<p> </p>
<p>Many machine learning models exist for battery management systems to utilize. Few have been shown to work. This work focuses on gathering data from cycling battery packs and sending this data directly to machine learning models built off robust datasets for applying the resulting predicted values and outputs directly on top of real-time systems. A parasitic sensor network was created composed of a main microcontroller, a host CPU, and various sensors including resistance temperature detection devices (RTDs), a voltage measurement circuit, current measurement circuit, and an accelerometer/gyroscope. The resulting network was integrated parasitically with a 4-cell 18650 SONY VTC6 battery pack, then tested both on-ground and in-flight with a commercial quadcopter. Real-time data for the battery pack with four cells in series was gathered. This real-time data stream was then integrated with data-driven neural network algorithms trained on various 18650 datasets and a real physical model to finalize the “AI BMS”. Using the power of non-linear models to infer battery health impacts not normally considered in battery management systems, the “AI BMS” was able to use low-fidelity real-time data in conjunction with a powerful multi-faceted model to make predictive decisions about battery health characteristics on top of normal system operations.</p>
| Identifer | oai:union.ndltd.org:purdue.edu/oai:figshare.com:article/23535369 |
| Date | 16 June 2023 |
| Creators | Alexey Y Serov (16385037) |
| Source Sets | Purdue University |
| Detected Language | English |
| Type | Text, Thesis |
| Rights | CC BY 4.0 |
| Relation | https://figshare.com/articles/thesis/_strong_DEVELOPMENT_OF_A_BATTERY_MONITORING_SYSTEM_FOR_DATA-DRIVEN_AI_DETECTION_OF_ACCELERATED_LITHIUM-ION_DEGRADATION_strong_Untitled_Item/23535369 |
Page generated in 0.0022 seconds