Battery Management System

• https://www.coursera.org/specializations/algorithms-for-battery-management-systems

Challenges -

• Different battery sizes requires different terminal current
• Cross charging
• Balancing
• Peak power handling

Big Ideas -

• VSYS voltage monitor comparetor should be inside PMIC or inside the GreenPak
• Data based DoU characterization method - benefits
  • Direct comparable among different products/models
  • Standardized characterization procedure
  • Indisputable marketing DoU claim
  • Easy translation to what-if DoU
  • User oriented (vs engineering oriented) DoU communications

• Battery swells (some reversable, some permanent). It is found to be very beneficial to reduce aging by applying 5 psi pressure to physically limiting the swelling. Too much pressure can also be problematic. This information was provided by professor Anna G. Stefanopoulou from U. Michigan.
• P2D physical model
• PCA correlation analysis
• Parallel R*C benefits
• SOC physical model not temperature, c-rate dependent
• Symbolic regression.
• DFN (doyle fuller newman) model
• Single particle model (simple model has limitations)
• POD - reduced order model ROM
• ESPM
• Adaptive Ensemble Sparse Identification (AESI)
• SPCI method
• SYNDy approach
• SPM → ESPM
• acc-plett-advanced_battery_management-perspectives_on_the_role_of_machine_learning.zip

• PCA - Principal Component Analysis (PCA) and neural networks are powerful tools in data analysis, and they can be used together in various ways. PCA is often used as a preprocessing step for neural networks, particularly with high-dimensional data, to reduce dimensionality and improve model performance. Additionally, neural networks can be designed to perform PCA-like operations, effectively learning the principal components during training
• Sys ID - surrogate AI model types and how to choose - simple feedforward, long-short (LSTM) memory model. Delayed input…
• Model Conversion, Training methods - PyTorch, TensorFlow, MATLA
• RNN vs LSTM vs GRU vs Transformers - https://www.geeksforgeeks.org/deep-learning/rnn-vs-lstm-vs-gru-vs-transformers/
• Virtual sensor - How to identify if the import has correlation to output to reduce the order of the system - perturbation sensitivity analysis.
• Imitation learning
• NLP/QP solving for MPC is expensive
• Neural state space model
• Reinforcement learning (RL)
• Pruning and projection (structure compression), quantization (data compression) to deploy for low memory low computing power applications.