Multi-Class Electrical and Mechanical Fault Classification Using Random Convolutional Kernels

Mouhamadou Mansour Lo; Mouad Talbaoui; Gildas Morvan; Mathieu Rossi; Fabrice Morganti; David Mercier

cs.LGAug 19, 2026

Multi-Class Electrical and Mechanical Fault Classification Using Random Convolutional Kernels

Authors: Mouhamadou Mansour Lo, Mouad Talbaoui, Gildas Morvan, Mathieu Rossi, Fabrice Morganti, David Mercier

Organizations: UR 4025, LSEE University of Artois F-62400 B´ethune, France · UR 3926, LGI2A University of Artois F-62400 B´ethune, France

Abstract

Diagnosing faults in rotating machinery is essential for ensuring the reliability of industrial processes. Random convolutional kernel-based Time Series Classification (TSC) methods, such as ROCKET and its variants, provide an attractive trade-off between predictive performance and computational efficiency. In this work, we evaluate SelF-Rocket for the multi-class diagnosis of both mechanical and electrical faults and introduce, as a new contribution, a multivariate extension of the original method. The proposed approach is compared with leading ROCKET-based methods on two public benchmark datasets, MaFaulDa (mechanical faults) and ITSC-UDG (stator inter-turn short circuits), under both univariate and multivariate settings. Experimental results show that SelF-Rocket achieves the best overall accuracy-latency trade-off among the evaluated methods, obtaining the highest classification performance on MaFaulDa while remaining highly competitive on the more challenging ITSC-UDG dataset.

Figures & tables

Panel of Figure 1 — Figure 1: Multivariate SelF-Rocket Training Architecture

Figure 2: Proposed pipeline for multi-class fault classification using multivariate time series sensor data from a rotating machine

Classifiers	Hyperparameter Setting
MiniRocket *	MiniRocketClassifier(n_jobs=-1)
MultiRocket *	MultiRocketClassifier(n_jobs=-1)
Hydra *	HydraClassifier(n_jobs=-1)
Hydra + MR *	MultiRocketHydraClassifier(n_jobs=-1)
SelF-Rocket	SelFRocket(num_kernels=5000)

Table I : Hyperparameter settings for each ROCKET-based classifier. Methods marked with * are implemented in the AEON Python Toolkit.

Panel of Figure 3 — Figure 3: MaFaulDa multivariate t-SNE Projection of Feature Representations

Panel of Figure 4 — Figure 4: MaFaulDa univariate and multivariate Balanced Accuracy vs. Total Inference Time

Panel of Figure 5 — Figure 5: ITSC-UDG multivariate UMAP Projection of Feature Representations

Panel of Figure 6 — Figure 6: ITSC-UDG univariate and multivariate Balanced Accuracy vs. Total Inference Time

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

UNIVARIATE
$w$	M.	MiniR	MultiR	Hydra	H+MR	SelFR
500	Acc.	$72.83{\pm}0.26$	$72.23{\pm}0.21$	$70.33{\pm}0.22$	$74.01{\pm}0.20$	$\bm{77.95{\pm}0.38}$
	BAcc.	$66.40{\pm}0.22$	$66.73{\pm}0.17$	$63.93{\pm}0.19$	$68.37{\pm}0.19$	$\bm{72.21{\pm}0.41}$
	F1	$67.01{\pm}0.23$	$67.18{\pm}0.18$	$64.65{\pm}0.21$	$68.92{\pm}0.21$	$\bm{73.23{\pm}0.44}$
1000	Acc.	$82.05{\pm}0.25$	$82.66{\pm}0.15$	$82.58{\pm}0.18$	$84.72{\pm}0.18$	$\bm{87.13{\pm}0.45}$
	BAcc.	$76.00{\pm}0.37$	$77.37{\pm}0.19$	$76.00{\pm}0.19$	$79.52{\pm}0.27$	$\bm{81.70{\pm}0.53}$

Appendix

Table II : Average classification performance (mean

\pm

standard error) over 10 cross-validation folds on the MaFaulDa dataset for the univariate and multivariate settings.

w

denotes the window size, and M. denotes the evaluation metric (Accuracy, Balanced Accuracy, and

F1_{\mathrm{macro}}

). MiniR, MultiR, H+MR, and SelFR denote MiniRocket, MultiRocket, Hydra+MultiRocket, and SelF-Rocket, respectively; the remaining column (Hydra) is shown under its full name.

