Description

This Master Prompt establishes a framework for integrating signal processing with machine learning, deep learning and physics-informed modelling. It addresses regression, classification, clustering, anomaly detection and forecasting through conventional algorithms, convolutional networks, recurrent architectures, attention mechanisms and representation learning. Drawing on the supplied studies of ultrasound contrastive learning, physiological signal cleansing, adaptive model selection, tool-wear transfer prediction, variable-speed machinery diagnosis and physics-guided manufacturing data augmentation, it examines the relationships among signal representation, model design, training objectives and evaluation. Particular attention is given to independent data partitioning, limited labels, class imbalance, domain shift, hyperparameter selection, uncertainty and interpretability. The framework distinguishes predictive performance from physical explanation, model complexity from substantive contribution, and retrospective testing from demonstrated deployment. It supports reproducible, evidence-based development of models whose conclusions remain appropriate to the data and validation conditions.