A thorough analysis of unsupervised learning techniques used in the health monitoring of industrial robots explores significant trends and key algorithms, such as clustering, autoencoders, and generative models, assessing their effectiveness in identifying faults and performance degradation.
Abstract
As industrial robots become increasingly essential to modern manufacturing and automation systems, ensuring their durability and operational integrity has emerged as a key concern. Traditional defect detection methods typically depend on labeled datasets and supervised learning techniques, which can be difficult and impractical to implement in real-world industries. In contrast, unsupervised learning presents a compelling alternative by facilitating anomaly detection and fault diagnosis without the need for labeled data. This article offers a thorough analysis of unsupervised learning techniques used in the health monitoring of industrial robots. We explore significant trends and key algorithms, such as clustering, autoencoders, and generative models, assessing their effectiveness in identifying faults and performance degradation. The research addresses the unique challenges associated with high-dimensional sensor data, variable operating conditions, and the lack of ground truth labels. Additionally, we highlight unresolved research questions and potential future directions, emphasizing the need for scalable, interpretable, and real-time solutions. This survey serves as a foundational reference for researchers and practitioners aiming to develop resilient and autonomous health monitoring systems for industrial robots.
Industrial robotic systems are essential to modern manufacturing, but their reliability is threatened by progressive mechanical and electrical degradation. Traditional reactive and preventive maintenance strategies are inadequate for the complex, high-dimensional sensor environments of contemporary industrial robots. This study developed and evaluated AE-LSTM-ATT (Attention-Enhanced Hybrid LSTM Autoencoder), an LSTM encoder-decoder architecture with Bahdanau-style additive attention, for unsupervised anomaly detection in industrial robotic systems. The model was evaluated on the public Industrial Robot Anomaly Detection (IndRAD) dataset (robot joint positions, velocities, torques, and motor currents) against an attention-free LSTM Autoencoder and an Isolation Forest baseline, across five independent training runs and four synthetically injected anomaly types (spike, step, freeze-to-zero, and freeze-to-last-value). AE-LSTM-ATT and the attention-free baseline performed comparably on abrupt anomalies (spike, step; AUC-ROC (Area Under the Receiver Operating Characteristic Curve) = 1.000 for both). For the subtler, gradually manifesting anomaly types, attention provided a small, directionally positive advantage: for freeze-to-last-value, AE-LSTM-ATT achieved AUC-ROC = 0.518 ± 0.059 and F1-score = 0.682 ± 0.020 (95% CI, 5 seeds), versus 0.491 ± 0.046 and 0.670 ± 0.017 for the baseline; freeze-to-zero showed overlapping confidence intervals. An independent replication of this comparison with a paired significance test across five matched seeds (section "Ablation: Contribution of the Attention Mechanism") did not find the difference to be statistically significant, and the direction of the per-seed difference was not uniform. These results indicate that any attention-related benefit in this architecture is, at most, small and anomaly-type dependent, and is not established as statistically significant at the sample sizes evaluated here; attention showed no distinguishable advantage for abrupt, high-amplitude faults already near-perfectly separable by reconstruction error alone. The framework operates in an unsupervised paradigm requiring no labelled fault data, supporting deployment in settings where run-to-failure data are scarce. Overall, this study finds no evidence that attention-enhanced temporal autoencoders offer a universal advantage over simpler deep learning baselines for industrial robotic predictive maintenance; any benefit appears concentrated in detecting subtle, temporally extended fault signatures, is small in magnitude, and was not confirmed as statistically significant in the present sample.
The rise of low-power, affordable sensing technologies and machine learning algorithms has sparked a growing interest in using data-driven approaches to monitor industrial assets. In particular, applying machine learning to analyze the sounds produced by manufacturing tools is becoming an effective method for quickly detecting deviations from standard operating conditions. However, the implementation of these systems encounters a major challenge due to the difficulty in obtaining suitable training data. Self-supervised learning offers a promising solution to this issue. It enables the training of anomaly detection models exclusively with signals representing normal conditions, which are more accessible than anomalous signals. Despite its potential, achieving robust and consistent performance across machines of varying models and types remains critical. Existing methods generally follow one of two approaches. The first, single-machine training, involves training a dedicated model using local data from each machine. Although straightforward, this approach often yields suboptimal performance. The second, multi-machine training, aims to enhance detection capability by aggregating data from multiple machines to train a shared model. This strategy requires transmitting data from geographically dispersed manufacturing sites—often belonging to different clients—to centralized facilities, raising concerns about data transmission costs and risks of exposing sensitive production information. We address these issues by introducing a novel self-supervised strategy for effective single-machine training based on classifying the distance between the monitored machine (i.e., the sound source) and each microphone sensor of a multi-channel recording system. Our method bypasses the need for data aggregation, providing a cost-efficient and privacy-preserving alternative while delivering competitive detection performance. Experiments on the MIMII dataset highlight the effectiveness of our approach, with performance improvements up to 14.53% over conventional single-machine training strategies and 18.58% over multi-machine training strategies.