UNIVARIATE
$w$	M.	MiniR	MultiR	Hydra	H+MR	SelFR
500	TrFG	$\bm{0.48{\pm}0.01}$	$4.11{\pm}0.17$	$6.82{\pm}0.61$	$12.21{\pm}0.86$	$23.50{\pm}1.04$
	ClTr	$4.51{\pm}0.12$	$8.46{\pm}0.67$	$\bm{4.32{\pm}0.20}$	$9.31{\pm}0.55$	$4.99{\pm}0.26$
	TeFG	$\bm{0.38{\pm}0.00}$	$4.61{\pm}0.45$	$6.31{\pm}0.56$	$12.03{\pm}0.93$	$0.95{\pm}0.12$
	ClTe	$0.09{\pm}0.00$	$0.42{\pm}0.01$	$\bm{0.05{\pm}0.00}$	$0.48{\pm}0.01$	$0.09{\pm}0.01$
1000	TrFG	$\bm{0.95{\pm}0.10}$	$8.84{\pm}0.92$	$14.28{\pm}0.77$	$22.83{\pm}0.93$	$27.23{\pm}0.97$

Appendix

Table III : Average time performance (mean

\pm

standard error) over 10 cross-validation folds on the MaFaulDa dataset for the univariate and multivariate settings.

w

denotes the window size, and M. denotes the timing metric (TrFG = Training Feature Generation Time, ClTr = Classifier Training Time, TeFG = Test Feature Generation Time, and ClTe = Classifier Inference Time).

UNIVARIATE
$w$	M.	MiniR	MultiR	Hydra	H+MR	SelFR
500	Acc.	$40.48{\pm}1.19$	$37.91{\pm}1.31$	$\mathbf{41.70{\pm}1.27}$	$39.04{\pm}1.29$	$41.28{\pm}1.53$
	BAcc.	$40.96{\pm}1.25$	$38.70{\pm}1.39$	$\mathbf{42.22{\pm}1.14}$	$39.64{\pm}1.42$	$42.16{\pm}1.59$
	F1	$40.61{\pm}1.06$	$38.31{\pm}1.12$	$\mathbf{41.72{\pm}0.90}$	$39.34{\pm}1.16$	$41.54{\pm}1.21$
1000	Acc.	$51.26{\pm}1.33$	$48.69{\pm}1.95$	$\mathbf{53.40{\pm}1.51}$	$49.82{\pm}1.66$	$52.68{\pm}2.01$
	BAcc.	$52.35{\pm}1.67$	$50.19{\pm}2.17$	$\mathbf{54.40{\pm}1.84}$	$51.18{\pm}1.89$	$53.74{\pm}2.32$

Appendix

Table IV : Average classification performance (mean

\pm

standard error) over 10 cross-validation folds on the ITSC-UDG dataset for univariate and multivariate settings.

UNIVARIATE
$w$	M.	MiniR	MultiR	Hydra	H+MR	SelFR
500	TrFG	$\mathbf{0.18{\pm}0.05}$	$0.46{\pm}0.02$	$0.33{\pm}0.02$	$0.68{\pm}0.03$	$2.42{\pm}0.05$
	ClTr	$0.04{\pm}0.00$	$0.15{\pm}0.01$	$0.04{\pm}0.00$	$0.15{\pm}0.01$	$\mathbf{0.03{\pm}0.00}$
	TeFG	$\mathbf{0.02{\pm}0.00}$	$0.35{\pm}0.02$	$0.34{\pm}0.01$	$0.51{\pm}0.05$	$0.07{\pm}0.02$
	ClTe	$0.01{\pm}0.00$	$0.03{\pm}0.00$	$0.01{\pm}0.00$	$0.02{\pm}0.00$	$\mathbf{0.00{\pm}0.00}$
1000	TrFG	$\mathbf{0.20{\pm}0.01}$	$0.84{\pm}0.04$	$0.68{\pm}0.02$	$1.65{\pm}0.05$	$2.93{\pm}0.08$

Appendix

Table V : Average time performance (mean

\pm

standard error) over 10 cross-validation folds on the ITSC-UDG dataset for univariate and multivariate settings.

