Noise-Robust Quantum-Inspired Synthetic Minority Oversampling for Condition Monitoring and Fault Classification in Industrial and Energy Systems

The gist The study presents a comprehensive investigation into Quantum Synthetic Minority Oversampling Technique (QSMOTE) under quantum-inspired noise for industrial and energy-system fault classification, providing a systematic framework for evaluating class-imbalance mitigation and robustness against structured perturbations.

How it works

1. The QSMOTE framework utilizes a three-stage hybrid pipeline that blends classical clustering and quantum-inspired geometric reasoning to construct an augmented dataset (Xout, yout) that restores approximate balance while preserving the intrinsic manifold structure of the feature space.

2. The process involves:

Cluster Formation: The K-means algorithm partitions the feature matrix X into K clusters, resulting in centroids µj K j=1.

Quantum similarity estimation: The normalized feature representations of each minority instance x and centroid c(x) are treated as amplitude-encoded quantum states.

3. Synthetic sample generation is governed by an angle-based mechanism where the quantum-inspired angle α(x, c) quantifies the alignment between x and c(x) in a normalized feature space. The step length r is sampled from Unif 0, α(x, c) κ to ensure adaptivity to cluster geometry.

Key Contributions and Methodology

The paper introduces a novel noisy-QSMOTE analysis framework where quantum-inspired perturbations are injected directly into the compact-swap-test-based similarity estimation process used during synthetic sample generation. This method systematically investigates how perturbations affect overlap estimation, angle computation, synthetic sample generation, and ultimately downstream classification performance.

The study evaluates the effectiveness of QSMOTE across four multi-class industrial datasets: the Solar Panel Image Dataset (SPID), the CWRU Bearing Dataset (CWRUBD), the Engine Failure Detection Dataset (EFDD), and the Industrial Fault Detection Dataset (IFDD). Performance is assessed using Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), Logistic Regression (LR), and Naive Bayes (NB) classifiers.

Robustness Analysis under Noise

The analysis considers six quantum-inspired noise channels: bit-flip, phase-flip, bit-phase-flip, depolarizing, amplitude damping, and phase damping channels. These perturbations are injected into the compact swap test circuit before ancilla measurement to modify the measured probability p0. The resulting noisy-QSMOTE-generated dataset is then evaluated using classical ML models only as a downstream assessment tool.

Performance Findings

The results demonstrate that QSMOTE consistently mitigates class imbalance and significantly improves the performance of non-linear classifiers, yielding improvements of up to 170% on EFDD and achieving near-perfect accuracy (≥ 0.99) on IFDD. Specifically, QSMOTE improves the performance of all classical algorithms for the hexa classification task of SPID. In the noisy-QSMOTE experiments, RF and SVM remain highly resilient to noisy-QSMOTE-generated samples, whereas LR and NB exhibited greater sensitivity. The findings suggest that ensemble methods and margin-based classifiers are more resilient to variations introduced through noisy-QSMOTE-generated samples.

Conclusion

The proposed framework establishes a systematic methodology for jointly evaluating baseline performance, QSMOTE effectiveness, and noisy-QSMOTE robustness within a unified setting. The study confirms that the effectiveness of quantum-inspired oversampling depends not only on imbalance correction but also on the robustness of the underlying similarity-estimation mechanism. Future work will focus on developing noise-aware quantum-inspired oversampling techniques, more robust similarity-estimation strategies, and implementations on emerging quantum computing platforms for real-world industrial fault-diagnosis applications. The proposed framework provides a practical way to study both class-imbalance mitigation and the effects of noise on quantum-inspired oversampling methods in industrial AI applications.

--- Page 48 ---

The results from all four datasets show that RF and SVM are the most resilient against quantum-inspired noise, maintaining constant performance even under extreme perturbations. In contrast, NB and LR show significant vulnerability, particularly in noise-sensitive datasets like EFDD and IFDD. The persistence of these trends across datasets reveals a common pattern: ensemble methods and margin-based classifiers are more resilient to variations introduced through noisy-QSMOTE-generated samples.

--- Page 49 ---

The results in Table A8 show that QSMOTE consistently improves most classifiers for the penta-class dataset, with the exception of NB. RF achieves near-perfect classification results, while DT also benefits substantially.

--- Page 50 ---

The noisy-QSMOTE experiments provide insights into how quantum-inspired perturbations affect the synthetic sample generation process. Unlike conventional robustness studies, where noise is introduced directly into a learning model or input data, the noise channels considered in this work are injected into the compactswap-test-based similarity estimation process used during QSMOTE. Consequently, the perturbations first affect the ancilla measurement probability, which alters the overlap estimation between minority samples and cluster centroids. These changes propagate to the angle computation used for synthetic sample generation, resulting in modified synthetic samples and altered balanced datasets.