Erich Malan, Valentino Peluso, A. Calimera et al.· IEEE Access· 0 citations
Fault detection and diagnosis is a crucial task for modern mobile robots, as it permits their correct functioning with a positive impact on availability, autonomy and safety. Despite several approaches of fault detection and diagnosis of sensor faults in mobile robots, there exists a lack of scientific studies addressing the low-cost LiDAR faults which are commonly adopted in indoor robots. This paper is a preliminary attempt to fill this research gap by proposing a data-driven approach for developing a diagnostic module for mobile robots, focusing on the specific case study of the Stretch robot.Several LiDAR faults are first modeled, then simulated in ROS and Gazebo environments to generate a high-quality dataset. The dataset is used for training a machine learning model, which can diagnose such faults as well as distinguish them from other typical faults, such as IMU pose drift.Our proposed model shows high detection and isolation accuracies across the injected fault scenarios, thus paving the way for the deployment on a real robotic system for further evaluation and analysis.
Abdalmotalib Alghoul, A. Freddi, A. Monteriù· International Conference on...· 0 citations
This work proposes a self-supervised masking scheme that simulates common sensor failures and explicitly trains the model to recover the original signal, and demonstrates that the resulting representations significantly improve the robustness of predictions to seen and unseen sensor failures on a vehicle dynamics dataset.
Jens U. Brandt, N. Pütz, M. Greiff et al.· Neural Information Processin...· 1 citation
Smart manufacturing processes are often installed with a large number of sensors, imaging devices and computers, which not only enable instant communication across various modules of a production system but also aid in intelligent manufacturing management. In this paper, we introduce MODERN, a deep learning framework for quality monitoring and fault isolation, which integrates these enhanced capabilities into the practice of industrial quality control. Using the architecture of an inception residual neural network, we develop a control chart that monitors the likelihood of a product containing defects. We also propose a faulty region estimator that identifies the defective area using transfer learning. To extend our framework to cases where there are not sufficient training data, we suggest a transfer monitoring technique that requires only a small sample size and a hypothesis testing approach for quantitatively assessing the applicability of our method. Theoretically, we establish the minimax optimal convergence rate for both our defect likelihood estimation and fault diagnosis. Our results lead to a seemingly counter-intuitive managerial implication - it may not always be in a manufacturer's best interests to keep upgrading its monitoring equipment regardless of the cost. Empirically, we demonstrate the superior performance of our method in comparison with a state-of-the-art approach using both simulated experiments and real data.
Yicheng Kang, Yuling Jiao, Xin Geng et al.· 0 citations
Rotating machinery is essential to energy, aerospace, manufacturing, and other safety-critical industries. Although deep-learning-based fault diagnosis has achieved high accuracy, its black-box nature limits trust, verification, and industrial deployment. This review examines explainable artificial intelligence for rotating machinery fault diagnosis and classifies existing methods into ante hoc and post hoc approaches according to their integration with model architectures. Their physical interpretability, applicable fault scenarios, explanation quality, computational overhead, robustness, and edge-deployment potential are critically compared. Quantitative criteria, including fidelity, stability, robustness, localization, and physical consistency are discussed to support objective evaluation of explanations. The review further highlights the gap between laboratory validation and industrial operation, particularly under sensor degradation, electromagnetic interference, variable working conditions, limited computing resources, and scarce fault data. It also discusses how model-relative explanations can be mapped to calibrated vibration quantities, fault-characteristic frequencies, industrial diagnostic standards, and actionable maintenance decisions. The distinction between correlation-based attribution and causal root-cause analysis is clarified, together with the role of digital twins and human-in-the-loop decision support. Finally, future research priorities are identified in standardized benchmarking, robust lightweight models, causal reasoning, and human-centered industrial deployment.
Shengnan Tang, Zeng-Yu Ren, Lei-Qi Zheng et al.· Italian National Conference...· 0 citations