Fault Diagnosis

Fault diagnosis identifies, localizes, and explains failures in engineered and computational systems using sensor data, logs, models, and domain knowledge.

See all 143 papers

Motor compound fault diagnosis using an attention-based cross-talk architecture
Oct 19, 2026Wonjun Yi, Wonho Jung, Junho Kim +2Fault Diagnosis Industrial Fault Diagnosis
Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Republic of Korea
RocketPFN: Accurate Time Series Classification via In-Context Learning
Jun 19, 2026Franco Martino O'Rourke, Ana Trisovic, Dimitris BertsimasTime Series Classification Classification
Leakage-Robust Evaluation and Data-Scale Sensitivity of Attention-Enhanced Multi-Task Learning for Joint Fault Diagnosis and Remaining Useful Life Estimation
Jul 17, 2026Md Mahamudur Rahaman Shamim, Md. Nuruzzaman, Zannatul Ferdus +5Data Leakage RUL Estimation
Department of Manufacturing Engineering Technology, Western Illinois University, Macomb, IL-61455, USA · Department of Mechanical and Production Engineering, Ahsanullah University of Science and Technology, 141-142 Love Road, Tejgaon Industrial Area, Dhaka, Bangladesh · Master of Science in Industrial/Engineering Management, Trine University, 720 Thunder Dr, Angola, IN 46703, USA +3

Oct 19, 2026J/K move · Enter open · S save

Motor compound fault diagnosis using an attention-based cross-talk architecture

Wonjun Yi, Wonho Jung, Junho Kim, Rismaya Kumar Mishra+1

Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Republic of Korea

Fault Diagnosis Industrial Fault Diagnosis Multi-Task Learning Bearing Fault Diagnosis

Open paper AskPDF arXiv

As physical AI increasingly relies on sensing to understand physical systems, reliable condition monitoring of industrial machinery has become more important. Motors are ubiquitous in industrial systems, and increasing system complexity raises the likelihood of simultaneous faults. Accordingly, compound fault diagnosis that identifies multiple fault components and their severities is increasingly required. Because different fault components jointly affect the same vibration signal, their severity classification can be formulated as related tasks, making multi-task learning (MTL) suitable for exploiting shared information. However, conventional shared-trunk architectures force tasks to share a common representation, potentially causing negative transfer because fault components are characterized by different time-frequency regions. To address this issue, this study proposes a Residual Neural Dimension Reductor with Time-Frequency Channel Attention (RNDR-TFCA), an attention-enhanced cross-talk architecture for multi-output compound fault diagnosis. RNDR-TFCA preserves task-specific feature extractors while selectively exchanging information across tasks. Time-frequency attention emphasizes informative intra-task regions, while channel attention enhances useful inter-task information after nonlinear feature fusion. Experiments on a rotor-bearing system with inner-race fault, outer-race fault, misalignment, and unbalance showed that RNDR-TFCA achieved a macro F1 score of 0.898, outperforming single-task, multi-class, shared-trunk, and cross-talk baselines. Ablation experiments confirmed the benefit of jointly applying both attention mechanisms. RNDR-TFCA required 713 K parameters and 170.42 MMACs, only slightly higher than RNDR. Channel-expanded baselines still underperformed RNDR-TFCA, indicating that the gain was not simply attributable to increased model capacity.

Abstract

Fault Diagnosis

Explore similar work

Motor compound fault diagnosis using an attention-based cross-talk architecture

RocketPFN: Accurate Time Series Classification via In-Context Learning

Leakage-Robust Evaluation and Data-Scale Sensitivity of Attention-Enhanced Multi-Task Learning for Joint Fault Diagnosis and Remaining Useful Life Estimation