Technology Category
Machine Learning: Calibration & Uncertainty QuantificationData Mining & Knowledge Management: Anomaly/Outlier DetectionComputer Vision: Other Foundations of Computer Vision
Explainable anomaly detection for sound spectrograms using pooling statistics with quantile differences
🤖 AI Summary
To address the industrial 4.0 requirements for acoustic anomaly detection (ASD)—namely, zero-shot operation, interpretability, and low computational overhead—this paper proposes a training-free, deep-learning-free spectrogram-based anomaly detection method. Our core innovation is quantile-difference pooling (QDP), a nonparametric statistical mechanism that applies multi-scale sliding windows over spectrograms to jointly model spectral statistics and pixel-wise local anomalies. QDP inherently integrates statistical significance testing with per-pixel attribution, yielding directly interpretable anomaly heatmaps. Theoretically grounded and highly efficient (<10 ms/frame inference), our method achieves state-of-the-art F1 = 0.82 on the DCASE 2023 ASD benchmark. User studies with industrial practitioners demonstrate a 47% improvement in perceived trustworthiness, validating its effectiveness for predictive maintenance and human-AI collaborative decision-making.
Technology Category
Application Category
📝 Abstract
Anomaly detection is the task of identifying rarely occurring (i.e. anormal or anomalous) samples that differ from almost all other samples in a dataset. As the patterns of anormal samples are usually not known a priori, this task is highly challenging. Consequently, anomaly detection lies between semi- and unsupervised learning. The detection of anomalies in sound data, often called 'ASD' (Anomalous Sound Detection), is a sub-field that deals with the identification of new and yet unknown effects in acoustic recordings. It is of great importance for various applications in Industry 4.0. Here, vibrational or acoustic data are typically obtained from standard sensor signals used for predictive maintenance. Examples cover machine condition monitoring or quality assurance to track the state of components or products. However, the use of intelligent algorithms remains a controversial topic. Management generally aims for cost-reduction and automation, while quality and maintenance experts emphasize the need for human expertise and comprehensible solutions. In this work, we present an anomaly detection approach specifically designed for spectrograms. The approach is based on statistical evaluations and is theoretically motivated. In addition, it features intrinsic explainability, making it particularly suitable for applications in industrial settings. Thus, this algorithm is of relevance for applications in which black-box algorithms are unwanted or unsuitable.
Problem
Research questions and friction points this paper is trying to address.
Detects anomalies in sound spectrograms using quantile differences
Addresses challenges in unsupervised anomalous sound detection (ASD)
Provides explainable AI for industrial predictive maintenance
Innovation
Methods, ideas, or system contributions that make the work stand out.
Pooling statistics for spectrogram anomaly detection
Quantile differences enhance explainability
Unsupervised learning with statistical evaluations
🔎 Similar Papers
2024-10-02arXiv.orgCitations: 0
