OSELM Code Implementation: Online Sequential Extreme Learning Machine

This document introduces the two operational phases of Online Sequential Extreme Learning Machine (OS-ELM): Initialization Phase and Online Learning Phase. During initialization, the algorithm employs Extreme Learning Machine (ELM) methodology to train on limited fault data samples, then discards the training data while preserving the learned parameters within the network architecture. In the online learning phase, real-time parameter updates for matrices H (hidden layer output) and β (output weights) continuously reinforce the network's classification capability and generalization performance, thereby improving fault diagnosis accuracy. The trained OS-ELM parameters stored in network nodes support cross-platform migration - the weight parameter β can be deployed to new computational environments including PCs, DSP processors, and ARM-based embedded systems. For new fault test data, the system only requires regeneration of the corresponding hidden layer output matrix H through forward propagation, enabling seamless fault pattern re-learning and classification functionality. This approach provides robust adaptation to diverse fault test data scenarios while enhancing both efficiency and accuracy in fault diagnosis systems. Key implementation aspects include: - Matrix-based computation for efficient parameter updates - Sequential learning algorithm eliminating retraining on historical data - Modular parameter storage enabling hardware-independent deployment - Real-time hidden layer output regeneration for new test instances