--- Page 55 ---

The results in Table A9 show that QSMOTE considerably improves classifier performance for the hepta-class dataset, with the exception of NB. RF achieves near-perfect classification results, while DT also benefits substantially.

--- Page 58 ---

The noisy-QSMOTE results shown in Fig. 7 provide a detailed assessment of the robustness of classical ML models under multiple quantum-inspired noise channels, namely BF, PF, BPF, DP, AD, and PD across four fault-diagnostic datasets.

--- Page 60 ---

The noisy-QSMOTE results shown in Fig. 7 provide a detailed assessment of the robustness of classical ML models under multiple quantum-inspired noise channels, namely BF, PF, BPF, DP, AD, and PD across four fault-diagnostic datasets.

--- Page 63 ---

The results in Table A7 highlight the quantitative improvements achieved through QSMOTE in the tetra-class classification task. RF achieves near-perfect classification results, while DT also benefits substantially.

--- Page 65 ---

The results in Table A8 show that QSMOTE consistently improves most classifiers for the penta-class dataset, with the exception of NB. RF achieves near-perfect classification, while DT also benefits substantially.

--- Page 68 ---

The results in Table A9 show that QSMOTE considerably improves classifier performance for the hepta-class dataset, with the exception of NB. RF achieves near-perfect classification results, while DT also benefits substantially.

--- Page 70 ---

The results in Table A8 show that QSMOTE consistently improves most classifiers for the octa-class dataset, with the exception of NB. RF achieves near-perfect classification results, while DT also benefits substantially.

1. Robust synthetic sample generation by injecting quantum-inspired noise directly into the compact swap test circuit, which affects overlap estimation, angle computation, and synthetic samples. This allows systems to generate balanced datasets that maintain geometric fidelity even when faced with structured quantum perturbations.

2. Enhanced fault classification performance across diverse industrial datasets (SPID, CWRUBD, EFDD, IFDD) by systematically evaluating models through three stages: baseline evaluation on original data, QSMOTE-based mitigation, and noisy-QSMOTE robustness analysis. This provides a comprehensive understanding of how noise propagate[s] through synthetic sample generation and influence downstream classification performance.

3. Targeted model selection based on noise resilience by identifying which classifiers are most stable under quantum corruption. The paper shows that RF and SVM remain highly resilient to noisy-QSMOTE-generated samples, suggesting these models are preferred when deploying systems in environments with quantum noise, while NB and LR show significant vulnerability.

4. Optimized decision boundary learning for non-linear models by leveraging QSMOTE's ability to improve performance for tree-based learners. The analysis demonstrates that DT benefits substantially from QSMOTE, increasing its accuracy from 0.5724 to 0.7439 on EFDD, confirming that class balancing strongly helps weaker models.

5. Noise-aware similarity estimation for real-time monitoring systems by using the compact swap test as a core mechanism where noise is injected before measurement. This enables systems to estimate noisy overlap estimate and compute a noisy quantum-inspired angle, allowing for synthetic sample creation based on corrupted but realistic geometric information.

Class imbalance remains a major challenge in industrial condition monitoring and fault classification, where minority fault patterns can be poorly represented during machine learning. This work develops a noise-aware quantum-inspired oversampling framework and systematically evaluates its effectiveness and robustness for industrial and energy-system fault diagnosis. The principal artificial-intelligence contribution is a noise-aware extension of the Quantum Synthetic Minority Oversampling Technique (QSMOTE), in which quantum-inspired perturbations are introduced directly into the compact-swap-test-based similarity estimation used during synthetic minority-sample generation. Consequently, noise propagates through overlap estimation and angle computation to the generated samples, enabling direct assessment of the robustness of the oversampling mechanism rather than perturbing only the downstream classifier. The engineering application is evaluated on four public datasets covering solar-panel faults, bearing condition monitoring, engine failures, and industrial fault detection. Five classical classifiers are used to compare performance on the original imbalanced data, after QSMOTE-based balancing, and under noisy-QSMOTE. The results show substantial classifier-dependent benefits from QSMOTE, including improvements of up to 170% on the engine-failure dataset and accuracy exceeding 0.99 on the industrial fault-detection dataset. Robustness experiments across six quantum-inspired noise channels further demonstrate that perturbations in similarity estimation can produce markedly different downstream effects depending on the noise model and classifier. The proposed framework therefore provides a reproducible approach for jointly assessing class-imbalance mitigation and noise robustness of quantum-inspired oversampling in practical industrial artificial-intelligence applications.