This paper presents a novel multi-output classification (MOC) framework for domain adaptation on partially labeled target datasets to address the challenges of complex fault diagnosis under diverse operating conditions, such as rotational speed and torque variations. Unlike conventional multi-class classification (MCC), the MOC framework simultaneously classifies the severity levels of complex faults. Various single-task and multi-task architectures, including shared trunk and crosstalk-based designs, are applied to the MOC formulation to perform complex fault diagnosis under partially labeled conditions. Specifically, we propose a novel crosstalk architecture, the Residual Neural Dimensionality Reducer (RNDR), which enables selective information sharing, to improve classification performance under complex fault scenarios. We incorporate frequency hierarchical normalization to enhance domain adaptation performance for motor vibration data. We evaluate the implemented complex fault conditions under six domain adaptation scenarios using a motor-based test setup. Experimental results demonstrate superior macro F1 performance compared to baseline models, demonstrating that the structural advantages of RNDR are more pronounced in complex fault settings than in single-fault settings. We also verified that frequency hierarchical normalization is more suitable for fault diagnosis tasks than existing methods, and analyzed RNDR and other models with increased number of parameters under various conditions, and compared them with the pruned RNDR structure.
