YOTOnet: Zero-Shot Cross-Domain Fault Diagnosis via Domain-Conditioned Mixture of Experts

2026-05-06 · Source: Takara TLDR - Daily AI Papers · Field: Manufacturing & Industrial — Smart Manufacturing & Industry 4.0, Automation & Robotics, Artificial Intelligence & Machine Learning · Depth: Expert, medium

Summary

YOTOnet (You Only Train Once) is a novel architecture designed for zero-shot cross-domain fault diagnosis in mechanical equipment, addressing the generalization limitations of deep learning models across varying equipment and operating conditions. Proposed by Zesen Wang et al. on May 6, 2026, YOTOnet integrates three core components: a physics-aware Invariant Feature Distiller using multi-scale dilated convolutions and FFT-based time-frequency fusion for domain-agnostic representations; Domain-Conditioned Sparse Experts (DC-MoE) for adaptive input routing without external metadata; and a dual-head classification system with auxiliary supervision. Extensive validation across 30 cross-dataset protocols on five public bearing datasets (CWRU, MFPT, XJTU, OTTAWA, HUST) demonstrates YOTOnet's superior performance. The model exhibits a clear scaling effect, with average test F1 improving from 0.5339 (1 training dataset) to 0.705 (4 datasets), indicating that foundation model principles can enable robust, train-once deployment for industrial fault diagnosis.

Key takeaway

For AI Scientists and Machine Learning Engineers developing industrial fault diagnosis systems, YOTOnet offers a robust approach to overcome domain shift challenges. Its "train once" capability and demonstrated scaling effect across multiple datasets suggest a path to more generalized and efficient deployment. You should consider integrating its core architectural principles, such as physics-aware feature distillation and domain-conditioned sparse experts, to enhance model adaptability and reduce retraining efforts in diverse operational environments.

Key insights

YOTOnet enables zero-shot cross-domain fault diagnosis in mechanical equipment using a domain-conditioned mixture of experts.

Principles

• Domain shift limits generalization in fault diagnosis.
• Foundation model principles enable robust, train-once deployment.
• Scaling training data improves fault diagnosis F1 scores.

Method

YOTOnet extracts domain-agnostic features via a physics-aware distiller, routes inputs through Domain-Conditioned Sparse Experts, and uses a dual-head classification system with auxiliary supervision.

In practice

• Apply YOTOnet for zero-shot fault diagnosis.
• Utilize multi-scale dilated convolutions for feature extraction.
• Integrate FFT-based time-frequency fusion.

Topics

• YOTOnet
• Zero-Shot Fault Diagnosis
• Domain-Conditioned MoE
• Invariant Feature Distillation
• Mechanical Fault Diagnosis

Code references

• giangdip2410/Domain-specific-Experts

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Domain-Aware Hierarchical Contrastive Learning for Semi-Supervised Generalization Fault Diagnosis — DAHCL improves semi-supervised fault diagnosis by correcting pseudo-label bias and leveraging uncertain unlabeled data through domain-aware and hierarchical contrastive learning.
• Rethinking Generalization in Reasoning SFT: A Conditional Analysis on Optimization, Data, and Model Capability — Reasoning SFT can generalize cross-domain, but it is conditional on optimization, data quality, and model capability.
• Leveraging systems' non-linearity to tackle the scarcity of data in the design of Intelligent Fault Diagnosis Systems — Leveraging system non-linearity and novel visualization creates data for DTL-based Intelligent Fault Diagnosis Systems despite scarcity.
• Algebraic Dead Directions in LayerNorm Transformers: A Forward-Pass-Only Diagnostic at LLM Scale — The LayerNorm inverse-scale parameter directly reveals a transformer's algebraic dead direction without complex computation.
• Frequency-Enhanced Dual-Subspace Networks for Few-Shot Fine-Grained Image Classification — FEDSNet improves few-shot fine-grained image classification by integrating frequency domain features to enhance structural stability and reduce overfitting.
• ROBUST-WT: Robust Uncertainty-aware Segmentation Transform via Whitening and Training Enhancements — ROBUST-WT demonstrates that targeted training enhancements can substantially improve medical image segmentation robustness without altering core architecture.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.