Unsupervised Anomaly Detection in Process-Complex Industrial Time Series: A Real-World Case Study

Sergej Krasnikov; Lukas Meitz; Samineh Bagheri; Michael Heider; Thorsten Schöler; Jörg Hähner

cs.LGApr 15, 2026

Unsupervised Anomaly Detection in Process-Complex Industrial Time Series: A Real-World Case Study

Authors: Sergej Krasnikov, Lukas Meitz, Samineh Bagheri, Michael Heider, Thorsten Schöler, Jörg Hähner

Organizations: Universität Augsburg Augsburg, Germany · Technische Hochschule Augsburg Augsburg, Germany · Independent Researcher Karlsruhe, Germany

Abstract

Industrial time-series data from real production environments exhibits substantially higher complexity than commonly used benchmark datasets, primarily due to heterogeneous, multi-stage operational processes. As a result, anomaly detection methods validated under simplified conditions often fail to generalize to industrial settings. This work presents an empirical study on a unique dataset collected from fully operational industrial machinery, explicitly capturing pronounced process-induced variability. We evaluate which model classes are capable of capturing this complexity, starting with a classical Isolation Forest baseline and extending to multiple autoencoder architectures. Experimental results show that Isolation Forest is insufficient for modeling the non-periodic, multi-scale dynamics present in the data, whereas autoencoders consistently perform better. Among them, temporal convolutional autoencoders achieve the most robust performance, while recurrent and variational variants require more careful tuning.

Figures & tables

Fig. 1 : Plot of a process sample from the dataset, showing readings from four selected sensors captured during a single run executed on one machine. Each sensor records different physical quantities over time, exhibiting distinct behaviors including stable operational phases, gradual drifts, and abrupt changes indicative of process transitions or anomalies. The view of the data has been deliberately perturbed to protect sensitive operational information, while preserving the underlying structure and complexity.

Model	Encoder Configuration	Decoder Configuration
TCN-AE	Conv1D + TCN blocks + Avg Pooling	Inverse TCN blocks + Conv1D
TCN-VAE	Conv1D + TCN blocks + Avg Pooling + Normal ( $\mu$ , $\sigma$ )	Inverse TCN blocks + Conv1D
LSTM-AE	2 LSTM layers	2 LSTM layers
LSTM-VAE	2 LSTM layers + Normal ( $\mu$ , $\sigma$ )	2 LSTM layers
GRU-AE	2 GRU layers	2 GRU layers
GRU-VAE	2 GRU layers + Normal ( $\mu$ , $\sigma$ )	2 GRU layers

TABLE I : Architectural configurations of the AE models used in the study. VAE models include an encoded normal distribution signified by mean (

\mu

) and standard deviation (

\sigma

Hyperparameter Description	Values
Gradient Descent Learning Rate	(0.001, 0.0001)
Batch Size	{64, 128, 256, 512}
Input Sequence Length	{480, 600, 720, 840, 960, 1080}
Hidden Layer Size	{32, 64}
Number of encoder/decoder Layers	(1, 3)
Latent Dimension (VAE)	{16, 32, 64}

TABLE II : Hyperparameter search space for model training. Values represent ranges (min, max) for continuous parameters and discrete sets {…} for categorical parameters. Architecture-specific parameters are shown for VAE and TCN models.

	Anomaly Detection	Reconstruction
Model	F1 Score	MSE / ELBO ( $\times 10^{-4}$ )
IF	0.120 $\pm$ 0.126	-
TCN-AE	0.991 $\pm$ 0.009	0.22 $\pm$ 0.06
LSTM-AE	0.853 $\pm$ 0.102	1.23 $\pm$ 0.25
GRU-AE	0.918 $\pm$ 0.066	0.84 $\pm$ 0.15
TCN-VAE	0.968 $\pm$ 0.010	-7.29 $\pm$ 0.47

TABLE III : Average performance metrics for the evaluated models in the presented industrial case study. Reported values are means

\pm

standard deviation (MSE for standard AE, ELBO for VAE).

