1000mA), measured in different ways (for example, load regulation over 10-100% or 25%-100% load), or with incomplete information (for example, 2kV isolation instead of 2kVDC for 60 seconds isolation). The software must not only search through and recognize similar information that is presented in different parts of the various datasheets but also understand the context in which it is given, to be able to judge if the data is genuinely equivalent.

Such a task is today the job of a skilled and experienced component engineer. Artificial intelligence will not replace that skill but will greatly speed up the process – ideally replacing the long slog of trawling through hundreds of datasheets to make a shortlist of a few compatible candidates.

Optimized distributed power architecture for an automated guided vehicle system

Illustration of AGV from the outside and its power supply from the inside
Fig. 2: Typical AGV subsystems (source: Hirose Electric)
Many circuit designs need multiple board-level power supply voltages. Take, for example, an autonomous or automated guided vehicle (AGV) that can drive itself along the pavements of a city to deliver packages or food parcels. The AGV will need various functional blocks – sensor inputs such as video cameras, object detection via ultrasonic, radar, or lidar rangefinders, a GPS orientation system, a microcontroller, and several motor drivers – all powered from a single rechargeable battery, typically a 48V Lithium-Ion battery pack (Figure 2).

All these functional subsystems need power, which will be supplied by the 48V battery pack. The battery voltage will vary according to load and state of charge, from 24V when depleted up to 54V when freshly charged, with a transient input range of 20V-60V1 . The regulated board-level voltages could be +15V for the motor controllers, +12V for the camera and ultrasonic sensors, +300V for the LiDAR system, 3.4V for the GPS subsystem, and +5V, +3.3V, 1.8V, and 1.2V for the microcontroller, interface, and memory ICs. The wheel motor drives would be powered directly from the battery. Obviously, these board-level power supplies will be supplied by various voltage regulators and DC/DC converters, but which is the optimum arrangement? Individually powered from the 48V battery, or via a high current DC/DC converter to reduce and regulate the supply with LDO regulators for each system, or a combination thereof? (Figure 3).
Graphics of two power supply solutions for the AGV board
Fig. 3: Two possible distributed power solutions for the AGV board: centralized and distributed
Both power supply power tree solutions shown in Figure 3 deliver the same end results. The top distributed solution uses more DC/DC converter modules which is highly efficient but not cheap. The bottom centralized solution uses more cost-effective low-dropout (LDO) regulators than the upper solution but has higher conversion losses. Is maximum overall efficiency the most important factor to give the longest possible battery life? Or is cost the most important criterion? Or is there another power tree arrangement that would give a more optimal solution delivering good efficiency at a reasonable cost? There are CAD software programs available that can help experienced designers select the right sockets in the power supply tree based on given technical specifications, but artificial intelligence can improve this process to allow less experienced power engineers to find the most optimal solution with the best performance at the lowest cost.

RECOM is partnering with an A.I. software company that is developing such an automated design platform to offer targeted solutions. The A.I. algorithm will intuitively capture the functional requirements of the application, automatically generate the project design, schematics, and documentation, and select the optimum components to streamline the design to best meet the project parameters.

A.I. in marketing

Video still from an A.I. generated avatar with audio and video commentary
Fig. 4: Video still from an A.I. generated synthetic avatar with audio and video commentary. Production time was a few hours
Most will have heard of ChatGPT, a ‘large language model’ that uses machine learning to give intelligent replies to written questions. When released at the end of 2022, it became the fastest-growing software application ever – gaining over 100 million users within a few months. Building on the success of the original free-to-use version, the latest pay-for-use versions have expanded the A.I.’s capability to generate images as well as text based on a user’s prompts. Other A.I. platforms can now create avatar videos with spoken commentaries based on text and image inputs.

RECOM has started to use these A.I. tools to create simple blog posts and short videos to promote new product releases (Figure 4). RECOM is a highly innovative company, releasing on average a new product every week. We need to communicate these new developments to our large customer base (>100k customers) quickly and simply without bombarding them with weekly texts and press releases, so a short blog, podcast, or YouTube video is an ideal way to keep our customers informed without being overbearing.

However, traditional video shooting, audio recording, or image animation is a very expensive and time-consuming process, so we have had to be very selective in choosing just the main products that were worth the additional investment in time and resources.

Automated text, audio, and video generation using A.I has allowed us to generate informative video bites, podcasts, and blogs that are available on demand, freeing up the time and in-boxes of our customers, but still offering up-to-date information on our ever-expanding product portfolio. It also means that we can quickly create new videos in different languages to support local markets and rapidly update our customer communications with any last-minute corrections or new features. Needless to say, RECOM experts always proofread the material before publication.

As the capabilities of A.I.-generated videos, texts, and audio tracks continue to advance, there is a growing opportunity to enhance our creativity in marketing efforts. For instance, we can provide more comprehensive information in the local language of our customers or implement layered product marketing strategies. This approach includes a succinct introduction to the main benefits for a broader audience and a detailed exploration of technological advances and features tailored to more sophisticated users. Being a technology company focused on delivering solutions, not just products, artificial intelligence software plays a crucial role in shaping our future. While it cannot replace the profound knowledge, skills, and experience of our engineers and sales support teams, it serves as a valuable tool to augment our services for customers. By leveraging A.I., we aim to elevate the overall quality of our offerings, ensuring a more personalized and efficient experience for our clients.

Note: the last paragraph was written by A.I.

1Specification according to LV148 automotive standard
Applications