batubaco

Inspiration

In the current AI era, data centers have become a fundamental necessity in the technology sector. However, a new challenge has emerged regarding their construction: the energy required to power them. While numerous renewable energy alternatives—such as hydroelectric, solar, and wind power—exist and have been extensively researched and tested, they all require a means to store the energy they generate.

Initial issues—such as a lack of wind, the onset of night, or insufficient water flow—have already been addressed through energy storage solutions. But what happens when the problem lies within the storage system itself? As a researcher, I encountered this firsthand: procuring new storage is costly, and existing units often suffer from wear and tear, making their remaining lifespan uncertain. My concern stemmed from an experience with my home lab server; it initially ran smoothly on solar power, but eventually failed because I hadn't accurately tracked the storage system's age, performance, or durability.

This became my focus: how could I enable my battery storage system to "communicate" its capabilities and health status to me, eliminating the need for traditional manual checks and calculations? Furthermore, the widespread availability of Large Language Models (LLMs) and forecasting algorithms offered opportunities to enhance the capabilities of my existing battery storage system.

What it does

The name "batubaco" is formed from two words originating from two different Indonesian languages/cultures. In Javanese, the local term for "battery" is "batu", while in Minangkabau, "baco" means "read." Consequently, this system reads battery parameters—specifically voltage and current—adhering to the following standards:

• Readings must be precise, accurate, and rapid.
• Readings must be performed in real-time, 24/7.
• Data transmission must occur continuously in real-time (24/7) without interruption, utilizing a strict protocol switch-over scheme.
• Forecasting will be employed for advanced mathematical calculations to predict battery lifespan, capacity, and the timing of performance degradation.
• Forecasting results will be passed to an LLM to automate agent-based decision-making.

How we built it

We built Batubaco using a dual-microcontroller setup—featuring the STM32H7 and ESP32—installed on the battery with the following configuration:

• The STM32H7 handles high-precision monitoring of battery current and voltage.
• The ESP32 manages signal transmission to the server via Wi-Fi or eth, ensuring data integrity through redundant communication protocols.

Batubaco relies on high-level computing performed on a server capable of processing data using LLMs and forecasting models, following this workflow:

• After the STM32H7 reads the voltage and current data, the ESP32 transmits it via WebSocket (WS) for faster communication.
• The server receives the data in real-time, and the raw data is processed by the forecasting module.
• The forecasting output is fed into an LLM, which makes further decisions—crucial since the system monitors multiple batteries.
• Once the LLM determines the outcome based on the forecast, an LLM-driven agent issues the appropriate alerts or recommendations.
• This entire process must be completed before the battery fails or degrades.

Challenges we ran into

The challenges we faced at this stage included ensuring stable readings—free from electrical noise interference—and managing heat. We needed to find the right sealing solution to prevent signal interference without compromising the MCU's reading capabilities.

Furthermore, regarding data transmission to the server, we had to devise a switchover or failover communication scheme to handle potential failures in either wireless or wired connections, as maintaining high-quality time-series data was essential.

Accomplishments that we're proud of

With this system in place, we can predict battery health at an early stage. This allows us to plan for battery replacements more efficiently and optimize our budget, given that the investment in batteries is substantial and critical.

What we learned

Given the high current and voltage levels from the battery, we recognize the critical importance of noise management, as this affects the data and power entering the MCU. Since different materials influence this factor, we must be prudent in our material selection to minimize noise.

What's next for batubaco

Batubaco will continue to refine its algorithms to make them superior and more efficient, enabling them to run directly on edge devices. This will result in faster, lighter, and more efficient data transmission to the server. Furthermore, the system will only require an API call to send notifications, as all processing will take place locally on the Batubaco device itself.

Built With

• c
• c++
• esp32
• json
• llm
• machine-learning
• postgresql
• stm32
• tensorflow
• ubuntu