Fig. 2 : Differences in anomaly detection performance across models with TCN variants achieving the most desirable results. Shown are the F1-scores of the top 5 models for each architecture. Note: IF is excluded from this visualization due to its significantly lower performance (cf. Table III ), which would distort the scale and clarity of the plot.

Explore similar work

Detecting the Undetectable: Enhancing Unsupervised time series Anomaly Detection via Active Learning
Jul 1, 2026Seung Hun Han, Hyeongwon Kang, Jinwoo Park +1Industrial Anomaly Detection Active Learning
LG CNS, 71 Magokjungang 8-ro, Gangseo-gu, Seoul, Republic of Korea · Department of Industrial & Management Engineering, Korea University, 126-16 Anam-dong 5-ga, Seongbuk-gu, Seoul, Republic of Korea · Department of Industrial Engineering, Seoul National University, Gwanak-ro 1, Gwanak-gu, Seoul, Republic of Korea
ChronosAD: Leveraging Time Series Foundation Models for Accurate Anomaly Detection
May 31, 2026Uzair Khan, Luigi Capogrosso, Francesco Biondani +4Industrial Anomaly Detection Time-Series Anomaly Detection
University of Verona · Interdisciplinary Transformation University of Austria · ETH Zurich
Optimal Transport for Efficient, Unsupervised Anomaly Detection on Industrial Data
Sep 12, 2026Abigail Langbridge, Fearghal O'Donncha, James T Rayfield +1Anomaly Detection Industrial Anomaly Detection
Dyson School of Design Engineering Imperial College London* United Kingdom · AI for Sustainable Industry IBM Research Europe Dublin, Ireland · IBM T.J. Watson Research Center IBM Research NY, USA

Jul 1, 2026J/K move · Enter open · S save

Detecting the Undetectable: Enhancing Unsupervised time series Anomaly Detection via Active Learning

Seung Hun Han, Hyeongwon Kang, Jinwoo Park, Pilsung Kang

LG CNS, 71 Magokjungang 8-ro, Gangseo-gu, Seoul, Republic of Korea · Department of Industrial & Management Engineering, Korea University, 126-16 Anam-dong 5-ga, Seongbuk-gu, Seoul, Republic of Korea · Department of Industrial Engineering, Seoul National University, Gwanak-ro 1, Gwanak-gu, Seoul, Republic of Korea

Industrial Anomaly Detection Active Learning Reconstruction-Based Anomaly Detection Unsupervised Anomaly Detection Time-Series Anomaly Detection

Open paper AskPDF arXiv

Despite the increasing sophistication of industrial AI systems, the ability to reliably detect subtle and noisy anomalies in complex time series data remains a critical yet unresolved challenge. In large-scale industrial applications, labeling time series data is often prohibitively expensive and time-consuming, making unsupervised learning a practical and widely adopted approach. However, existing unsupervised methods frequently struggle to distinguish near-normal anomalies from normal patterns and are vulnerable to noise contamination within normal samples. To address these limitations, we propose a novel framework that leverages active learning to iteratively enhance the performance of unsupervised models. Our framework's core contributions are (1) a masked time-series reconstruction feedback strategy that forces the model to learn robust temporal dependencies, and (2) a minimax learning strategy that promotes robustness by differentially treating normal and abnormal samples. This process encourages the model to better capture the dynamics of subtle and noisy patterns. The proposed framework is evaluated across 28 test cases involving four multivariate time-series datasets and seven unsupervised backbone models. Experimental results demonstrate a 12.39% improvement in AUC compared to the original models, confirming that our method can be readily integrated into existing unsupervised reconstruction-based anomaly detection systems to significantly enhance their performance.

Abstract

Explore similar work

Detecting the Undetectable: Enhancing Unsupervised time series Anomaly Detection via Active Learning

ChronosAD: Leveraging Time Series Foundation Models for Accurate Anomaly Detection

Optimal Transport for Efficient, Unsupervised Anomaly Detection on Industrial Data