Variable operating conditions fault diagnosis Based on Meta-learning-based Multimodal Large Model

Deep learning models have achieved remarkable success in fault diagnosis recently. However, related fields of research are mainly focused on single-modal, single-task, and limited-data. To embrace the new AI paradigms of multi-modal, multi-task, massive-data, and super-large model, a novel approach Meta-learning based Multimodal Large Model (MMLM) that integrates the larger multimodal model into fault diagnosis by infusing domain knowledge via meta-learning. By augmenting a robust base model with a low learnable parameters method and calibrating the domain in a cross-dataset meta-learning framework, the method achieves superior performance in few-shot classification, significantly outperforming existing methods. The model demonstrates an impressive ability to rapidly incorporate new information without extensive retraining, ensuring optimal efficiency. It also exhibits remarkable adaptability to various distribution shifts, seamlessly accommodating tasks across a wide array of machines, working conditions, and environments.